faqs.org - Internet FAQ Archives

RFC 4695 - RTP Payload Format for MIDI


Or Display the document by number




Network Working Group                                         J. Lazzaro
Request for Comments: 4695                                  J. Wawrzynek
Category: Standards Track                                    UC Berkeley
                                                           November 2006

                      RTP Payload Format for MIDI

Status of This Memo

   This document specifies an Internet standards track protocol for the
   Internet community, and requests discussion and suggestions for
   improvements.  Please refer to the current edition of the "Internet
   Official Protocol Standards" (STD 1) for the standardization state
   and status of this protocol.  Distribution of this memo is unlimited.

Copyright Notice

   Copyright (C) The IETF Trust (2006).

Abstract

   This memo describes a Real-time Transport Protocol (RTP) payload
   format for the MIDI (Musical Instrument Digital Interface) command
   language.  The format encodes all commands that may legally appear on
   a MIDI 1.0 DIN cable.  The format is suitable for interactive
   applications (such as network musical performance) and content-
   delivery applications (such as file streaming).  The format may be
   used over unicast and multicast UDP and TCP, and it defines tools for
   graceful recovery from packet loss.  Stream behavior, including the
   MIDI rendering method, may be customized during session setup.  The
   format also serves as a mode for the mpeg4-generic format, to support
   the MPEG 4 Audio Object Types for General MIDI, Downloadable Sounds
   Level 2, and Structured Audio.

Table of Contents

   1. Introduction ....................................................4
      1.1. Terminology ................................................5
      1.2. Bitfield Conventions .......................................6
   2. Packet Format ...................................................6
      2.1. RTP Header .................................................7
      2.2. MIDI Payload ..............................................11
   3. MIDI Command Section ...........................................12
      3.1.  Timestamps ...............................................14
      3.2.  Command Coding ...........................................16

   4. The Recovery Journal System ....................................22
   5. Recovery Journal Format ........................................24
   6. Session Description Protocol ...................................28
      6.1. Session Descriptions for Native Streams ...................29
      6.2. Session Descriptions for mpeg4-generic Streams ............30
      6.3. Parameters ................................................33
   7. Extensibility ..................................................34
   8. Congestion Control .............................................35
   9. Security Considerations ........................................35
   10. Acknowledgements ..............................................36
   11. IANA Considerations ...........................................37
      11.1. rtp-midi Media Type Registration .........................37
           11.1.1. Repository Request for "audio/rtp-midi" ...........40
      11.2. mpeg4-generic Media Type Registration ....................41
           11.2.1. Repository Request for Mode rtp-midi for
                   mpeg4-generic .....................................44
      11.3. asc Media Type Registration ..............................46
   A. The Recovery Journal Channel Chapters ..........................48
      A.1. Recovery Journal Definitions ..............................48
      A.2. Chapter P: MIDI Program Change ............................52
      A.3. Chapter C: MIDI Control Change ............................53
           A.3.1. Log Inclusion Rules ................................54
           A.3.2. Controller Log Format ..............................55
           A.3.3. Log List Coding Rules ..............................57
           A.3.4. The Parameter System ...............................60
      A.4. Chapter M: MIDI Parameter System ..........................62
           A.4.1. Log Inclusion Rules ................................64
           A.4.2. Log Coding Rules ...................................65
                 A.4.2.1. The Value Tool .............................67
                 A.4.2.2. The Count Tool .............................70
      A.5. Chapter W: MIDI Pitch Wheel ...............................71
      A.6. Chapter N: MIDI NoteOff and NoteOn ........................71
           A.6.1. Header Structure ...................................73
           A.6.2. Note Structures ....................................74
      A.7. Chapter E: MIDI Note Command Extras .......................75
           A.7.1. Note Log Format ....................................76
           A.7.2. Log Inclusion Rules ................................76
      A.8. Chapter T: MIDI Channel Aftertouch ........................77
      A.9. Chapter A: MIDI Poly Aftertouch ...........................78
   B. The Recovery Journal System Chapters ...........................79
      B.1. System Chapter D: Simple System Commands ..................79
           B.1.1. Undefined System Commands ..........................80
      B.2. System Chapter V: Active Sense Command ....................83
      B.3. System Chapter Q: Sequencer State Commands ................83
           B.3.1. Non-compliant Sequencers ...........................85
      B.4. System Chapter F: MIDI Time Code Tape Position ............86
           B.4.1. Partial Frames .....................................88

      B.5. System Chapter X: System Exclusive ........................89
           B.5.1. Chapter Format .....................................90
           B.5.2. Log Inclusion Semantics ............................92
           B.5.3. TCOUNT and COUNT Fields ............................95
   C. Session Configuration Tools ....................................95
      C.1. Configuration Tools: Stream Subsetting ....................97
      C.2. Configuration Tools: The Journalling System ..............101
           C.2.1. The j_sec Parameter ...............................102
           C.2.2. The j_update Parameter ............................103
                 C.2.2.1. The anchor Sending Policy .................104
                 C.2.2.2. The closed-loop Sending Policy ............104
                 C.2.2.3. The open-loop Sending Policy ..............108
           C.2.3. Recovery Journal Chapter Inclusion Parameters .....110
      C.3. Configuration Tools: Timestamp Semantics .................115
           C.3.1. The comex Algorithm ...............................115
           C.3.2. The async Algorithm ...............................116
           C.3.3. The buffer Algorithm ..............................117
      C.4. Configuration Tools: Packet Timing Tools .................118
           C.4.1. Packet Duration Tools .............................119
           C.4.2. The guardtime Parameter ...........................120
      C.5. Configuration Tools: Stream Description ..................121
      C.6. Configuration Tools: MIDI Rendering ......................128
           C.6.1. The multimode Parameter ...........................129
           C.6.2. Renderer Specification ............................129
           C.6.3. Renderer Initialization ...........................131
           C.6.4. MIDI Channel Mapping ..............................133
                 C.6.4.1. The smf_info Parameter ....................134
                 C.6.4.2. The smf_inline, smf_url, and smf_cid
                          Parameters ................................136
                 C.6.4.3. The chanmask Parameter ....................136
           C.6.5. The audio/asc Media Type ..........................137
      C.7. Interoperability .........................................139
           C.7.1. MIDI Content Streaming Applications ...............139
           C.7.2. MIDI Network Musical Performance Applications .....142
   D. Parameter Syntax Definitions ..................................150
   E. A MIDI Overview for Networking Specialists ....................156
      E.1. Commands Types ...........................................159
      E.2. Running Status ...........................................159
      E.3. Command Timing ...........................................160
      E.4. AudioSpecificConfig Templates for MMA Renderers ..........160
   References .......................................................165
   Normative References .............................................165
   Informative References ...........................................166

1.  Introduction

   The Internet Engineering Task Force (IETF) has developed a set of
   focused tools for multimedia networking ([RFC3550] [RFC4566]
   [RFC3261] [RFC2326]).  These tools can be combined in different ways
   to support a variety of real-time applications over Internet Protocol
   (IP) networks.

   For example, a telephony application might use the Session Initiation
   Protocol (SIP, [RFC3261]) to set up a phone call.  Call setup would
   include negotiations to agree on a common audio codec [RFC3264].
   Negotiations would use the Session Description Protocol (SDP,
   [RFC4566]) to describe candidate codecs.

   After a call is set up, audio data would flow between the parties
   using the Real Time Protocol (RTP, [RFC3550]) under any applicable
   profile (for example, the Audio/Visual Profile (AVP, [RFC3551])).
   The tools used in this telephony example (SIP, SDP, RTP) might be
   combined in a different way to support a content streaming
   application, perhaps in conjunction with other tools, such as the
   Real Time Streaming Protocol (RTSP, [RFC2326]).

   The MIDI (Musical Instrument Digital Interface) command language
   [MIDI] is widely used in musical applications that are analogous to
   the examples described above.  On stage and in the recording studio,
   MIDI is used for the interactive remote control of musical
   instruments, an application similar in spirit to telephony.  On web
   pages, Standard MIDI Files (SMFs, [MIDI]) rendered using the General
   MIDI standard [MIDI] provide a low-bandwidth substitute for audio
   streaming.

   This memo is motivated by a simple premise: if MIDI performances
   could be sent as RTP streams that are managed by IETF session tools,
   a hybridization of the MIDI and IETF application domains may occur.

   For example, interoperable MIDI networking may foster network music
   performance applications, in which a group of musicians, located at
   different physical locations, interact over a network to perform as
   they would if they were located in the same room [NMP].  As a second
   example, the streaming community may begin to use MIDI for low-
   bitrate audio coding, perhaps in conjunction with normative sound
   synthesis methods [MPEGSA].

   To enable MIDI applications to use RTP, this memo defines an RTP
   payload format and its media type.  Sections 2-5 and Appendices A-B
   define the RTP payload format.  Section 6 and Appendices C-D define
   the media types identifying the payload format, the parameters needed
   for configuration, and how the parameters are utilized in SDP.

   Appendix C also includes interoperability guidelines for the example
   applications described above: network musical performance using SIP
   (Appendix C.7.2) and content-streaming using RTSP (Appendix C.7.1).

   Another potential application area for RTP MIDI is MIDI networking
   for professional audio equipment and electronic musical instruments.
   We do not offer interoperability guidelines for this application in
   this memo.  However, RTP MIDI has been designed with stage and studio
   applications in mind, and we expect that efforts to define a stage
   and studio framework will rely on RTP MIDI for MIDI transport
   services.

   Some applications may require MIDI media delivery at a certain
   service quality level (latency, jitter, packet loss, etc).  RTP
   itself does not provide service guarantees.  However, applications
   may use lower-layer network protocols to configure the quality of the
   transport services that RTP uses.  These protocols may act to reserve
   network resources for RTP flows [RFC2205] or may simply direct RTP
   traffic onto a dedicated "media network" in a local installation.
   Note that RTP and the MIDI payload format do provide tools that
   applications may use to achieve the best possible real-time
   performance at a given service level.

   This memo normatively defines the syntax and semantics of the MIDI
   payload format.  However, this memo does not define algorithms for
   sending and receiving packets.  An ancillary document [RFC4696]
   provides informative guidance on algorithms.  Supplemental
   information may be found in related conference publications [NMP]
   [GRAME].

   Throughout this memo, the phrase "native stream" refers to a stream
   that uses the rtp-midi media type.  The phrase "mpeg4-generic stream"
   refers to a stream that uses the mpeg4-generic media type (in mode
   rtp-midi) to operate in an MPEG 4 environment [RFC3640].  Section 6
   describes this distinction in detail.

1.1.  Terminology

   In this document, the key words "MUST", "MUST NOT", "REQUIRED",
   "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY",
   and "OPTIONAL" are to be interpreted as described in BCP 14, RFC 2119
   [RFC2119].

1.2.  Bitfield Conventions

   In this document, the packet bitfields that share a common name often
   have identical semantics.  As most of these bitfields appear in
   Appendices A-B, we define the common bitfield names in Appendix A.1.

   However, a few of these common names also appear in the main text of
   this document.  For convenience, we list these definitions below:

     o R flag bit.  R flag bits are reserved for future use.  Senders
       MUST set R bits to 0.  Receivers MUST ignore R bit values.

     o LENGTH field.  All fields named LENGTH (as distinct from LEN)
       code the number of octets in the structure that contains it,
       including the header it resides in and all hierarchical levels
       below it.  If a structure contains a LENGTH field, a receiver
       MUST use the LENGTH field value to advance past the structure
       during parsing, rather than use knowledge about the internal
       format of the structure.

2.  Packet Format

   In this section, we introduce the format of RTP MIDI packets.  The
   description includes some background information on RTP, for the
   benefit of MIDI implementors new to IETF tools.  Implementors should
   consult [RFC3550] for an authoritative description of RTP.

   This memo assumes that the reader is familiar with MIDI syntax and
   semantics.  Appendix E provides a MIDI overview, at a level of detail
   sufficient to understand most of this memo.  Implementors should
   consult [MIDI] for an authoritative description of MIDI.

   The MIDI payload format maps a MIDI command stream (16 voice channels
   + systems) onto an RTP stream.  An RTP media stream is a sequence of
   logical packets that share a common format.  Each packet consists of
   two parts: the RTP header and the MIDI payload.  Figure 1 shows this
   format (vertical space delineates the header and payload).

   We describe RTP packets as "logical" packets to highlight the fact
   that RTP itself is not a network-layer protocol.  Instead, RTP
   packets are mapped onto network protocols (such as unicast UDP,
   multicast UDP, or TCP) by an application [ALF].  The interleaved mode
   of the Real Time Streaming Protocol (RTSP, [RFC2326]) is an example
   of an RTP mapping to TCP transport, as is [RFC4571].

2.1.  RTP Header

   [RFC3550] provides a complete description of the RTP header fields.
   In this section, we clarify the role of a few RTP header fields for
   MIDI applications.  All fields are coded in network byte order (big-
   endian).

       0                   1                   2                   3
       0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
      | V |P|X|  CC   |M|     PT      |        Sequence number        |
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
      |                           Timestamp                           |
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
      |                             SSRC                              |
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
      |                     MIDI command section ...                  |
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
      |                       Journal section ...                     |
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

                         Figure 1 -- Packet format

   The behavior of the 1-bit M field depends on the media type of the
   stream.  For native streams, the M bit MUST be set to 1 if the MIDI
   command section has a non-zero LEN field, and MUST be set to 0
   otherwise.  For mpeg4-generic streams, the M bit MUST be set to 1 for
   all packets (to conform to [RFC3640]).

   In an RTP MIDI stream, the 16-bit sequence number field is
   initialized to a randomly chosen value and is incremented by one
   (modulo 2^16) for each packet sent in the stream.  A related
   quantity, the 32-bit extended packet sequence number, may be computed
   by tracking rollovers of the 16-bit sequence number.  Note that
   different receivers of the same stream may compute different extended
   packet sequence numbers, depending on when the receiver joined the
   session.

   The 32-bit timestamp field sets the base timestamp value for the
   packet.  The payload codes MIDI command timing relative to this
   value.  The timestamp units are set by the clock rate parameter.  For
   example, if the clock rate has a value of 44100 Hz, two packets whose
   base timestamp values differ by 2 seconds have RTP timestamp fields
   that differ by 88200.

   Note that the clock rate parameter is not encoded within each RTP
   MIDI packet.  A receiver of an RTP MIDI stream becomes aware of the
   clock rate as part of the session setup process.  For example, if a
   session management tool uses the Session Description Protocol (SDP,
   [RFC4566]) to describe a media session, the clock rate parameter is
   set using the rtpmap attribute.  We show examples of session setup in
   Section 6.

   For RTP MIDI streams destined to be rendered into audio, the clock
   rate SHOULD be an audio sample rate of 32 KHz or higher.  This
   recommendation is due to the sensitivity of human musical perception
   to small timing errors in musical note sequences, and due to the
   timbral changes that occur when two near-simultaneous MIDI NoteOns
   are rendered with a different timing than that desired by the content
   author due to clock rate quantization.  RTP MIDI streams that are not
   destined for audio rendering (such as MIDI streams that control stage
   lighting) MAY use a lower clock rate but SHOULD use a clock rate high
   enough to avoid timing artifacts in the application.

   For RTP MIDI streams destined to be rendered into audio, the clock
   rate SHOULD be chosen from rates in common use in professional audio
   applications or in consumer audio distribution.  At the time of this
   writing, these rates include 32 KHz, 44.1 KHz, 48 KHz, 64 KHz, 88.2
   KHz, 96 KHz, 176.4 KHz, and 192 KHz.  If the RTP MIDI session is a
   part of a synchronized media session that includes another (non-MIDI)
   RTP audio stream with a clock rate of 32 KHz or higher, the RTP MIDI
   stream SHOULD use a clock rate that matches the clock rate of the
   other audio stream.  However, if the RTP MIDI stream is destined to
   be rendered into audio, the RTP MIDI stream SHOULD NOT use a clock
   rate lower than 32 KHz, even if this second stream has a clock rate
   less than 32 KHz.

   Timestamps of consecutive packets do not necessarily increment at a
   fixed rate, because RTP MIDI packets are not necessarily sent at a
   fixed rate.  The degree of packet transmission regularity reflects
   the underlying application dynamics.  Interactive applications may
   vary the packet sending rate to track the gestural rate of a human
   performer, whereas content-streaming applications may send packets at
   a fixed rate.

   Therefore, the timestamps for two sequential RTP packets may be
   identical, or the second packet may have a timestamp arbitrarily
   larger than the first packet (modulo 2^32).  Section 3 places
   additional restrictions on the RTP timestamps for two sequential RTP
   packets, as does the guardtime parameter (Appendix C.4.2).

   We use the term "media time" to denote the temporal duration of the
   media coded by an RTP packet.  The media time coded by a packet is

   computed by subtracting the last command timestamp in the MIDI
   command section from the RTP timestamp (modulo 2^32).  If the MIDI
   list of the MIDI command section of a packet is empty, the media time
   coded by the packet is 0 ms.  Appendix C.4.1 discusses media time
   issues in detail.

   We now define RTP session semantics, in the context of sessions
   specified using the session description protocol [RFC4566].  A
   session description media line ("m=") specifies an RTP session.  An
   RTP session has an independent space of 2^32 synchronization sources.
   Synchronization source identifiers are coded in the SSRC header field
   of RTP session packets.  The payload types that may appear in the PT
   header field of RTP session packets are listed at the end of the
   media line.

   Several RTP MIDI streams may appear in an RTP session.  Each stream
   is distinguished by a unique SSRC value and has a unique sequence
   number and RTP timestamp space.  Multiple streams in the RTP session
   may be sent by a single party.  Multiple parties may send streams in
   the RTP session.  An RTP MIDI stream encodes data for a single MIDI
   command name space (16 voice channels + Systems).

   Streams in an RTP session may use different payload types, or they
   may use the same payload type.  However, each party may send, at
   most, one RTP MIDI stream for each payload type mapped to an RTP MIDI
   payload format in an RTP session.  Recall that dynamic binding of
   payload type numbers in [RFC4566] lets a party map many payload type
   numbers to the RTP MIDI payload format; thus a party may send many
   RTP MIDI streams in a single RTP session.  Pairs of streams (unicast
   or multicast) that communicate between two parties in an RTP session
   and that share a payload type have the same association as a MIDI
   cable pair that cross-connects two devices in a MIDI 1.0 DIN network.

   The RTP session architecture described above is efficient in its use
   of network ports, as one RTP session (using a port pair per party)
   supports the transport of many MIDI name spaces (16 MIDI channels +
   systems).  We define tools for grouping and labelling MIDI name
   spaces across streams and sessions in Appendix C.5 of this memo.

   The RTP header timestamps for each stream in an RTP session have
   separately and randomly chosen initialization values.  Receivers use
   the timing fields encoded in the RTP control protocol (RTCP,
   [RFC3550]) sender reports to synchronize the streams sent by a party.
   The SSRC values for each stream in an RTP session are also separately
   and randomly chosen, as described in [RFC3550].  Receivers use the
   CNAME field encoded in RTCP sender reports to verify that streams
   were sent by the same party, and to detect SSRC collisions, as
   described in [RFC3550].

   In some applications, a receiver renders MIDI commands into audio (or
   into control actions, such as the rewind of a tape deck or the
   dimming of stage lights).  In other applications, a receiver presents
   a MIDI stream to software programs via an Application Programmer
   Interface (API).  Appendix C.6 defines session configuration tools to
   specify what receivers should do with a MIDI command stream.

   If a multimedia session uses different RTP MIDI streams to send
   different classes of media, the streams MUST be sent over different
   RTP sessions.  For example, if a multimedia session uses one MIDI
   stream for audio and a second MIDI stream to control a lighting
   system, the audio and lighting streams MUST be sent over different
   RTP sessions, each with its own media line.

   Session description tools defined in Appendix C.5 let a sending party
   split a single MIDI name space (16 voice channels + systems) over
   several RTP MIDI streams.  Split transport of a MIDI command stream
   is a delicate task, because correct command stream reconstruction by
   a receiver depends on exact timing synchronization across the
   streams.

   To support split name spaces, we define the following requirements:

     o  A party MUST NOT send several RTP MIDI streams that share a MIDI
        name space in the same RTP session.  Instead, each stream MUST
        be sent from a different RTP session.

     o  If several RTP MIDI streams sent by a party share a MIDI name
        space, all streams MUST use the same SSRC value and MUST use the
        same randomly chosen RTP timestamp initialization value.

   These rules let a receiver identify streams that share a MIDI name
   space (by matching SSRC values) and also let a receiver accurately
   reconstruct the source MIDI command stream (by using RTP timestamps
   to interleave commands from the two streams).  Care MUST be taken by
   senders to ensure that SSRC changes due to collisions are reflected
   in both streams.  Receivers MUST regularly examine the RTCP CNAME
   fields associated with the linked streams, to ensure that the assumed
   link is legitimate and not the result of an SSRC collision by another
   sender.

   Except for the special cases described above, a party may send many
   RTP MIDI streams in the same session.  However, it is sometimes
   advantageous for two RTP MIDI streams to be sent over different RTP
   sessions.  For example, two streams may need different values for RTP
   session-level attributes (such as the sendonly and recvonly
   attributes).  As a second example, two RTP sessions may be needed to
   send two unicast streams in a multimedia session that originate on

   different computers (with different IP numbers).  Two RTP sessions
   are needed in this case because transport addresses are specified on
   the RTP-session or multimedia-session level, not on a payload type
   level.

   On a final note, in some uses of MIDI, parties send bidirectional
   traffic to conduct transactions (such as file exchange).  These
   commands were designed to work over MIDI 1.0 DIN cable networks may
   be configured in a multicast topology, which use pure "party-line"
   signalling.  Thus, if a multimedia session ensures a multicast
   connection between all parties, bidirectional MIDI commands will work
   without additional support from the RTP MIDI payload format.

2.2. MIDI Payload

   The payload (Figure 1) MUST begin with the MIDI command section.  The
   MIDI command section codes a (possibly empty) list of timestamped
   MIDI commands, and provides the essential service of the payload
   format.

   The payload MAY also contain a journal section.  The journal section
   provides resiliency by coding the recent history of the stream.  A
   flag in the MIDI command section codes the presence of a journal
   section in the payload.

   Section 3 defines the MIDI command section.  Sections 4-5 and
   Appendices A-B define the recovery journal, the default format for
   the journal section.  Here, we describe how these payload sections
   operate in a stream in an RTP session.

   The journalling method for a stream is set at the start of a session
   and MUST NOT be changed thereafter.  A stream may be set to use the
   recovery journal, to use an alternative journal format (none are
   defined in this memo), or not to use a journal.

   The default journalling method of a stream is inferred from its
   transport type.  Streams that use unreliable transport (such as UDP)
   default to using the recovery journal.  Streams that use reliable
   transport (such as TCP) default to not using a journal.  Appendix
   C.2.1 defines session configuration tools for overriding these
   defaults.  For all types of transport, a sender MUST transmit an RTP
   packet stream with consecutive sequence numbers (modulo 2^16).

   If a stream uses the recovery journal, every payload in the stream
   MUST include a journal section.  If a stream does not use
   journalling, a journal section MUST NOT appear in a stream payload.
   If a stream uses an alternative journal format, the specification for
   the journal format defines an inclusion policy.

   If a stream is sent over UDP transport, the Maximum Transmission Unit
   (MTU) of the underlying network limits the practical size of the
   payload section (for example, an Ethernet MTU is 1500 octets), for
   applications where predictable and minimal packet transmission
   latency is critical.  A sender SHOULD NOT create RTP MIDI UDP packets
   whose size exceeds the MTU of the underlying network.  Instead, the
   sender SHOULD take steps to keep the maximum packet size under the
   MTU limit.

   These steps may take many forms.  The default closed-loop recovery
   journal sending policy (defined in Appendix C.2.2.2) uses RTP control
   protocol (RTCP, [RFC3550]) feedback to manage the RTP MIDI packet
   size.  In addition, Section 3.2 and Appendix B.5.2 provide specific
   tools for managing the size of packets that code MIDI System
   Exclusive (0xF0) commands.  Appendix C.5 defines session
   configuration tools that may be used to split a dense MIDI name space
   into several UDP streams (each sent in a different RTP session, per
   Section 2.1) so that the payload fits comfortably into an MTU.
   Another option is to use TCP.  Section 4.3 of [RFC4696] provides
   non-normative advice for packet size management.

3.  MIDI Command Section

   Figure 2 shows the format of the MIDI command section.

       0                   1                   2                   3
       0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
      |B|J|Z|P|LEN... |  MIDI list ...                                |
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

                      Figure 2 -- MIDI command section

   The MIDI command section begins with a variable-length header.

   The header field LEN codes the number of octets in the MIDI list that
   follow the header.  If the header flag B is 0, the header is one
   octet long, and LEN is a 4-bit field, supporting a maximum MIDI list
   length of 15 octets.

   If B is 1, the header is two octets long, and LEN is a 12-bit field,
   supporting a maximum MIDI list length of 4095 octets.  LEN is coded
   in network byte order (big-endian): the 4 bits of LEN that appear in
   the first header octet code the most significant 4 bits of the 12-bit
   LEN value.

   A LEN value of 0 is legal, and it codes an empty MIDI list.

   If the J header bit is set to 1, a journal section MUST appear after
   the MIDI command section in the payload.  If the J header bit is set
   to 0, the payload MUST NOT contain a journal section.

   We define the semantics of the P header bit in Section 3.2.

   If the LEN header field is nonzero, the MIDI list has the structure
   shown in Figure 3.

      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
      |  Delta Time 0     (1-4 octets long, or 0 octets if Z = 1)     |
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
      |  MIDI Command 0   (1 or more octets long)                     |
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
      |  Delta Time 1     (1-4 octets long)                           |
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
      |  MIDI Command 1   (1 or more octets long)                     |
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
      |                              ...                              |
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
      |  Delta Time N     (1-4 octets long)                           |
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
      |  MIDI Command N   (0 or more octets long)                     |
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

                       Figure 3 -- MIDI list structure

   If the header flag Z is 1, the MIDI list begins with a complete MIDI
   command (coded in the MIDI Command 0 field, in Figure 3) preceded by
   a delta time (coded in the Delta Time 0 field).  If Z is 0, the Delta
   Time 0 field is not present in the MIDI list, and the command coded
   in the MIDI Command 0 field has an implicit delta time of 0.

   The MIDI list structure may also optionally encode a list of N
   additional complete MIDI commands, each coded in a MIDI Command K
   field.  Each additional command MUST be preceded by a Delta Time K
   field, which codes the command's delta time.  We discuss exceptions
   to the "command fields code complete MIDI commands" rule in Section
   3.2.

   The final MIDI command field (i.e., the MIDI Command N field, shown
   in Figure 3) in the MIDI list MAY be empty.  Moreover, a MIDI list
   MAY consist a single delta time (encoded in the Delta Time 0 field)
   without an associated command (which would have been encoded in the
   MIDI Command 0 field).  These rules enable MIDI coding features that
   are explained in Section 3.1.  We delay the explanations because an
   understanding of RTP MIDI timestamps is necessary to describe the
   features.

3.1.  Timestamps

   In this section, we describe how RTP MIDI encodes a timestamp for
   each MIDI list command.  Command timestamps have the same units as
   RTP packet header timestamps (described in Section 2.1 and
   [RFC3550]).  Recall that RTP timestamps have units of seconds, whose
   scaling is set during session configuration (see Section 6.1 and
   [RFC4566]).

   As shown in Figure 3, the MIDI list encodes time using a compact
   delta-time format.  The RTP MIDI delta time syntax is a modified form
   of the MIDI File delta time syntax [MIDI].  RTP MIDI delta times use
   1-4 octet fields to encode 32-bit unsigned integers.  Figure 4 shows
   the encoded and decoded forms of delta times.  Note that delta time
   values may be legally encoded in multiple formats; for example, there
   are four legal ways to encode the zero delta time (0x00, 0x8000,
   0x808000, 0x80808000).

   RTP MIDI uses delta times to encode a timestamp for each MIDI
   command.  The timestamp for MIDI Command K is the summation (modulo
   2^32) of the RTP timestamp and decoded delta times 0 through K.  This
   cumulative coding technique, borrowed from MIDI File delta time
   coding, is efficient because it reduces the number of multi-octet
   delta times.

   All command timestamps in a packet MUST be less than or equal to the
   RTP timestamp of the next packet in the stream (modulo 2^32).

   This restriction ensures that a particular RTP MIDI packet in a
   stream is uniquely responsible for encoding time starting at the
   moment after the RTP timestamp encoded in the RTP packet header, and
   ending at the moment before the final command timestamp encoded in
   the MIDI list.  The "moment before" and "moment after" qualifiers
   acknowledge the "less than or equal" semantics (as opposed to
   "strictly less than") in the sentence above this paragraph.

   Note that it is possible to "pad" the end of an RTP MIDI packet with
   time that is guaranteed to be void of MIDI commands, by setting the
   "Delta Time N" field of the MIDI list to the end of the void time,
   and by omitting its corresponding "MIDI Command N" field (a syntactic
   construction the preamble of Section 3 expressly made legal).

   In addition, it is possible to code an RTP MIDI packet to express
   that a period of time in the stream is void of MIDI commands.  The
   RTP timestamp in the header would code the start of the void time.
   The MIDI list of this packet would consist of a "Delta Time 0" field

   that coded the end of the void time.  No other fields would be
   present in the MIDI list (a syntactic construction the preamble of
   Section 3 also expressly made legal).

   By default, a command timestamp indicates the execution time for the
   command.  The difference between two timestamps indicates the time
   delay between the execution of the commands.  This difference may be
   zero, coding simultaneous execution.  In this memo, we refer to this
   interpretation of timestamps as "comex" (COMmand EXecution)
   semantics.  We formally define comex semantics in Appendix C.3.

   The comex interpretation of timestamps works well for transcoding a
   Standard MIDI File (SMF) into an RTP MIDI stream, as SMFs code a
   timestamp for each MIDI command stored in the file.  To transcode an
   SMF that uses metric time markers, use the SMF tempo map (encoded in
   the SMF as meta-events) to convert metric SMF timestamp units into
   seconds-based RTP timestamp units.

   The comex interpretation also works well for MIDI hardware
   controllers that are coding raw sensor data directly onto an RTP MIDI
   stream.  Note that this controller design is preferable to a design
   that converts raw sensor data into a MIDI 1.0 cable command stream
   and then transcodes the stream onto an RTP MIDI stream.

   The comex interpretation of timestamps is usually not the best
   timestamp interpretation for transcoding a MIDI source that uses
   implicit command timing (such as MIDI 1.0 DIN cables) into an RTP
   MIDI stream.  Appendix C.3 defines alternatives to comex semantics
   and describes session configuration tools for selecting the timestamp
   interpretation semantics for a stream.

        One-Octet Delta Time:

           Encoded form: 0ddddddd
           Decoded form: 00000000 00000000 00000000 0ddddddd

        Two-Octet Delta Time:

           Encoded form: 1ccccccc 0ddddddd
           Decoded form: 00000000 00000000 00cccccc cddddddd

        Three-Octet Delta Time:

           Encoded form: 1bbbbbbb 1ccccccc 0ddddddd
           Decoded form: 00000000 000bbbbb bbcccccc cddddddd

        Four-Octet Delta Time:

           Encoded form: 1aaaaaaa 1bbbbbbb 1ccccccc 0ddddddd
           Decoded form: 0000aaaa aaabbbbb bbcccccc cddddddd

                  Figure 4 -- Decoding delta time formats

3.2.  Command Coding

   Each non-empty MIDI Command field in the MIDI list codes one of the
   MIDI command types that may legally appear on a MIDI 1.0 DIN cable.
   Standard MIDI File meta-events do not fit this definition and MUST
   NOT appear in the MIDI list.  As a rule, each MIDI Command field
   codes a complete command, in the binary command format defined in
   [MIDI].  In the remainder of this section, we describe exceptions to
   this rule.

   The first MIDI channel command in the MIDI list MUST include a status
   octet.  Running status coding, as defined in [MIDI], MAY be used for
   all subsequent MIDI channel commands in the list.  As in [MIDI],
   System Common and System Exclusive messages (0xF0 ... 0xF7) cancel
   the running status state, but System Real-time messages (0xF8 ...
   0xFF) do not affect the running status state.  All System commands in
   the MIDI list MUST include a status octet.

   As we note above, the first channel command in the MIDI list MUST
   include a status octet.  However, the corresponding command in the
   original MIDI source data stream might not have a status octet (in
   this case, the source would be coding the command using running
   status).  If the status octet of the first channel command in the
   MIDI list does not appear in the source data stream, the P (phantom)
   header bit MUST be set to 1.  In all other cases, the P bit MUST be
   set to 0.

   Note that the P bit describes the MIDI source data stream, not the
   MIDI list encoding; regardless of the state of the P bit, the MIDI
   list MUST include the status octet.

   As receivers MUST be able to decode running status, sender
   implementors should feel free to use running status to improve
   bandwidth efficiency.  However, senders SHOULD NOT introduce timing
   jitter into an existing MIDI command stream through an inappropriate
   use or removal of running status coding.  This warning primarily
   applies to senders whose RTP MIDI streams may be transcoded onto a
   MIDI 1.0 DIN cable [MIDI] by the receiver: both the timestamps and
   the command coding (running status or not) must comply with the
   physical restrictions of implicit time coding over a slow serial
   line.

   On a MIDI 1.0 DIN cable [MIDI], a System Real-time command may be
   embedded inside of another "host" MIDI command.  This syntactic
   construction is not supported in the payload format: a MIDI Command
   field in the MIDI list codes exactly one MIDI command (partially or
   completely).

   To encode an embedded System Real-time command, senders MUST extract
   the command from its host and code it in the MIDI list as a separate
   command.  The host command and System Real-time command SHOULD appear
   in the same MIDI list.  The delta time of the System Real-time
   command SHOULD result in a command timestamp that encodes the System
   Real-time command placement in its original embedded position.

   Two methods are provided for encoding MIDI System Exclusive (SysEx)
   commands in the MIDI list.  A SysEx command may be encoded in a MIDI
   Command field verbatim: a 0xF0 octet, followed by an arbitrary number
   of data octets, followed by a 0xF7 octet.

   Alternatively, a SysEx command may be encoded as multiple segments.
   The command is divided into two or more SysEx command segments; each
   segment is encoded in its own MIDI Command field in the MIDI list.

   The payload format supports segmentation in order to encode SysEx
   commands that encode information in the temporal pattern of data
   octets.  By encoding these commands as a series of segments, each
   data octet may be associated with a distinct delta time.
   Segmentation also supports the coding of large SysEx commands across
   several packets.

   To segment a SysEx command, first partition its data octet list into
   two or more sublists.  The last sublist MAY be empty (i.e., contain
   no octets); all other sublists MUST contain at least one data octet.
   To complete the segmentation, add the status octets defined in Figure

   5 to the head and tail of the first, last, and any "middle" sublists.
   Figure 6 shows example segmentations of a SysEx command.

   A sender MAY cancel a segmented SysEx command transmission that is in
   progress, by sending the "cancel" sublist shown in Figure 5.  A
   "cancel" sublist MAY follow a "first" or "middle" sublist in the
   transmission, but MUST NOT follow a "last" sublist.  The cancel MUST
   be empty (thus, 0xF7 0xF4 is the only legal cancel sublist).

   The cancellation feature is needed because Appendix C.1 defines
   configuration tools that let session parties exclude certain SysEx
   commands in the stream.  Senders that transcode a MIDI source onto an
   RTP MIDI stream under these constraints have the responsibility of
   excluding undesired commands from the RTP MIDI stream.

   The cancellation feature lets a sender start the transmission of a
   command before the MIDI source has sent the entire command.  If a
   sender determines that the command whose transmission is in progress
   should not appear on the RTP stream, it cancels the command.  Without
   a method for cancelling a SysEx command transmission, senders would
   be forced to use a high-latency store-and-forward approach to
   transcoding SysEx commands onto RTP MIDI packets, in order to
   validate each SysEx command before transmission.

   The recommended receiver reaction to a cancellation depends on the
   capabilities of the receiver.  For example, a sound synthesizer that
   is directly parsing RTP MIDI packets and rendering them to audio will
   be aware of the fact that SysEx commands may be cancelled in RTP
   MIDI.  These receivers SHOULD detect a SysEx cancellation in the MIDI
   list and act as if they had never received the SysEx command.

   As a second example, a synthesizer may be receiving MIDI data from an
   RTP MIDI stream via a MIDI DIN cable (or a software API emulation of
   a MIDI DIN cable).  In this case, an RTP-MIDI-aware system receives
   the RTP MIDI stream and transcodes it onto the MIDI DIN cable (or its
   emulation).  Upon the receipt of the cancel sublist, the RTP-MIDI-
   aware transcoder might have already sent the first part of the SysEx
   command on the MIDI DIN cable to the receiver.

   Unfortunately, the MIDI DIN cable protocol cannot directly code
   "cancel SysEx in progress" semantics.  However, MIDI DIN cable
   receivers begin SysEx processing after the complete command arrives.
   The receiver checks to see if it recognizes the command (coded in the
   first few octets) and then checks to see if the command is the
   correct length.  Thus, in practice, a transcoder can cancel a SysEx
   command by sending an 0xF7 to (prematurely) end the SysEx command --
   the receiver will detect the incorrect command length and discard the
   command.

   Appendix C.1 defines configuration tools that may be used to prohibit
   SysEx command cancellation.

   The relative ordering of SysEx command segments in a MIDI list must
   match the relative ordering of the sublists in the original SysEx
   command.  By default, commands other than System Real-time MIDI
   commands MUST NOT appear between SysEx command segments (Appendix C.1
   defines configuration tools to change this default, to let other
   commands types appear between segments).  If the command segments of
   a SysEx command are placed in the MIDI lists of two or more RTP
   packets, the segment ordering rules apply to the concatenation of all
   affected MIDI lists.

          -----------------------------------------------------------
         | Sublist Position |  Head Status Octet | Tail Status Octet |
         |-----------------------------------------------------------|
         |    first         |       0xF0         |       0xF0        |
         |-----------------------------------------------------------|
         |    middle        |       0xF7         |       0xF0        |
         |-----------------------------------------------------------|
         |    last          |       0xF7         |       0xF7        |
         |-----------------------------------------------------------|
         |    cancel        |       0xF7         |       0xF4        |
          -----------------------------------------------------------

               Figure 5 -- Command segmentation status octets

   [MIDI] permits 0xF7 octets that are not part of a (0xF0, 0xF7) pair
   to appear on a MIDI 1.0 DIN cable.  Unpaired 0xF7 octets have no
   semantic meaning in MIDI, apart from cancelling running status.

   Unpaired 0xF7 octets MUST NOT appear in the MIDI list of the MIDI
   Command section.  We impose this restriction to avoid interference
   with the command segmentation coding defined in Figure 5.

   SysEx commands carried on a MIDI 1.0 DIN cable may use the "dropped
   0xF7" construction [MIDI].  In this coding method, the 0xF7 octet is
   dropped from the end of the SysEx command, and the status octet of
   the next MIDI command acts both to terminate the SysEx command and
   start the next command.  To encode this construction in the payload
   format, follow these steps:

     o  Determine the appropriate delta times for the SysEx command and
        the command that follows the SysEx command.

     o  Insert the "dropped" 0xF7 octet at the end of the SysEx command,
        to form the standard SysEx syntax.

     o  Code both commands into the MIDI list using the rules above.

     o  Replace the 0xF7 octet that terminates the verbatim SysEx
        encoding or the last segment of the segmented SysEx encoding
        with a 0xF5 octet.  This substitution informs the receiver of
        the original dropped 0xF7 coding.

   [MIDI] reserves the undefined System Common commands 0xF4 and 0xF5
   and the undefined System Real-time commands 0xF9 and 0xFD for future
   use.  By default, undefined commands MUST NOT appear in a MIDI
   Command field in the MIDI list, with the exception of the 0xF5 octets
   used to code the "dropped 0xF7" construction and the 0xF4 octets used
   by SysEx "cancel" sublists.

   During session configuration, a stream may be customized to transport
   undefined commands (Appendix C.1).  For this case, we now define how
   senders encode undefined commands in the MIDI list.

   An undefined System Real-time command MUST be coded using the System
   Real-time rules.

   If the undefined System Common commands are put to use in a future
   version of [MIDI], the command will begin with an 0xF4 or 0xF5 status
   octet, followed by an arbitrary number of data octets (i.e., zero or
   more data bytes).  To encode these commands, senders MUST terminate
   the command with an 0xF7 octet and place the modified command into
   the MIDI Command field.

   Unfortunately, non-compliant uses of the undefined System Common
   commands may appear in MIDI implementations.  To model these
   commands, we assume that the command begins with an 0xF4 or 0xF5
   status octet, followed by zero or more data octets, followed by zero
   or more trailing 0xF7 status octets.  To encode the command, senders
   MUST first remove all trailing 0xF7 status octets from the command.
   Then, senders MUST terminate the command with an 0xF7 octet and place
   the modified command into the MIDI Command field.

   Note that we include the trailing octets in our model as a cautionary
   measure: if such commands appeared in a non-compliant use of an
   undefined System Common command, an RTP MIDI encoding of the command
   that did not remove trailing octets could be mistaken for an encoding
   of "middle" or "last" sublist of a segmented SysEx commands (Figure
   5) under certain packet loss conditions.

          Original SysEx command:

              0xF0 0x01 0x02 0x03 0x04 0x05 0x06 0x07 0x08 0xF7

          A two-segment segmentation:

              0xF0 0x01 0x02 0x03 0x04 0xF0

              0xF7 0x05 0x06 0x07 0x08 0xF7

          A different two-segment segmentation:

              0xF0 0x01 0xF0

              0xF7 0x02 0x03 0x04 0x05 0x06 0x07 0x08 0xF7

          A three-segment segmentation:

              0xF0 0x01 0x02 0xF0

              0xF7 0x03 0x04 0xF0

              0xF7 0x05 0x06 0x07 0x08 0xF7

         The segmentation with the largest number of segments:

              0xF0 0x01 0xF0

              0xF7 0x02 0xF0

              0xF7 0x03 0xF0

              0xF7 0x04 0xF0

              0xF7 0x05 0xF0

              0xF7 0x06 0xF0

              0xF7 0x07 0xF0

              0xF7 0x08 0xF0

              0xF7 0xF7

                     Figure 6 -- Example segmentations

4.  The Recovery Journal System

   The recovery journal is the default resiliency tool for unreliable
   transport.  In this section, we normatively define the roles that
   senders and receivers play in the recovery journal system.

   MIDI is a fragile code.  A single lost command in a MIDI command
   stream may produce an artifact in the rendered performance.  We
   normatively classify rendering artifacts into two categories:

     o Transient artifacts.  Transient artifacts produce immediate but
       short-term glitches in the performance.  For example, a lost
       NoteOn (0x9) command produces a transient artifact: one note
       fails to play, but the artifact does not extend beyond the end of
       that note.

     o Indefinite artifacts.  Indefinite artifacts produce long-lasting
       errors in the rendered performance.  For example, a lost NoteOff
       (0x8) command may produce an indefinite artifact: the note that
       should have been ended by the lost NoteOff command may sustain
       indefinitely.  As a second example, the loss of a Control Change
       (0xB) command for controller number 7 (Channel Volume) may
       produce an indefinite artifact: after the loss, all notes on the
       channel may play too softly or too loudly.

   The purpose of the recovery journal system is to satisfy the recovery
   journal mandate: the MIDI performance rendered from an RTP MIDI
   stream sent over unreliable transport MUST NOT contain indefinite
   artifacts.

   The recovery journal system does not use packet retransmission to
   satisfy this mandate.  Instead, each packet includes a special
   section, called the recovery journal.

   The recovery journal codes the history of the stream, back to an
   earlier packet called the checkpoint packet.  The range of coverage
   for the journal is called the checkpoint history.  The recovery
   journal codes the information necessary to recover from the loss of
   an arbitrary number of packets in the checkpoint history.  Appendix
   A.1 normatively defines the checkpoint packet and the checkpoint
   history.

   When a receiver detects a packet loss, it compares its own knowledge
   about the history of the stream with the history information coded in
   the recovery journal of the packet that ends the loss event.  By
   noting the differences in these two versions of the past, a receiver
   is able to transform all indefinite artifacts in the rendered

   performance into transient artifacts, by executing MIDI commands to
   repair the stream.

   We now state the normative role for senders in the recovery journal
   system.

   Senders prepare a recovery journal for every packet in the stream.
   In doing so, senders choose the checkpoint packet identity for the
   journal.  Senders make this choice by applying a sending policy.
   Appendix C.2.2 normatively defines three sending policies: "closed-
   loop", "open-loop", and "anchor".

   By default, senders MUST use the closed-loop sending policy.  If the
   session description overrides this default policy, by using the
   parameter j_update defined in Appendix C.2.2, senders MUST use the
   specified policy.

   After choosing the checkpoint packet identity for a packet, the
   sender creates the recovery journal.  By default, this journal MUST
   conform to the normative semantics in Section 5 and Appendices A-B in
   this memo.  In Appendix C.2.3, we define parameters that modify the
   normative semantics for recovery journals.  If the session
   description uses these parameters, the journal created by the sender
   MUST conform to the modified semantics.

   Next, we state the normative role for receivers in the recovery
   journal system.

   A receiver MUST detect each RTP sequence number break in a stream.
   If the sequence number break is due to a packet loss event (as
   defined in [RFC3550]), the receiver MUST repair all indefinite
   artifacts in the rendered MIDI performance caused by the loss.  If
   the sequence number break is due to an out-of-order packet (as
   defined in [RFC3550]), the receiver MUST NOT take actions that
   introduce indefinite artifacts (ignoring the out-of-order packet is a
   safe option).

   Receivers take special precautions when entering or exiting a
   session.  A receiver MUST process the first received packet in a
   stream as if it were a packet that ends a loss event.  Upon exiting a
   session, a receiver MUST ensure that the rendered MIDI performance
   does not end with indefinite artifacts.

   Receivers are under no obligation to perform indefinite artifact
   repairs at the moment a packet arrives.  A receiver that uses a
   playout buffer may choose to wait until the moment of rendering
   before processing the recovery journal, as the "lost" packet may be a
   late packet that arrives in time to use.

   Next, we state the normative role for the creator of the session
   description in the recovery journal system.  Depending on the
   application, the sender, the receivers, and other parties may take
   part in creating or approving the session description.

   A session description that specifies the default closed-loop sending
   policy and the default recovery journal semantics satisfies the
   recovery journal mandate.  However, these default behaviors may not
   be appropriate for all sessions.  If the creators of a session
   description use the parameters defined in Appendix C.2 to override
   these defaults, the creators MUST ensure that the parameters define a
   system that satisfies the recovery journal mandate.

   Finally, we note that this memo does not specify sender or receiver
   recovery journal algorithms.  Implementations are free to use any
   algorithm that conforms to the requirements in this section.  The
   non-normative [RFC4696] discusses sender and receiver algorithm
   design.

5.  Recovery Journal Format

   This section introduces the structure of the recovery journal and
   defines the bitfields of recovery journal headers.  Appendices A-B
   complete the bitfield definition of the recovery journal.

   The recovery journal has a three-level structure:

     o Top-level header.

     o Channel and system journal headers.  These headers encode
       recovery information for a single voice channel (channel journal)
       or for all systems commands (system journal).

     o Chapters.  Chapters describe recovery information for a single
       MIDI command type.

   Figure 7 shows the top-level structure of the recovery journal.  The
   recovery journals consists of a 3-octet header, followed by an
   optional system journal (labeled S-journal in Figure 7) and an
   optional list of channel journals.  Figure 8 shows the recovery
   journal header format.

       0                   1                   2                   3
       0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
      |            Recovery journal header            | S-journal ... |
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
      |                      Channel journals ...                     |
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

                Figure 7 -- Top-level recovery journal format

              0                   1                   2
              0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3
             +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
             |S|Y|A|H|TOTCHAN|   Checkpoint Packet Seqnum    |
             +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

                    Figure 8 -- Recovery journal header

   If the Y header bit is set to 1, the system journal appears in the
   recovery journal, directly following the recovery journal header.

   If the A header bit is set to 1, the recovery journal ends with a
   list of (TOTCHAN + 1) channel journals (the 4-bit TOTCHAN header
   field is interpreted as an unsigned integer).

   A MIDI channel MAY be represented by (at most) one channel journal in
   a recovery journal.  Channel journals MUST appear in the recovery
   journal in ascending channel-number order.

   If A and Y are both zero, the recovery journal only contains its 3-
   octet header and is considered to be an "empty" journal.

   The S (single-packet loss) bit appears in most recovery journal
   structures, including the recovery journal header.  The S bit helps
   receivers efficiently parse the recovery journal in the common case
   of the loss of a single packet.  Appendix A.1 defines S bit
   semantics.

   The H bit indicates if MIDI channels in the stream have been
   configured to use the enhanced Chapter C encoding (Appendix A.3.3).

   By default, the payload format does not use enhanced Chapter C
   encoding.  In this default case, the H bit MUST be set to 0 for all
   packets in the stream.

   If the stream has been configured so that controller numbers for one
   or more MIDI channels use enhanced Chapter C encoding, the H bit MUST
   be set to 1 in all packets in the stream.  In Appendix C.2.3, we show
   how to configure a stream to use enhanced Chapter C encoding.

   The 16-bit Checkpoint Packet Seqnum header field codes the sequence
   number of the checkpoint packet for this journal, in network byte
   order (big-endian).  The choice of the checkpoint packet sets the
   depth of the checkpoint history for the journal (defined in Appendix
   A.1).

   Receivers may use the Checkpoint Packet Seqnum field of the packet
   that ends a loss event to verify that the journal checkpoint history
   covers the entire loss event.  The checkpoint history covers the loss
   event if the Checkpoint Packet Seqnum field is less than or equal to
   one plus the highest RTP sequence number previously received on the
   stream (modulo 2^16).

       0                   1                   2                   3
       0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
      |S| CHAN  |H|      LENGTH       |P|C|M|W|N|E|T|A|  Chapters ... |
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

                     Figure 9 -- Channel journal format

   Figure 9 shows the structure of a channel journal: a 3-octet header,
   followed by a list of leaf elements called channel chapters.  A
   channel journal encodes information about MIDI commands on the MIDI
   channel coded by the 4-bit CHAN header field.  Note that CHAN uses
   the same bit encoding as the channel nibble in MIDI Channel Messages
   (the cccc field in Figure E.1 of Appendix E).

   The 10-bit LENGTH field codes the length of the channel journal.  The
   semantics for LENGTH fields are uniform throughout the recovery
   journal, and are defined in Appendix A.1.

   The third octet of the channel journal header is the Table of
   Contents (TOC) of the channel journal.  The TOC is a set of bits that
   encode the presence of a chapter in the journal.  Each chapter
   contains information about a certain class of MIDI channel command:

      o  Chapter P: MIDI Program Change (0xC)
      o  Chapter C: MIDI Control Change (0xB)
      o  Chapter M: MIDI Parameter System (part of 0xB)
      o  Chapter W: MIDI Pitch Wheel (0xE)
      o  Chapter N: MIDI NoteOff (0x8), NoteOn (0x9)
      o  Chapter E: MIDI Note Command Extras (0x8, 0x9)

      o  Chapter T: MIDI Channel Aftertouch (0xD)
      o  Chapter A: MIDI Poly Aftertouch (0xA)

   Chapters appear in a list following the header, in order of their
   appearance in the TOC.  Appendices A.2-9 describe the bitfield format
   for each chapter, and define the conditions under which a chapter
   type MUST appear in the recovery journal.  If any chapter types are
   required for a channel, an associated channel journal MUST appear in
   the recovery journal.

   The H bit indicates if controller numbers on a MIDI channel have been
   configured to use the enhanced Chapter C encoding (Appendix A.3.3).

   By default, controller numbers on a MIDI channel do not use enhanced
   Chapter C encoding.  In this default case, the H bit MUST be set to 0
   for all channel journal headers for the channel in the recovery
   journal, for all packets in the stream.

   However, if at least one controller number for a MIDI channel has
   been configured to use the enhanced Chapter C encoding, the H bit for
   its channel journal MUST be set to 1, for all packets in the stream.

   In Appendix C.2.3, we show how to configure a controller number to
   use enhanced Chapter C encoding.

       0                   1                   2                   3
       0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
      |S|D|V|Q|F|X|      LENGTH       |  System chapters ...          |
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

                     Figure 10 -- System journal format

   Figure 10 shows the structure of the system journal: a 2-octet
   header, followed by a list of system chapters.  Each chapter codes
   information about a specific class of MIDI Systems command:

      o  Chapter D: Song Select (0xF3), Tune Request (0xF6), Reset
                    (0xFF), undefined System commands (0xF4, 0xF5, 0xF9,
                    0xFD)
      o  Chapter V: Active Sense (0xFE)
      o  Chapter Q: Sequencer State (0xF2, 0xF8, 0xF9, 0xFA, 0xFB, 0xFC)
      o  Chapter F: MTC Tape Position (0xF1, 0xF0 0x7F 0xcc 0x01 0x01)
      o  Chapter X: System Exclusive (all other 0xF0)

   The 10-bit LENGTH field codes the size of the system journal and
   conforms to semantics described in Appendix A.1.

   The D, V, Q, F, and X header bits form a Table of Contents (TOC) for
   the system journal.  A TOC bit that is set to 1 codes the presence of
   a chapter in the journal.  Chapters appear in a list following the
   header, in the order of their appearance in the TOC.

   Appendix B describes the bitfield format for the system chapters and
   defines the conditions under which a chapter type MUST appear in the
   recovery journal.  If any system chapter type is required to appear
   in the recovery journal, the system journal MUST appear in the
   recovery journal.

6.  Session Description Protocol

   RTP does not perform session management.  Instead, RTP works together
   with session management tools, such as the Session Initiation
   Protocol (SIP, [RFC3261]) and the Real Time Streaming Protocol (RTSP,
   [RFC2326]).

   RTP payload formats define media type parameters for use in session
   management (for example, this memo defines "rtp-midi" as the media
   type for native RTP MIDI streams).

   In most cases, session management tools use the media type parameters
   via another standard, the Session Description Protocol (SDP,
   [RFC4566]).

   SDP is a textual format for specifying session descriptions.  Session
   descriptions specify the network transport and media encoding for RTP
   sessions.  Session management tools coordinate the exchange of
   session descriptions between participants ("parties").

   Some session management tools use SDP to negotiate details of media
   transport (network addresses, ports, etc.).  We refer to this use of
   SDP as "negotiated usage".  One example of negotiated usage is the
   Offer/Answer protocol ([RFC3264] and Appendix C.7.2 in this memo) as
   used by SIP.

   Other session management tools use SDP to declare the media encoding
   for the session but use other techniques to negotiate network
   transport.  We refer to this use of SDP as "declarative usage".  One
   example of declarative usage is RTSP ([RFC2326] and Appendix C.7.1 in
   this memo).

   Below, we show session description examples for native (Section 6.1)
   and mpeg4-generic (Section 6.2) streams.  In Section 6.3, we
   introduce session configuration tools that may be used to customize
   streams.

6.1.  Session Descriptions for Native Streams

   The session description below defines a unicast UDP RTP session (via
   a media ("m=") line) whose sole payload type (96) is mapped to a
   minimal native RTP MIDI stream.

   v=0
   o=lazzaro 2520644554 2838152170 IN IP4 first.example.net
   s=Example
   t=0 0
   m=audio 5004 RTP/AVP 96
   c=IN IP4 192.0.2.94
   a=rtpmap:96 rtp-midi/44100

   The rtpmap attribute line uses the "rtp-midi" media type to specify
   an RTP MIDI native stream.  The clock rate specified on the rtpmap
   line (in the example above, 44100 Hz) sets the scaling for the RTP
   timestamp header field (see Section 2.1, and also [RFC3550]).

   Note that this document does not specify a default clock rate value
   for RTP MIDI.  When RTP MIDI is used with SDP, parties MUST use the
   rtpmap line to communicate the clock rate.  Guidance for selecting
   the RTP MIDI clock rate value appears in Section 2.1.

   We consider the RTP MIDI stream shown above to be "minimal" because
   the session description does not customize the stream with
   parameters.  Without such customization, a native RTP MIDI stream has
   these characteristics:

     1. If the stream uses unreliable transport (unicast UDP, multicast
        UDP, etc.), the recovery journal system is in use, and the RTP
        payload contains both the MIDI command section and the journal
        section.  If the stream uses reliable transport (such as TCP),
        the stream does not use journalling, and the payload contains
        only the MIDI command section (Section 2.2).

     2. If the stream uses the recovery journal system, the recovery
        journal system uses the default sending policy and the default
        journal semantics (Section 4).

     3. In the MIDI command section of the payload, command timestamps
        use the default "comex" semantics (Section 3).

     4. The recommended temporal duration ("media time") of an RTP
        packet ranges from 0 to 200 ms, and the RTP timestamp difference
        between sequential packets in the stream may be arbitrarily
        large (Section 2.1).

     5. If more than one minimal rtp-midi stream appears in a session,
        the MIDI name spaces for these streams are independent: channel
        1 in the first stream does not reference the same MIDI channel
        as channel 1 in the second stream (see Appendix C.5 for a
        discussion of the independence of minimal rtp-midi streams).

     6. The rendering method for the stream is not specified.  What the
        receiver "does" with a minimal native MIDI stream is "out of
        scope" of this memo.  For example, in content creation
        environments, a user may manually configure client software to
        render the stream with a specific software package.

   As in standard in RTP, RTP sessions managed by SIP are sendrecv by
   default (parties send and receive MIDI), and RTP sessions managed by
   RTSP are recvonly by default (server sends and client receives).

   In sendrecv RTP MIDI sessions for the session description shown
   above, the 16 voice channel + systems MIDI name space is unique for
   each sender.  Thus, in a two-party session, the voice channel 0 sent
   by one party is distinct from the voice channel 0 sent by the other
   party.

   This behavior corresponds to what occurs when two MIDI 1.0 DIN
   devices are cross-connected with two MIDI cables (one cable routing
   MIDI Out from the first device into MIDI In of the second device, a
   second cable routing MIDI In from the first device into MIDI Out of
   the second device).  We define this "association" formally in Section
   2.1.

   MIDI 1.0 DIN networks may be configured in a "party-line" multicast
   topology.  For these networks, the MIDI protocol itself provides
   tools for addressing specific devices in transactions on a multicast
   network, and for device discovery.  Thus, apart from providing a 1-
   to-many forward path and a many-to-1 reverse path, IETF protocols do
   not need to provide any special support for MIDI multicast
   networking.

6.2.  Session Descriptions for mpeg4-generic Streams

   An mpeg4-generic [RFC3640] RTP MIDI stream uses an MPEG 4 Audio
   Object Type to render MIDI into audio.  Three Audio Object Types
   accept MIDI input:

     o General MIDI (Audio Object Type ID 15), based on the General MIDI
       rendering standard [MIDI].

     o Wavetable Synthesis (Audio Object Type ID 14), based on the
       Downloadable Sounds Level 2 (DLS 2) rendering standard [DLS2].

     o Main Synthetic (Audio Object Type ID 13), based on Structured
       Audio and the programming language SAOL [MPEGSA].

   The primary service of an mpeg4-generic stream is to code Access
   Units (AUs).  We define the mpeg4-generic RTP MIDI AU as the MIDI
   payload shown in Figure 1 of Section 2.1 of this memo: a MIDI command
   section optionally followed by a journal section.

   Exactly one RTP MIDI AU MUST be mapped to one mpeg4-generic RTP MIDI
   packet.  The mpeg4-generic options for placing several AUs in an RTP
   packet MUST NOT be used with RTP MIDI.  The mpeg4-generic options for
   fragmenting and interleaving AUs MUST NOT be used with RTP MIDI.  The
   mpeg4-generic RTP packet payload (Figure 1 in [RFC3640]) MUST contain
   empty AU Header and Auxiliary sections.  These rules yield mpeg4-
   generic packets that are structurally identical to native RTP MIDI
   packets, an essential property for the correct operation of the
   payload format.

   The session description that follows defines a unicast UDP RTP
   session (via a media ("m=") line) whose sole payload type (96) is
   mapped to a minimal mpeg4-generic RTP MIDI stream.  This example uses
   the General MIDI Audio Object Type under Synthesis Profile @ Level 2.

   v=0
   o=lazzaro 2520644554 2838152170 IN IP6 first.example.net
   s=Example
   t=0 0
   m=audio 5004 RTP/AVP 96
   c=IN IP6 2001:DB80::7F2E:172A:1E24
   a=rtpmap:96 mpeg4-generic/44100
   a=fmtp:96 streamtype=5; mode=rtp-midi; profile-level-id=12;
   config=7A0A0000001A4D546864000000060000000100604D54726B0000
   000600FF2F000

   (The a=fmtp line has been wrapped to fit the page to accommodate memo
   formatting restrictions; it comprises a single line in SDP.)

   The fmtp attribute line codes the four parameters (streamtype, mode,
   profile-level-id, and config) that are required in all mpeg4-generic
   session descriptions [RFC3640].  For RTP MIDI streams, the streamtype
   parameter MUST be set to 5, the "mode" parameter MUST be set to
   "rtp-midi", and the "profile-level-id" parameter MUST be set to the
   MPEG-4 Profile Level for the stream.  For the Synthesis Profile,
   legal profile-level-id values are 11, 12, and 13, coding low (11),
   medium (12), or high (13) decoder computational complexity, as
   defined by MPEG conformance tests.

   In a minimal RTP MIDI session description, the config value MUST be a
   hexadecimal encoding [RFC3640] of the AudioSpecificConfig data block
   [MPEGAUDIO] for the stream.  AudioSpecificConfig encodes the Audio
   Object Type for the stream and also encodes initialization data (SAOL
   programs, DLS 2 wave tables, etc.).  Standard MIDI Files encoded in
   AudioSpecificConfig in a minimal session description MUST be ignored
   by the receiver.

   Receivers determine the rendering algorithm for the session by
   interpreting the first 5 bits of AudioSpecificConfig as an unsigned
   integer that codes the Audio Object Type.  In our example above, the
   leading config string nibbles "7A" yield the Audio Object Type 15
   (General MIDI).  In Appendix E.4, we derive the config string value
   in the session description shown above; the starting point of the
   derivation is the MPEG bitstreams defined in [MPEGSA] and
   [MPEGAUDIO].

   We consider the stream to be "minimal" because the session
   description does not customize the stream through the use of
   parameters, other than the 4 required mpeg4-generic parameters
   described above.  In Section 6.1, we describe the behavior of a
   minimal native stream, as a numbered list of characteristics.  Items
   1-4 on that list also describe the minimal mpeg4-generic stream, but
   items 5 and 6 require restatements, as listed below:

     5. If more than one minimal mpeg4-generic stream appears in a
        session, each stream uses an independent instance of the Audio
        Object Type coded in the config parameter value.

     6. A minimal mpeg4-generic stream encodes the AudioSpecificConfig
        as an inline hexadecimal constant.  If a session description is
        sent over UDP, it may be impossible to transport large
        AudioSpecificConfig blocks within the Maximum Transmission Size
        (MTU) of the underlying network (for Ethernet, the MTU is 1500
        octets).  In some cases, the AudioSpecificConfig block may
        exceed the maximum size of the UDP packet itself.

   The comments in Section 6.1 on SIP and RTSP stream directional
   defaults, sendrecv MIDI channel usage, and MIDI 1.0 DIN multicast
   networks also apply to mpeg4-generic RTP MIDI sessions.

   In sendrecv sessions, each party's session description MUST use
   identical values for the mpeg4-generic parameters (including the
   required streamtype, mode, profile-level-id, and config parameters).
   As a consequence, each party uses an identically configured MPEG 4
   Audio Object Type to render MIDI commands into audio.  The preamble
   to Appendix C discusses a way to create "virtual sendrecv" sessions
   that do not have this restriction.

6.3.  Parameters

   This section introduces parameters for session configuration for RTP
   MIDI streams.  In session descriptions, parameters modify the
   semantics of a payload type.  Parameters are specified on an fmtp
   attribute line.  See the session description example in Section 6.2
   for an example of a fmtp attribute line.

   The parameters add features to the minimal streams described in
   Sections 6.1-2, and support several types of services:

     o  Stream subsetting.  By default, all MIDI commands that are legal
        to appear on a MIDI 1.0 DIN cable may appear in an RTP MIDI
        stream.  The cm_unused parameter overrides this default by
        prohibiting certain commands from appearing in the stream.  The
        cm_used parameter is used in conjunction with cm_unused, to
        simplify the specification of complex exclusion rules.  We
        describe cm_unused and cm_used in Appendix C.1.

     o  Journal customization.  The j_sec and j_update parameters
        configure the use of the journal section.  The ch_default,
        ch_never, and ch_anchor parameters configure the semantics of
        the recovery journal chapters.  These parameters are described
        in Appendix C.2 and override the default stream behaviors 1 and
        2, listed in Section 6.1 and referenced in Section 6.2.

     o  MIDI command timestamp semantics.  The tsmode, octpos, mperiod,
        and linerate parameters customize the semantics of timestamps in
        the MIDI command section.  These parameters let RTP MIDI
        accurately encode the implicit time coding of MIDI 1.0 DIN
        cables.  These parameters are described in Appendix C.3 and
        override default stream behavior 3, listed in Section 6.1 and
        referenced in Section 6.2

     o  Media time.  The rtp_ptime and rtp_maxptime parameters define
        the temporal duration ("media time") of an RTP MIDI packet.  The
        guardtime parameter sets the minimum sending rate of stream
        packets.  These parameters are described in Appendix C.4 and
        override default stream behavior 4, listed in Section 6.1 and
        referenced in Section 6.2.

     o  Stream description.  The musicport parameter labels the MIDI
        name space of RTP streams in a multimedia session.  Musicport is
        described in Appendix C.5.  The musicport parameter overrides
        default stream behavior 5, in Sections 6.1 and 6.2.

     o  MIDI rendering.  Several parameters specify the MIDI rendering
        method of a stream.  These parameters are described in Appendix
        C.6 and override default stream behavior 6, in Sections 6.1 and
        6.2.

   In Appendix C.7, we specify interoperability guidelines for two RTP
   MIDI application areas: content-streaming using RTSP (Appendix C.7.1)
   and network musical performance using SIP (Appendix C.7.2).

7.  Extensibility

   The payload format defined in this memo exclusively encodes all
   commands that may legally appear on a MIDI 1.0 DIN cable.

   Many worthy uses of MIDI over RTP do not fall within the narrow scope
   of the payload format.  For example, the payload format does not
   support the direct transport of Standard MIDI File (SMF) meta-event
   and metric timing data.  As a second example, the payload format does
   not define transport tools for user-defined commands (apart from
   tools to support System Exclusive commands [MIDI]).

   The payload format does not provide an extension mechanism to support
   new features of this nature, by design.  Instead, we encourage the
   development of new payload formats for specialized musical
   applications.  The IETF session management tools [RFC3264] [RFC2326]
   support codec negotiation, to facilitate the use of new payload
   formats in a backward-compatible way.

   However, the payload format does provide several extensibility tools,
   which we list below:

     o  Journalling.  As described in Appendix C.2, new token values for
        the j_sec and j_update parameters may be defined in IETF
        standards-track documents.  This mechanism supports the design
        of new journal formats and the definition of new journal sending
        policies.

     o  Rendering.  The payload format may be extended to support new
        MIDI renderers (Appendix C.6.2).  Certain general aspects of the
        RTP MIDI rendering process may also be extended, via the
        definition of new token values for the render (Appendix C.6) and
        smf_info (Appendix C.6.4.1) parameters.

     o  Undefined commands.  [MIDI] reserves 4 MIDI System commands for
        future use (0xF4, 0xF5, 0xF9, 0xFD).  If updates to [MIDI]
        define the reserved commands, IETF standards-track documents may
        be defined to provide resiliency support for the commands.

        Opaque LEGAL fields appear in System Chapter D for this purpose
        (Appendix B.1.1).

   A final form of extensibility involves the inclusion of the payload
   format in framework documents.  Framework documents describe how to
   combine protocols to form a platform for interoperable applications.
   For example, a stage and studio framework might define how to use SIP
   [RFC3261], RTSP [RFC2326], SDP [RFC4566], and RTP [RFC3550] to
   support media networking for professional audio equipment and
   electronic musical instruments.

8.  Congestion Control

   The RTP congestion control requirements defined in [RFC3550] apply to
   RTP MIDI sessions, and implementors should carefully read the
   congestion control section in [RFC3550].  As noted in [RFC3550], all
   transport protocols used on the Internet need to address congestion
   control in some way, and RTP is not an exception.

   In addition, the congestion control requirements defined in [RFC3551]
   applies to RTP MIDI sessions run under applicable profiles.  The
   basic congestion control requirement defined in [RFC3551] is that RTP
   sessions that use UDP transport should monitor packet loss (via RTCP
   or other means) to ensure that the RTP stream competes fairly with
   TCP flows that share the network.

   Finally, RTP MIDI has congestion control issues that are unique for
   an audio RTP payload format.  In applications such as network musical
   performance [NMP], the packet rate is linked to the gestural rate of
   a human performer.  Senders MUST monitor the MIDI command source for
   patterns that result in excessive packet rates and take actions
   during RTP transcoding to reduce the RTP packet rate.  [RFC4696]
   offers implementation guidance on this issue.

9.  Security Considerations

   Implementors should carefully read the Security Considerations
   sections of the RTP [RFC3550], AVP [RFC3551], and other RTP profile
   documents, as the issues discussed in these sections directly apply
   to RTP MIDI streams.  Implementors should also review the Secure
   Real-time Transport Protocol (SRTP, [RFC3711]), an RTP profile that
   addresses the security issues discussed in [RFC3550] and [RFC3551].

   Here, we discuss security issues that are unique to the RTP MIDI
   payload format.

   When using RTP MIDI, authentication of incoming RTP and RTCP packets
   is RECOMMENDED.  Per-packet authentication may be provided by SRTP or

   by other means.  Without the use of authentication, attackers could
   forge MIDI commands into an ongoing stream, damaging speakers and
   eardrums.  An attacker could also craft RTP and RTCP packets to
   exploit known bugs in the client and take effective control of a
   client machine.

   Session management tools (such as SIP [RFC3261]) SHOULD use
   authentication during the transport of all session descriptions
   containing RTP MIDI media streams.  For SIP, the Security
   Considerations section in [RFC3261] provides an overview of possible
   authentication mechanisms.  RTP MIDI session descriptions should use
   authentication because the session descriptions may code
   initialization data using the parameters described in Appendix C.  If
   an attacker inserts bogus initialization data into a session
   description, he can corrupt the session or forge an client attack.

   Session descriptions may also code renderer initialization data by
   reference, via the url (Appendix C.6.3) and smf_url (Appendix
   C.6.4.2) parameters.  If the coded URL is spoofed, both session and
   client are open to attack, even if the session description itself is
   authenticated.  Therefore, URLs specified in url and smf_url
   parameters SHOULD use [RFC2818].

   Section 2.1 allows streams sent by a party in two RTP sessions to
   have the same SSRC value and the same RTP timestamp initialization
   value, under certain circumstances.  Normally, these values are
   randomly chosen for each stream in a session, to make plaintext
   guessing harder to do if the payloads are encrypted.  Thus, Section
   2.1 weakens this aspect of RTP security.

10.  Acknowledgements

   We thank the networking, media compression, and computer music
   community members who have commented or contributed to the effort,
   including Kurt B, Cynthia Bruyns, Steve Casner, Paul Davis, Robin
   Davies, Joanne Dow, Tobias Erichsen, Nicolas Falquet, Dominique
   Fober, Philippe Gentric, Michael Godfrey, Chris Grigg, Todd Hager,
   Michel Jullian, Phil Kerr, Young-Kwon Lim, Jessica Little, Jan van
   der Meer, Colin Perkins, Charlie Richmond, Herbie Robinson, Larry
   Rowe, Eric Scheirer, Dave Singer, Martijn Sipkema, William Stewart,
   Kent Terry, Magnus Westerlund, Tom White, Jim Wright, Doug Wyatt, and
   Giorgio Zoia.  We also thank the members of the San Francisco Bay
   Area music and audio community for creating the context for the work,
   including Don Buchla, Chris Chafe, Richard Duda, Dan Ellis, Adrian
   Freed, Ben Gold, Jaron Lanier, Roger Linn, Richard Lyon, Dana Massie,
   Max Mathews, Keith McMillen, Carver Mead, Nelson Morgan, Tom
   Oberheim, Malcolm Slaney, Dave Smith, Julius Smith, David Wessel, and
   Matt Wright.

11.  IANA Considerations

   This section makes a series of requests to IANA.  The IANA has
   completed registration/assignments of the below requests.

   The sub-sections that follow hold the actual, detailed requests.  All
   registrations in this section are in the IETF tree and follow the
   rules of [RFC4288] and [RFC3555], as appropriate.

   In Section 11.1, we request the registration of a new media type:
   "audio/rtp-midi".  Paired with this request is a request for a
   repository for new values for several parameters associated with
   "audio/rtp-midi".  We request this repository in Section 11.1.1.

   In Section 11.2, we request the registration of a new value ("rtp-
   midi") for the "mode" parameter of the "mpeg4-generic" media type.
   The "mpeg4-generic" media type is defined in [RFC3640], and [RFC3640]
   defines a repository for the "mode" parameter.  However, we believe
   we are the first to request the registration of a "mode" value, so we
   believe the registry for "mode" has not yet been created by IANA.

   Paired with our "mode" parameter value request for "mpeg4-generic" is
   a request for a repository for new values for several parameters we
   have defined for use with the "rtp-midi" mode value.  We request this
   repository in Section 11.2.1.

   In Section 11.3, we request the registration of a new media type:
   "audio/asc".  No repository request is associated with this request.

11.1.  rtp-midi Media Type Registration

   This section requests the registration of the "rtp-midi" subtype for
   the "audio" media type.  We request the registration of the
   parameters listed in the "optional parameters" section below (both
   the "non-extensible parameters" and the "extensible parameters"
   lists).  We also request the creation of repositories for the
   "extensible parameters"; the details of this request appear in
   Section 11.1.1, below.

   Media type name:

       audio

   Subtype name:

       rtp-midi

   Required parameters:

       rate: The RTP timestamp clock rate.  See Sections 2.1 and 6.1
       for usage details.

   Optional parameters:

       Non-extensible parameters:

          ch_anchor:    See Appendix C.2.3 for usage details.
          ch_default:   See Appendix C.2.3 for usage details.
          ch_never:     See Appendix C.2.3 for usage details.
          cm_unused:    See Appendix C.1 for usage details.
          cm_used:      See Appendix C.1 for usage details.
          chanmask:     See Appendix C.6.4.3 for usage details.
          cid:          See Appendix C.6.3 for usage details.
          guardtime:    See Appendix C.4.2 for usage details.
          inline:       See Appendix C.6.3 for usage details.
          linerate:     See Appendix C.3 for usage details.
          mperiod:      See Appendix C.3 for usage details.
          multimode:    See Appendix C.6.1 for usage details.
          musicport:    See Appendix C.5 for usage details.
          octpos:       See Appendix C.3 for usage details.
          rinit:        See Appendix C.6.3 for usage details.
          rtp_maxptime: See Appendix C.4.1 for usage details.
          rtp_ptime:    See Appendix C.4.1 for usage details.
          smf_cid:      See Appendix C.6.4.2 for usage details.
          smf_inline:   See Appendix C.6.4.2 for usage details.
          smf_url:      See Appendix C.6.4.2 for usage details.
          tsmode:       See Appendix C.3 for usage details.
          url:          See Appendix C.6.3 for usage details.

       Extensible parameters:

          j_sec:        See Appendix C.2.1 for usage details.  See
                        Section 11.1.1 for repository details.
          j_update:     See Appendix C.2.2 for usage details.  See
                        Section 11.1.1 for repository details.
          render:       See Appendix C.6 for usage details.  See
                        Section 11.1.1 for repository details.
          subrender:    See Appendix C.6.2 for usage details.  See
                        Section 11.1.1 for repository details.
          smf_info:     See Appendix C.6.4.1 for usage details.  See
                        Section 11.1.1 for repository details.

   Encoding considerations:

       The format for this type is framed and binary.

   Restrictions on usage:

       This type is only defined for real-time transfers of MIDI
       streams via RTP.  Stored-file semantics for rtp-midi may
       be defined in the future.

   Security considerations:

       See Section 9 of this memo.

   Interoperability considerations:

       None.

   Published specification:

       This memo and [MIDI] serve as the normative specification.  In
       addition, references [NMP], [GRAME], and [RFC4696] provide
       non-normative implementation guidance.

   Applications that use this media type:

       Audio content-creation hardware, such as MIDI controller piano
       keyboards and MIDI audio synthesizers.  Audio content-creation
       software, such as music sequencers, digital audio workstations,
       and soft synthesizers.  Computer operating systems, for network
       support of MIDI Application Programmer Interfaces.  Content
       distribution servers and terminals may use this media type for
       low bit-rate music coding.

   Additional information:

       None.

   Person & email address to contact for further information:

       John Lazzaro <lazzaro@cs.berkeley.edu>

   Intended usage:

       COMMON.

   Author:

       John Lazzaro <lazzaro@cs.berkeley.edu>

   Change controller:

       IETF Audio/Video Transport Working Group delegated
       from the IESG.

11.1.1.  Repository Request for "audio/rtp-midi"

   For the "rtp-midi" subtype, we request the creation of repositories
   for extensions to the following parameters (which are those listed as
   "extensible parameters" in Section 11.1).

      j_sec:

         Registrations for this repository may only occur
         via an IETF standards-track document.  Appendix C.2.1
         of this memo describes appropriate registrations for this
         repository.

         Initial values for this repository appear below:

         "none":  Defined in Appendix C.2.1 of this memo.
         "recj":  Defined in Appendix C.2.1 of this memo.

      j_update:

         Registrations for this repository may only occur
         via an IETF standards-track document.  Appendix C.2.2
         of this memo describes appropriate registrations for this
         repository.

         Initial values for this repository appear below:

         "anchor":  Defined in Appendix C.2.2 of this memo.
         "open-loop":  Defined in Appendix C.2.2 of this memo.
         "closed-loop":  Defined in Appendix C.2.2 of this memo.

      render:

         Registrations for this repository MUST include a
         specification of the usage of the proposed value.
         See text in the preamble of Appendix C.6 for details
         (the paragraph that begins "Other render token ...").

         Initial values for this repository appear below:

         "unknown":  Defined in Appendix C.6 of this memo.
         "synthetic":  Defined in Appendix C.6 of this memo.
         "api":  Defined in Appendix C.6 of this memo.
         "null":  Defined in Appendix C.6 of this memo.

      subrender:

         Registrations for this repository MUST include a
         specification of the usage of the proposed value.
         See text Appendix C.6.2 for details (the paragraph
         that begins "Other subrender token ...").

         Initial values for this repository appear below:

         "default":  Defined in Appendix C.6.2 of this memo.

      smf_info:

         Registrations for this repository MUST include a
         specification of the usage of the proposed value.
         See text in Appendix C.6.4.1 for details (the
         paragraph that begins "Other smf_info token ...").

         Initial values for this repository appear below:

         "ignore":  Defined in Appendix C.6.4.1 of this memo.
         "sdp_start":  Defined in Appendix C.6.4.1 of this memo.
         "identity":  Defined in Appendix C.6.4.1 of this memo.

11.2.  mpeg4-generic Media Type Registration

   This section requests the registration of the "rtp-midi" value for
   the "mode" parameter of the "mpeg4-generic" media type.  The "mpeg4-
   generic" media type is defined in [RFC3640], and [RFC3640] defines a
   repository for the "mode" parameter.  We are registering mode rtp-
   midi to support the MPEG Audio codecs [MPEGSA] that use MIDI.

   In conjunction with this registration request, we request the
   registration of the parameters listed in the "optional parameters"
   section below (both the "non-extensible parameters" and the
   "extensible parameters" lists).  We also request the creation of
   repositories for the "extensible parameters"; the details of this
   request appear in Appendix 11.2.1, below.

   Media type name:

       audio

   Subtype name:

       mpeg4-generic

   Required parameters:

       The "mode" parameter is required by [RFC3640].  [RFC3640]
       requests a repository for "mode", so that new values for mode
       may be added.  We request that the value "rtp-midi" be
       added to the "mode" repository.

       In mode rtp-midi, the mpeg4-generic parameter rate is
       a required parameter.  Rate specifies the RTP timestamp
       clock rate.  See Sections 2.1 and 6.2 for usage details
       of rate in mode rtp-midi.

   Optional parameters:

       We request registration of the following parameters
       for use in mode rtp-midi for mpeg4-generic.

       Non-extensible parameters:

          ch_anchor:    See Appendix C.2.3 for usage details.
          ch_default:   See Appendix C.2.3 for usage details.
          ch_never:     See Appendix C.2.3 for usage details.
          cm_unused:    See Appendix C.1 for usage details.
          cm_used:      See Appendix C.1 for usage details.
          chanmask:     See Appendix C.6.4.3 for usage details.
          cid:          See Appendix C.6.3 for usage details.
          guardtime:    See Appendix C.4.2 for usage details.
          inline:       See Appendix C.6.3 for usage details.
          linerate:     See Appendix C.3 for usage details.
          mperiod:      See Appendix C.3 for usage details.
          multimode:    See Appendix C.6.1 for usage details.
          musicport:    See Appendix C.5 for usage details.
          octpos:       See Appendix C.3 for usage details.
          rinit:        See Appendix C.6.3 for usage details.
          rtp_maxptime: See Appendix C.4.1 for usage details.
          rtp_ptime:    See Appendix C.4.1 for usage details.
          smf_cid:      See Appendix C.6.4.2 for usage details.
          smf_inline:   See Appendix C.6.4.2 for usage details.

          smf_url:      See Appendix C.6.4.2 for usage details.
          tsmode:       See Appendix C.3 for usage details.
          url:          See Appendix C.6.3 for usage details.

       Extensible parameters:

          j_sec:        See Appendix C.2.1 for usage details.  See
                        Section 11.2.1 for repository details.
          j_update:     See Appendix C.2.2 for usage details.  See
                        Section 11.2.1 for repository details.
          render:       See Appendix C.6 for usage details.  See
                        Section 11.2.1 for repository details.
          subrender:    See Appendix C.6.2 for usage details.  See
                        Section 11.2.1 for repository details.
          smf_info:     See Appendix C.6.4.1 for usage details.  See
                        Section 11.2.1 for repository details.

   Encoding considerations:

       The format for this type is framed and binary.

   Restrictions on usage:

       Only defined for real-time transfers of audio/mpeg4-generic
       RTP streams with mode=rtp-midi.

   Security considerations:

       See Section 9 of this memo.

   Interoperability considerations:

       Except for the marker bit (Section 2.1), the packet formats
       for audio/rtp-midi and audio/mpeg4-generic (mode rtp-midi)
       are identical.  The formats differ in use: audio/mpeg4-generic
       is for MPEG work, and audio/rtp-midi is for all other work.

   Published specification:

       This memo, [MIDI], and [MPEGSA] are the normative references.
       In addition, references [NMP], [GRAME], and [RFC4696] provide
       non-normative implementation guidance.

   Applications that use this media type:

       MPEG 4 servers and terminals that support [MPEGSA].

   Additional information:

       None.

   Person & email address to contact for further information:

       John Lazzaro <lazzaro@cs.berkeley.edu>

   Intended usage:

       COMMON.

   Author:

       John Lazzaro <lazzaro@cs.berkeley.edu>

   Change controller:

       IETF Audio/Video Transport Working Group delegated
       from the IESG.

11.2.1.  Repository Request for Mode rtp-midi for mpeg4-generic

   For mode rtp-midi of the mpeg4-generic subtype, we request the
   creation of repositories for extensions to the following parameters
   (which are those listed as "extensible parameters" in Section 11.2).

      j_sec:

         Registrations for this repository may only occur
         via an IETF standards-track document.  Appendix C.2.1
         of this memo describes appropriate registrations for this
         repository.

         Initial values for this repository appear below:

         "none":  Defined in Appendix C.2.1 of this memo.
         "recj":  Defined in Appendix C.2.1 of this memo.

      j_update:

         Registrations for this repository may only occur
         via an IETF standards-track document.  Appendix C.2.2
         of this memo describes appropriate registrations for this
         repository.

         Initial values for this repository appear below:

         "anchor":  Defined in Appendix C.2.2 of this memo.
         "open-loop":  Defined in Appendix C.2.2 of this memo.
         "closed-loop":  Defined in Appendix C.2.2 of this memo.

      render:

         Registrations for this repository MUST include a
         specification of the usage of the proposed value.
         See text in the preamble of Appendix C.6 for details
         (the paragraph that begins "Other render token ...").

         Initial values for this repository appear below:

         "unknown":  Defined in Appendix C.6 of this memo.
         "synthetic":  Defined in Appendix C.6 of this memo.
         "null":  Defined in Appendix C.6 of this memo.

      subrender:

         Registrations for this repository MUST include a
         specification of the usage of the proposed value.
         See text Appendix C.6.2 for details (the paragraph
         that begins "Other subrender token ..." and
         subsequent paragraphs).  Note that the text in
         Appendix C.6.2 contains restrictions on subrender
         registrations for mpeg4-generic ("Registrations
         for mpeg4-generic subrender values ...").

         Initial values for this repository appear below:

         "default":  Defined in Appendix C.6.2 of this memo.

      smf_info:

         Registrations for this repository MUST include a
         specification of the usage of the proposed value.
         See text in Appendix C.6.4.1 for details (the
         paragraph that begins "Other smf_info token ...").

         Initial values for this repository appear below:

         "ignore":  Defined in Appendix C.6.4.1 of this memo.
         "sdp_start":  Defined in Appendix C.6.4.1 of this memo.
         "identity":  Defined in Appendix C.6.4.1 of this memo.

11.3.  asc Media Type Registration

   This section registers "asc" as a subtype for the "audio" media type.
   We register this subtype to support the remote transfer of the
   "config" parameter of the mpeg4-generic media type [RFC3640] when it
   is used with mpeg4-generic mode rtp-midi (registered in Appendix 11.2
   above).  We explain the mechanics of using "audio/asc" to set the
   config parameter in Section 6.2 and Appendix C.6.5 of this document.

   Note that this registration is a new subtype registration and is not
   an addition to a repository defined by MPEG-related memos (such as
   [RFC3640]).  Also note that this request for "audio/asc" does not
   register parameters, and does not request the creation of a
   repository.

   Media type name:

       audio

   Subtype name:

       asc

   Required parameters:

       None.

   Optional parameters:

       None.

   Encoding considerations:

       The native form of the data object is binary data,
       zero-padded to an octet boundary.

   Restrictions on usage:

       This type is only defined for data object (stored file)
       transfer.  The most common transports for the type are
       HTTP and SMTP.

   Security considerations:

       See Section 9 of this memo.

   Interoperability considerations:

       None.

   Published specification:

       The audio/asc data object is the AudioSpecificConfig
       binary data structure, which is normatively defined in
       [MPEGAUDIO].

   Applications that use this media type:

       MPEG 4 Audio servers and terminals that support
       audio/mpeg4-generic RTP streams for mode rtp-midi.

   Additional information:

       None.

   Person & email address to contact for further information:

       John Lazzaro <lazzaro@cs.berkeley.edu>

   Intended usage:

       COMMON.

   Author:

       John Lazzaro <lazzaro@cs.berkeley.edu>

   Change controller:

       IETF Audio/Video Transport Working Group delegated
       from the IESG.

A.  The Recovery Journal Channel Chapters

A.1.  Recovery Journal Definitions

   This appendix defines the terminology and the coding idioms that are
   used in the recovery journal bitfield descriptions in Section 5
   (journal header structure), Appendices A.2 to A.9 (channel journal
   chapters) and Appendices B.1 to B.5 (system journal chapters).

   We assume that the recovery journal resides in the journal section of
   an RTP packet with sequence number I ("packet I") and that the
   Checkpoint Packet Seqnum field in the top-level recovery journal
   header refers to a previous packet with sequence number C (an
   exception is the self-referential C = I case).  Unless stated
   otherwise, algorithms are assumed to use modulo 2^16 arithmetic for
   calculations on 16-bit sequence numbers and modulo 2^32 arithmetic
   for calculations on 32-bit extended sequence numbers.

   Several bitfield coding idioms appear throughout the recovery journal
   system, with consistent semantics.  Most recovery journal elements
   begin with an "S" (Single-packet loss) bit.  S bits are designed to
   help receivers efficiently parse through the recovery journal
   hierarchy in the common case of the loss of a single packet.

   As a rule, S bits MUST be set to 1.  However, an exception applies if
   a recovery journal element in packet I encodes data about a command
   stored in the MIDI command section of packet I - 1.  In this case,
   the S bit of the recovery journal element MUST be set to 0.  If a
   recovery journal element has its S bit set to 0, all higher-level
   recovery journal elements that contain it MUST also have S bits that
   are set to 0, including the top-level recovery journal header.

   Other consistent bitfield coding idioms are described below:

     o R flag bit.  R flag bits are reserved for future use.  Senders
       MUST set R bits to 0.  Receivers MUST ignore R bit values.

     o LENGTH field.  All fields named LENGTH (as distinct from LEN)
       code the number of octets in the structure that contains it,
       including the header it resides in and all hierarchical levels
       below it.  If a structure contains a LENGTH field, a receiver
       MUST use the LENGTH field value to advance past the structure
       during parsing, rather than use knowledge about the internal
       format of the structure.

   We now define normative terms used to describe recovery journal
   semantics.

     o Checkpoint history.  The checkpoint history of a recovery journal
       is the concatenation of the MIDI command sections of packets C
       through I - 1.  The final command in the MIDI command section for
       packet I - 1 is considered the most recent command; the first
       command in the MIDI command section for packet C is the oldest
       command.  If command X is less recent than command Y, X is
       considered to be "before Y".  A checkpoint history with no
       commands is considered to be empty.  The checkpoint history never
       contains the MIDI command section of packet I (the packet
       containing the recovery journal), so if C == I, the checkpoint
       history is empty by definition.

     o Session history.  The session history of a recovery journal is
       the concatenation of MIDI command sections from the first packet
       of the session up to packet I - 1.  The definitions of command
       recency and history emptiness follow those in the checkpoint
       history.  The session history never contains the MIDI command
       section of packet I, and so the session history of the first
       packet in the session is empty by definition.

     o Finished/unfinished commands.  If all octets of a MIDI command
       appear in the session history, the command is defined as being
       finished.  If some but not all octets of a command appear in the
       session history, the command is defined as being unfinished.
       Unfinished commands occur if segments of a SysEx command appear
       in several RTP packets.  For example, if a SysEx command is coded
       as 3 segments, with segment 1 in packet K, segment 2 in packet K
       + 1, and segment 3 in packet K + 2, the session histories for
       packets K + 1 and K + 2 contain unfinished versions of the
       command.  A session history contains a finished version of a
       cancelled SysEx command if the history contains the cancel
       sublist for the command.

     o Reset State commands.  Reset State (RS) commands reset renderers
       to an initialized "powerup" condition.  The RS commands are:
       System Reset (0xFF), General MIDI System Enable (0xF0 0x7E 0xcc
       0x09 0x01 0xF7), General MIDI 2 System Enable (0xF0 0x7E 0xcc
       0x09 0x03 0xF7), General MIDI System Disable (0xF0 0x7E 0xcc 0x09
       0x00 0xF7), Turn DLS On (0xF0 0x7E 0xcc 0x0A 0x01 0xF7), and Turn
       DLS Off (0xF0 0x7E 0xcc 0x0A 0x02 0xF7).  Registrations of
       subrender parameter token values (Appendix C.6.2) and IETF
       standards-track documents MAY specify additional RS commands.

     o Active commands.  Active command are MIDI commands that do not
       appear before a Reset State command in the session history.

     o N-active commands.  N-active commands are MIDI commands that do
       not appear before one of the following commands in the session
       history:  MIDI Control Change numbers 123-127 (numbers with All
       Notes Off semantics) or 120 (All Sound Off), and any Reset State
       command.

     o C-active commands.  C-active commands are MIDI commands that do
       not appear before one of the following commands in the session
       history:  MIDI Control Change number 121 (Reset All Controllers)
       and any Reset State command.

     o Oldest-first ordering rule.  Several recovery journal chapters
       contain a list of elements, where each element is associated with
       a MIDI command that appears in the session history.  In most
       cases, the chapter definition requires that list elements be
       ordered in accordance with the "oldest-first ordering rule".
       Below, we normatively define this rule:

       Elements associated with the most recent command in the session
       history coded in the list MUST appear at the end of the list.

       Elements associated with the oldest command in the session
       history coded in the list MUST appear at the start of the list.

       All other list elements MUST be arranged with respect to these
       boundary elements, to produce a list ordering that strictly
       reflects the relative session history recency of the commands
       coded by the elements in the list.

     o Parameter system.  A MIDI feature that provides two sets of
       16,384 parameters to expand the 0-127 controller number space.
       The Registered Parameter Names (RPN) system and the Non-
       Registered Parameter Names (NRPN) system each provides 16,384
       parameters.

     o Parameter system transaction.  The value of RPNs and NRPNs are
       changed by a series of Control Change commands that form a
       parameter system transaction.  A canonical transaction begins
       with two Control Change commands to set the parameter number
       (controller numbers 99 and 98 for NRPNs, controller numbers 101
       and 100 for RPNs).  The transaction continues with an arbitrary
       number of Data Entry (controller numbers 6 and 38), Data
       Increment (controller number 96), and Data Decrement (controller
       number 97) Control Change commands to set the parameter value.
       The transaction ends with a second pair of (99, 98) or (101, 100)
       Control Change commands that specify the null parameter (MSB
       value 0x7F, LSB value 0x7F).

       Several variants of the canonical transaction sequence are
       possible.  Most commonly, the terminal pair of (99, 98) or (101,
       100) Control Change commands may specify a parameter other than
       the null parameter.  In this case, the command pair terminates
       the first transaction and starts a second transaction.  The
       command pair is considered to be a part of both transactions.
       This variant is legal and recommended in [MIDI].  We refer to
       this variant as a "type 1 variant".

       Less commonly, the MSB (99 or 101) or LSB (98 or 100) command of
       a (99, 98) or (101, 100) Control Change pair may be omitted.

       If the MSB command is omitted, the transaction uses the MSB value
       of the most recent C-active Control Change command for controller
       number 99 or 101 that appears in the session history.  We refer
       to this variant as a "type 2 variant".

       If the LSB command is omitted, the LSB value 0x00 is assumed.  We
       refer to this variant as a "type 3 variant".  The type 2 and type
       3 variants are defined as legal, but are not recommended, in
       [MIDI].

       System real-time commands may appear at any point during a
       transaction (even between octets of individual commands in the
       transaction).  More generally, [MIDI] does not forbid the
       appearance of unrelated MIDI commands during an open transaction.
       As a rule, these commands are considered to be "outside" the
       transaction and do not affect the status of the transaction in
       any way.  Exceptions to this rule are commands whose semantics
       act to terminate transactions:  Reset State commands, and Control
       Change (0xB) for controller number 121 (Reset All Controllers)
       [RP015].

     o Initiated parameter system transaction.  A canonical parameter
       system transaction whose (99, 98) or (101, 100) initial Control
       Change command pair appears in the session history is considered
       to be an initiated parameter system transaction.  This definition
       also holds for type 1 variants.  For type 2 variants (dropped
       MSB), a transaction whose initial LSB Control Change command
       appears in the session history is an initiated transaction.  For
       type 3 variants (dropped LSB), a transaction is considered to be
       initiated if at least one transaction command follows the initial
       MSB (99 or 101) Control Change command in the session history.
       The completion of a transaction does not nullify its "initiated"
       status.

     o Session history reference counts.  Several recovery journal
       chapters include a reference count field, which codes the total
       number of commands of a type that appear in the session history.
       Examples include the Reset and Tune Request command logs (Chapter
       D, Appendix B.1) and the Active Sense command (Chapter V,
       Appendix B.2).  Upon the detection of a loss event, reference
       count fields let a receiver deduce if any instances of the
       command have been lost, by comparing the journal reference count
       with its own reference count.  Thus, a reference count field
       makes sense, even for command types in which knowing the NUMBER
       of lost commands is irrelevant (as is true with all of the
       example commands mentioned above).

   The chapter definitions in Appendices A.2 to A.9 and B.1 to B.5
   reflect the default recovery journal behavior.  The ch_default,
   ch_never, and ch_anchor parameters modify these definitions, as
   described in Appendix C.2.3.

   The chapter definitions specify if data MUST be present in the
   journal.  Senders MAY also include non-required data in the journal.
   This optional data MUST comply with the normative chapter definition.
   For example, if a chapter definition states that a field codes data
   from the most recent active command in the session history, the
   sender MUST NOT code inactive commands or older commands in the
   field.

   Finally, we note that a channel journal only encodes information
   about MIDI commands appearing on the MIDI channel the journal
   protects.  All references to MIDI commands in Appendices A.2 to A.9
   should be read as "MIDI commands appearing on this channel."

A.2.  Chapter P: MIDI Program Change

   A channel journal MUST contain Chapter P if an active Program Change
   (0xC) command appears in the checkpoint history.  Figure A.2.1 shows
   the format for Chapter P.

                0                   1                   2
                0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3
               +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
               |S|   PROGRAM   |B|   BANK-MSB  |X|  BANK-LSB   |
               +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

                      Figure A.2.1 -- Chapter P format

   The chapter has a fixed size of 24 bits.  The PROGRAM field indicates
   the data value of the most recent active Program Change command in
   the session history.  By default, the B, BANK-MSB, X, and BANK-LSB

   fields MUST be set to 0.  Below, we define exceptions to this default
   condition.

   If an active Control Change (0xB) command for controller number 0
   (Bank Select MSB) appears before the Program Change command in the
   session history, the B bit MUST be set to 1, and the BANK-MSB field
   MUST code the data value of the Control Change command.

   If B is set to 1, the BANK-LSB field MUST code the data value of the
   most recent Control Change command for controller number 32 (Bank
   Select LSB) that preceded the Program Change command coded in the
   PROGRAM field and followed the Control Change command coded in the
   BANK-MSB field.  If no such Control Change command exists, the BANK-
   LSB field MUST be set to 0.

   If B is set to 1, and if a Control Change command for controller
   number 121 (Reset All Controllers) appears in the MIDI stream between
   the Control Change command coded by the BANK-MSB field and the
   Program Change command coded by the PROGRAM field, the X bit MUST be
   set to 1.

   Note that [RP015] specifies that Reset All Controllers does not reset
   the values of controller numbers 0 (Bank Select MSB) and 32 (Bank
   Select LSB).  Thus, the X bit does not effect how receivers will use
   the BANK-LSB and BANK-MSB values when recovering from a lost Program
   Change command.  The X bit serves to aid recovery in MIDI
   applications where controller numbers 0 and 32 are used in a non-
   standard way.

A.3.  Chapter C: MIDI Control Change

   Figure A.3.1 shows the format for Chapter C.

       0                   1                   2                   3
       0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 8 0 1
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
      |S|     LEN     |S|   NUMBER    |A|  VALUE/ALT  |S|   NUMBER    |
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
      |A|  VALUE/ALT  |  ....                                         |
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

                       Figure A.3.1 -- Chapter C format

   The chapter consists of a 1-octet header, followed by a variable
   length list of 2-octet controller logs.  The list MUST contain at
   least one controller log.  The 7-bit LEN field codes the number of
   controller logs in the list, minus one.  We define the semantics of
   the controller log fields in Appendix A.3.2.

   A channel journal MUST contain Chapter C if the rules defined in this
   appendix require that one or more controller logs appear in the list.

A.3.1.  Log Inclusion Rules

   A controller log encodes information about a particular Control
   Change command in the session history.

   In the default use of the payload format, list logs MUST encode
   information about the most recent active command in the session
   history for a controller number.  Logs encoding earlier commands MUST
   NOT appear in the list.

   Also, as a rule, the list MUST contain a log for the most recent
   active command for a controller number that appears in the checkpoint
   history.  Below, we define exceptions to this rule:

     o  MIDI streams may transmit 14-bit controller values using paired
        Most Significant Byte (MSB, controller numbers 0-31, 99, 101)
        and Least Significant Byte (LSB, controller numbers 32-63, 98,
        100) Control Change commands [MIDI].

        If the most recent active Control Change command in the session
        history for a 14-bit controller pair uses the MSB number,
        Chapter C MAY omit the controller log for the most recent active
        Control Change command for the associated LSB number, as the
        command ordering makes this LSB value irrelevant.  However, this
        exception MUST NOT be applied if the sender is not certain that
        the MIDI source uses 14-bit semantics for the controller number
        pair.  Note that some MIDI sources ignore 14-bit controller
        semantics and use the LSB controller numbers as independent 7-
        bit controllers.

     o  If active Control Change commands for controller numbers 0 (Bank
        Select MSB) or 32 (Bank Select LSB) appear in the checkpoint
        history, and if the command instances are also coded in the
        BANK-MSB and BANK-LSB fields of the Chapter P (Appendix A.2),
        Chapter C MAY omit the controller logs for the commands.

     o  Several controller number pairs are defined to be mutually
        exclusive.  Controller numbers 124 (Omni Off) and 125 (Omni On)
        form a mutually exclusive pair, as do controller numbers 126
        (Mono) and 127 (Poly).

        If active Control Change commands for one or both members of a
        mutually exclusive pair appear in the checkpoint history, a log
        for the controller number of the most recent command for the
        pair in the checkpoint history MUST appear in the controller

        list.  However, the list MAY omit the controller log for the
        most recent active command for the other number in the pair.

        If active Control Change commands for one or both members of a
        mutually exclusive pair appear in the session history, and if a
        log for the controller number of the most recent command for the
        pair does not appear in the controller list, a log for the most
        recent command for the other number of the pair MUST NOT appear
        in the controller list.

     o  If an active Control Change command for controller number 121
        (Reset All Controllers) appears in the session history, the
        controller list MAY omit logs for Control Change commands that
        precede the Reset All Controllers command in the session
        history, under certain conditions.

        Namely, a log MAY be omitted if the sender is certain that a
        command stream follows the Reset All Controllers semantics
        defined in [RP015], and if the log codes a controller number for
        which [RP015] specifies a reset value.

        For example, [RP015] specifies that controller number 1
        (Modulation Wheel) is reset to the value 0, and thus a
        controller log for Modulation Wheel MAY be omitted from the
        controller log list.  In contrast, [RP015] specifies that
        controller number 7 (Channel Volume) is not reset, and thus a
        controller log for Channel Volume MUST NOT be omitted from the
        controller log list.

     o  Appendix A.3.4 defines exception rules for the MIDI Parameter
        System controller numbers 6, 38, and 96-101.

A.3.2.  Controller Log Format

   Figure A.3.2 shows the controller log structure of Chapter C.

                       0                   1
                       0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5
                      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
                      |S|    NUMBER   |A|  VALUE/ALT  |
                      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

                  Figure A.3.2 -- Chapter C controller log

   The 7-bit NUMBER field identifies the controller number of the coded
   command.  The 7-bit VALUE/ALT field codes recovery information for
   the command.  The A bit sets the format of the VALUE/ALT field.

   A log encodes recovery information using one of the following tools:
   the value tool, the toggle tool, or the count tool.

   A log uses the value tool if the A bit is set to 0.  The value tool
   codes the 7-bit data value of a command in the VALUE/ALT field.  The
   value tool works best for controllers that code a continuous
   quantity, such as number 1 (Modulation Wheel).

   The A bit is set to 1 to code the toggle or count tool.  These tools
   work best for controllers that code discrete actions.  Figure A.3.3
   shows the controller log for these tools.

                       0                   1
                       0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5
                      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
                      |S|    NUMBER   |1|T|    ALT    |
                      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

                 Figure A.3.3 -- Controller log for ALT tools

   A log uses the toggle tool if the T bit is set to 0.  A log uses the
   count tool if the T bit is set to 1.  Both methods use the 6-bit ALT
   field as an unsigned integer.

   The toggle tool works best for controllers that act as on/off
   switches, such as 64 (Damper Pedal (Sustain)).  These controllers
   code the "off" state with control values 0-63 and the "on" state with
   64-127.

   For the toggle tool, the ALT field codes the total number of toggles
   (off->on and on->off) due to Control Change commands in the session
   history, up to and including a toggle caused by the command coded by
   the log.  The toggle count includes toggles caused by Control Change
   commands for controller number 121 (Reset All Controllers).

   Toggle counting is performed modulo 64.  The toggle count is reset at
   the start of a session, and whenever a Reset State command (Appendix
   A.1) appears in the session history.  When these reset events occur,
   the toggle count for a controller is set to 0 (for controllers whose
   default value is 0-63) or 1 (for controllers whose default value is
   64-127).

   The Damper Pedal (Sustain) controller illustrates the benefits of the
   toggle tool over the value tool for switch controllers.  As often
   used in piano applications, the "on" state of the controller lets
   notes resonate, while the "off" state immediately damps notes to
   silence.  The loss of the "off" command in an "on->off->on" sequence
   results in ringing notes that should have been damped silent.  The

   toggle tool lets receivers detect this lost "off" command, but the
   value tool does not.

   The count tool is conceptually similar to the toggle tool.  For the
   count tool, the ALT field codes the total number of Control Change
   commands in the session history, up to and including the command
   coded by the log.  Command counting is performed modulo 64.  The
   command count is set to 0 at the start of the session and is reset to
   0 whenever a Reset State command (Appendix A.1) appears in the
   session history.

   Because the count tool ignores the data value, it is a good match for
   controllers whose controller value is ignored, such as number 123
   (All Notes Off).  More generally, the count tool may be used to code
   a (modulo 64) identification number for a command.

A.3.3.  Log List Coding Rules

   In this section, we describe the organization of controller logs in
   the Chapter C log list.

   A log encodes information about a particular Control Change command
   in the session history.  In most cases, a command SHOULD be coded by
   a single tool (and, thus, a single log).  If a number is coded with a
   single tool and this tool is the count tool, recovery Control Change
   commands generated by a receiver SHOULD use the default control value
   for the controller.

   However, a command MAY be coded by several tool types (and, thus,
   several logs, each using a different tool).  This technique may
   improve recovery performance for controllers with complex semantics,
   such as controller number 84 (Portamento Control) or controller
   number 121 (Reset All Controllers) when used with a non-zero data
   octet (with the semantics described in [DLS2]).

   If a command is encoded by multiple tools, the logs MUST be placed in
   the list in the following order: count tool log (if any), followed by
   value tool log (if any), followed by toggle tool log (if any).

   The Chapter C log list MUST obey the oldest-first ordering rule
   (defined in Appendix A.1).  Note that this ordering preserves the
   information necessary for the recovery of 14-bit controller values,
   without precluding the use of MSB and LSB controller pairs as
   independent 7-bit controllers.

   In the default use of the payload format, all logs that appear in the
   list for a controller number encode information about one Control
   Change command -- namely, the most recent active Control Change
   command in the session history for the number.

   This coding scheme provides good recovery performance for the
   standard uses of Control Change commands defined in [MIDI].  However,
   not all MIDI applications restrict the use of Control Change commands
   to those defined in [MIDI].

   For example, consider the common MIDI encoding of rotary encoders
   ("infinite" rotation knobs).  The mixing console MIDI convention
   defined in [LCP] codes the position of rotary encoders as a series of
   Control Change commands.  Each command encodes a relative change of
   knob position from the last update (expressed as a clockwise or
   counter-clockwise knob turning angle).

   As the knob position is encoded incrementally over a series of
   Control Change commands, the best recovery performance is obtained if
   the log list encodes all Control Change commands for encoder
   controller numbers that appear in the checkpoint history, not only
   the most recent command.

   To support application areas that use Control Change commands in this
   way, Chapter C may be configured to encode information about several
   Control Change commands for a controller number.  We use the term
   "enhanced" to describe this encoding method, which we describe below.

   In Appendix C.2.3, we show how to configure a stream to use enhanced
   Chapter C encoding for specific controller numbers.  In Section 5 in
   the main text, we show how the H bits in the recovery journal header
   (Figure 8) and in the channel journal header (Figure 9) indicate the
   use of enhanced Chapter C encoding.

   Here, we define how to encode a Chapter C log list that uses the
   enhanced encoding method.

   Senders that use the enhanced encoding method for a controller number
   MUST obey the rules below.  These rules let a receiver determine
   which logs in the list correspond to lost commands.  Note that these
   rules override the exceptions listed in Appendix A.3.1.

     o  If N commands for a controller number are encoded in the list,
        the commands MUST be the N most recent commands for the
        controller number in the session history.  For example, for N =
        2, the sender MUST encode the most recent command and the second
        most recent command, not the most recent command and the third
        most recent command.

     o  If a controller number uses enhanced encoding, the encoding of
        the least-recent command for the controller number in the log
        list MUST include a count tool log.  In addition, if commands
        are encoded for the controller number whose logs have S bits set
        to 0, the encoding of the least-recent command with S = 0 logs
        MUST include a count tool log.

        The count tool is OPTIONAL for the other commands for the
        controller number encoded in the list, as a receiver is able to
        efficiently deduce the count tool value for these commands, for
        both single-packet and multi-packet loss events.

     o  The use of the value and toggle tools MUST be identical for all
        commands for a controller number encoded in the list.  For
        example, a value tool log either MUST appear for all commands
        for the controller number coded in the list, or alternatively,
        value tool logs for the controller number MUST NOT appear in the
        list.  Likewise, a toggle tool log either MUST appear for all
        commands for the controller number coded in the list, or
        alternatively, toggle tool logs for the controller number MUST
        NOT appear in the list.

     o  If a command is encoded by multiple tools, the logs MUST be
        placed in the list in the following order: count tool log (if
        any), followed by value tool log (if any), followed by toggle
        tool log (if any).

   These rules permit a receiver recovering from a packet loss to use
   the count tool log to match the commands encoded in the list with its
   own history of the stream, as we describe below.  Note that the text
   below describes a non-normative algorithm; receivers are free to use
   any algorithm to match its history with the log list.

   In a typical implementation of the enhanced encoding method, a
   receiver computes and stores count, value, and toggle tool data field
   values for the most recent Control Change command it has received for
   a controller number.

   After a loss event, a receiver parses the Chapter C list and
   processes list logs for a controller number that uses enhanced
   encoding as follows.

   The receiver compares the count tool ALT field for the least-recent
   command for the controller number in the list against its stored
   count data for the controller number, to determine if recovery is
   necessary for the command coded in the list.  The value and toggle
   tool logs (if any) that directly follow the count tool log are
   associated with this least-recent command.

   To check more-recent commands for the controller, the receiver
   detects additional value and/or toggle tool logs for the controller
   number in the list and infers count tool data for the command coded
   by these logs.  This inferred data is used to determine if recovery
   is necessary for the command coded by the value and/or toggle tool
   logs.

   In this way, a receiver is able to execute only lost commands,
   without executing a command twice.  While recovering from a single
   packet loss, a receiver may skip through S = 1 logs in the list, as
   the first S = 0 log for an enhanced controller number is always a
   count tool log.

   Note that the requirements in Appendix C.2.2.2 for protective sender
   and receiver actions during session startup for multicast operation
   are of particular importance for enhanced encoding, as receivers need
   to initialize its count tool data structures with recovery journal
   data in order to match commands correctly after a loss event.

   Finally, we note in passing that in some applications of rotary
   encoders, a good user experience may be possible without the use of
   enhanced encoding.  These applications are distinguished by visual
   feedback of encoding position that is driven by the post-recovery
   rotary encoding stream, and relatively low packet loss.  In these
   domains, recovery performance may be acceptable for rotary encoders
   if the log list encodes only the most recent command, if both count
   and value logs appear for the command.

A.3.4.  The Parameter System

   Readers may wish to review the Appendix A.1 definitions of "parameter
   system", "parameter system transaction", and "initiated parameter
   system transaction" before reading this section.

   Parameter system transactions update a MIDI Registered Parameter
   Number (RPN) or Non-Registered Parameter Number (NRPN) value.  A
   parameter system transaction is a sequence of Control Change commands
   that may use the following controllers numbers:

     o  Data Entry MSB (6)
     o  Data Entry LSB (38)
     o  Data Increment (96)
     o  Data Decrement (97)
     o  Non-Registered Parameter Number (NRPN) LSB (98)
     o  Non-Registered Parameter Number (NRPN) MSB (99)
     o  Registered Parameter Number (RPN) LSB (100)
     o  Registered Parameter Number (RPN) MSB (101)

   Control Change commands that are a part of a parameter system
   transaction MUST NOT be coded in Chapter C controller logs.  Instead,
   these commands are coded in Chapter M, the MIDI Parameter chapter
   defined in Appendix A.4.

   However, Control Change commands that use the listed controllers as
   general-purpose controllers (i.e., outside of a parameter system
   transaction) MUST NOT be coded in Chapter M.

   Instead, the controllers are coded in Chapter C controller logs.  The
   controller logs follow the coding rules stated in Appendix A.3.2 and
   A.3.3.  The rules for coding paired LSB and MSB controllers, as
   defined in Appendix A.3.1, apply to the pairs (6, 38), (99, 98), and
   (101, 100) when coded in Chapter C.

   If active Control Change commands for controller numbers 6, 38, or
   96-101 appear in the checkpoint history, and these commands are used
   as general-purpose controllers, the most recent general-purpose
   command instance for these controller numbers MUST appear as entries
   in the Chapter C controller list.

   MIDI syntax permits a source to use controllers 6, 38, 96, and 97 as
   parameter-system controllers and general-purpose controllers in the
   same stream.  An RTP MIDI sender MUST deduce the role of each Control
   Change command for these controller numbers by noting the placement
   of the command in the stream and MUST use this information to code
   the command in Chapter C or Chapter M, as appropriate.

   Specifically, active Control Change commands for controllers 6, 38,
   96, and 97 act in a general-purpose way when

     o  no active Control Change commands that set an RPN or NRPN
        parameter number appear in the session history, or

     o  the most recent active Control Change commands in the session
        history that set an RPN or NRPN parameter number code the null
        parameter (MSB value 0x7F, LSB value 0x7F), or

     o  a Control Change command for controller number 121 (Reset All
        Controllers) appears more recently in the session history than
        all active Control Change commands that set an RPN or NRPN
        parameter number (see [RP015] for details).

   Finally, we note that a MIDI source that follows the recommendations
   of [MIDI] exclusively uses numbers 98-101 as parameter system
   controllers.  Alternatively, a MIDI source may exclusively use 98-101
   as general-purpose controllers and lose the ability perform parameter
   system transactions in a stream.

   In the language of [MIDI], the general-purpose use of controllers
   98-101 constitutes a non-standard controller assignment.  As most
   real-world MIDI sources use the standard controller assignment for
   controller numbers 98-101, an RTP MIDI sender SHOULD assume these
   controllers act as parameter system controllers, unless it knows that
   a MIDI source uses controller numbers 98-101 in a general-purpose
   way.

A.4.  Chapter M: MIDI Parameter System

   Readers may wish to review the Appendix A.1 definitions for
   "C-active", "parameter system", "parameter system transaction", and
   "initiated parameter system transaction" before reading this
   appendix.

   Chapter M protects parameter system transactions for Registered
   Parameter Number (RPN) and Non-Registered Parameter Number (NRPN)
   values.  Figure A.4.1 shows the format for Chapter M.

       0                   1                   2                   3
       0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
      |S|P|E|U|W|Z|      LENGTH       |Q|  PENDING    |  Log list ... |
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

                  Figure A.4.1 -- Top-level Chapter M format

   Chapter M begins with a 2-octet header.  If the P header bit is set
   to 1, a 1-octet field follows the header, coding the 7-bit PENDING
   value and its associated Q bit.

   The 10-bit LENGTH field codes the size of Chapter M and conforms to
   semantics described in Appendix A.1.

   Chapter M ends with a list of zero or more variable-length parameter
   logs.  Appendix A.4.2 defines the bitfield format of a parameter log.
   Appendix A.4.1 defines the inclusion semantics of the log list.

   A channel journal MUST contain Chapter M if the rules defined in
   Appendix A.4.1 require that one or more parameter logs appear in the
   list.

   A channel journal also MUST contain Chapter M if the most recent
   C-active Control Change command involved in a parameter system
   transaction in the checkpoint history is

     o  an RPN MSB (101) or NRPN MSB (99) controller, or

     o  an RPN LSB (100) or NRPN LSB (98) controller that completes the
        coding of the null parameter (MSB value 0x7F, LSB value 0x7F).

   This rule provides loss protection for partially transmitted
   parameter numbers and for the null parameter numbers.

   If the most recent C-active Control Change command involved in a
   parameter system transaction in the session history is for the RPN
   MSB or NRPN MSB controller, the P header bit MUST be set to 1, and
   the PENDING field (and its associated Q bit) MUST follow the Chapter
   M header.  Otherwise, the P header bit MUST be set to 0, and the
   PENDING field and Q bit MUST NOT appear in Chapter M.

   If PENDING codes an NRPN MSB, the Q bit MUST be set to 1.  If PENDING
   codes an RPN MSB, the Q bit MUST be set to 0.

   The E header bit codes the current transaction state of the MIDI
   stream.  If E = 1, an initiated transaction is in progress.  Below,
   we define the rules for setting the E header bit:

     o  If no C-active parameter system transaction Control Change
        commands appear in the session history, the E bit MUST be set to
        0.

     o  If the P header bit is set to 1, the E bit MUST be set to 0.

     o  If the most recent C-active parameter system transaction Control
        Change command in the session history is for the NRPN LSB or RPN
        LSB controller number, and if this command acts to complete the
        coding of the null parameter (MSB value 0x7F, LSB value 0x7F),
        the E bit MUST be set to 0.

     o  Otherwise, an initiated transaction is in progress, and the E
        bit MUST be set to 1.

   The U, W, and Z header bits code properties that are shared by all
   parameter logs in the list.  If these properties are set, parameter
   logs may be coded with improved efficiency (we explain how in A.4.1).

   By default, the U, W, and Z bits MUST be set to 0.  If all parameter
   logs in the list code RPN parameters, the U bit MAY be set to 1.  If
   all parameter logs in the list code NRPN parameters, the W bit MAY be
   set to 1.  If the parameter numbers of all RPN and NRPN logs in the
   list lie in the range 0-127 (and thus have an MSB value of 0), the Z
   bit MAY be set to 1.

   Note that C-active semantics appear in the preceding paragraphs
   because [RP015] specifies that pending Parameter System transactions
   are closed by a Control Change command for controller number 121
   (Reset All Controllers).

A.4.1.  Log Inclusion Rules

   Parameter logs code recovery information for a specific RPN or NRPN
   parameter.

   A parameter log MUST appear in the list if an active Control Change
   command that forms a part of an initiated transaction for the
   parameter appears in the checkpoint history.

   An exception to this rule applies if the checkpoint history only
   contains transaction Control Change commands for controller numbers
   98-101 that act to terminate the transaction.  In this case, a log
   for the parameter MAY be omitted from the list.

   A log MAY appear in the list if an active Control Change command that
   forms a part of an initiated transaction for the parameter appears in
   the session history.  Otherwise, a log for the parameter MUST NOT
   appear in the list.

   Multiple logs for the same RPN or NRPN parameter MUST NOT appear in
   the log list.

   The parameter log list MUST obey the oldest-first ordering rule
   (defined in Appendix A.1), with the phrase "parameter transaction"
   replacing the word "command" in the rule definition.

   Parameter logs associated with the RPN or NRPN null parameter (LSB =
   0x7F, MSB = 0x7F) MUST NOT appear in the log list.  Chapter M uses
   the E header bit (Figure A.4.1) and the log list ordering rules to
   code null parameter semantics.

   Note that "active" semantics (rather than "C-active" semantics)
   appear in the preceding paragraphs because [RP015] specifies that
   pending Parameter System transactions are not reset by a Control
   Change command for controller number 121 (Reset All Controllers).
   However, the rule that follows uses C-active semantics, because it
   concerns the protection of the transaction system itself, and [RP015]
   specifies that Reset All Controllers acts to close a transaction in
   progress.

   In most cases, parameter logs for RPN and NRPN parameters that are
   assigned to the ch_never parameter (Appendix C.2.3) MAY be omitted
   from the list.  An exception applies if

     o  the log codes the most recent initiated transaction in the
        session history, and

     o  a C-active command that forms a part of the transaction appears
        in the checkpoint history, and

     o  the E header bit for the top-level Chapter M header (Figure
        A.4.1) is set to 1.

   In this case, a log for the parameter MUST appear in the list.  This
   log informs receivers recovering from a loss that a transaction is in
   progress, so that the receiver is able to correctly interpret RPN or
   NRPN Control Change commands that follow the loss event.

A.4.2.  Log Coding Rules

   Figure A.4.2 shows the parameter log structure of Chapter M.

       0                   1                   2                   3
       0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 8 0 1
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
      |S|  PNUM-LSB   |Q|  PNUM-MSB   |J|K|L|M|N|T|V|R|   Fields ...  |
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

                      Figure A.4.2 -- Parameter log format

   The log begins with a header, whose default size (as shown in Figure
   A.4.2) is 3 octets.  If the Q header bit is set to 0, the log encodes
   an RPN parameter.  If Q = 1, the log encodes an NRPN parameter.  The
   7-bit PNUM-MSB and PNUM-LSB fields code the parameter number and
   reflect the Control Change command data values for controllers 99 and
   98 (for NRPNs) or 101 and 100 (for RPNs).

   The J, K, L, M, and N header bits form a Table of Contents (TOC) for
   the log and signal the presence of fixed-sized fields that follow the
   header.  A header bit that is set to 1 codes the presence of a field
   in the log.  The ordering of fields in the log follows the ordering
   of the header bits in the TOC.  Appendices A.4.2.1-2 define the
   fields associated with each TOC header bit.

   The T and V header bits code information about the parameter log but
   are not part of the TOC.  A set T or V bit does not signal the
   presence of any parameter log field.

   If the rules in Appendix A.4.1 state that a log for a given parameter
   MUST appear in Chapter M, the log MUST code sufficient information to
   protect the parameter from the loss of active parameter transaction
   Control Change commands in the checkpoint history.

   This rule does not apply if the parameter coded by the log is
   assigned to the ch_never parameter (Appendix C.2.3).  In this case,
   senders MAY choose to set the J, K, L, M, and N TOC bits to 0, coding
   a parameter log with no fields.

   Note that logs to protect parameters that are assigned to ch_never
   are REQUIRED under certain conditions (see Appendix A.4.1).  The
   purpose of the log is to inform receivers recovering from a loss that
   a transaction is in progress, so that the receiver is able to
   correctly interpret RPN or NRPN Control Change commands that follow
   the loss event.

   Parameter logs provide two tools for parameter protection: the value
   tool and the count tool.  Depending on the semantics of the
   parameter, senders may use either tool, both tools, or neither tool
   to protect a given parameter.

   The value tool codes information a receiver may use to determine the
   current value of an RPN or NRPN parameter.  If a parameter log uses
   the value tool, the V header bit MUST be set to 1, and the semantics
   defined in Appendices A.4.2.1 for setting the J, K, L, and M TOC bits
   MUST be followed.  If a parameter log does not use the value tool,
   the V bit MUST be set to 0, and the J, K, L, and M TOC bits MUST also
   be set to 0.

   The count tool codes the number of transactions for an RPN or NRPN
   parameter.  If a parameter log uses the count tool, the T header bit
   MUST be set to 1, and the semantics defined in Appendices A.4.2.2 for
   setting the N TOC bit MUST be followed.  If a parameter log does not
   use the count tool, the T bit and the N TOC bit MUST be set to 0.

   Note that V and T are set if the sender uses value (V) or count (T)
   tool for the log on an ongoing basis.  Thus, V may be set even if J =
   K = L = M = 0, and T may be set even if N = 0.

   In many cases, all parameters coded in the log list are of one type
   (RPN and NRPN), and all parameter numbers lie in the range 0-127.  As
   described in Appendix A.4.1, senders MAY signal this condition by
   setting the top-level Chapter M header bit Z to 1 (to code the
   restricted range) and by setting the U or W bit to 1 (to code the
   parameter type).

   If the top-level Chapter M header codes Z = 1 and either U = 1 or
   W = 1, all logs in the parameter log list MUST use a modified header
   format.  This modification deletes bits 8-15 of the bitfield shown in
   Figure A.4.2, to yield a 2-octet header.  The values of the deleted
   PNUM-MSB and Q fields may be inferred from the U, W, and Z bit
   values.

A.4.2.1.  The Value Tool

   The value tool uses several fields to track the value of an RPN or
   NRPN parameter.

   The J TOC bit codes the presence of the octet shown in Figure A.4.3
   in the field list.

                             0
                             0 1 2 3 4 5 6 7
                            +-+-+-+-+-+-+-+-+
                            |X|  ENTRY-MSB  |
                            +-+-+-+-+-+-+-+-+

                      Figure A.4.3 -- ENTRY-MSB field

   The 7-bit ENTRY-MSB field codes the data value of the most recent
   active Control Change command for controller number 6 (Data Entry
   MSB) in the session history that appears in a transaction for the log
   parameter.

   The X bit MUST be set to 1 if the command coded by ENTRY-MSB precedes
   the most recent Control Change command for controller 121 (Reset All
   Controllers) in the session history.  Otherwise, the X bit MUST be
   set to 0.

   A parameter log that uses the value tool MUST include the ENTRY-MSB
   field if an active Control Change command for controller number 6
   appears in the checkpoint history.

   Note that [RP015] specifies that Control Change commands for
   controller 121 (Reset All Controllers) do not reset RPN and NRPN
   values, and thus the X bit would not play a recovery role for MIDI
   systems that comply with [RP015].

   However, certain renderers (such as DLS 2 [DLS2]) specify that
   certain RPN values are reset for some uses of Reset All Controllers.
   The X bit (and other bitfield features of this nature in this
   appendix) plays a role in recovery for renderers of this type.

   The K TOC bit codes the presence of the octet shown in Figure A.4.4
   in the field list.

                             0
                             0 1 2 3 4 5 6 7
                            +-+-+-+-+-+-+-+-+
                            |X|  ENTRY-LSB  |
                            +-+-+-+-+-+-+-+-+

                      Figure A.4.4 -- ENTRY-LSB field

   The 7-bit ENTRY-LSB field codes the data value of the most recent
   active Control Change command for controller number 38 (Data Entry
   LSB) in the session history that appears in a transaction for the log
   parameter.

   The X bit MUST be set to 1 if the command coded by ENTRY-LSB precedes
   the most recent Control Change command for controller 121 (Reset All
   Controllers) in the session history.  Otherwise, the X bit MUST be
   set to 0.

   As a rule, a parameter log that uses the value tool MUST include the
   ENTRY-LSB field if an active Control Change command for controller
   number 38 appears in the checkpoint history.  However, the ENTRY-LSB
   field MUST NOT appear in a parameter log if the Control Change
   command associated with the ENTRY-LSB precedes a Control Change
   command for controller number 6 (Data Entry MSB) that appears in a
   transaction for the log parameter in the session history.

   The L TOC bit codes the presence of the octets shown in Figure A.4.5
   in the field list.

                       0                   1
                       0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5
                      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
                      |G|X|       A-BUTTON            |
                      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

                       Figure A.4.5 -- A-BUTTON field

   The 14-bit A-BUTTON field codes a count of the number of active
   Control Change commands for controller numbers 96 and 97 (Data
   Increment and Data Decrement) in the session history that appear in a
   transaction for the log parameter.

   The M TOC bit codes the presence of the octets shown in Figure A.4.6
   in the field list.

                       0                   1
                       0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5
                      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
                      |G|R|       C-BUTTON            |
                      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

                       Figure A.4.6 -- C-BUTTON field

   The 14-bit C-BUTTON field has semantics identical to A-BUTTON, except
   that Data Increment and Data Decrement Control Change commands that
   precede the most recent Control Change command for controller 121
   (Reset All Controllers) in the session history are not counted.

   For both A-BUTTON and C-BUTTON, Data Increment and Data Decrement
   Control Change commands are not counted if they precede Control
   Changes commands for controller numbers 6 (Data Entry MSB) or 38
   (Data Entry LSB) that appear in a transaction for the log parameter
   in the session history.

   The A-BUTTON and C-BUTTON fields are interpreted as unsigned
   integers, and the G bit associated the field codes the sign of the
   integer (G = 0 for positive or zero, G = 1 for negative).

   To compute and code the count value, initialize the count value to 0,
   add 1 for each qualifying Data Increment command, and subtract 1 for
   each qualifying Data Decrement command.  After each add or subtract,
   limit the count magnitude to 16383.  The G bit codes the sign of the
   count, and the A-BUTTON or C-BUTTON field codes the count magnitude.

   For the A-BUTTON field, if the most recent qualified Data Increment
   or Data Decrement command precedes the most recent Control Change
   command for controller 121 (Reset All Controllers) in the session
   history, the X bit associated with A-BUTTON field MUST be set to 1.
   Otherwise, the X bit MUST be set to 0.

   A parameter log that uses the value tool MUST include the A-BUTTON
   and C-BUTTON fields if an active Control Change command for
   controller numbers 96 or 97 appears in the checkpoint history.
   However, to improve coding efficiency, this rule has several
   exceptions:

     o  If the log includes the A-BUTTON field, and if the X bit of the
        A-BUTTON field is set to 1, the C-BUTTON field (and its
        associated R and G bits) MAY be omitted from the log.

     o  If the log includes the A-BUTTON field, and if the A-BUTTON and
        C-BUTTON fields (and their associated G bits) code identical
        values, the C-BUTTON field (and its associated R and G bits) MAY
        be omitted from the log.

A.4.2.2.  The Count Tool

   The count tool tracks the number of transactions for an RPN or NRPN
   parameter.  The N TOC bit codes the presence of the octet shown in
   Figure A.4.7 in the field list.

                          0
                          0 1 2 3 4 5 6 7
                         +-+-+-+-+-+-+-+-+
                         |X|    COUNT    |
                         +-+-+-+-+-+-+-+-+

                     Figure A.4.7 -- COUNT field

   The 7-bit COUNT codes the number of initiated transactions for the
   log parameter that appear in the session history.  Initiated
   transactions are counted if they contain one or more active Control
   Change commands, including commands for controllers 98-101 that
   initiate the parameter transaction.

   If the most recent counted transaction precedes the most recent
   Control Change command for controller 121 (Reset All Controllers) in
   the session history, the X bit associated with the COUNT field MUST
   be set to 1.  Otherwise, the X bit MUST be set to 0.

   Transaction counting is performed modulo 128.  The transaction count
   is set to 0 at the start of a session and is reset to 0 whenever a
   Reset State command (Appendix A.1) appears in the session history.

   A parameter log that uses the count tool MUST include the COUNT field
   if an active command that increments the transaction count (modulo
   128) appears in the checkpoint history.

A.5.  Chapter W: MIDI Pitch Wheel

   A channel journal MUST contain Chapter W if a C-active MIDI Pitch
   Wheel (0xE) command appears in the checkpoint history.  Figure A.5.1
   shows the format for Chapter W.

                       0                   1
                       0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5
                      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
                      |S|     FIRST   |R|    SECOND   |
                      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

                      Figure A.5.1 -- Chapter W format

   The chapter has a fixed size of 16 bits.  The FIRST and SECOND fields
   are the 7-bit values of the first and second data octets of the most
   recent active Pitch Wheel command in the session history.

   Note that Chapter W encodes C-active commands and thus does not
   encode active commands that are not C-active (see the second-to-last
   paragraph of Appendix A.1 for an explanation of chapter inclusion
   text in this regard).

   Chapter W does not encode "active but not C-active" commands because
   [RP015] declares that Control Change commands for controller number
   121 (Reset All Controllers) act to reset the Pitch Wheel value to 0.
   If Chapter W encoded "active but not C-active" commands, a repair
   operation following a Reset All Controllers command could incorrectly
   repair the stream with a stale Pitch Wheel value.

A.6.  Chapter N: MIDI NoteOff and NoteOn

   In this appendix, we consider NoteOn commands with zero velocity to
   be NoteOff commands.  Readers may wish to review the Appendix A.1
   definition of "N-active commands" before reading this appendix.

   Chapter N completely protects note commands in streams that alternate
   between NoteOn and NoteOff commands for a particular note number.
   However, in rare applications, multiple overlapping NoteOn commands
   may appear for a note number.  Chapter E, described in Appendix A.7,
   augments Chapter N to completely protect these streams.

   A channel journal MUST contain Chapter N if an N-active MIDI NoteOn
   (0x9) or NoteOff (0x8) command appears in the checkpoint history.
   Figure A.6.1 shows the format for Chapter N.

       0                   1                   2                   3
       0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 8 0 1
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
      |B|     LEN     |  LOW  | HIGH  |S|   NOTENUM   |Y|  VELOCITY   |
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
      |S|   NOTENUM   |Y|  VELOCITY   |             ....              |
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
      |    OFFBITS    |    OFFBITS    |     ....      |    OFFBITS    |
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

                      Figure A.6.1 -- Chapter N format

   Chapter N consists of a 2-octet header, followed by at least one of
   the following data structures:

      o A list of note logs to code NoteOn commands.
      o A NoteOff bitfield structure to code NoteOff commands.

   We define the header bitfield semantics in Appendix A.6.1.  We define
   the note log semantics and the NoteOff bitfield semantics in Appendix
   A.6.2.

   If one or more N-active NoteOn or NoteOff commands in the checkpoint
   history reference a note number, the note number MUST be coded in
   either the note log list or the NoteOff bitfield structure.

   The note log list MUST contain an entry for all note numbers whose
   most recent checkpoint history appearance is in an N-active NoteOn
   command.  The NoteOff bitfield structure MUST contain a set bit for
   all note numbers whose most recent checkpoint history appearance is
   in an N-active NoteOff command.

   A note number MUST NOT be coded in both structures.

   All note logs and NoteOff bitfield set bits MUST code the most recent
   N-active NoteOn or NoteOff reference to a note number in the session
   history.

   The note log list MUST obey the oldest-first ordering rule (defined
   in Appendix A.1).

A.6.1.  Header Structure

   The header for Chapter N, shown in Figure A.6.2, codes the size of
   the note list and bitfield structures.

                       0                   1
                       0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5
                      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
                      |B|     LEN     |  LOW  | HIGH  |
                      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

                      Figure A.6.2 -- Chapter N header

   The LEN field, a 7-bit integer value, codes the number of 2-octet
   note logs in the note list.  Zero is a valid value for LEN and codes
   an empty note list.

   The 4-bit LOW and HIGH fields code the number of OFFBITS octets that
   follow the note log list.  LOW and HIGH are unsigned integer values.
   If LOW <= HIGH, there are (HIGH - LOW + 1) OFFBITS octets in the
   chapter.  The value pairs (LOW = 15, HIGH = 0) and (LOW = 15, HIGH =
   1) code an empty NoteOff bitfield structure (i.e., no OFFBITS
   octets).  Other (LOW > HIGH) value pairs MUST NOT appear in the
   header.

   The B bit provides S-bit functionality (Appendix A.1) for the NoteOff
   bitfield structure.  By default, the B bit MUST be set to 1.
   However, if the MIDI command section of the previous packet (packet I
   - 1, with I as defined in Appendix A.1) includes a NoteOff command
   for the channel, the B bit MUST be set to 0.  If the B bit is set to
   0, the higher-level recovery journal elements that contain Chapter N
   MUST have S bits that are set to 0, including the top-level journal
   header.

   The LEN value of 127 codes a note list length of 127 or 128 note
   logs, depending on the values of LOW and HIGH.  If LEN = 127, LOW =
   15, and HIGH = 0, the note list holds 128 note logs, and the NoteOff
   bitfield structure is empty.  For other values of LOW and HIGH, LEN =
   127 codes that the note list contains 127 note logs.  In this case,
   the chapter has (HIGH - LOW + 1) NoteOff OFFBITS octets if LOW <=
   HIGH and has no OFFBITS octets if LOW = 15 and HIGH = 1.

A.6.2.  Note Structures

   Figure A.6.3 shows the 2-octet note log structure.

                       0                   1
                       0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5
                      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
                      |S|   NOTENUM   |Y|  VELOCITY   |
                      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

                     Figure A.6.3 -- Chapter N note log

   The 7-bit NOTENUM field codes the note number for the log.  A note
   number MUST NOT be represented by multiple note logs in the note
   list.

   The 7-bit VELOCITY field codes the velocity value for the most recent
   N-active NoteOn command for the note number in the session history.
   Multiple overlapping NoteOns for a given note number may be coded
   using Chapter E, as discussed in Appendix A.7.

   VELOCITY is never zero; NoteOn commands with zero velocity are coded
   as NoteOff commands in the NoteOff bitfield structure.

   The note log does not code the execution time of the NoteOn command.
   However, the Y bit codes a hint from the sender about the NoteOn
   execution time.  The Y bit codes a recommendation to play (Y = 1) or
   skip (Y = 0) the NoteOn command recovered from the note log.  See
   Section 4.2 of [RFC4696] for non-normative guidance on the use of the
   Y bit.

   Figure A.6.1 shows the NoteOff bitfield structure, as the list of
   OFFBITS octets at the end of the chapter.  A NoteOff OFFBITS octet
   codes NoteOff information for eight consecutive MIDI note numbers,
   with the most-significant bit representing the lowest note number.
   The most-significant bit of the first OFFBITS octet codes the note
   number 8*LOW; the most-significant bit of the last OFFBITS octet
   codes the note number 8*HIGH.

   A set bit codes a NoteOff command for the note number.  In the most
   efficient coding for the NoteOff bitfield structure, the first and
   last octets of the structure contain at least one set bit.  Note that
   Chapter N does not code NoteOff velocity data.

   Note that in the general case, the recovery journal does not code the
   relative placement of a NoteOff command and a Change Control command
   for controller 64 (Damper Pedal (Sustain)).  In many cases, a
   receiver processing a loss event may deduce this relative placement

   from the history of the stream and thus determine if a NoteOff note
   is sustained by the pedal.  If such a determination is not possible,
   receivers SHOULD err on the side of silencing pedal sustains, as
   erroneously sustained notes may produce unpleasant (albeit transient)
   artifacts.

A.7.  Chapter E: MIDI Note Command Extras

   Readers may wish to review the Appendix A.1 definition of "N-active
   commands" before reading this appendix.  In this appendix, a NoteOn
   command with a velocity of 0 is considered to be a NoteOff command
   with a release velocity value of 64.

   Chapter E encodes recovery information about MIDI NoteOn (0x9) and
   NoteOff (0x8) command features that rarely appear in MIDI streams.
   Receivers use Chapter E to reduce transient artifacts for streams
   where several NoteOn commands appear for a note number without an
   intervening NoteOff.  Receivers also use Chapter E to reduce
   transient artifacts for streams that use NoteOff release velocity.
   Chapter E supplements the note information coded in Chapter N
   (Appendix A.6).

   Figure A.7.1 shows the format for Chapter E.

       0                   1                   2                   3
       0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 8 0 1
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
      |S|     LEN     |S|   NOTENUM   |V|  COUNT/VEL  |S|  NOTENUM    |
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
      |V|  COUNT/VEL  |  ....                                         |
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

                       Figure A.7.1 -- Chapter E format

   The chapter consists of a 1-octet header, followed by a variable-
   length list of 2-octet note logs.  Appendix A.7.1 defines the
   bitfield format for a note log.

   The log list MUST contain at least one note log.  The 7-bit LEN
   header field codes the number of note logs in the list, minus one.  A
   channel journal MUST contain Chapter E if the rules defined in this
   appendix require that one or more note logs appear in the list.  The
   note log list MUST obey the oldest-first ordering rule (defined in
   Appendix A.1).

A.7.1.  Note Log Format

   Figure A.7.2 reproduces the note log structure of Chapter E.

                       0                   1
                       0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5
                      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
                      |S|   NOTENUM   |V|  COUNT/VEL  |
                      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

                      Figure A.7.2 -- Chapter E note log

   A note log codes information about the MIDI note number coded by the
   7-bit NOTENUM field.  The nature of the information depends on the
   value of the V flag bit.

   If the V bit is set to 1, the COUNT/VEL field codes the release
   velocity value for the most recent N-active NoteOff command for the
   note number that appears in the session history.

   If the V bit is set to 0, the COUNT/VEL field codes a reference count
   of the number of NoteOn and NoteOff commands for the note number that
   appear in the session history.

   The reference count is set to 0 at the start of the session.  NoteOn
   commands increment the count by 1.  NoteOff commands decrement the
   count by 1.  However, a decrement that generates a negative count
   value is not performed.

   If the reference count is in the range 0-126, the 7-bit COUNT/VEL
   field codes an unsigned integer representation of the count.  If the
   count is greater than or equal to 127, COUNT/VEL is set to 127.

   By default, the count is reset to 0 whenever a Reset State command
   (Appendix A.1) appears in the session history, and whenever MIDI
   Control Change commands for controller numbers 123-127 (numbers with
   All Notes Off semantics) or 120 (All Sound Off) appear in the session
   history.

A.7.2.  Log Inclusion Rules

   If the most recent N-active NoteOn or NoteOff command for a note
   number in the checkpoint history is a NoteOff command with a release
   velocity value other than 64, a note log whose V bit is set to 1 MUST
   appear in Chapter E for the note number.

   If the most recent N-active NoteOn or NoteOff command for a note
   number in the checkpoint history is a NoteOff command, and if the

   reference count for the note number is greater than 0, a note log
   whose V bit is set to 0 MUST appear in Chapter E for the note number.

   If the most recent N-active NoteOn or NoteOff command for a note
   number in the checkpoint history is a NoteOn command, and if the
   reference count for the note number is greater than 1, a note log
   whose V bit is set to 0 MUST appear in Chapter E for the note number.

   At most, two note logs MAY appear in Chapter E for a note number: one
   log whose V bit is set to 0, and one log whose V bit is set to 1.

   Chapter E codes a maximum of 128 note logs.  If the log inclusion
   rules yield more than 128 REQUIRED logs, note logs whose V bit is set
   to 1 MUST be dropped from Chapter E in order to reach the 128-log
   limit.  Note logs whose V bit is set to 0 MUST NOT be dropped.

   Most MIDI streams do not use NoteOn and NoteOff commands in ways that
   would trigger the log inclusion rules.  For these streams, Chapter E
   would never be REQUIRED to appear in a channel journal.

   The ch_never parameter (Appendix C.2.3) may be used to configure the
   log inclusion rules for Chapter E.

A.8.  Chapter T: MIDI Channel Aftertouch

   A channel journal MUST contain Chapter T if an N-active and C-active
   MIDI Channel Aftertouch (0xD) command appears in the checkpoint
   history.  Figure A.8.1 shows the format for Chapter T.

                             0
                             0 1 2 3 4 5 6 7
                            +-+-+-+-+-+-+-+-+
                            |S|   PRESSURE  |
                            +-+-+-+-+-+-+-+-+

                      Figure A.8.1 -- Chapter T format

   The chapter has a fixed size of 8 bits.  The 7-bit PRESSURE field
   holds the pressure value of the most recent N-active and C-active
   Channel Aftertouch command in the session history.

   Chapter T only encodes commands that are C-active and N-active.  We
   define a C-active restriction because [RP015] declares that a Control
   Change command for controller 121 (Reset All Controllers) acts to
   reset the channel pressure to 0 (see the discussion at the end of
   Appendix A.5 for a more complete rationale).

   We define an N-active restriction on the assumption that aftertouch
   commands are linked to note activity, and thus Channel Aftertouch
   commands that are not N-active are stale and should not be used to
   repair a stream.

A.9.  Chapter A: MIDI Poly Aftertouch

   A channel journal MUST contain Chapter A if a C-active Poly
   Aftertouch (0xA) command appears in the checkpoint history.  Figure
   A.9.1 shows the format for Chapter A.

       0                   1                   2                   3
       0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 8 0 1
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
      |S|    LEN      |S|   NOTENUM   |X|  PRESSURE   |S|   NOTENUM   |
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
      |X|  PRESSURE   |  ....                                         |
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

                       Figure A.9.1 -- Chapter A format

   The chapter consists of a 1-octet header, followed by a variable-
   length list of 2-octet note logs.  A note log MUST appear for a note
   number if a C-active Poly Aftertouch command for the note number
   appears in the checkpoint history.  A note number MUST NOT be
   represented by multiple note logs in the note list.  The note log
   list MUST obey the oldest-first ordering rule (defined in Appendix
   A.1).

   The 7-bit LEN field codes the number of note logs in the list, minus
   one.  Figure A.9.2 reproduces the note log structure of Chapter A.

                       0                   1
                       0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5
                      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
                      |S|   NOTENUM   |X|  PRESSURE   |
                      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

                      Figure A.9.2 -- Chapter A note log

   The 7-bit PRESSURE field codes the pressure value of the most recent
   C-active Poly Aftertouch command in the session history for the MIDI
   note number coded in the 7-bit NOTENUM field.

   As a rule, the X bit MUST be set to 0.  However, the X bit MUST be
   set to 1 if the command coded by the log appears before one of the
   following commands in the session history: MIDI Control Change

   numbers 123-127 (numbers with All Notes Off semantics) or 120 (All
   Sound Off).

   We define C-active restrictions for Chapter A because [RP015]
   declares that a Control Change command for controller 121 (Reset All
   Controllers) acts to reset the polyphonic pressure to 0 (see the
   discussion at the end of Appendix A.5 for a more complete rationale).

B.  The Recovery Journal System Chapters

B.1.  System Chapter D: Simple System Commands

   The system journal MUST contain Chapter D if an active MIDI Reset
   (0xFF), MIDI Tune Request (0xF6), MIDI Song Select (0xF3), undefined
   MIDI System Common (0xF4 and 0xF5), or undefined MIDI System Real-
   time (0xF9 and 0xFD) command appears in the checkpoint history.

   Figure B.1.1 shows the variable-length format for Chapter D.

       0                   1                   2                   3
       0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
      |S|B|G|H|J|K|Y|Z|  Command logs ...                             |
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

                  Figure B.1.1 -- System Chapter D format

   The chapter consists of a 1-octet header, followed by one or more
   command logs.  Header flag bits indicate the presence of command logs
   for the Reset (B = 1), Tune Request (G = 1), Song Select (H = 1),
   undefined System Common 0xF4 (J = 1), undefined System Common 0xF5 (K
   = 1), undefined System Real-time 0xF9 (Y = 1), or undefined System
   Real-time 0xFD (Z = 1) commands.

   Command logs appear in a list following the header, in the order that
   the flag bits appear in the header.

   Figure B.1.2 shows the 1-octet command log format for the Reset and
   Tune Request commands.

                            0
                            0 1 2 3 4 5 6 7
                           +-+-+-+-+-+-+-+-+
                           |S|    COUNT    |
                           +-+-+-+-+-+-+-+-+

             Figure B.1.2 -- Command log for Reset and Tune Request

   Chapter D MUST contain the Reset command log if an active Reset
   command appears in the checkpoint history.  The 7-bit COUNT field
   codes the total number of Reset commands (modulo 128) present in the
   session history.

   Chapter D MUST contain the Tune Request command log if an active Tune
   Request command appears in the checkpoint history.  The 7-bit COUNT
   field codes the total number of Tune Request commands (modulo 128)
   present in the session history.

   For these commands, the COUNT field acts as a reference count.  See
   the definition of "session history reference counts" in Appendix A.1
   for more information.

   Figure B.1.3 shows the 1-octet command log format for the Song Select
   command.

                               0
                               0 1 2 3 4 5 6 7
                              +-+-+-+-+-+-+-+-+
                              |S|    VALUE    |
                              +-+-+-+-+-+-+-+-+

                 Figure B.1.3 -- Song Select command log format

   Chapter D MUST contain the Song Select command log if an active Song
   Select command appears in the checkpoint history.  The 7-bit VALUE
   field codes the song number of the most recent active Song Select
   command in the session history.

B.1.1.  Undefined System Commands

   In this section, we define the Chapter D command logs for the
   undefined System commands.  [MIDI] reserves the undefined System
   commands 0xF4, 0xF5, 0xF9, and 0xFD for future use.  At the time of
   this writing, any MIDI command stream that uses these commands is
   non-compliant with [MIDI].  However, future versions of [MIDI] may
   define these commands, and a few products do use these commands in a
   non-compliant manner.

   Figure B.1.4 shows the variable-length command log format for the
   undefined System Common commands (0xF4 and 0xF5).

       0                   1                   2                   3
       0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
      |S|C|V|L|DSZ|      LENGTH       |    COUNT      |  VALUE ...    |
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
      |  LEGAL ...                                                    |
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

          Figure B.1.4 -- Undefined System Common command log format

   The command log codes a single command type (0xF4 or 0xF5, not both).
   Chapter D MUST contain a command log if an active 0xF4 command
   appears in the checkpoint history and MUST contain an independent
   command log if an active 0xF5 command appears in the checkpoint
   history.

   Chapter D consists of a two-octet header followed by a variable
   number of data fields.  Header flag bits indicate the presence of the
   COUNT field (C = 1), the VALUE field (V = 1), and the LEGAL field (L
   = 1).  The 10-bit LENGTH field codes the size of the command log and
   conforms to semantics described in Appendix A.1.

   The 2-bit DSZ field codes the number of data octets in the command
   instance that appears most recently in the session history.  If DSZ =
   0-2, the command has 0-2 data octets.  If DSZ = 3, the command has 3
   or more command data octets.

   We now define the default rules for the use of the COUNT, VALUE, and
   LEGAL fields.  The session configuration tools defined in Appendix
   C.2.3 may be used to override this behavior.

   By default, if the DSZ field is set to 0, the command log MUST
   include the COUNT field.  The 8-bit COUNT field codes the total
   number of commands of the type coded by the log (0xF4 or 0xF5)
   present in the session history, modulo 256.

   By default, if the DSZ field is set to 1-3, the command log MUST
   include the VALUE field.  The variable-length VALUE field codes a
   verbatim copy the data octets for the most recent use of the command
   type coded by the log (0xF4 or 0xF5) in the session history.  The
   most-significant bit of the final data octet MUST be set to 1, and
   the most-significant bit of all other data octets MUST be set to 0.

   The LEGAL field is reserved for future use.  If an update to [MIDI]
   defines the 0xF4 or 0xF5 command, an IETF standards-track document
   may define the LEGAL field.  Until such a document appears, senders
   MUST NOT use the LEGAL field, and receivers MUST use the LENGTH field
   to skip over the LEGAL field.  The LEGAL field would be defined by
   the IETF if the semantics of the new 0xF4 or 0xF5 command could not
   be protected from packet loss via the use of the COUNT and VALUE
   fields.

   Figure B.1.5 shows the variable-length command log format for the
   undefined System Real-time commands (0xF9 and 0xFD).

       0                   1                   2                   3
       0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
      |S|C|L| LENGTH  |     COUNT     |  LEGAL ...                    |
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

        Figure B.1.5 -- Undefined System Real-time command log format

   The command log codes a single command type (0xF9 or 0xFD, not both).
   Chapter D MUST contain a command log if an active 0xF9 command
   appears in the checkpoint history and MUST contain an independent
   command log if an active 0xFD command appears in the checkpoint
   history.

   Chapter D consists of a one-octet header followed by a variable
   number of data fields.  Header flag bits indicate the presence of the
   COUNT field (C = 1) and the LEGAL field (L = 1).  The 5-bit LENGTH
   field codes the size of the command log and conforms to semantics
   described in Appendix A.1.

   We now define the default rules for the use of the COUNT and LEGAL
   fields.  The session configuration tools defined in Appendix C.2.3
   may be used to override this behavior.

   The 8-bit COUNT field codes the total number of commands of the type
   coded by the log present in the session history, modulo 256.  By
   default, the COUNT field MUST be present in the command log.

   The LEGAL field is reserved for future use.  If an update to [MIDI]
   defines the 0xF9 or 0xFD command, an IETF standards-track document
   may define the LEGAL field to protect the command.  Until such a
   document appears, senders MUST NOT use the LEGAL field, and receivers
   MUST use the LENGTH field to skip over the LEGAL field.  The LEGAL
   field would be defined by the IETF if the semantics of the new 0xF9
   or 0xFD command could not be protected from packet loss via the use
   of the COUNT field.

   Finally, we note that some non-standard uses of the undefined System
   Real-time commands act to implement non-compliant variants of the
   MIDI sequencer system.  In Appendix B.3.1, we describe resiliency
   tools for the MIDI sequencer system that provide some protection in
   this case.

B.2.  System Chapter V: Active Sense Command

   The system journal MUST contain Chapter V if an active MIDI Active
   Sense (0xFE) command appears in the checkpoint history.  Figure B.2.1
   shows the format for Chapter V.

                               0
                               0 1 2 3 4 5 6 7
                              +-+-+-+-+-+-+-+-+
                              |S|    COUNT    |
                              +-+-+-+-+-+-+-+-+

                     Figure B.2.1 -- System Chapter V format

   The 7-bit COUNT field codes the total number of Active Sense commands
   (modulo 128) present in the session history.  The COUNT field acts as
   a reference count.  See the definition of "session history reference
   counts" in Appendix A.1 for more information.

B.3.  System Chapter Q: Sequencer State Commands

   This appendix describes Chapter Q, the system chapter for the MIDI
   sequencer commands.

   The system journal MUST contain Chapter Q if an active MIDI Song
   Position Pointer (0xF2), MIDI Clock (0xF8), MIDI Start (0xFA), MIDI
   Continue (0xFB), or MIDI Stop (0xFC) command appears in the
   checkpoint history, and if the rules defined in this appendix require
   a change in the Chapter Q bitfield contents because of the command
   appearance.

   Figure B.3.1 shows the variable-length format for Chapter Q.

       0                   1                   2                   3
       0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
      |S|N|D|C|T| TOP |            CLOCK              | TIMETOOLS ... |
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
      |              ...              |
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

                    Figure B.3.1 -- System Chapter Q format

   Chapter Q consists of a 1-octet header followed by several optional
   fields, in the order shown in Figure B.3.1.

   Header flag bits signal the presence of the 16-bit CLOCK field (C =
   1) and the 24-bit TIMETOOLS field (T = 1).  The 3-bit TOP header
   field is interpreted as an unsigned integer, as are CLOCK and
   TIMETOOLS.  We describe the TIMETOOLS field in Appendix B.3.1.

   Chapter Q encodes the most recent state of the sequencer system.
   Receivers use the chapter to re-synchronize the sequencer after a
   packet loss episode.  Chapter fields encode the on/off state of the
   sequencer, the current position in the song, and the downbeat.

   The N header bit encodes the relative occurrence of the Start, Stop,
   and Continue commands in the session history.  If an active Start or
   Continue command appears most recently, the N bit MUST be set to 1.
   If an active Stop appears most recently, or if no active Start, Stop,
   or Continue commands appear in the session history, the N bit MUST be
   set to 0.

   The C header flag, the TOP header field, and the CLOCK field act to
   code the current position in the sequence:

     o  If C = 1, the 3-bit TOP header field and the 16-bit CLOCK field
        are combined to form the 19-bit unsigned quantity 65536*TOP +
        CLOCK.  This value encodes the song position in units of MIDI
        Clocks (24 clocks per quarter note), modulo 524288.  Note that
        the maximum song position value that may be coded by the Song
        Position Pointer command is 98303 clocks (which may be coded
        with 17 bits), and that MIDI-coded songs are generally
        constructed to avoid durations longer than this value.  However,
        the 19-bit size may be useful for real-time applications, such
        as a drum machine MIDI output that is sending clock commands for
        long periods of time.

     o  If C = 0, the song position is the start of the song.  The C = 0
        position is identical to the position coded by C = 1, TOP = 0,
        and CLOCK = 0, for the case where the song position is less than
        524288 MIDI clocks.  In certain situations (defined later in
        this section), normative text may require the C = 0 or the C =
        1, TOP = 0, CLOCK = 0 encoding of the start of the song.

   The C, TOP, and CLOCK fields MUST be set to code the current song
   position, for both N = 0 and N = 1 conditions.  If C = 0, the TOP
   field MUST be set to 0.  See [MIDI] for a precise definition of a
   song position.

   The D header bit encodes information about the downbeat and acts to
   qualify the song position coded by the C, TOP, and CLOCK fields.

   If the D bit is set to 1, the song position represents the most
   recent position in the sequence that has played.  If D = 1, the next
   Clock command (if N = 1) or the next (Continue, Clock) pair (if
   N = 0) acts to increment the song position by one clock, and to play
   the updated position.

   If the D bit is set to 0, the song position represents a position in
   the sequence that has not yet been played.  If D = 0, the next Clock
   command (if N = 1) or the next (Continue, Clock) pair (if N = 0) acts
   to play the point in the song coded by the song position.  The song
   position is not incremented.

   An example of a stream that uses D = 0 coding is one whose most
   recent sequence command is a Start or Song Position Pointer command
   (both N = 1 conditions).  However, it is also possible to construct
   examples where D = 0 and N = 0.  A Start command immediately followed
   by a Stop command is coded in Chapter Q by setting C = 0, D = 0,
   N = 0, TOP = 0.

   If N = 1 (coding Start or Continue), D = 0 (coding that the downbeat
   has yet to be played), and the song position is at the start of the
   song, the C = 0 song position encoding MUST be used if a Start
   command occurs more recently than a Continue command in the session
   history, and the C = 1, TOP = 0, CLOCK = 0 song position encoding
   MUST be used if a Continue command occurs more recently than a Start
   command in the session history.

B.3.1.  Non-compliant Sequencers

   The Chapter Q description in this appendix assumes that the sequencer
   system counts off time with Clock commands, as mandated in [MIDI].
   However, a few non-compliant products do not use Clock commands to
   count off time, but instead use non-standard methods.

   Chapter Q uses the TIMETOOLS field to provide resiliency support for
   these non-standard products.  By default, the TIMETOOLS field MUST
   NOT appear in Chapter Q, and the T header bit MUST be set to 0.  The
   session configuration tools described in Appendix C.2.3 may be used
   to select TIMETOOLS coding.

   Figure B.3.2 shows the format of the 24-bit TIMETOOLS field.

                0                   1                   2
                0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3
               +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
               |                   TIME                        |
               +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

                       Figure B.3.2 -- TIMETOOLS format

   The TIME field is a 24-bit unsigned integer quantity, with units of
   milliseconds.  TIME codes an additive correction term for the song
   position coded by the TOP, CLOCK, and C fields.  TIME is coded in
   network byte order (big-endian).

   A receiver computes the correct song position by converting TIME into
   units of MIDI clocks and adding it to 65536*TOP + CLOCK (assuming
   C = 1).  Alternatively, a receiver may convert 65536*TOP + CLOCK into
   milliseconds (assuming C = 1) and add it to TIME.  The downbeat (D
   header bit) semantics defined in Appendix B.3 apply to the corrected
   song position.

B.4.  System Chapter F: MIDI Time Code Tape Position

   This appendix describes Chapter F, the system chapter for the MIDI
   Time Code (MTC) commands.  Readers may wish to review the Appendix
   A.1 definition of "finished/unfinished commands" before reading this
   appendix.

   The system journal MUST contain Chapter F if an active System Common
   Quarter Frame command (0xF1) or an active finished System Exclusive
   (Universal Real Time) MTC Full Frame command (F0 7F cc 01 01 hr mn sc
   fr F7) appears in the checkpoint history.  Otherwise, the system
   journal MUST NOT contain Chapter F.

   Figure B.4.1 shows the variable-length format for Chapter F.

       0                   1                   2                   3
       0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
      |S|C|P|Q|D|POINT|  COMPLETE ...                                 |
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
      |     ...       |  PARTIAL  ...                                 |
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
      |     ...       |
      +-+-+-+-+-+-+-+-+

                    Figure B.4.1 -- System Chapter F format

   Chapter F holds information about recent MTC tape positions coded in
   the session history.  Receivers use Chapter F to re-synchronize the
   MTC system after a packet loss episode.

   Chapter F consists of a 1-octet header followed by several optional
   fields, in the order shown in Figure B.4.1.  The C and P header bits
   form a Table of Contents (TOC) and signal the presence of the 32-bit
   COMPLETE field (C = 1) and the 32-bit PARTIAL field (P = 1).

   The Q header bit codes information about the COMPLETE field format.
   If Chapter F does not contain a COMPLETE field, Q MUST be set to 0.

   The D header bit codes the tape movement direction.  If the tape is
   moving forward, or if the tape direction is indeterminate, the D bit
   MUST be set to 0.  If the tape is moving in the reverse direction,
   the D bit MUST be set to 1.  In most cases, the ordering of commands
   in the session history clearly defines the tape direction.  However,
   a few command sequences have an indeterminate direction (such as a
   session history consisting of one Full Frame command).

   The 3-bit POINT header field is interpreted as an unsigned integer.
   Appendix B.4.1 defines how the POINT field codes information about
   the contents of the PARTIAL field.  If Chapter F does not contain a
   PARTIAL field, POINT MUST be set to 7 (if D = 0) or 0 (if D = 1).

   Chapter F MUST include the COMPLETE field if an active finished Full
   Frame command appears in the checkpoint history, or if an active
   Quarter Frame command that completes the encoding of a frame value
   appears in the checkpoint history.

   The COMPLETE field encodes the most recent active complete MTC frame
   value that appears in the session history.  This frame value may take
   the form of a series of 8 active Quarter Frame commands (0xF1 0x0n
   through 0xF1 0x7n for forward tape movement, 0xF1 0x7n through 0xF1
   0x0n for reverse tape movement) or may take the form of an active
   finished Full Frame command.

   If the COMPLETE field encodes a Quarter Frame command series, the Q
   header bit MUST be set to 1, and the COMPLETE field MUST have the
   format shown in Figure B.4.2.  The 4-bit fields MT0 through MT7 code
   the data (lower) nibble for the Quarter Frame commands for Message
   Type 0 through Message Type 7 [MIDI].  These nibbles encode a
   complete frame value, in addition to fields reserved for future use
   by [MIDI].

       0                   1                   2                   3
       0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
      |  MT0  |  MT1  |  MT2  |  MT3  |  MT4  |  MT5  |  MT6  |  MT7  |
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

              Figure B.4.2 -- COMPLETE field format, Q = 1

   In this usage, the frame value encoded in the COMPLETE field MUST be
   offset by 2 frames (relative to the frame value encoded in the
   Quarter Frame commands) if the frame value codes a 0xF1 0x0n through
   0xF1 0x7n command sequence.  This offset compensates for the two-
   frame latency of the Quarter Frame encoding for forward tape
   movement.  No offset is applied if the frame value codes a 0xF1 0x7n
   through 0xF1 0x0n Quarter Frame command sequence.

   The most recent active complete MTC frame value may alternatively be
   encoded by an active finished Full Frame command.  In this case, the
   Q header bit MUST be set to 0, and the COMPLETE field MUST have
   format shown in Figure B.4.3.  The HR, MN, SC, and FR fields
   correspond to the hr, mn, sc, and fr data octets of the Full Frame
   command.

       0                   1                   2                   3
       0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
      |      HR       |      MN       |      SC       |      FR       |
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

              Figure B.4.3 -- COMPLETE field format, Q = 0

B.4.1.  Partial Frames

   The most recent active session history command that encodes MTC frame
   value data may be a Quarter Frame command other than a forward-moving
   0xF1 0x7n command (which completes a frame value for forward tape
   movement) or a reverse-moving 0xF1 0x1n command (which completes a
   frame value for reverse tape movement).

   We consider this type of Quarter Frame command to be associated with
   a partial frame value.  The Quarter Frame sequence that defines a
   partial frame value MUST either start at Message Type 0 and increment
   contiguously to an intermediate Message Type less than 7, or start at
   Message Type 7 and decrement contiguously to an intermediate Message
   type greater than 0.  A Quarter Frame command sequence that does not
   follow this pattern is not associated with a partial frame value.

   Chapter F MUST include a PARTIAL field if the most recent active
   command in the checkpoint history that encodes MTC frame value data
   is a Quarter Frame command that is associated with a partial frame
   value.  Otherwise, Chapter F MUST NOT include a PARTIAL field.

   The partial frame value consists of the data (lower) nibbles of the
   Quarter Frame command sequence.  The PARTIAL field codes the partial
   frame value, using the format shown in Figure B.4.2.  Message Type
   fields that are not associated with a Quarter Frame command MUST be
   set to 0.

   The POINT header field indicates the Message Type fields in the
   PARTIAL field code valid data.  If P = 1, the POINT field MUST encode
   the unsigned integer value formed by the lower 3 bits of the upper
   nibble of the data value of the most recent active Quarter Frame
   command in the session history.  If D = 0 and P = 1, POINT MUST take
   on a value in the range 0-6.  If D = 1 and P = 1, POINT MUST take on
   a value in the range 1-7.

   If D = 0, MT fields (Figure B.4.2) in the inclusive range from 0 up
   to and including the POINT value encode the partial frame value.  If
   D = 1, MT fields in the inclusive range from 7 down to and including
   the POINT value encode the partial frame value.  Note that, unlike
   the COMPLETE field encoding, senders MUST NOT add a 2-frame offset to
   the partial frame value encoded in PARTIAL.

   For the default semantics, if a recovery journal contains Chapter F,
   and if the session history codes a legal [MIDI] series of Quarter
   Frame and Full Frame commands, the chapter always contains a COMPLETE
   or a PARTIAL field (and may contain both fields).  Thus, a one-octet
   Chapter F (C = P = 0) always codes the presence of an illegal command
   sequence in the session history (under some conditions, the C = 1,
   P = 0 condition may also code the presence of an illegal command
   sequence).  The illegal command sequence conditions are transient in
   nature and usually indicate that a Quarter Frame command sequence
   began with an intermediate Message Type.

B.5.  System Chapter X: System Exclusive

   This appendix describes Chapter X, the system chapter for MIDI System
   Exclusive (SysEx) commands (0xF0).  Readers may wish to review the
   Appendix A.1 definition of "finished/unfinished commands" before
   reading this appendix.

   Chapter X consists of a list of one or more command logs.  Each log
   in the list codes information about a specific finished or unfinished
   SysEx command that appears in the session history.  The system

   journal MUST contain Chapter X if the rules defined in Appendix B.5.2
   require that one or more logs appear in the list.

   The log list is not preceded by a header.  Instead, each log
   implicitly encodes its own length.  Given the length of the N'th list
   log, the presence of the (N+1)'th list log may be inferred from the
   LENGTH field of the system journal header (Figure 10 in Section 5 of
   the main text).  The log list MUST obey the oldest-first ordering
   rule (defined in Appendix A.1).

B.5.1.  Chapter Format

   Figure B.5.1 shows the bitfield format for the Chapter X command log.

       0                   1                   2                   3
       0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
      |S|T|C|F|D|L|STA|    TCOUNT     |     COUNT     |  FIRST ...    |
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
      |  DATA ...                                                     |
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

                 Figure B.5.1 -- Chapter X command log format

   A Chapter X command log consists of a 1-octet header, followed by the
   optional TCOUNT, COUNT, FIRST, and DATA fields.

   The T, C, F, and D header bits act as a Table of Contents (TOC) for
   the log.  If T is set to 1, the 1-octet TCOUNT field appears in the
   log.  If C is set to 1, the 1-octet COUNT field appears in the log.
   If F is set to 1, the variable-length FIRST field appears in the log.
   If D is set to 1, the variable-length DATA field appears in the log.

   The L header bit sets the coding tool for the log.  We define the log
   coding tools in Appendix B.5.2.

   The STA field codes the status of the command coded by the log.  The
   2-bit STA value is interpreted as an unsigned integer.  If STA is 0,
   the log codes an unfinished command.  Non-zero STA values code
   different classes of finished commands.  An STA value of 1 codes a
   cancelled command, an STA value of 2 codes a command that uses the
   "dropped F7" construction, and an STA value of 3 codes all other
   finished commands.  Section 3.2 in the main text describes cancelled
   and "dropped F7" commands.

   The S bit (Appendix A.1) of the first log in the list acts as the S
   bit for Chapter X.  For the other logs in the list, the S bit refers

   to the log itself.  The value of the "phantom" S bit associated with
   the first log is defined by the following rules:

     o  If the list codes one log, the phantom S-bit value is the same
        as the Chapter X S-bit value.

     o  If the list codes multiple logs, the phantom S-bit value is the
        logical OR of the S-bit value of the first and second command
        logs in the list.

   In all other respects, the S bit follows the semantics defined in
   Appendix A.1.

   The FIRST field (present if F = 1) encodes a variable-length unsigned
   integer value that sets the coverage of the DATA field.

   The FIRST field (present if F = 1) encodes a variable-length unsigned
   integer value that specifies which SysEx data bytes are encoded in
   the DATA field of the log.  The FIRST field consists of an octet
   whose most-significant bit is set to 0, optionally preceded by one or
   more octets whose most-significant bit is set to 1.  The algorithm
   shown in Figure B.5.2 decodes this format into an unsigned integer,
   to yield the value dec(FIRST).  FIRST uses a variable-length encoding
   because dec(FIRST) references a data octet in a SysEx command, and a
   SysEx command may contain an arbitrary number of data octets.

        One-Octet FIRST value:

           Encoded form: 0ddddddd
           Decoded form: 00000000 00000000 00000000 0ddddddd

        Two-Octet FIRST value:

           Encoded form: 1ccccccc 0ddddddd
           Decoded form: 00000000 00000000 00cccccc cddddddd

        Three-Octet FIRST value:

           Encoded form: 1bbbbbbb 1ccccccc 0ddddddd
           Decoded form: 00000000 000bbbbb bbcccccc cddddddd

        Four-Octet FIRST value:

           Encoded form: 1aaaaaaa 1bbbbbbb 1ccccccc 0ddddddd
           Decoded form: 0000aaaa aaabbbbb bbcccccc cddddddd

                Figure B.5.2 -- Decoding FIRST field formats

   The DATA field (present if D = 1) encodes a modified version of the
   data octets of the SysEx command coded by the log.  Status octets
   MUST NOT be coded in the DATA field.

   If F = 0, the DATA field begins with the first data octet of the
   SysEx command and includes all subsequent data octets for the command
   that appear in the session history.  If F = 1, the DATA field begins
   with the (dec(FIRST) + 1)'th data octet of the SysEx command and
   includes all subsequent data octets for the command that appear in
   the session history.  Note that the word "command" in the
   descriptions above refers to the original SysEx command as it appears
   in the source MIDI data stream, not to a particular MIDI list SysEx
   command segment.

   The length of the DATA field is coded implicitly, using the most-
   significant bit of each octet.  The most-significant bit of the final
   octet of the DATA field MUST be set to 1.  The most-significant bit
   of all other DATA octets MUST be set to 0.  This coding method relies
   on the fact that the most-significant bit of a MIDI data octet is 0
   by definition.  Apart from this length-coding modification, the DATA
   field encodes a verbatim copy of all data octets it encodes.

B.5.2.  Log Inclusion Semantics

   Chapter X offers two tools to protect SysEx commands: the "recency"
   tool and the "list" tool.  The tool definitions use the concept of
   the "SysEx type" of a command, which we now define.

   Each SysEx command instance in a session, excepting MTC Full Frame
   commands, is said to have a "SysEx type".  Types are used in equality
   comparisons: two SysEx commands in a session are said to have "the
   same SysEx type" or "different SysEx types".

   If efficiency is not a concern, a sender may follow a simple typing
   rule: every SysEx command in the session history has a different
   SysEx type, and thus no two commands in the session have the same
   type.

   To improve efficiency, senders MAY implement exceptions to this rule.
   These exceptions declare that certain sets of SysEx command instances
   have the same SysEx type.  Any command not covered by an exception
   follows the simple rule.  We list exceptions below:

     o  All commands with identical data octet fields (same number of
        data octets, same value for each data octet) have the same type.
        This rule MUST be applied to all SysEx commands in the session,
        or not at all.  Note that the implementation of this exception
        requires no sender knowledge of the format and semantics of the

        SysEx commands in the stream, merely the ability to count and
        compare octets.

     o  Two instances of the same command whose semantics set or report
        the value of the same "parameter" have the same type.  The
        implementation of this exception requires specific knowledge of
        the format and semantics of SysEx commands.  In practice, a
        sender implementation chooses to support this exception for
        certain classes of commands (such as the Universal System
        Exclusive commands defined in [MIDI]).  If a sender supports
        this exception for a particular command in a class (for example,
        the Universal Real Time System Exclusive message for Master
        Volume, F0 F7 cc 04 01 vv vv F7, defined in [MIDI]), it MUST
        support the exception to all instances of this particular
        command in the session.

   We now use this definition of "SysEx type" to define the "recency"
   tool and the "list" tool for Chapter X.

   By default, the Chapter X log list MUST code sufficient information
   to protect the rendered MIDI performance from indefinite artifacts
   caused by the loss of all finished or unfinished active SysEx
   commands that appear in the checkpoint history (excluding finished
   MTC Full Frame commands, which are coded in Chapter F (Appendix
   B.4)).

   To protect a command of a specific SysEx type with the recency tool,
   senders MUST code a log in the log list for the most recent finished
   active instance of the SysEx type that appears in the checkpoint
   history.  Additionally, if an unfinished active instance of the SysEx
   type appears in the checkpoint history, senders MUST code a log in
   the log list for the unfinished command instance.  The L header bit
   of both command logs MUST be set to 0.

   To protect a command of a specific SysEx type with the list tool,
   senders MUST code a log in the Chapter X log list for each finished
   or unfinished active instance of the SysEx type that appears in the
   checkpoint history.  The L header bit of list tool command logs MUST
   be set to 1.

   As a rule, a log REQUIRED by the list or recency tool MUST include a
   DATA field that codes all data octets that appear in the checkpoint
   history for the SysEx command instance associated with the log.  The
   FIRST field MAY be used to configure a DATA field that minimally
   meets this requirement.

   An exception to this rule applies to cancelled commands (defined in
   Section 3.2).  REQUIRED command logs associated with cancelled
   commands MAY be coded with no DATA field.  However, if DATA appears
   in the log, DATA MUST code all data octets that appear in the
   checkpoint history for the command associated with the log.

   As defined by the preceding text in this section, by default all
   finished or unfinished active SysEx commands that appear in the
   checkpoint history (excluding finished MTC Full Frame commands) MUST
   be protected by the list tool or the recency tool.

   For some MIDI source streams, this default yields a Chapter X whose
   size is too large.  For example, imagine that a sender begins to
   transcode a SysEx command with 10,000 data octets onto a UDP RTP
   stream "on the fly", by sending SysEx command segments as soon as
   data octets are delivered by the MIDI source.  After 1000 octets have
   been sent, the expansion of Chapter X yields an RTP packet that is
   too large to fit in the Maximum Transmission Unit (MTU) for the
   stream.

   In this situation, if a sender uses the closed-loop sending policy
   for SysEx commands, the RTP packet size may always be capped by
   stalling the stream.  In a stream stall, once the packet reaches a
   maximum size, the sender refrains from sending new packets with non-
   empty MIDI Command Sections until receiver feedback permits the
   trimming of Chapter X.  If the stream permits arbitrary commands to
   appear between SysEx segments (selectable during configuration using
   the tools defined in Appendix C.1), the sender may stall the SysEx
   segment stream but continue to code other commands in the MIDI list.

   Stalls are a workable but sub-optimal solution to Chapter X size
   issues.  As an alternative to stalls, senders SHOULD take preemptive
   action during session configuration to reduce the anticipated size of
   Chapter X, using the methods described below:

     o  Partitioned transport.  Appendix C.5 provides tools for sending
        a MIDI name space over several RTP streams.  Senders may use
        these tools to map a MIDI source into a low-latency UDP RTP
        stream (for channel commands and short SysEx commands) and a
        reliable [RFC4571] TCP stream (for bulk-data SysEx commands).
        The cm_unused and cm_used parameters (Appendix C.1) may be used
        to communicate the nature of the SysEx command partition.  As
        TCP is reliable, the RTP MIDI TCP stream would not use the
        recovery journal.  To minimize transmission latency for short
        SysEx commands, senders may begin segmental transmission for all
        SysEx commands over the UDP stream and then cancel the UDP
        transmission of long commands (using tools described in Section
        3.2) and resend the commands over the TCP stream.

     o  Selective protection.  Journal protection may not be necessary
        for all SysEx commands in a stream.  The ch_never parameter
        (Appendix C.2) may be used to communicate which SysEx commands
        are excluded from Chapter X.

B.5.3.  TCOUNT and COUNT Fields

   If the T header bit is set to 1, the 8-bit TCOUNT field appears in
   the command log.  If the C header bit is set to 1, the 8-bit COUNT
   field appears in the command log.  TCOUNT and COUNT are interpreted
   as unsigned integers.

   The TCOUNT field codes the total number of SysEx commands of the
   SysEx type coded by the log that appear in the session history, at
   the moment after the (finished or unfinished) command coded by the
   log enters the session history.

   The COUNT field codes the total number of SysEx commands that appear
   in the session history, excluding commands that are excluded from
   Chapter X via the ch_never parameter (Appendix C.2), at the moment
   after the (finished or unfinished) command coded by the log enters
   the session history.

   Command counting for TCOUNT and COUNT uses modulo-256 arithmetic.
   MTC Full Frame command instances (Appendix B.4) are included in
   command counting if the TCOUNT and COUNT definitions warrant their
   inclusion, as are cancelled commands (Section 3.2).

   Senders use the TCOUNT and COUNT fields to track the identity and
   (for TCOUNT) the sequence position of a command instance.  Senders
   MUST use the TCOUNT or COUNT fields if identity or sequence
   information is necessary to protect the command type coded by the
   log.

   If a sender uses the COUNT field in a session, the final command log
   in every Chapter X in the stream MUST code the COUNT field.  This
   rule lets receivers resynchronize the COUNT value after a packet
   loss.

C.  Session Configuration Tools

   In Sections 6.1-2 of the main text, we show session descriptions for
   minimal native and mpeg4-generic RTP MIDI streams.  Minimal streams
   lack the flexibility to support some applications.  In this appendix,
   we describe how to customize stream behavior through the use of the
   payload format parameters.

   The appendix begins with 6 sections, each devoted to parameters that
   affect a particular aspect of stream behavior:

     o  Appendix C.1 describes the stream subsetting system (cm_unused
        and cm_used).

     o  Appendix C.2 describes the journalling system (ch_anchor,
        ch_default, ch_never, j_sec, j_update).

     o  Appendix C.3 describes MIDI command timestamp semantics
        (linerate, mperiod, octpos, tsmode).

     o  Appendix C.4 describes the temporal duration ("media time") of
        an RTP MIDI packet (guardtime, rtp_maxptime, rtp_ptime).

     o  Appendix C.5 concerns stream description (musicport).

     o  Appendix C.6 describes MIDI rendering (chanmask, cid, inline,
        multimode, render, rinit, subrender, smf_cid, smf_info,
        smf_inline, smf_url, url).

   The parameters listed above may optionally appear in session
   descriptions of RTP MIDI streams.  If these parameters are used in an
   SDP session description, the parameters appear on an fmtp attribute
   line.  This attribute line applies to the payload type associated
   with the fmtp line.

   The parameters listed above add extra functionality ("features") to
   minimal RTP MIDI streams.  In Appendix C.7, we show how to use these
   features to support two classes of applications: content-streaming
   using RTSP (Appendix C.7.1) and network musical performance using SIP
   (Appendix C.7.2).

   The participants in a multimedia session MUST share a common view of
   all of the RTP MIDI streams that appear in an RTP session, as defined
   by a single media (m=) line.  In some RTP MIDI applications, the
   "common view" restriction makes it difficult to use sendrecv streams
   (all parties send and receive), as each party has its own
   requirements.  For example, a two-party network musical performance
   application may wish to customize the renderer on each host to match
   the CPU performance of the host [NMP].

   We solve this problem by using two RTP MIDI streams -- one sendonly,
   one recvonly -- in lieu of one sendrecv stream.  The data flows in
   the two streams travel in opposite directions, to control receivers
   configured to use different renderers.  In the third example in
   Appendix C.5, we show how the musicport parameter may be used to
   define virtual sendrecv streams.

   As a general rule, the RTP MIDI protocol does not handle parameter
   changes during a session well, because the parameters describe
   heavyweight or stateful configuration that is not easily changed once
   a session has begun.  Thus, parties SHOULD NOT expect that parameter
   change requests during a session will be accepted by other parties.
   However, implementors SHOULD support in-session parameter changes
   that are easy to handle (for example, the guardtime parameter defined
   in Appendix C.4) and SHOULD be capable of accepting requests for
   changes of those parameters, as received by its session management
   protocol (for example, re-offers in SIP [RFC3264]).

   Appendix D defines the Augmented Backus-Naur Form (ABNF, [RFC4234])
   syntax for the payload parameters.  Section 11 provides information
   to the Internet Assigned Numbers Authority (IANA) on the media types
   and parameters defined in this document.

   Appendix C.6.5 defines the media type "audio/asc", a stored object
   for initializing mpeg4-generic renderers.  As described in Appendix
   C.6, the audio/asc media type is assigned to the "rinit" parameter to
   specify an initialization data object for the default mpeg4-generic
   renderer.  Note that RTP stream semantics are not defined for
   "audio/asc".  Therefore, the "asc" subtype MUST NOT appear on the
   rtpmap line of a session description.

C.1.  Configuration Tools: Stream Subsetting

   As defined in Section 3.2 in the main text, the MIDI list of an RTP
   MIDI packet may encode any MIDI command that may legally appear on a
   MIDI 1.0 DIN cable.

   In this appendix, we define two parameters (cm_unused and cm_used)
   that modify this default condition, by excluding certain types of
   MIDI commands from the MIDI list of all packets in a stream.  For
   example, if a multimedia session partitions a MIDI name space into
   two RTP MIDI streams, the parameters may be used to define which
   commands appear in each stream.

   In this appendix, we define a simple language for specifying MIDI
   command types.  If a command type is assigned to cm_unused, the
   commands coded by the string MUST NOT appear in the MIDI list.  If a
   command type is assigned to cm_used, the commands coded by the string
   MAY appear in the MIDI list.

   The parameter list may code multiple assignments to cm_used and
   cm_unused.  Assignments have a cumulative effect and are applied in
   the order of appearance in the parameter list.  A later assignment of
   a command type to the same parameter expands the scope of the earlier
   assignment.  A later assignment of a command type to the opposite

   parameter cancels (partially or completely) the effect of an earlier
   assignment.

   To initialize the stream subsetting system, "implicit" assignments to
   cm_unused and cm_used are processed before processing the actual
   assignments that appear in the parameter list.  The System Common
   undefined commands (0xF4, 0xF5) and the System Real-Time Undefined
   commands (0xF9, 0xFD) are implicitly assigned to cm_unused.  All
   other command types are implicitly assigned to cm_used.

   Note that the implicit assignments code the default behavior of an
   RTP MIDI stream as defined in Section 3.2 in the main text (namely,
   that all commands that may legally appear on a MIDI 1.0 DIN cable may
   appear in the stream).  Also note that assignments of the System
   Common undefined commands (0xF4, 0xF5) apply to the use of these
   commands in the MIDI source command stream, not the special use of
   0xF4 and 0xF5 in SysEx segment encoding defined in Section 3.2 in the
   main text.

   As a rule, parameter assignments obey the following syntax (see
   Appendix D for ABNF):

     <parameter> = [channel list]<command-type list>[field list]

   The command-type list is mandatory; the channel and field lists are
   optional.

   The command-type list specifies the MIDI command types for which the
   parameter applies.  The command-type list is a concatenated sequence
   of one or more of the letters (ABCFGHJKMNPQTVWXYZ).  The letters code
   the following command types:

      o  A: Poly Aftertouch (0xA)
      o  B: System Reset (0xFF)
      o  C: Control Change (0xB)
      o  F: System Time Code (0xF1)
      o  G: System Tune Request (0xF6)
      o  H: System Song Select (0xF3)
      o  J: System Common Undefined (0xF4)
      o  K: System Common Undefined (0xF5)
      o  N: NoteOff (0x8), NoteOn (0x9)
      o  P: Program Change (0xC)
      o  Q: System Sequencer (0xF2, 0xF8, 0xF9, 0xFA, 0xFB, 0xFC)
      o  T: Channel Aftertouch (0xD)
      o  V: System Active Sense (0xFE)
      o  W: Pitch Wheel (0xE)

      o  X: SysEx (0xF0)
      o  Y: System Real-Time Undefined (0xF9)
      o  Z: System Real-Time Undefined (0xFD)

   In addition to the letters above, the letter M may also appear in the
   command-type list.  The letter M refers to the MIDI parameter system
   (see definition in Appendix A.1 and in [MIDI]).  An assignment of M
   to cm_unused codes that no RPN or NRPN transactions may appear in the
   MIDI list.

   Note that if cm_unused is assigned the letter M, Control Change (0xB)
   commands for the controller numbers in the standard controller
   assignment might still appear in the MIDI list.  For an explanation,
   see Appendix A.3.4 for a discussion of the "general-purpose" use of
   parameter system controller numbers.

   In the text below, rules that apply to "MIDI voice channel commands"
   also apply to the letter M.

   The letters in the command-type list MUST be uppercase and MUST
   appear in alphabetical order.  Letters other than
   (ABCFGHJKMNPQTVWXYZ) that appear in the list MUST be ignored.

   For MIDI voice channel commands, the channel list specifies the MIDI
   channels for which the parameter applies.  If no channel list is
   provided, the parameter applies to all MIDI channels (0-15).  The
   channel list takes the form of a list of channel numbers (0 through
   15) and dash-separated channel number ranges (i.e., 0-5, 8-12, etc).
   Dots (i.e., "." characters) separate elements in the channel list.

   Recall that System commands do not have a MIDI channel associated
   with them.  Thus, for most command-type letters that code System
   commands (B, F, G, H, J, K, Q, V, Y, and Z), the channel list is
   ignored.

   For the command-type letter X, the appearance of certain numbers in
   the channel list codes special semantics.

     o  The digit 0 codes that SysEx "cancel" sublists (Section 3.2 in
        the main text) MUST NOT appear in the MIDI list.

     o  The digit 1 codes that cancel sublists MAY appear in the MIDI
        list (the default condition).

     o  The digit 2 codes that commands other than System Real-time MIDI
        commands MUST NOT appear between SysEx command segments in the
        MIDI list (the default condition).

     o  The digit 3 codes that any MIDI command type may appear between
        SysEx command segments in the MIDI list, with the exception of
        the segmented encoding of a second SysEx command (verbatim SysEx
        commands are OK).

   For command-type X, the channel list MUST NOT contain both digits 0
   and 1, and it MUST NOT contain both digits 2 and 3.  For command-type
   X, channel list numbers other than the numbers defined above are
   ignored.  If X does not have a channel list, the semantics marked
   "the default condition" in the list above apply.

   The syntax for field lists in a parameter assignment follows the
   syntax for channel lists.  If no field list is provided, the
   parameter applies to all controller or note numbers.

   For command-type C (Control Change), the field list codes the
   controller numbers (0-255) for which the parameter applies.

   For command-type M (Parameter System), the field list codes the
   Registered Parameter Numbers (RPNs) and Non-Registered Parameter
   Numbers (NRPNs) for which the parameter applies.  The number range
   0-16383 specifies RPNs, the number range 16384-32767 specifies NRPNs
   (16384 corresponds to NRPN 0, 32767 corresponds to NRPN 16383).

   For command-types N (NoteOn and NoteOff) and A (Poly Aftertouch), the
   field list codes the note numbers for which the parameter applies.

   For command-types J and K (System Common Undefined), the field list
   consists of a single digit, which specifies the number of data octets
   that follow the command octet.

   For command-type X (SysEx), the field list codes the number of data
   octets that may appear in a SysEx command.  Thus, the field list
   0-255 specifies SysEx commands with 255 or fewer data octets, the
   field list 256-4294967295 specifies SysEx commands with more than 255
   data octets but excludes commands with 255 or fewer data octets, and
   the field list 0 excludes all commands.

   A secondary parameter assignment syntax customizes command-type X
   (see Appendix D for complete ABNF):

     <parameter> = "__" <h-list> ["_" <h-list>] "__"

   The assignment defines the class of SysEx commands that obeys the
   semantics of the assigned parameter.  The command class is specified
   by listing the permitted values of the first N data octets that
   follow the SysEx 0xF0 command octet.  Any SysEx command whose first N
   data octets match the list is a member of the class.

   Each <h-list> defines a data octet of the command, as a dot-separated
   (".") list of one or more hexadecimal constants (such as "7F") or
   dash-separated hexadecimal ranges (such as "01-1F").  Underscores
   ("_") separate each <h-list>.  Double-underscores ("__") delineate
   the data octet list.

   Using this syntax, each assignment specifies a single SysEx command
   class.  Session descriptions may use several assignments to cm_used
   and cm_unused to specify complex behaviors.

   The example session description below illustrates the use of the
   stream subsetting parameters:

   v=0
   o=lazzaro 2520644554 2838152170 IN IP6 first.example.net
   s=Example
   t=0 0
   m=audio 5004 RTP/AVP 96
   c=IN IP6 2001:DB80::7F2E:172A:1E24
   a=rtpmap:96 rtp-midi/44100
   a=fmtp:96 cm_unused=ACGHJKNMPTVWXYZ; cm_used=__7F_00-7F_01_01__

   The session description configures the stream for use in clock
   applications.  All voice channels are unused, as are all System
   Commands except those used for MIDI Time Code (command-type F, and
   the Full Frame SysEx command that is matched by the string assigned
   to cm_used), the System Sequencer commands (command-type Q), and
   System Reset (command-type B).

C.2.  Configuration Tools: The Journalling System

   In this appendix, we define the payload format parameters that
   configure stream journalling and the recovery journal system.

   The j_sec parameter (Appendix C.2.1) sets the journalling method for
   the stream.  The j_update parameter (Appendix C.2.2) sets the
   recovery journal sending policy for the stream.  Appendix C.2.2 also
   defines the sending policies of the recovery journal system.

   Appendix C.2.3 defines several parameters that modify the recovery
   journal semantics.  These parameters change the default recovery
   journal semantics as defined in Section 5 and Appendices A-B.

   The journalling method for a stream is set at the start of a session
   and MUST NOT be changed thereafter.  This requirement forbids changes
   to the j_sec parameter once a session has begun.

   A related requirement, defined in the appendix sections below,
   forbids the acceptance of parameter values that would violate the
   recovery journal mandate.  In many cases, a change in one of the
   parameters defined in this appendix during an ongoing session would
   result in a violation of the recovery journal mandate for an
   implementation; in this case, the parameter change MUST NOT be
   accepted.

C.2.1.  The j_sec Parameter

   Section 2.2 defines the default journalling method for a stream.
   Streams that use unreliable transport (such as UDP) default to using
   the recovery journal.  Streams that use reliable transport (such as
   TCP) default to not using a journal.

   The parameter j_sec may be used to override this default.  This memo
   defines two symbolic values for j_sec: "none", to indicate that all
   stream payloads MUST NOT contain a journal section, and "recj", to
   indicate that all stream payloads MUST contain a journal section that
   uses the recovery journal format.

   For example, the j_sec parameter might be set to "none" for a UDP
   stream that travels between two hosts on a local network that is
   known to provide reliable datagram delivery.

   The session description below configures a UDP stream that does not
   use the recovery journal:

   v=0
   o=lazzaro 2520644554 2838152170 IN IP4 first.example.net
   s=Example
   t=0 0
   m=audio 5004 RTP/AVP 96
   c=IN IP4 192.0.2.94
   a=rtpmap:96 rtp-midi/44100
   a=fmtp:96 j_sec=none

   Other IETF standards-track documents may define alternative journal
   formats.  These documents MUST define new symbolic values for the
   j_sec parameter to signal the use of the format.

   Parties MUST NOT accept a j_sec value that violates the recovery
   journal mandate (see Section 4 for details).  If a session
   description uses a j_sec value unknown to the recipient, the
   recipient MUST NOT accept the description.

   Special j_sec issues arise when sessions are managed by session
   management tools (like RTSP, [RFC2326]) that use SDP for "declarative
   usage" purposes (see the preamble of Section 6 for details).  For
   these session management tools, SDP does not code transport details
   (such as UDP or TCP) for the session.  Instead, server and client
   negotiate transport details via other means (for RTSP, the SETUP
   method).

   In this scenario, the use of the j_sec parameter may be ill-advised,
   as the creator of the session description may not yet know the
   transport type for the session.  In this case, the session
   description SHOULD configure the journalling system using the
   parameters defined in the remainder of Appendix C.2, but it SHOULD
   NOT use j_sec to set the journalling status.  Recall that if j_sec
   does not appear in the session description, the default method for
   choosing the journalling method is in effect (no journal for reliable
   transport, recovery journal for unreliable transport).

   However, in declarative usage situations where the creator of the
   session description knows that journalling is always required or
   never required, the session description SHOULD use the j_sec
   parameter.

C.2.2.  The j_update Parameter

   In Section 4, we use the term "sending policy" to describe the method
   a sender uses to choose the checkpoint packet identity for each
   recovery journal in a stream.  In the sub-sections that follow, we
   normatively define three sending policies: anchor, closed-loop, and
   open-loop.

   As stated in Section 4, the default sending policy for a stream is
   the closed-loop policy.  The j_update parameter may be used to
   override this default.

   We define three symbolic values for j_update: "anchor", to indicate
   that the stream uses the anchor sending policy, "open-loop", to
   indicate that the stream uses the open-loop sending policy, and
   "closed-loop", to indicate that the stream uses the closed-loop
   sending policy.  See Appendix C.2.3 for examples session descriptions
   that use the j_update parameter.

   Parties MUST NOT accept a j_update value that violates the recovery
   journal mandate (Section 4).

   Other IETF standards-track documents may define additional sending
   policies for the recovery journal system.  These documents MUST
   define new symbolic values for the j_update parameter to signal the

   use of the new policy.  If a session description uses a j_update
   value unknown to the recipient, the recipient MUST NOT accept the
   description.

C.2.2.1.  The anchor Sending Policy

   In the anchor policy, the sender uses the first packet in the stream
   as the checkpoint packet for all packets in the stream.  The anchor
   policy satisfies the recovery journal mandate (Section 4), as the
   checkpoint history always covers the entire stream.

   The anchor policy does not require the use of the RTP control
   protocol (RTCP, [RFC3550]) or other feedback from receiver to sender.
   Senders do not need to take special actions to ensure that received
   streams start up free of artifacts, as the recovery journal always
   covers the entire history of the stream.  Receivers are relieved of
   the responsibility of tracking the changing identity of the
   checkpoint packet, because the checkpoint packet never changes.

   The main drawback of the anchor policy is bandwidth efficiency.
   Because the checkpoint history covers the entire stream, the size of
   the recovery journals produced by this policy usually exceeds the
   journal size of alternative policies.  For single-channel MIDI data
   streams, the bandwidth overhead of the anchor policy is often
   acceptable (see Appendix A.4 of [NMP]).  For dense streams, the
   closed-loop or open-loop policies may be more appropriate.

C.2.2.2.  The closed-loop Sending Policy

   The closed-loop policy is the default policy of the recovery journal
   system.  For each packet in the stream, the policy lets senders
   choose the smallest possible checkpoint history that satisfies the
   recovery journal mandate.  As smaller checkpoint histories generally
   yield smaller recovery journals, the closed-loop policy reduces the
   bandwidth of a stream, relative to the anchor policy.

   The closed-loop policy relies on feedback from receiver to sender.
   The policy assumes that a receiver periodically informs the sender of
   the highest sequence number it has seen so far in the stream, coded
   in the 32-bit extension format defined in [RFC3550].  For RTCP,
   receivers transmit this information in the Extended Highest Sequence
   Number Received (EHSNR) field of Receiver Reports.  RTCP Sender or
   Receiver Reports MUST be sent by any participant in a session with
   closed loop sending policy, unless another feedback mechanism has
   been agreed upon.

   The sender may safely use receiver sequence number feedback to guide
   checkpoint history management, because Section 4 requires that
   receivers repair indefinite artifacts whenever a packet loss event
   occur.

   We now normatively define the closed-loop policy.  At the moment a
   sender prepares an RTP packet for transmission, the sender is aware
   of R >= 0 receivers for the stream.  Senders may become aware of a
   receiver via RTCP traffic from the receiver, via RTP packets from a
   paired stream sent by the receiver to the sender, via messages from a
   session management tool, or by other means.  As receivers join and
   leave a session, the value of R changes.

   Each known receiver k (1 <= k <= R) is associated with a 32-bit
   extended packet sequence number M(k), where the extension reflects
   the sequence number rollover count of the sender.

   If the sender has received at least one feedback report from receiver
   k, M(k) is the most recent report of the highest RTP packet sequence
   number seen by the receiver, normalized to reflect the rollover count
   of the sender.

   If the sender has not received a feedback report from the receiver,
   M(k) is the extended sequence number of the last packet the sender
   transmitted before it became aware of the receiver.  If the sender
   became aware of this receiver before it sent the first packet in the
   stream, M(k) is the extended sequence number of the first packet in
   the stream.

   Given this definition of M(), we now state the closed-loop policy.
   When preparing a new packet for transmission, a sender MUST choose a
   checkpoint packet with extended sequence number N, such that M(k) >=
   (N - 1) for all k, 1 <= k <= R, where R >= 1.  The policy does not
   restrict sender behavior in the R == 0 (no known receivers) case.

   Under the closed-loop policy as defined above, a sender may transmit
   packets whose checkpoint history is shorter than the session history
   (as defined in Appendix A.1).  In this event, a new receiver that
   joins the stream may experience indefinite artifacts.

   For example, if a Control Change (0xB) command for Channel Volume
   (controller number 7) was sent early in a stream, and later a new
   receiver joins the session, the closed-loop policy may permit all
   packets sent to the new receiver to use a checkpoint history that
   does not include the Channel Volume Control Change command.  As a
   result, the new receiver experiences an indefinite artifact, and
   plays all notes on a channel too loudly or too softly.

   To address this issue, the closed-loop policy states that whenever a
   sender becomes aware of a new receiver, the sender MUST determine if
   the receiver would be subject to indefinite artifacts under the
   closed-loop policy.  If so, the sender MUST ensure that the receiver
   starts the session free of indefinite artifacts.

   For example, to solve the Channel Volume issue described above, the
   sender may code the current state of the Channel Volume controller
   numbers in the recovery journal Chapter C, until it receives the
   first RTCP RR report that signals that a packet containing this
   Chapter C has been received.

   In satisfying this requirement, senders MAY infer the initial MIDI
   state of the receiver from the session description.  For example, the
   stream example in Section 6.2 has the initial state defined in [MIDI]
   for General MIDI.

   In a unicast RTP session, a receiver may safely assume that the
   sender is aware of its presence of a receiver from the first packet
   sent in the RTP stream.  However, in other types of RTP sessions
   (multicast, conference focus, RTP translator/mixer), a receiver is
   often not able to determine if the sender is initially aware of its
   presence as a receiver.

   To address this issue, the closed-loop policy states that if a
   receiver participates in a session where it may have access to a
   stream whose sender is not aware of the receiver, the receiver MUST
   take actions to ensure that its rendered MIDI performance does not
   contain indefinite artifacts.  These protections will be necessarily
   incomplete.  For example, a receiver may monitor the Checkpoint
   Packet Seqnum for uncovered loss events, and "err on the side of
   caution" with respect to handling stuck notes due to lost MIDI
   NoteOff commands, but the receiver is not able to compensate for the
   lack of Channel Volume initialization data in the recovery journal.

   The receiver MUST NOT discontinue these protective actions until it
   is certain that the sender is aware of its presence.  If a receiver
   is not able to ascertain sender awareness, the receiver MUST continue
   these protective actions for the duration of the session.

   Note that in a multicast session where all parties are expected to
   send and receive, the reception of RTCP receiver reports from the
   sender about the RTP stream a receiver is multicasting is evidence of
   the sender's awareness that the RTP stream multicast by the sender is
   being monitored by the receiver.  Receivers may also obtain sender
   awareness evidence from session management tools, or by other means.
   In practice, ongoing observation of the Checkpoint Packet Seqnum to
   determine if the sender is taking actions to prevent loss events for

   a receiver is a good indication of sender awareness, as is the sudden
   appearance of recovery journal chapters with numerous Control Change
   controller data that was not foreshadowed by recent commands coded in
   the MIDI list shortly after sending an RTCP RR.

   The final set of normative closed-loop policy requirements concern
   how senders and receivers handle unplanned disruptions of RTCP
   feedback from a receiver to a sender.  By "unplanned", we refer to
   disruptions that are not due to the signalled termination of an RTP
   stream, via an RTCP BYE or via session management tools.

   As defined earlier in this section, the closed-loop policy states
   that a sender MUST choose a checkpoint packet with extended sequence
   number N, such that M(k) >= (N - 1) for all k, 1 <= k <= R, where R
   >= 1.  If the sender has received at least one feedback report from
   receiver k, M(k) is the most recent report of the highest RTP packet
   sequence number seen by the receiver, normalized to reflect the
   rollover count of the sender.

   If this receiver k stops sending feedback to the sender, the M(k)
   value used by the sender reflects the last feedback report from the
   receiver.  As time progresses without feedback from receiver k, this
   fixed M(k) value forces the sender to increase the size of the
   checkpoint history, and thus increases the bandwidth of the stream.

   At some point, the sender may need to take action in order to limit
   the bandwidth of the stream.  In most envisioned uses of RTP MIDI,
   long before this point is reached, the SSRC time-out mechanism
   defined in [RFC3550] will remove the uncooperative receiver from the
   session (note that the closed-loop policy does not suggest or require
   any special sender behavior upon an SSRC time-out, other than the
   sender actions related to changing R, described earlier in this
   section).

   However, in rare situations, the bandwidth of the stream (due to a
   lack of feedback reports from the sender) may become too large to
   continue sending the stream to the receiver before the SSRC time-out
   occurs for the receiver.  In this case, the closed-loop policy states
   that the sender should invoke the SSRC time-out for the receiver
   early.

   We now discuss receiver responsibilities in the case of unplanned
   disruptions of RTCP feedback from receiver to sender.

   In the unicast case, if a sender invokes the SSRC time-out mechanism
   for a receiver, the receiver stops receiving packets from the sender.
   The sender behavior imposed by the guardtime parameter (Appendix

   C.4.2) lets the receiver conclude that an SSRC time-out has occurred
   in a reasonable time period.

   In this case of a time-out, a receiver MUST keep sending RTCP
   feedback, in order to re-establish the RTP flow from the sender.
   Unless the receiver expects a prompt recovery of the RTP flow, the
   receiver MUST take actions to ensure that the rendered MIDI
   performance does not exhibit "very long transient artifacts" (for
   example, by silencing NoteOns to prevent stuck notes) while awaiting
   reconnection of the flow.

   In the multicast case, if a sender invokes the SSRC time-out
   mechanism for a receiver, the receiver may continue to receive
   packets, but the sender will no longer be using the M(k) feedback
   from the receiver to choose each checkpoint packet.  If the receiver
   does not have additional information that precludes an SSRC time-out
   (such as RTCP Receiver Reports from the sender about an RTP stream
   the receiver is multicasting back to the sender), the receiver MUST
   monitor the Checkpoint Packet Seqnum to detect an SSRC time-out.  If
   an SSRC time-out is detected, the receiver MUST follow the
   instructions for SSRC time-outs described for the unicast case above.

   Finally, we note that the closed-loop policy is suitable for use in
   RTP/RTCP sessions that use multicast transport.  However, aspects of
   the closed-loop policy do not scale well to sessions with large
   numbers of participants.  The sender state scales linearly with the
   number of receivers, as the sender needs to track the identity and
   M(k) value for each receiver k.  The average recovery journal size is
   not independent of the number of receivers, as the RTCP reporting
   interval backoff slows down the rate of a full update of M(k) values.
   The backoff algorithm may also increase the amount of ancillary state
   used by implementations of the normative sender and receiver
   behaviors defined in Section 4.

C.2.2.3.  The open-loop Sending Policy

   The open-loop policy is suitable for sessions that are not able to
   implement the receiver-to-sender feedback required by the closed-loop
   policy, and that are also not able to use the anchor policy because
   of bandwidth constraints.

   The open-loop policy does not place constraints on how a sender
   chooses the checkpoint packet for each packet in the stream.  In the
   absence of such constraints, a receiver may find that the recovery
   journal in the packet that ends a loss event has a checkpoint history
   that does not cover the entire loss event.  We refer to loss events
   of this type as uncovered loss events.

   To ensure that uncovered loss events do not compromise the recovery
   journal mandate, the open-loop policy assigns specific recovery tasks
   to senders, receivers, and the creators of session descriptions.  The
   underlying premise of the open-loop policy is that the indefinite
   artifacts produced during uncovered loss events fall into two
   classes.

   One class of artifacts is recoverable indefinite artifacts.
   Receivers are able to repair recoverable artifacts that occur during
   an uncovered loss event without intervention from the sender, at the
   potential cost of unpleasant transient artifacts.

   For example, after an uncovered loss event, receivers are able to
   repair indefinite artifacts due to NoteOff (0x8) commands that may
   have occurred during the loss event, by executing NoteOff commands
   for all active NoteOns commands.  This action causes a transient
   artifact (a sudden silent period in the performance), but ensures
   that no stuck notes sound indefinitely.  We refer to MIDI commands
   that are amenable to repair in this fashion as recoverable MIDI
   commands.

   A second class of artifacts is unrecoverable indefinite artifacts.
   If this class of artifact occurs during an uncovered loss event, the
   receiver is not able to repair the stream.

   For example, after an uncovered loss event, receivers are not able to
   repair indefinite artifacts due to Control Change (0xB) Channel
   Volume (controller number 7) commands that have occurred during the
   loss event.  A repair is impossible because the receiver has no way
   of determining the data value of a lost Channel Volume command.  We
   refer to MIDI commands that are fragile in this way as unrecoverable
   MIDI commands.

   The open-loop policy does not specify how to partition the MIDI
   command set into recoverable and unrecoverable commands.  Instead, it
   assumes that the creators of the session descriptions are able to
   come to agreement on a suitable recoverable/unrecoverable MIDI
   command partition for an application.

   Given these definitions, we now state the normative requirements for
   the open-loop policy.

   In the open-loop policy, the creators of the session description MUST
   use the ch_anchor parameter (defined in Appendix C.2.3) to protect
   all unrecoverable MIDI command types from indefinite artifacts, or
   alternatively MUST use the cm_unused parameter (defined in Appendix

   C.1) to exclude the command types from the stream.  These options act
   to shield command types from artifacts during an uncovered loss
   event.

   In the open-loop policy, receivers MUST examine the Checkpoint Packet
   Seqnum field of the recovery journal header after every loss event,
   to check if the loss event is an uncovered loss event.  Section 5
   shows how to perform this check.  If an uncovered loss event has
   occurred, a receiver MUST perform indefinite artifact recovery for
   all MIDI command types that are not shielded by ch_anchor and
   cm_unused parameter assignments in the session description.

   The open-loop policy does not place specific constraints on the
   sender.  However, the open-loop policy works best if the sender
   manages the size of the checkpoint history to ensure that uncovered
   losses occur infrequently, by taking into account the delay and loss
   characteristics of the network.  Also, as each checkpoint packet
   change incurs the risk of an uncovered loss, senders should only move
   the checkpoint if it reduces the size of the journal.

C.2.3.  Recovery Journal Chapter Inclusion Parameters

   The recovery journal chapter definitions (Appendices A-B) specify
   under what conditions a chapter MUST appear in the recovery journal.
   In most cases, the definition states that if a certain command
   appears in the checkpoint history, a certain chapter type MUST appear
   in the recovery journal to protect the command.

   In this section, we describe the chapter inclusion parameters.  These
   parameters modify the conditions under which a chapter appears the
   journal.  These parameters are essential to the use of the open-loop
   policy (Appendix C.2.2.3) and may also be used to simplify
   implementations of the closed-loop (Appendix C.2.2.2) and anchor
   (Appendix C.2.2.1) policies.

   Each parameter represents a type of chapter inclusion semantics.  An
   assignment to a parameter declares which chapters (or chapter
   subsets) obey the inclusion semantics.  We describe the assignment
   syntax for these parameters later in this section.

   A party MUST NOT accept chapter inclusion parameter values that
   violate the recovery journal mandate (Section 4).  All assignments of
   the subsetting parameters (cm_used and cm_unused) MUST precede the
   first assignment of a chapter inclusion parameter in the parameter
   list.

   Below, we normatively define the semantics of the chapter inclusion
   parameters.  For clarity, we define the action of parameters on
   complete chapters.  If a parameter is assigned a subset of a chapter,
   the definition applies only to the chapter subset.

     o  ch_never.  A chapter assigned to the ch_never parameter MUST NOT
        appear in the recovery journal (Appendix A.4.1-2 defines
        exceptions to this rule for Chapter M).  To signal the exclusion
        of a chapter from the journal, an assignment to ch_never MUST be
        made, even if the commands coded by the chapter are assigned to
        cm_unused.  This rule simplifies the handling of commands types
        that may be coded in several chapters.

     o  ch_default.  A chapter assigned to the ch_default parameter MUST
        follow the default semantics for the chapter, as defined in
        Appendices A-B.

     o  ch_anchor.  A chapter assigned to the ch_anchor MUST obey a
        modified version of the default chapter semantics.  In the
        modified semantics, all references to the checkpoint history are
        replaced with references to the session history, and all
        references to the checkpoint packet are replaced with references
        to the first packet sent in the stream.

   Parameter assignments obey the following syntax (see Appendix D for
   ABNF):

     <parameter> = [channel list]<chapter list>[field list]

   The chapter list is mandatory; the channel and field lists are
   optional.  Multiple assignments to parameters have a cumulative
   effect and are applied in the order of parameter appearance in a
   media description.

   To determine the semantics of a list of chapter inclusion parameter
   assignments, we begin by assuming an implicit assignment of all
   channel and system chapters to the ch_default parameter, with the
   default values for the channel list and field list for each chapter
   that are defined below.

   We then interpret the semantics of the actual parameter assignments,
   using the rules below.

   A later assignment of a chapter to the same parameter expands the
   scope of the earlier assignment.  In most cases, a later assignment
   of a chapter to a different parameter cancels (partially or
   completely) the effect of an earlier assignment.

   The chapter list specifies the channel or system chapters for which
   the parameter applies.  The chapter list is a concatenated sequence
   of one or more of the letters corresponding to the chapter types
   (ACDEFMNPQTVWX).  In addition, the list may contain one or more of
   the letters for the sub-chapter types (BGHJKYZ) of System Chapter D.

   The letters in a chapter list MUST be uppercase and MUST appear in
   alphabetical order.  Letters other than (ABCDEFGHJKMNPQTVWXYZ) that
   appear in the chapter list MUST be ignored.

   The channel list specifies the channel journals for which this
   parameter applies; if no channel list is provided, the parameter
   applies to all channel journals.  The channel list takes the form of
   a list of channel numbers (0 through 15) and dash-separated channel
   number ranges (i.e., 0-5, 8-12, etc.).  Dots (i.e., "." characters)
   separate elements in the channel list.

   Several of the systems chapters may be configured to have special
   semantics.  Configuration occurs by specifying a channel list for the
   systems channel, using the coding described below (note that MIDI
   Systems commands do not have a "channel", and thus the original
   purpose of the channel list does not apply to systems chapters).  The
   expression "the digit N" in the text below refers to the inclusion of
   N as a "channel" in the channel list for a systems chapter.

   For the J and K Chapter D sub-chapters (undefined System Common), the
   digit 0 codes that the parameter applies to the LEGAL field of the
   associated command log (Figure B.1.4 of Appendix B.1), the digit 1
   codes that the parameter applies to the VALUE field of the command
   log, and the digit 2 codes that the parameter applies to the COUNT
   field of the command log.

   For the Y and Z Chapter D sub-chapters (undefined System Real-time),
   the digit 0 codes that the parameter applies to the LEGAL field of
   the associated command log (Figure B.1.5 of Appendix B.1) and the
   digit 1 codes that the parameter applies to the COUNT field of the
   command log.

   For Chapter Q (Sequencer State Commands), the digit 0 codes that the
   parameter applies to the default Chapter Q definition, which forbids
   the TIME field.  The digit 1 codes that the parameter applies to the
   optional Chapter Q definition, which supports the TIME field.

   The syntax for field lists follows the syntax for channel lists.  If
   no field list is provided, the parameter applies to all controller or
   note numbers.  For Chapter C, if no field list is provided, the
   controller numbers do not use enhanced Chapter C encoding (Appendix
   A.3.3).

   For Chapter C, the field list may take on values in the range 0 to
   255.  A field value X in the range 0-127 refers to a controller
   number X, and indicates that the controller number does not use
   enhanced Chapter C encoding.  A field value X in the range 128-255
   refers to a controller number "X minus 128" and indicates the
   controller number does use the enhanced Chapter C encoding.

   Assignments made to configure the Chapter C encoding method for a
   controller number MUST be made to the ch_default or ch_anchor
   parameters, as assignments to ch_never act to exclude the number from
   the recovery journal (and thus the indicated encoding method is
   irrelevant).

   A Chapter C field list MUST NOT encode conflicting information about
   the enhanced encoding status of a particular controller number.  For
   example, values 0 and 128 MUST NOT both be coded by a field list.

   For Chapter M, the field list codes the Registered Parameter Numbers
   (RPNs) and Non-Registered Parameter Numbers (NRPNs) for which the
   parameter applies.  The number range 0-16383 specifies RPNs, the
   number range 16384-32767 specifies NRPNs (16384 corresponds to NRPN
   0, 32767 corresponds to NRPN 16383).

   For Chapters N and A, the field list codes the note numbers for which
   the parameter applies.  The note number range specified for Chapter N
   also applies to Chapter E.

   For Chapter E, the digit 0 codes that the parameter applies to
   Chapter E note logs whose V bit is set to 0, and the digit 1 codes
   that the parameter applies to note logs whose V bit is set to 1.

   For Chapter X, the field list codes the number of data octets that
   may appear in a SysEx command that is coded in the chapter.  Thus,
   the field list 0-255 specifies SysEx commands with 255 or fewer data
   octets, the field list 256-4294967295 specifies SysEx commands with
   more than 255 data octets but excludes commands with 255 or fewer
   data octets, and the field list 0 excludes all commands.

   A secondary parameter assignment syntax customizes Chapter X (see
   Appendix D for complete ABNF):

     <parameter> = "__" <h-list> ["_" <h-list>] "__"

   The assignment defines a class of SysEx commands whose Chapter X
   coding obeys the semantics of the assigned parameter.  The command
   class is specified by listing the permitted values of the first N

   data octets that follow the SysEx 0xF0 command octet.  Any SysEx
   command whose first N data octets match the list is a member of the
   class.

   Each <h-list> defines a data octet of the command, as a dot-separated
   (".") list of one or more hexadecimal constants (such as "7F") or
   dash-separated hexadecimal ranges (such as "01-1F").  Underscores
   ("_") separate each <h-list>.  Double-underscores ("__") delineate
   the data octet list.

   Using this syntax, each assignment specifies a single SysEx command
   class.  Session descriptions may use several assignments to the same
   (or different) parameters to specify complex Chapter X behaviors.
   The ordering behavior of multiple assignments follows the guidelines
   for chapter parameter assignments described earlier in this section.

   The example session description below illustrates the use of the
   chapter inclusion parameters:

   v=0
   o=lazzaro 2520644554 2838152170 IN IP6 first.example.net
   s=Example
   t=0 0
   m=audio 5004 RTP/AVP 96
   c=IN IP6 2001:DB80::7F2E:172A:1E24
   a=rtpmap:96 rtp-midi/44100
   a=fmtp:96 j_update=open-loop; cm_unused=ABCFGHJKMQTVWXYZ;
   cm_used=__7E_00-7F_09_01.02.03__;
   cm_used=__7F_00-7F_04_01.02__; cm_used=C7.64;
   ch_never=ABCDEFGHJKMQTVWXYZ; ch_never=4.11-13N;
   ch_anchor=P; ch_anchor=C7.64;
   ch_anchor=__7E_00-7F_09_01.02.03__;
   ch_anchor=__7F_00-7F_04_01.02__

   (The a=fmtp line has been wrapped to fit the page to accommodate
    memo formatting restrictions; it comprises a single line in SDP.)

   The j_update parameter codes that the stream uses the open-loop
   policy.  Most MIDI command-types are assigned to cm_unused and thus
   do not appear in the stream.  As a consequence, the assignments to
   the first ch_never parameter reflect that most chapters are not in
   use.

   Chapter N for several MIDI channels is assigned to ch_never.  Chapter
   N for MIDI channels other than 4, 11, 12, and 13 may appear in the
   recovery journal, using the (default) ch_default semantics.  In
   practice, this assignment pattern would reflect knowledge about a
   resilient rendering method in use for the excluded channels.

   The MIDI Program Change command and several MIDI Control Change
   controller numbers are assigned to ch_anchor.  Note that the ordering
   of the ch_anchor chapter C assignment after the ch_never command acts
   to override the ch_never assignment for the listed controller numbers
   (7 and 64).

   The assignment of command-type X to cm_unused excludes most SysEx
   commands from the stream.  Exceptions are made for General MIDI
   System On/Off commands and for the Master Volume and Balance
   commands, via the use of the secondary assignment syntax.  The
   cm_used assignment codes the exception, and the ch_anchor assignment
   codes how these commands are protected in Chapter X.

C.3.  Configuration Tools: Timestamp Semantics

   The MIDI command section of the payload format consists of a list of
   commands, each with an associated timestamp.  The semantics of
   command timestamps may be set during session configuration, using the
   parameters we describe in this section

   The parameter "tsmode" specifies the timestamp semantics for a
   stream.  The parameter takes on one of three token values: "comex",
   "async", or "buffer".

   The default "comex" value specifies that timestamps code the
   execution time for a command (Appendix C.3.1) and supports the
   accurate transcoding Standard MIDI Files (SMFs, [MIDI]).  The "comex"
   value is also RECOMMENDED for new MIDI user-interface controller
   designs.  The "async" value specifies an asynchronous timestamp
   sampling algorithm for time-of-arrival sources (Appendix C.3.2).  The
   "buffer" value specifies a synchronous timestamp sampling algorithm
   (Appendix C.3.3) for time-of-arrival sources.

   Ancillary parameters MAY follow tsmode in a media description.  We
   define these parameters in Appendices C.3.2-3 below.

C.3.1.  The comex Algorithm

   The default "comex" (COMmand EXecution) tsmode value specifies the
   execution time for the command.  With comex, the difference between
   two timestamps indicates the time delay between the execution of the
   commands.  This difference may be zero, coding simultaneous
   execution.

   The comex interpretation of timestamps works well for transcoding a
   Standard MIDI File (SMF, [MIDI]) into an RTP MIDI stream, as SMFs
   code a timestamp for each MIDI command stored in the file.  To
   transcode an SMF that uses metric time markers, use the SMF tempo map

   (encoded in the SMF as meta-events) to convert metric SMF timestamp
   units into seconds-based RTP timestamp units.

   New MIDI controller designs (piano keyboard, drum pads, etc.) that
   support RTP MIDI and that have direct access to sensor data SHOULD
   use comex interpretation for timestamps, so that simultaneous
   gestural events may be accurately coded by RTP MIDI.

   Comex is a poor choice for transcoding MIDI 1.0 DIN cables [MIDI],
   for a reason that we will now explain.  A MIDI DIN cable is an
   asynchronous serial protocol (320 microseconds per MIDI byte).  MIDI
   commands on a DIN cable are not tagged with timestamps.  Instead,
   MIDI DIN receivers infer command timing from the time of arrival of
   the bytes.  Thus, two two-byte MIDI commands that occur at a source
   simultaneously are encoded on a MIDI 1.0 DIN cable with a 640
   microsecond time offset.  A MIDI DIN receiver is unable to tell if
   this time offset existed in the source performance or is an artifact
   of the serial speed of the cable.  However, the RTP MIDI comex
   interpretation of timestamps declares that a timestamp offset between
   two commands reflects the timing of the source performance.

   This semantic mismatch is the reason that comex is a poor choice for
   transcoding MIDI DIN cables.  Note that the choice of the RTP
   timestamp rate (Section 6.1-2 in the main text) cannot fix this
   inaccuracy issue.  In the sections that follow, we describe two
   alternative timestamp interpretations ("async" and "buffer") that are
   a better match to MIDI 1.0 DIN cable timing, and to other MIDI time-
   of-arrival sources.

   The "octpos", "linerate", and "mperiod" ancillary parameters (defined
   below) SHOULD NOT be used with comex.

C.3.2.  The async Algorithm

   The "async" tsmode value specifies the asynchronous sampling of a
   MIDI time-of-arrival source.  In asynchronous sampling, the moment an
   octet is received from a source, it is labelled with a wall-clock
   time value.  The time value has RTP timestamp units.

   The "octpos" ancillary parameter defines how RTP command timestamps
   are derived from octet time values.  If octpos has the token value
   "first", a timestamp codes the time value of the first octet of the
   command.  If octpos has the token value "last", a timestamp codes the
   time value of the last octet of the command.  If the octpos parameter
   does not appear in the media description, the sender does not know
   which octet of the command the timestamp references (for example, the
   sender may be relying on an operating system service that does not
   specify this information).

   The octpos semantics refer to the first or last octet of a command as
   it appears on a time-of-arrival MIDI source, not as it appears in an
   RTP MIDI packet.  This distinction is significant because the RTP
   coding may contain octets that are not present in the source.  For
   example, the status octet of the first MIDI command in a packet may
   have been added to the MIDI stream during transcoding, to comply with
   the RTP MIDI running status requirements (Section 3.2).

   The "linerate" ancillary parameter defines the timespan of one MIDI
   octet on the transmission medium of the MIDI source to be sampled
   (such as a MIDI 1.0 DIN cable).  The parameter has units of
   nanoseconds, and takes on integral values.  For MIDI 1.0 DIN cables,
   the correct linerate value is 320000 (this value is also the default
   value for the parameter).

   We now show a session description example for the async algorithm.
   Consider a sender that is transcoding a MIDI 1.0 DIN cable source
   into RTP.  The sender runs on a computing platform that assigns time
   values to every incoming octet of the source, and the sender uses the
   time values to label the first octet of each command in the RTP
   packet.  This session description describes the transcoding:

   v=0
   o=lazzaro 2520644554 2838152170 IN IP4 first.example.net
   s=Example
   t=0 0
   m=audio 5004 RTP/AVP 96
   c=IN IP4 192.0.2.94
   a=rtpmap:96 rtp-midi/44100
   a=sendonly
   a=fmtp:96 tsmode=async; linerate=320000; octpos=first

C.3.3.  The buffer Algorithm

   The "buffer" tsmode value specifies the synchronous sampling of a
   MIDI time-of-arrival source.

   In synchronous sampling, octets received from a source are placed in
   a holding buffer upon arrival.  At periodic intervals, the RTP sender
   examines the buffer.  The sender removes complete commands from the
   buffer and codes those commands in an RTP packet.  The command
   timestamp codes the moment of buffer examination, expressed in RTP
   timestamp units.  Note that several commands may have the same
   timestamp value.

   The "mperiod" ancillary parameter defines the nominal periodic
   sampling interval.  The parameter takes on positive integral values
   and has RTP timestamp units.

   The "octpos" ancillary parameter, defined in Appendix C.3.1 for
   asynchronous sampling, plays a different role in synchronous
   sampling.  In synchronous sampling, the parameter specifies the
   timestamp semantics of a command whose octets span several sampling
   periods.

   If octpos has the token value "first", the timestamp reflects the
   arrival period of the first octet of the command.  If octpos has the
   token value "last", the timestamp reflects the arrival period of the
   last octet of the command.  The octpos semantics refer to the first
   or last octet of the command as it appears on a time-of-arrival
   source, not as it appears in the RTP packet.

   If the octpos parameter does not appear in the media description, the
   timestamp MAY reflect the arrival period of any octet of the command;
   senders use this option to signal a lack of knowledge about the
   timing details of the buffering process at sub-command granularity.

   We now show a session description example for the buffer algorithm.
   Consider a sender that is transcoding a MIDI 1.0 DIN cable source
   into RTP.  The sender runs on a computing platform that places source
   data into a buffer upon receipt.  The sender polls the buffer 1000
   times a second, extracts all complete commands from the buffer, and
   places the commands in an RTP packet.  This session description
   describes the transcoding:

   v=0
   o=lazzaro 2520644554 2838152170 IN IP6 first.example.net
   s=Example
   t=0 0
   m=audio 5004 RTP/AVP 96
   c=IN IP6 2001:DB80::7F2E:172A:1E24
   a=rtpmap:96 rtp-midi/44100
   a=sendonly
   a=fmtp:96 tsmode=buffer; linerate=320000; octpos=last; mperiod=44

   The mperiod value of 44 is derived by dividing the clock rate
   specified by the rtpmap attribute (44100 Hz) by the 1000 Hz buffer
   sampling rate and rounding to the nearest integer.  Command
   timestamps might not increment by exact multiples of 44, as the
   actual sampling period might not precisely match the nominal mperiod
   value.

C.4.  Configuration Tools: Packet Timing Tools

   In this appendix, we describe session configuration tools for
   customizing the temporal behavior of MIDI stream packets.

C.4.1.  Packet Duration Tools

   Senders control the granularity of a stream by setting the temporal
   duration ("media time") of the packets in the stream.  Short media
   times (20 ms or less) often imply an interactive session.  Longer
   media times (100 ms or more) usually indicate a content streaming
   session.  The RTP AVP profile [RFC3551] recommends audio packet media
   times in a range from 0 to 200 ms.

   By default, an RTP receiver dynamically senses the media time of
   packets in a stream and chooses the length of its playout buffer to
   match the stream.  A receiver typically sizes its playout buffer to
   fit several audio packets and adjusts the buffer length to reflect
   the network jitter and the sender timing fidelity.

   Alternatively, the packet media time may be statically set during
   session configuration.  Session descriptions MAY use the RTP MIDI
   parameter "rtp_ptime" to set the recommended media time for a packet.
   Session descriptions MAY also use the RTP MIDI parameter
   "rtp_maxptime" to set the maximum media time for a packet permitted
   in a stream.  Both parameters MAY be used together to configure a
   stream.

   The values assigned to the rtp_ptime and rtp_maxptime parameters have
   the units of the RTP timestamp for the stream, as set by the rtpmap
   attribute (see Section 6.1).  Thus, if rtpmap sets the clock rate of
   a stream to 44100 Hz, a maximum packet media time of 10 ms is coded
   by setting rtp_maxptime=441.  As stated in the Appendix C preamble,
   the senders and receivers of a stream MUST agree on common values for
   rtp_ptime and rtp_maxptime if the parameters appear in the media
   description for the stream.

   0 ms is a reasonable media time value for MIDI packets and is often
   used in low-latency interactive applications.  In a packet with a 0
   ms media time, all commands execute at the instant they are coded by
   the packet timestamp.  The session description below configures all
   packets in the stream to have 0 ms media time:

   v=0
   o=lazzaro 2520644554 2838152170 IN IP4 first.example.net
   s=Example
   t=0 0
   m=audio 5004 RTP/AVP 96
   c=IN IP4 192.0.2.94
   a=rtpmap:96 rtp-midi/44100
   a=fmtp:96 rtp_ptime=0; rtp_maxptime=0

   The session attributes ptime and maxptime [RFC4566] MUST NOT be used
   to configure an RTP MIDI stream.  Sessions MUST use rtp_ptime in lieu
   of ptime and MUST use rtp_maxptime in lieu of maxptime.  RTP MIDI
   defines its own parameters for media time configuration because 0 ms
   values for ptime and maxptime are forbidden by [RFC3264] but are
   essential for certain applications of RTP MIDI.

   See the Appendix C.7 examples for additional discussion about using
   rtp_ptime and rtp_maxptime for session configuration.

C.4.2.  The guardtime Parameter

   RTP permits a sender to stop sending audio packets for an arbitrary
   period of time during a session.  When sending resumes, the RTP
   sequence number series continues unbroken, and the RTP timestamp
   value reflects the media time silence gap.

   This RTP feature has its roots in telephony, but it is also well
   matched to interactive MIDI sessions, as players may fall silent for
   several seconds during (or between) songs.

   Certain MIDI applications benefit from a slight enhancement to this
   RTP feature.  In interactive applications, receivers may use on-line
   network models to guide heuristics for handling lost and late RTP
   packets.  These models may work poorly if a sender ceases packet
   transmission for long periods of time.

   Session descriptions may use the parameter "guardtime" to set a
   minimum sending rate for a media session.  The value assigned to
   guardtime codes the maximum separation time between two sequential
   packets, as expressed in RTP timestamp units.

   Typical guardtime values are 500-2000 ms.  This value range is not a
   normative bound, and parties SHOULD be prepared to process values
   outside this range.

   The congestion control requirements for sender implementations
   (described in Section 8 and [RFC3550]) take precedence over the
   guardtime parameter.  Thus, if the guardtime parameter requests a
   minimum sending rate, but sending at this rate would violate the
   congestion control requirements, senders MUST ignore the guardtime
   parameter value.  In this case, senders SHOULD use the lowest minimum
   sending rate that satisfies the congestion control requirements.

   Below, we show a session description that uses the guardtime
   parameter.

   v=0
   o=lazzaro 2520644554 2838152170 IN IP6 first.example.net
   s=Example
   t=0 0
   m=audio 5004 RTP/AVP 96
   c=IN IP6 2001:DB80::7F2E:172A:1E24
   a=rtpmap:96 rtp-midi/44100
   a=fmtp:96 guardtime=44100; rtp_ptime=0; rtp_maxptime=0

C.5.  Configuration Tools: Stream Description

   As we discussed in Section 2.1, a party may send several RTP MIDI
   streams in the same RTP session, and several RTP sessions that carry
   MIDI may appear in a multimedia session.

   By default, the MIDI name space (16 channels + systems) of each RTP
   stream sent by a party in a multimedia session is independent.  By
   independent, we mean three distinct things:

     o  If a party sends two RTP MIDI streams (A and B), MIDI voice
        channel 0 in stream A is a different "channel 0" than MIDI voice
        channel 0 in stream B.

     o  MIDI voice channel 0 in stream B is not considered to be
        "channel 16" of a 32-channel MIDI voice channel space whose
        "channel 0" is channel 0 of stream A.

     o  Streams sent by different parties over different RTP sessions,
        or over the same RTP session but with different payload type
        numbers, do not share the association that is shared by a MIDI
        cable pair that cross-connects two devices in a MIDI 1.0 DIN
        network.  By default, this association is only held by streams
        sent by different parties in the same RTP session that use the
        same payload type number.

   In this appendix, we show how to express that specific RTP MIDI
   streams in a multimedia session are not independent but instead are
   related in one of the three ways defined above.  We use two tools to
   express these relations:

     o  The musicport parameter.  This parameter is assigned a non-
        negative integer value between 0 and 4294967295.  It appears in
        the fmtp lines of payload types.

     o  The FID grouping attribute [RFC3388] signals that several RTP
        sessions in a multimedia session are using the musicport
        parameter to express an inter-session relationship.

   If a multimedia session has several payload types whose musicport
   parameters are assigned the same integer value, streams using these
   payload types share an "identity relationship" (including streams
   that use the same payload type).  Streams in an identity relationship
   share two properties:

     o  Identity relationship streams sent by the same party target the
        same MIDI name space.  Thus, if streams A and B share an
        identity relationship, voice channel 0 in stream A is the same
        "channel 0" as voice channel 0 in stream B.

     o  Pairs of identity relationship streams that are sent by
        different parties share the association that is shared by a MIDI
        cable pair that cross-connects two devices in a MIDI 1.0 DIN
        network.

   A party MUST NOT send two RTP MIDI streams that share an identity
   relationship in the same RTP session.  Instead, each stream MUST be
   in a separate RTP session.  As explained in Section 2.1, this
   restriction is necessary to support the RTP MIDI method for the
   synchronization of streams that share a MIDI name space.

   If a multimedia session has several payload types whose musicport
   parameters are assigned sequential values (i.e., i, i+1, ... i+k),
   the streams using the payload types share an "ordered relationship".
   For example, if payload type A assigns 2 to musicport and payload
   type B assigns 3 to musicport, A and B are in an ordered
   relationship.

   Streams in an ordered relationship that are sent by the same party
   are considered by renderers to form a single larger MIDI space.  For
   example, if stream A has a musicport value of 2 and stream B has a
   musicport value of 3, MIDI voice channel 0 in stream B is considered
   to be voice channel 16 in the larger MIDI space formed by the
   relationship.  Note that it is possible for streams to participate in
   both an identity relationship and an ordered relationship.

   We now state several rules for using musicport:

     o  If streams from several RTP sessions in a multimedia session use
        the musicport parameter, the RTP sessions MUST be grouped using
        the FID grouping attribute defined in [RFC3388].

     o  An ordered or identity relationship MUST NOT contain both native
        RTP MIDI streams and mpeg4-generic RTP MIDI streams.  An
        exception applies if a relationship consists of sendonly and
        recvonly (but not sendrecv) streams.  In this case, the sendonly
        streams MUST NOT contain both types of streams, and the recvonly
        streams MUST NOT contain both types of streams.

     o  It is possible to construct identity relationships that violate
        the recovery journal mandate (for example, sending NoteOns for a
        voice channel on stream A and NoteOffs for the same voice
        channel on stream B).  Parties MUST NOT generate (or accept)
        session descriptions that exhibit this flaw.

     o  Other payload formats MAY define musicport media type
        parameters.  Formats would define these parameters so that their
        sessions could be bundled into RTP MIDI name spaces.  The
        parameter definitions MUST be compatible with the musicport
        semantics defined in this appendix.

   As a rule, at most one payload type in a relationship may specify a
   MIDI renderer.  An exception to the rule applies to relationships
   that contain sendonly and recvonly streams but no sendrecv streams.
   In this case, one sendonly session and one recvonly session may each
   define a renderer.

   Renderer specification in a relationship may be done using the tools
   described in Appendix C.6.  These tools work for both native streams
   and mpeg4-generic streams.  An mpeg4-generic stream that uses the
   Appendix C.6 tools MUST set all "config" parameters to the empty
   string ("").

   Alternatively, for mpeg4-generic streams, renderer specification may
   be done by setting one "config" parameter in the relationship to the
   renderer configuration string, and all other config parameters to the
   empty string ("").

   We now define sender and receiver rules that apply when a party sends
   several streams that target the same MIDI name space.

   Senders MAY use the subsetting parameters (Appendix C.1) to predefine
   the partitioning of commands between streams, or they MAY use a
   dynamic partitioning strategy.

   Receivers that merge identity relationship streams into a single MIDI
   command stream MUST maintain the structural integrity of the MIDI
   commands coded in each stream during the merging process, in the same
   way that software that merges traditional MIDI 1.0 DIN cable flows is

   responsible for creating a merged command flow compatible with
   [MIDI].

   Senders MUST partition the name space so that the rendered MIDI
   performance does not contain indefinite artifacts (as defined in
   Section 4).  This responsibility holds even if all streams are sent
   over reliable transport, as different stream latencies may yield
   indefinite artifacts.  For example, stuck notes may occur in a
   performance split over two TCP streams, if NoteOn commands are sent
   on one stream and NoteOff commands are sent on the other.

   Senders MUST NOT split a Registered Parameter Name (RPN) or Non-
   Registered Parameter Name (NRPN) transaction appearing on a MIDI
   channel across multiple identity relationship sessions.  Receivers
   MUST assume that the RPN/NRPN transactions that appear on different
   identity relationship sessions are independent and MUST preserve
   transactional integrity during the MIDI merge.

   A simple way to safely partition voice channel commands is to place
   all MIDI commands for a particular voice channel into the same
   session.  Safe partitioning of MIDI Systems commands may be more
   complicated for sessions that extensively use System Exclusive.

   We now show several session description examples that use the
   musicport parameter.

   Our first session description example shows two RTP MIDI streams that
   drive the same General MIDI decoder.  The sender partitions MIDI
   commands between the streams dynamically.  The musicport values
   indicate that the streams share an identity relationship.

   v=0
   o=lazzaro 2520644554 2838152170 IN IP4 first.example.net
   s=Example
   t=0 0
   a=group:FID 1 2
   c=IN IP4 192.0.2.94
   m=audio 5004 RTP/AVP 96
   a=rtpmap:96 mpeg4-generic/44100
   a=mid:1
   a=fmtp:96 streamtype=5; mode=rtp-midi; profile-level-id=12;
   config=7A0A0000001A4D546864000000060000000100604D54726B0
   000000600FF2F000; musicport=12
   m=audio 5006 RTP/AVP 96
   a=rtpmap:96 mpeg4-generic/44100
   a=mid:2
   a=fmtp:96 streamtype=5; mode=rtp-midi; config="";
   profile-level-id=12; musicport=12

   (The a=fmtp lines have been wrapped to fit the page to accommodate
    memo formatting restrictions; they comprise single lines in SDP.)

   Recall that Section 2.1 defines rules for streams that target the
   same MIDI name space.  Those rules, implemented in the example above,
   require that each stream resides in a separate RTP session, and that
   the grouping mechanisms defined in [RFC3388] signal an inter-session
   relationship.  The "group" and "mid" attribute lines implement this
   grouping mechanism.

   A variant on this example, whose session description is not shown,
   would use two streams in an identity relationship driving the same
   MIDI renderer, each with a different transport type.  One stream
   would use UDP and would be dedicated to real-time messages.  A second
   stream would use TCP [RFC4571] and would be used for SysEx bulk data
   messages.

   In the next example, two mpeg4-generic streams form an ordered
   relationship to drive a Structured Audio decoder with 32 MIDI voice
   channels.  Both streams reside in the same RTP session.

   v=0
   o=lazzaro 2520644554 2838152170 IN IP6 first.example.net
   s=Example
   t=0 0
   m=audio 5006 RTP/AVP 96 97
   c=IN IP6 2001:DB80::7F2E:172A:1E24
   a=rtpmap:96 mpeg4-generic/44100
   a=fmtp:96 streamtype=5; mode=rtp-midi; config="";
   profile-level-id=13; musicport=5
   a=rtpmap:97 mpeg4-generic/44100
   a=fmtp:97 streamtype=5; mode=rtp-midi; config="";
   profile-level-id=13; musicport=6; render=synthetic;
   rinit="audio/asc";
   url="http://example.com/cardinal.asc";
   cid="azsldkaslkdjqpwojdkmsldkfpe"

   (The a=fmtp lines have been wrapped to fit the page to accommodate
    memo formatting restrictions; they comprise single lines in SDP.)

   The sequential musicport values for the two sessions establish the
   ordered relationship.  The musicport=5 session maps to Structured
   Audio extended channels range 0-15, the musicport=6 session maps to
   Structured Audio extended channels range 16-31.

   Both config strings are empty.  The configuration data is specified
   by parameters that appear in the fmtp line of the second media
   description.  We define this configuration method in Appendix C.6.

   The next example shows two RTP MIDI streams (one recvonly, one
   sendonly) that form a "virtual sendrecv" session.  Each stream
   resides in a different RTP session (a requirement because sendonly
   and recvonly are RTP session attributes).

   v=0
   o=lazzaro 2520644554 2838152170 IN IP4 first.example.net
   s=Example
   t=0 0
   a=group:FID 1 2
   c=IN IP4 192.0.2.94
   m=audio 5004 RTP/AVP 96
   a=sendonly
   a=rtpmap:96 mpeg4-generic/44100
   a=mid:1
   a=fmtp:96 streamtype=5; mode=rtp-midi; profile-level-id=12;
   config=7A0A0000001A4D546864000000060000000100604D54726B0
   000000600FF2F000; musicport=12
   m=audio 5006 RTP/AVP 96
   a=recvonly
   a=rtpmap:96 mpeg4-generic/44100
   a=mid:2
   a=fmtp:96 streamtype=5; mode=rtp-midi; profile-level-id=12;
   config=7A0A0000001A4D546864000000060000000100604D54726B0
   000000600FF2F000; musicport=12

   (The a=fmtp lines have been wrapped to fit the page to accommodate
    memo formatting restrictions; they comprise single lines in SDP.)

   To signal the "virtual sendrecv" semantics, the two streams assign
   musicport to the same value (12).  As defined earlier in this
   section, pairs of identity relationship streams that are sent by
   different parties share the association that is shared by a MIDI
   cable pair that cross-connects two devices in a MIDI 1.0 network.  We
   use the term "virtual sendrecv" because streams sent by different
   parties in a true sendrecv session also have this property.

   As discussed in the preamble to Appendix C, the primary advantage of
   the virtual sendrecv configuration is that each party can customize
   the property of the stream it receives.  In the example above, each
   stream defines its own "config" string that could customize the
   rendering algorithm for each party (in fact, the particular strings
   shown in this example are identical, because General MIDI is not a
   configurable MPEG 4 renderer).

C.6.  Configuration Tools: MIDI Rendering

   This appendix defines the session configuration tools for rendering.

   The "render" parameter specifies a rendering method for a stream.
   The parameter is assigned a token value that signals the top-level
   rendering class.  This memo defines four token values for render:
   "unknown", "synthetic", "api", and "null":

     o  An "unknown" renderer is a renderer whose nature is unspecified.
        It is the default renderer for native RTP MIDI streams.

     o  A "synthetic" renderer transforms the MIDI stream into audio
        output (or sometimes into stage lighting changes or other
        actions).  It is the default renderer for mpeg4-generic RTP MIDI
        streams.

     o  An "api" renderer presents the command stream to applications
        via an Application Programmer Interface (API).

     o  The "null" renderer discards the MIDI stream.

   The "null" render value plays special roles during Offer/Answer
   negotiations [RFC3264].  A party uses the "null" value in an answer
   to reject an offered renderer.  Note that rejecting a renderer is
   independent from rejecting a payload type (coded by removing the
   payload type from a media line) and rejecting a media stream (coded
   by zeroing the port of a media line that uses the renderer).

   Other render token values MAY be registered with IANA.  The token
   value MUST adhere to the ABNF for render tokens defined in Appendix
   D.  Registrations MUST include a complete specification of parameter
   value usage, similar in depth to the specifications that appear
   throughout Appendix C.6 for "synthetic" and "api" render values.  If
   a party is offered a session description that uses a render token
   value that is not known to the party, the party MUST NOT accept the
   renderer.  Options include rejecting the renderer (using the "null"
   value), the payload type, the media stream, or the session
   description.

   Other parameters MAY follow a render parameter in a parameter list.
   The additional parameters act to define the exact nature of the
   renderer.  For example, the "subrender" parameter (defined in
   Appendix C.6.2) specifies the exact nature of the renderer.

   Special rules apply to using the render parameter in an mpeg4-generic
   stream.  We define these rules in Appendix C.6.5.

C.6.1.  The multimode Parameter

   A media description MAY contain several render parameters.  By
   default, if a parameter list includes several render parameters, a
   receiver MUST choose exactly one renderer from the list to render the
   stream.  The "multimode" parameter may be used to override this
   default.  We define two token values for multimode: "one" and "all":

     o  The default "one" value requests rendering by exactly one of the
        listed renderers.

     o  The "all" value requests the synchronized rendering of the RTP
        MIDI stream by all listed renderers, if possible.

   If the multimode parameter appears in a parameter list, it MUST
   appear before the first render parameter assignment.

   Render parameters appear in the parameter list in order of decreasing
   priority.  A receiver MAY use the priority ordering to decide which
   renderer(s) to retain in a session.

   If the "offer" in an Offer/Answer-style negotiation [RFC3264]
   contains a parameter list with one or more render parameters, the
   "answer" MUST set the render parameters of all unchosen renderers to
   "null".

C.6.2.  Renderer Specification

   The render parameter (Appendix C.6 preamble) specifies, in a broad
   sense, what a renderer does with a MIDI stream.  In this appendix, we
   describe the "subrender" parameter.  The token value assigned to
   subrender defines the exact nature of the renderer.  Thus, "render"
   and "subrender" combine to define a renderer, in the same way as MIME
   types and MIME subtypes combine to define a type of media [RFC2045].

   If the subrender parameter is used for a renderer definition, it MUST
   appear immediately after the render parameter in the parameter list.
   At most one subrender parameter may appear in a renderer definition.

   This document defines one value for subrender: the value "default".
   The "default" token specifies the use of the default renderer for the
   stream type (native or mpeg4-generic).  The default renderer for
   native RTP MIDI streams is a renderer whose nature is unspecified
   (see point 6 in Section 6.1 for details).  The default renderer for
   mpeg4-generic RTP MIDI streams is an MPEG 4 Audio Object Type whose
   ID number is 13, 14, or 15 (see Section 6.2 for details).

   If a renderer definition does not use the subrender parameter, the
   value "default" is assumed for subrender.

   Other subrender token values may be registered with IANA.  We now
   discuss guidelines for registering subrender values.

   A subrender value is registered for a specific stream type (native or
   mpeg4-generic) and a specific render value (excluding "null" and
   "unknown").  Registrations for mpeg4-generic subrender values are
   restricted to new MPEG 4 Audio Object Types that accept MIDI input.
   The syntax of the token MUST adhere to the token definition in
   Appendix D.

   For "render=synthetic" renderers, a subrender value registration
   specifies an exact method for transforming the MIDI stream into audio
   (or sometimes into video or control actions, such as stage lighting).
   For standardized renderers, this specification is usually a pointer
   to a standards document, perhaps supplemented by RTP-MIDI-specific
   information.  For commercial products and open-source projects, this
   specification usually takes the form of instructions for interfacing
   the RTP MIDI stream with the product or project software.  A
   "render=synthetic" registration MAY specify additional Reset State
   commands for the renderer (Appendix A.1).

   A "render=api" subrender value registration specifies how an RTP MIDI
   stream interfaces with an API (Application Programmers Interface).
   This specification is usually a pointer to programmer's documentation
   for the API, perhaps supplemented by RTP-MIDI-specific information.

   A subrender registration MAY specify an initialization file (referred
   to in this document as an initialization data object) for the stream.
   The initialization data object MAY be encoded in the parameter list
   (verbatim or by reference) using the coding tools defined in Appendix
   C.6.3.  An initialization data object MUST have a registered
   [RFC4288] media type and subtype [RFC2045].

   For "render=synthetic" renderers, the data object usually encodes
   initialization data for the renderer (sample files, synthesis patch
   parameters, reverberation room impulse responses, etc.).

   For "render=api" renderers, the data object usually encodes data
   about the stream used by the API (for example, for an RTP MIDI stream
   generated by a piano keyboard controller, the manufacturer and model
   number of the keyboard, for use in GUI presentation).

   Usually, only one initialization object is encoded for a renderer.
   If a renderer uses multiple data objects, the correct receiver
   interpretation of multiple data objects MUST be defined in the
   subrender registration.

   A subrender value registration may also specify additional
   parameters, to appear in the parameter list immediately after
   subrender.  These parameter names MUST begin with the subrender
   value, followed by an underscore ("_"), to avoid name space
   collisions with future RTP MIDI parameter names (for example, a
   parameter "foo_bar" defined for subrender value "foo").

   We now specify guidelines for interpreting the subrender parameter
   during session configuration.

   If a party is offered a session description that uses a renderer
   whose subrender value is not known to the party, the party MUST NOT
   accept the renderer.  Options include rejecting the renderer (using
   the "null" value), the payload type, the media stream, or the session
   description.

   Receivers MUST be aware of the Reset State commands (Appendix A.1)
   for the renderer specified by the subrender parameter and MUST insure
   that the renderer does not experience indefinite artifacts due to the
   presence (or the loss) of a Reset State command.

C.6.3.  Renderer Initialization

   If the renderer for a stream uses an initialization data object, an
   "rinit" parameter MUST appear in the parameter list immediately after
   the "subrender" parameter.  If the renderer parameter list does not
   include a subrender parameter (recall the semantics for "default" in
   Appendix C.6.2), the "rinit" parameter MUST appear immediately after
   the "render" parameter.

   The value assigned to the rinit parameter MUST be the media
   type/subtype [RFC2045] for the initialization data object.  If an
   initialization object type is registered with several media types,
   including audio, the assignment to rinit MUST use the audio media
   type.

   RTP MIDI supports several parameters for encoding initialization data
   objects for renderers in the parameter list: "inline", "url", and
   "cid".

   If the "inline", "url", and/or "cid" parameters are used by a
   renderer, these parameters MUST immediately follow the "rinit"
   parameter.

   If a "url" parameter appears for a renderer, an "inline" parameter
   MUST NOT appear.  If an "inline" parameter appears for a renderer, a
   "url" parameter MUST NOT appear.  However, neither "url" or "inline"
   is required to appear.  If neither "url" or "inline" parameters
   follow "rinit", the "cid" parameter MUST follow "rinit".

   The "inline" parameter supports the inline encoding of the data
   object.  The parameter is assigned a double-quoted Base64 [RFC2045]
   encoding of the binary data object, with no line breaks.  Appendix
   E.4 shows an example that constructs an inline parameter value.

   The "url" parameter is assigned a double-quoted string representation
   of a Uniform Resource Locator (URL) for the data object.  The string
   MUST specify a HyperText Transport Protocol URL (HTTP, [RFC2616]).
   HTTP MAY be used over TCP or MAY be used over a secure network
   transport, such as the method described in [RFC2818].  The media
   type/subtype for the data object SHOULD be specified in the
   appropriate HTTP transport header.

   The "cid" parameter supports data object caching.  The parameter is
   assigned a double-quoted string value that encodes a globally unique
   identifier for the data object.

   A cid parameter MAY immediately follow an inline parameter, in which
   case the cid identifier value MUST be associated with the inline data
   object.

   If a url parameter is present, and if the data object for the URL is
   expected to be unchanged for the life of the URL, a cid parameter MAY
   immediately follow the url parameter.  The cid identifier value MUST
   be associated with the data object for the URL.  A cid parameter
   assigned to the same identifier value SHOULD be specified following
   the data object type/subtype in the appropriate HTTP transport
   header.

   If a url parameter is present, and if the data object for the URL is
   expected to change during the life of the URL, a cid parameter MUST
   NOT follow the url parameter.  A receiver interprets the presence of
   a cid parameter as an indication that it is safe to use a cached copy
   of the url data object; the absence of a cid parameter is an
   indication that it is not safe to use a cached copy, as it may
   change.

   Finally, the cid parameter MAY be used without the inline and url
   parameters.  In this case, the identifier references a local or
   distributed catalog of data objects.

   In most cases, only one data object is coded in the parameter list
   for each renderer.  For example, the default renderer for mpeg4-
   generic streams uses a single data object (see Appendix C.6.5 for
   example usage).

   However, a subrender registration MAY permit the use of multiple data
   objects for a renderer.  If multiple data objects are encoded for a
   renderer, each object encoding begins with an "rinit" parameter,
   followed by "inline", "url", and/or "cid" parameters.

   Initialization data objects MAY encapsulate a Standard MIDI File
   (SMF).  By default, the SMFs that are encapsulated in a data object
   MUST be ignored by an RTP MIDI receiver.  We define parameters to
   override this default in Appendix C.6.4.

   To end this section, we offer guidelines for registering media types
   for initialization data objects.  These guidelines are in addition to
   the information in [RFC4288] [RFC4289].

   Some initialization data objects are also capable of encoding MIDI
   note information and thus complete audio performances.  These objects
   SHOULD be registered using the "audio" media type, so that the
   objects may also be used for store-and-forward rendering, and
   "application" media type, to support editing tools.  Initialization
   objects without note storage, or initialization objects for non-audio
   renderers, SHOULD be registered only for an "application" media type.

C.6.4.  MIDI Channel Mapping

   In this appendix, we specify how to map MIDI name spaces (16 voice
   channels + systems) onto a renderer.

   In the general case:

     o  A session may define an ordered relationship (Appendix C.5) that
        presents more than one MIDI name space to a renderer.

     o  A renderer may accept an arbitrary number of MIDI name spaces,
        or it may expect a specific number of MIDI name spaces.

   A session description SHOULD provide a compatible MIDI name space to
   each renderer in the session.  If a receiver detects that a session
   description has too many or too few MIDI name spaces for a renderer,
   MIDI data from extra stream name spaces MUST be discarded, and extra
   renderer name spaces MUST NOT be driven with MIDI data (except as
   described in Appendix C.6.4.1, below).

   If a parameter list defines several renderers and assigns the "all"
   token value to the multimode parameter, the same name space is
   presented to each renderer.  However, the "chanmask" parameter may be
   used to mask out selected voice channels to each renderer.  We define
   "chanmask" and other MIDI management parameters in the sub-sections
   below.

C.6.4.1.  The smf_info Parameter

   The smf_info parameter defines the use of the SMFs encapsulated in
   renderer data objects (if any).  The smf_info parameter also defines
   the use of SMFs coded in the smf_inline, smf_url, and smf_cid
   parameters (defined in Appendix C.6.4.2).

   The smf_info parameter describes the "render" parameter that most
   recently precedes it in the parameter list.  The smf_info parameter
   MUST NOT appear in parameter lists that do not use the "render"
   parameter, and MUST NOT appear before the first use of "render" in
   the parameter list.

   We define three token values for smf_info: "ignore", "sdp_start", and
   "identity":

     o  The "ignore" value indicates that the SMFs MUST be discarded.
        This behavior is the default SMF rendering behavior.

     o  The "sdp_start" value codes that SMFs MUST be rendered, and that
        the rendering MUST begin upon the acceptance of the session
        description.  If a receiver is offered a session description
        with a renderer that uses an smf_info parameter set to
        sdp_start, and if the receiver does not support rendering SMFs,
        the receiver MUST NOT accept the renderer associated with the
        smf_info parameter.  Options include rejecting the renderer (by
        setting the "render" parameter to "null"), the payload type, the
        media stream, or the entire session description.

     o  The "identity" value indicates that the SMFs code the identity
        of the renderer.  The value is meant for use with the "unknown"
        renderer (see Appendix C.6 preamble).  The MIDI commands coded
        in the SMF are informational in nature and MUST NOT be presented
        to a renderer for audio presentation.  In typical use, the SMF
        would use SysEx Identity Reply commands (F0 7E nn 06 02, as
        defined in [MIDI]) to identify devices, and use device-specific
        SysEx commands to describe current state of the devices (patch
        memory contents, etc.).

   Other smf_info token values MAY be registered with IANA.  The token
   value MUST adhere to the ABNF for render tokens defined in Appendix

   D.  Registrations MUST include a complete specification of parameter
   usage, similar in depth to the specifications that appear in this
   appendix for "sdp_start" and "identity".

   If a party is offered a session description that uses an smf_info
   parameter value that is not known to the party, the party MUST NOT
   accept the renderer associated with the smf_info parameter.  Options
   include rejecting the renderer, the payload type, the media stream,
   or the entire session description.

   We now define the rendering semantics for the "sdp_start" token value
   in detail.

   The SMFs and RTP MIDI streams in a session description share the same
   MIDI name space(s).  In the simple case of a single RTP MIDI stream
   and a single SMF, the SMF MIDI commands and RTP MIDI commands are
   merged into a single name space and presented to the renderer.  The
   indefinite artifact responsibilities for merged MIDI streams defined
   in Appendix C.5 also apply to merging RTP and SMF MIDI data.

   If a payload type codes multiple SMFs, the SMF name spaces are
   presented as an ordered entity to the renderer.  To determine the
   ordering of SMFs for a renderer (which SMF is "first", which is
   "second", etc.), use the following rules:

     o  If the renderer uses a single data object, the order of
        appearance of the SMFs in the object's internal structure
        defines the order of the SMFs (the earliest SMF in the object is
        "first", the next SMF in the object is "second", etc.).

     o  If multiple data objects are encoded for a renderer, the
        appearance of each data object in the parameter list sets the
        relative order of the SMFs encoded in each data object (SMFs
        encoded in parameters that appear earlier in the list are
        ordered before SMFs encoded in parameters that appear later in
        the list).

     o  If SMFs are encoded in data objects parameters and in the
        parameters defined in C.6.4.2, the relative order of the data
        object parameters and C.6.4.2 parameters in the parameter list
        sets the relative order of SMFs (SMFs encoded in parameters that
        appear earlier in the list are ordered before SMFs in parameters
        that appear later in the list).

   Given this ordering of SMFs, we now define the mapping of SMFs to
   renderer name spaces.  The SMF that appears first for a renderer maps
   to the first renderer name space.  The SMF that appears second for a
   renderer maps to the second renderer name space, etc.  If the

   associated RTP MIDI streams also form an ordered relationship, the
   first SMF is merged with the first name space of the relationship,
   the second SMF is merged to the second name space of the
   relationship, etc.

   Unless the streams and the SMFs both use MIDI Time Code, the time
   offset between SMF and stream data is unspecified.  This restriction
   limits the use of SMFs to applications where synchronization is not
   critical, such as the transport of System Exclusive commands for
   renderer initialization, or human-SMF interactivity.

   Finally, we note that each SMF in the sdp_start discussion above
   encodes exactly one MIDI name space (16 voice channels + systems).
   Thus, the use of the Device Name SMF meta event to specify several
   MIDI name spaces in an SMF is not supported for sdp_start.

C.6.4.2.  The smf_inline, smf_url, and smf_cid Parameters

   In some applications, the renderer data object may not encapsulate
   SMFs, but an application may wish to use SMFs in the manner defined
   in Appendix C.6.4.1.

   The "smf_inline", "smf_url", and "smf_cid" parameters address this
   situation.  These parameters use the syntax and semantics of the
   inline, url, and cid parameters defined in Appendix C.6.3, except
   that the encoded data object is an SMF.

   The "smf_inline", "smf_url", and "smf_cid" parameters belong to the
   "render" parameter that most recently precedes it in the session
   description.  The "smf_inline", "smf_url", and "smf_cid" parameters
   MUST NOT appear in parameter lists that do not use the "render"
   parameter and MUST NOT appear before the first use of "render" in the
   parameter list.  If several "smf_inline", "smf_url", or "smf_cid"
   parameters appear for a renderer, the order of the parameters defines
   the SMF name space ordering.

C.6.4.3.  The chanmask Parameter

   The chanmask parameter instructs the renderer to ignore all MIDI
   voice commands for certain channel numbers.  The parameter value is a
   concatenated string of "1" and "0" digits.  Each string position maps
   to a MIDI voice channel number (system channels may not be masked).
   A "1" instructs the renderer to process the voice channel; a "0"
   instructs the renderer to ignore the voice channel.

   The string length of the chanmask parameter value MUST be 16 (for a
   single stream or an identity relationship) or a multiple of 16 (for
   an ordered relationship).

   The chanmask parameter describes the "render" parameter that most
   recently precedes it in the session description; chanmask MUST NOT
   appear in parameter lists that do not use the "render" parameter and
   MUST NOT appear before the first use of "render" in the parameter
   list.

   The chanmask parameter describes the final MIDI name spaces presented
   to the renderer.  The SMF and stream components of the MIDI name
   spaces may not be independently masked.

   If a receiver is offered a session description with a renderer that
   uses the chanmask parameter, and if the receiver does not implement
   the semantics of the chanmask parameter, the receiver MUST NOT accept
   the renderer unless the chanmask parameter value contains only "1"s.

C.6.5.  The audio/asc Media Type

   In Appendix 11.3, we register the audio/asc media type.  The data
   object for audio/asc is a binary encoding of the AudioSpecificConfig
   data block used to initialize mpeg4-generic streams (Section 6.2 and
   [MPEGAUDIO]).

   An mpeg4-generic parameter list MAY use the render, subrender, and
   rinit parameters with the audio/asc media type for renderer
   configuration.  Several restrictions apply to the use of these
   parameters in mpeg4-generic parameter lists:

     o  An mpeg4-generic media description that uses the render
        parameter MUST assign the empty string ("") to the mpeg4-generic
        "config" parameter.  The use of the streamtype, mode, and
        profile-level-id parameters MUST follow the normative text in
        Section 6.2.

     o  Sessions that use identity or ordered relationships MUST follow
        the mpeg4-generic configuration restrictions in Appendix C.5.

     o  The render parameter MUST be assigned the value "synthetic",
        "unknown", "null", or a render value that has been added to the
        IANA repository for use with mpeg4-generic RTP MIDI streams.
        The "api" token value for render MUST NOT be used.

     o  If a subrender parameter is present, it MUST immediately follow
        the render parameter, and it MUST be assigned the token value
        "default" or assigned a subrender value added to the IANA
        repository for use with mpeg4-generic RTP MIDI streams.  A
        subrender parameter assignment may be left out of the renderer
        configuration, in which case the implied value of subrender is
        the default value of "default".

     o  If the render parameter is assigned the value "synthetic" and
        the subrender parameter has the value "default" (assigned or
        implied), the rinit parameter MUST be assigned the value
        "audio/asc", and an AudioSpecificConfig data object MUST be
        encoded using the mechanisms defined in C.6.2-3.  The
        AudioSpecificConfig data MUST encode one of the MPEG 4 Audio
        Object Types defined for use with mpeg4-generic in Section 6.2.
        If the subrender value is other than "default", refer to the
        subrender registration for information on the use of "audio/asc"
        with the renderer.

     o  If the render parameter is assigned the value "null" or
        "unknown", the data object MAY be omitted.

   Several general restrictions apply to the use of the audio/asc media
   type in RTP MIDI:

     o  A native stream MUST NOT assign "audio/asc" to rinit.  The
        audio/asc media type is not intended to be a general-purpose
        container for rendering systems outside of MPEG usage.

     o  The audio/asc media type defines a stored object type; it does
        not define semantics for RTP streams.  Thus, audio/asc MUST NOT
        appear on an rtpmap line of a session description.

   Below, we show session description examples for audio/asc.  The
   session description below uses the inline parameter to code the
   AudioSpecificConfig block for a mpeg4-generic General MIDI stream.
   We derive the value assigned to the inline parameter in Appendix E.4.
   The subrender token value of "default" is implied by the absence of
   the subrender parameter in the parameter list.

   v=0
   o=lazzaro 2520644554 2838152170 IN IP4 first.example.net
   s=Example
   t=0 0
   m=audio 5004 RTP/AVP 96
   c=IN IP4 192.0.2.94
   a=rtpmap:96 mpeg4-generic/44100
   a=fmtp:96 streamtype=5; mode=rtp-midi; config="";
   profile-level-id=12; render=synthetic; rinit="audio/asc";
   inline="egoAAAAaTVRoZAAAAAYAAAABAGBNVHJrAAAABgD/LwAA"

   (The a=fmtp line has been wrapped to fit the page to accommodate
    memo formatting restrictions; it comprises a single line in SDP.)

   The session description below uses the url parameter to code the
   AudioSpecificConfig block for the same General MIDI stream:

   v=0
   o=lazzaro 2520644554 2838152170 IN IP4 first.example.net
   s=Example
   t=0 0
   m=audio 5004 RTP/AVP 96
   c=IN IP4 192.0.2.94
   a=rtpmap:96 mpeg4-generic/44100
   a=fmtp:96 streamtype=5; mode=rtp-midi; config="";
   profile-level-id=12; render=synthetic; rinit="audio/asc";
   url="http://example.net/oski.asc";
   cid="xjflsoeiurvpa09itnvlduihgnvet98pa3w9utnuighbuk"

   (The a=fmtp line has been wrapped to fit the page to accommodate
    memo formatting restrictions; it comprises a single line in SDP.)

C.7.  Interoperability

   In this appendix, we define interoperability guidelines for two
   application areas:

     o  MIDI content-streaming applications.  RTP MIDI is added to
        RTSP-based content-streaming servers, so that viewers may
        experience MIDI performances (produced by a specified client-
        side renderer) in synchronization with other streams (video,
        audio).

     o  Long-distance network musical performance applications.  RTP
        MIDI is added to SIP-based voice chat or videoconferencing
        programs, as an alternative, or as an addition, to audio and/or
        video RTP streams.

   For each application, we define a core set of functionality that all
   implementations MUST implement.

   The applications we address in this section are not an exhaustive
   list of potential RTP MIDI uses.  We expect framework documents for
   other applications to be developed, within the IETF or within other
   organizations.  We discuss other potential application areas for RTP
   MIDI in Section 1 of the main text of this memo.

C.7.1.  MIDI Content Streaming Applications

   In content-streaming applications, a user invokes an RTSP client to
   initiate a request to an RTSP server to view a multimedia session.
   For example, clicking on a web page link for an Internet Radio

   channel launches an RTSP client that uses the link's RTSP URL to
   contact the RTSP server hosting the radio channel.

   The content may be pre-recorded (for example, on-demand replay of
   yesterday's football game) or "live" (for example, football game
   coverage as it occurs), but in either case the user is usually an
   "audience member" as opposed to a "participant" (as the user would be
   in telephony).

   Note that these examples describe the distribution of audio content
   to an audience member.  The interoperability guidelines in this
   appendix address RTP MIDI applications of this nature, not
   applications such as the transmission of raw MIDI command streams for
   use in a professional environment (recording studio, performance
   stage, etc.).

   In an RTSP session, a client accesses a session description that is
   "declared" by the server, either via the RTSP DESCRIBE method, or via
   other means, such as HTTP or email.  The session description defines
   the session from the perspective of the client.  For example, if a
   media line in the session description contains a non-zero port
   number, it encodes the server's preference for the client's port
   numbers for RTP and RTCP reception.  Once media flow begins, the
   server sends an RTP MIDI stream to the client, which renders it for
   presentation, perhaps in synchrony with video or other audio streams.

   We now define the interoperability text for content-streaming RTSP
   applications.

   In most cases, server interoperability responsibilities are described
   in terms of limits on the "reference" session description a server
   provides for a performance if it has no information about the
   capabilities of the client.  The reference session is a "lowest
   common denominator" session that maximizes the odds that a client
   will be able to view the session.  If a server is aware of the
   capabilities of the client, the server is free to provide a session
   description customized for the client in the DESCRIBE reply.

   Clients MUST support unicast UDP RTP MIDI streams that use the
   recovery journal with the closed-loop or the anchor sending policies.
   Clients MUST be able to interpret stream subsetting and chapter
   inclusion parameters in the session description that qualify the
   sending policies.  Client support of enhanced Chapter C encoding is
   OPTIONAL.

   The reference session description offered by a server MUST send all
   RTP MIDI UDP streams as unicast streams that use the recovery journal
   and the closed-loop or anchor sending policies.  Servers SHOULD use

   the stream subsetting and chapter inclusion parameters in the
   reference session description, to simplify the rendering task of the
   client.  Server support of enhanced Chapter C encoding is OPTIONAL.

   Clients and servers MUST support the use of RTSP interleaved mode (a
   method for interleaving RTP onto the RTSP TCP transport).

   Clients MUST be able to interpret the timestamp semantics signalled
   by the "comex" value of the tsmode parameter (i.e., the timestamp
   semantics of Standard MIDI Files [MIDI]).  Servers MUST use the
   "comex" value for the "tsmode" parameter in the reference session
   description.

   Clients MUST be able to process an RTP MIDI stream whose packets
   encode an arbitrary temporal duration ("media time").  Thus, in
   practice, clients MUST implement a MIDI playout buffer.  Clients MUST
   NOT depend on the presence of rtp_ptime, rtp_maxtime, and guardtime
   parameters in the session description in order to process packets,
   but they SHOULD be able to use these parameters to improve packet
   processing.

   Servers SHOULD strive to send RTP MIDI streams in the same way media
   servers send conventional audio streams: a sequence of packets that
   either all code the same temporal duration (non-normative example: 50
   ms packets) or that code one of an integral number of temporal
   durations (non-normative example: 50 ms, 100 ms, 250 ms, or 500 ms
   packets).  Servers SHOULD encode information about the packetization
   method in the rtp_ptime and rtp_maxtime parameters in the session
   description.

   Clients MUST be able to examine the render and subrender parameter,
   to determine if a multimedia session uses a renderer it supports.
   Clients MUST be able to interpret the default "one" value of the
   "multimode" parameter, to identify supported renderers from a list of
   renderer descriptions.  Clients MUST be able to interpret the
   musicport parameter, to the degree that it is relevant to the
   renderers it supports.  Clients MUST be able to interpret the
   chanmask parameter.

   Clients supporting renderers whose data object (as encoded by a
   parameter value for "inline") could exceed 300 octets in size MUST
   support the url and cid parameters and thus must implement the HTTP
   protocol in addition to RTSP.

   Servers MUST specify complete rendering systems for RTP MIDI streams.
   Note that a minimal RTP MIDI native stream does not meet this
   requirement (Section 6.1), as the rendering method for such streams
   is "not specified".

   At the time of this memo, the only way for servers to specify a
   complete rendering system is to specify an mpeg4-generic RTP MIDI
   stream in mode rtp-midi (Section 6.2 and C.6.5).  As a consequence,
   the only rendering systems that may be presently used are General
   MIDI [MIDI], DLS 2 [DLS2], or Structured Audio [MPEGSA].  Note that
   the maximum inline value for General MIDI is well under 300 octets
   (and thus clients need not support the "url" parameter), and that the
   maximum inline values for DLS 2 and Structured Audio may be much
   larger than 300 octets (and thus clients MUST support the url
   parameter).

   We anticipate that the owners of rendering systems (both standardized
   and proprietary) will register subrender parameters for their
   renderers.  Once registration occurs, native RTP MIDI sessions may
   use render and subrender (Appendix C.6.2) to specify complete
   rendering systems for RTSP content-streaming multimedia sessions.

   Servers MUST NOT use the sdp_start value for the smf_info parameter
   in the reference session description, as this use would require that
   clients be able to parse and render Standard MIDI Files.

   Clients MUST support mpeg4-generic mode rtp-midi General MIDI (GM)
   sessions, at a polyphony limited by the hardware capabilities of the
   client.  This requirement provides a "lowest common denominator"
   rendering system for content providers to target.  Note that this
   requirement does not force implementors of a non-GM renderer (such as
   DLS 2 or Structured Audio) to add a second rendering engine.
   Instead, a client may satisfy the requirement by including a set of
   voice patches that implement the GM instrument set, and using this
   emulation for mpeg4-generic GM sessions.

   It is RECOMMENDED that servers use General MIDI as the renderer for
   the reference session description, because clients are REQUIRED to
   support it.  We do not require General MIDI as the reference
   renderer, because for normative applications it is an inappropriate
   choice.  Servers using General MIDI as a "lowest common denominator"
   renderer SHOULD use Universal Real-Time SysEx MIP message [SPMIDI] to
   communicate the priority of voices to polyphony-limited clients.

C.7.2.  MIDI Network Musical Performance Applications

   In Internet telephony and videoconferencing applications, parties
   interact over an IP network as they would face-to-face.  Good user
   experiences require low end-to-end audio latency and tight
   audiovisual synchronization (for "lip-sync").  The Session Initiation
   Protocol (SIP, [RFC3261]) is used for session management.

   In this appendix section, we define interoperability guidelines for
   using RTP MIDI streams in interactive SIP applications.  Our primary
   interest is supporting Network Musical Performances (NMP), where
   musicians in different locations interact over the network as if they
   were in the same room.  See [NMP] for background information on NMP,
   and see [RFC4696] for a discussion of low-latency RTP MIDI
   implementation techniques for NMP.

   Note that the goal of NMP applications is telepresence: the parties
   should hear audio that is close to what they would hear if they were
   in the same room.  The interoperability guidelines in this appendix
   address RTP MIDI applications of this nature, not applications such
   as the transmission of raw MIDI command streams for use in a
   professional environment (recording studio, performance stage, etc.).

   We focus on session management for two-party unicast sessions that
   specify a renderer for RTP MIDI streams.  Within this limited scope,
   the guidelines defined here are sufficient to let applications
   interoperate.  We define the REQUIRED capabilities of RTP MIDI
   senders and receivers in NMP sessions and define how session
   descriptions exchanged are used to set up network musical performance
   sessions.

   SIP lets parties negotiate details of the session, using the
   Offer/Answer protocol [RFC3264].  However, RTP MIDI has so many
   parameters that "blind" negotiations between two parties using
   different applications might not yield a common session
   configuration.

   Thus, we now define a set of capabilities that NMP parties MUST
   support.  Session description offers whose options lie outside the
   envelope of REQUIRED party behavior risk negotiation failure.  We
   also define session description idioms that the RTP MIDI part of an
   offer MUST follow, in order to structure the offer for simpler
   analysis.

   We use the term "offerer" for the party making a SIP offer, and
   "answerer" for the party answering the offer.  Finally, we note that
   unless it is qualified by the adjective "sender" or "receiver", a
   statement that a party MUST support X implies that it MUST support X
   for both sending and receiving.

   If an offerer wishes to define a "sendrecv" RTP MIDI stream, it may
   use a true sendrecv session or the "virtual sendrecv" construction
   described in the preamble to Appendix C and in Appendix C.5.  A true
   sendrecv session indicates that the offerer wishes to participate in
   a session where both parties use identically configured renderers.  A
   virtual sendrecv session indicates that the offerer is willing to

   participate in a session where the two parties may be using different
   renderer configurations.  Thus, parties MUST be prepared to see both
   real and virtual sendrecv sessions in an offer.

   Parties MUST support unicast UDP transport of RTP MIDI streams.
   These streams MUST use the recovery journal with the closed-loop or
   anchor sending policies.  These streams MUST use the stream
   subsetting and chapter inclusion parameters to declare the types of
   MIDI commands that will be sent on the stream (for sendonly streams)
   or will be processed (for recvonly streams), including the size
   limits on System Exclusive commands.  Support of enhanced Chapter C
   encoding is OPTIONAL.

   Note that both TCP and multicast UDP support are OPTIONAL.  We make
   TCP OPTIONAL because we expect NMP renderers to rely on data objects
   (signalled by "rinit" and associated parameters) for initialization
   at the start of the session, and only to use System Exclusive
   commands for interactive control during the session.  These
   interactive commands are small enough to be protected via the
   recovery journal mechanism of RTP MIDI UDP streams.

   We now discuss timestamps, packet timing, and packet sending
   algorithms.

   Recall that the tsmode parameter controls the semantics of command
   timestamps in the MIDI list of RTP packets.

   Parties MUST support clock rates of 44.1 kHz, 48 kHz, 88.2 kHz, and
   96 kHz.  Parties MUST support streams using the "comex", "async", and
   "buffer" tsmode values.  Recvonly offers MUST offer the default
   "comex".

   Parties MUST support a wide range of packet temporal durations: from
   rtp_ptime and rtp_maxptime values of 0, to rtp_ptime and rtp_maxptime
   values that code 100 ms.  Thus, receivers MUST be able to implement a
   playout buffer.

   Offers and answers MUST present rtp_ptime, rtp_maxptime, and
   guardtime values that support the latency that users would expect in
   the application, subject to bandwidth constraints.  As senders MUST
   abide by values set for these parameters in a session description, a
   receiver SHOULD use these values to size its playout buffer to
   produce the lowest reliable latency for a session.  Implementers
   should refer to [RFC4696] for information on packet sending
   algorithms for latency-sensitive applications.  Parties MUST be able
   to implement the semantics of the guardtime parameter, for times from
   5 ms to 5000 ms.

   We now discuss the use of the render parameter.

   Sessions MUST specify complete rendering systems for all RTP MIDI
   streams.  Note that a minimal RTP MIDI native stream does not meet
   this requirement (Section 6.1), as the rendering method for such
   streams is "not specified".

   At the time this writing, the only way for parties to specify a
   complete rendering system is to specify an mpeg4-generic RTP MIDI
   stream in mode rtp-midi (Section 6.2 and C.6.5).  We anticipate that
   the owners of rendering systems (both standardized and proprietary)
   will register subrender values for their renderers.  Once IANA
   registration occurs, native RTP MIDI sessions may use render and
   subrender (Appendix C.6.2) to specify complete rendering systems for
   SIP network musical performance multimedia sessions.

   All parties MUST support General MIDI (GM) sessions, at a polyphony
   limited by the hardware capabilities of the party.  This requirement
   provides a "lowest common denominator" rendering system, without
   which practical interoperability will be quite difficult.  When using
   GM, parties SHOULD use Universal Real-Time SysEx MIP message [SPMIDI]
   to communicate the priority of voices to polyphony-limited clients.

   Note that this requirement does not force implementors of a non-GM
   renderer (for mpeg4-generic sessions, DLS 2, or Structured Audio) to
   add a second rendering engine.  Instead, a client may satisfy the
   requirement by including a set of voice patches that implement the GM
   instrument set, and using this emulation for mpeg4-generic GM
   sessions.  We require GM support so that an offerer that wishes to
   maximize interoperability may do so by offering GM if its preferred
   renderer is not accepted by the answerer.

   Offerers MUST NOT present several renderers as options in a session
   description by listing several payload types on a media line, as
   Section 2.1 uses this construct to let a party send several RTP MIDI
   streams in the same RTP session.

   Instead, an offerer wishing to present rendering options SHOULD offer
   a single payload type that offers several renderers.  In this
   construct, the parameter list codes a list of render parameters (each
   followed by its support parameters).  As discussed in Appendix C.6.1,
   the order of renderers in the list declares the offerer's preference.
   The "unknown" and "null" values MUST NOT appear in the offer.  The
   answer MUST set all render values except the desired renderer to
   "null".  Thus, "unknown" MUST NOT appear in the answer.

   We use SHOULD instead of MUST in the first sentence in the paragraph
   above, because this technique does not work in all situations
   (example:  an offerer wishes to offer both mpeg4-generic renderers
   and native RTP MIDI renderers as options).  In this case, the offerer
   MUST present a series of session descriptions, each offering a single
   renderer, until the answerer accepts a session description.

   Parties MUST support the musicport, chanmask, subrender, rinit, and
   inline parameters.  Parties supporting renderers whose data object
   (as encoded by a parameter value for "inline") could exceed 300
   octets in size MUST support the url and cid parameters and thus must
   implement HTTP protocol.  Note that in mpeg4-generic, General MIDI
   data objects cannot exceed 300 octets, but DLS 2 and Structured Audio
   data objects may.  Support for the other rendering parameters
   (smf_cif, smf_info, smf_inline, smf_url) is OPTIONAL.

   Thus far in this document, our discussion has assumed that the only
   MIDI flows that drive a renderer are the network flows described in
   the session description.  In NMP applications, this assumption would
   require two rendering engines: one for local use by a party, a second
   for the remote party.

   In practice, applications may wish to have both parties share a
   single rendering engine.  In this case, the session description MUST
   use a virtual sendrecv session and MUST use the stream subsetting and
   chapter inclusion parameters to allocate which MIDI channels are
   intended for use by a party.  If two parties are sharing a MIDI
   channels, the application MUST ensure that appropriate MIDI merging
   occurs at the input to the renderer.

   We now discuss the use of (non-MIDI) audio streams in the session.

   Audio streams may be used for two purposes: as a "talkback" channel
   for parties to converse, or as a way to conduct a performance that
   includes MIDI and audio channels.  In the latter case, offers MUST
   use sample rates and the packet temporal durations for the audio and
   MIDI streams that support low-latency synchronized rendering.

   We now show an example of an offer/answer exchange in a network
   musical performance application (next page).  Below, we show an offer
   that complies with the interoperability text in this appendix
   section.

   v=0
   o=first 2520644554 2838152170 IN IP4 first.example.net
   s=Example
   t=0 0
   a=group:FID 1 2
   c=IN IP4 192.0.2.94
   m=audio 16112 RTP/AVP 96
   a=recvonly
   a=mid:1
   a=rtpmap:96 mpeg4-generic/44100
   a=fmtp:96 streamtype=5; mode=rtp-midi; config="";
   profile-level-id=12; cm_unused=ABCFGHJKMNPQTVWXYZ; cm_used=2NPTW;
   cm_used=2C0.1.7.10.11.64.121.123; cm_used=2M0.1.2
   cm_used=X0-16; ch_never=ABCDEFGHJKMNPQTVWXYZ;
   ch_default=2NPTW; ch_default=2C0.1.7.10.11.64.121.123;
   ch_default=2M0.1.2; cm_default=X0-16;
   rtp_ptime=0; rtp_maxptime=0; guardtime=44100;
   musicport=1; render=synthetic; rinit="audio/asc";
   inline="egoAAAAaTVRoZAAAAAYAAAABAGBNVHJrAAAABgD/LwAA"
   m=audio 16114 RTP/AVP 96
   a=sendonly
   a=mid:2
   a=rtpmap:96 mpeg4-generic/44100
   a=fmtp:96 streamtype=5; mode=rtp-midi; config="";
   profile-level-id=12; cm_unused=ABCFGHJKMNPQTVWXYZ; cm_used=1NPTW;
   cm_used=1C0.1.7.10.11.64.121.123; cm_used=1M0.1.2
   cm_used=X0-16; ch_never=ABCDEFGHJKMNPQTVWXYZ;
   ch_default=1NPTW; ch_default=1C0.1.7.10.11.64.121.123;
   ch_default=1M0.1.2; cm_default=X0-16;
   rtp_ptime=0; rtp_maxptime=0; guardtime=44100;
   musicport=1; render=synthetic; rinit="audio/asc";
   inline="egoAAAAaTVRoZAAAAAYAAAABAGBNVHJrAAAABgD/LwAA"

   (The a=fmtp lines have been wrapped to fit the page to accommodate
    memo formatting restrictions; it comprises a single line in SDP.)

   The owner line (o=) identifies the session owner as "first".

   The session description defines two MIDI streams: a recvonly stream
   on which "first" receives a performance, and a sendonly stream that
   "first" uses to send a performance.  The recvonly port number encodes
   the ports on which "first" wishes to receive RTP (16112) and RTCP
   (16113) media at IP4 address 192.0.2.94.  The sendonly port number

   encodes the port on which "first" wishes to receive RTCP for the
   stream (16115).

   The musicport parameters code that the two streams share and identity
   relationship and thus form a virtual sendrecv stream.

   Both streams are mpeg4-generic RTP MIDI streams that specify a
   General MIDI renderer.  The stream subsetting parameters code that
   the recvonly stream uses MIDI channel 1 exclusively for voice
   commands, and that the sendonly stream uses MIDI channel 2
   exclusively for voice commands.  This mapping permits the application
   software to share a single renderer for local and remote performers.

   We now show the answer to the offer.

   v=0
   o=second 2520644554 2838152170 IN IP4 second.example.net
   s=Example
   t=0 0
   a=group:FID 1 2
   c=IN IP4 192.0.2.105
   m=audio 5004 RTP/AVP 96
   a=sendonly
   a=mid:1
   a=rtpmap:96 mpeg4-generic/44100
   a=fmtp:96 streamtype=5; mode=rtp-midi; config="";
   profile-level-id=12; cm_unused=ABCFGHJKMNPQTVWXYZ; cm_used=2NPTW;
   cm_used=2C0.1.7.10.11.64.121.123; cm_used=2M0.1.2
   cm_used=X0-16; ch_never=ABCDEFGHJKMNPQTVWXYZ;
   ch_default=2NPTW; ch_default=2C0.1.7.10.11.64.121.123;
   ch_default=2M0.1.2; cm_default=X0-16;
   rtp_ptime=0; rtp_maxptime=882; guardtime=44100;
   musicport=1; render=synthetic; rinit="audio/asc";
   inline="egoAAAAaTVRoZAAAAAYAAAABAGBNVHJrAAAABgD/LwAA"
   m=audio 5006 RTP/AVP 96
   a=recvonly
   a=mid:2
   a=rtpmap:96 mpeg4-generic/44100
   a=fmtp:96 streamtype=5; mode=rtp-midi; config="";
   profile-level-id=12; cm_unused=ABCFGHJKMNPQTVWXYZ; cm_used=1NPTW;
   cm_used=1C0.1.7.10.11.64.121.123; cm_used=1M0.1.2
   cm_used=X0-16; ch_never=ABCDEFGHJKMNPQTVWXYZ;
   ch_default=1NPTW; ch_default=1C0.1.7.10.11.64.121.123;
   ch_default=1M0.1.2; cm_default=X0-16;
   rtp_ptime=0; rtp_maxptime=0; guardtime=88200;
   musicport=1; render=synthetic; rinit="audio/asc";
   inline="egoAAAAaTVRoZAAAAAYAAAABAGBNVHJrAAAABgD/LwAA"

   (The a=fmtp lines have been wrapped to fit the page to accommodate
    memo formatting restrictions; they comprise single lines in SDP.)

   The owner line (o=) identifies the session owner as "second".

   The port numbers for both media streams are non-zero; thus, "second"
   has accepted the session description.  The stream marked "sendonly"
   in the offer is marked "recvonly" in the answer, and vice versa,
   coding the different view of the session held by "session".  The IP4
   number (192.0.2.105) and the RTP (5004 and 5006) and RTCP (5005 and
   5007) have been changed by "second" to match its transport wishes.

   In addition, "second" has made several parameter changes:
   rtp_maxptime for the sendonly stream has been changed to code 2 ms
   (441 in clock units), and the guardtime for the recvonly stream has
   been doubled.  As these parameter modifications request capabilities
   that are REQUIRED to be implemented by interoperable parties,
   "second" can make these changes with confidence that "first" can
   abide by them.

D.  Parameter Syntax Definitions

   In this appendix, we define the syntax for the RTP MIDI media type
   parameters in Augmented Backus-Naur Form (ABNF, [RFC4234]).  When
   using these parameters with SDP, all parameters MUST appear on a
   single fmtp attribute line of an RTP MIDI media description.  For
   mpeg4-generic RTP MIDI streams, this line MUST also include any
   mpeg4-generic parameters (usage described in Section 6.2).  An fmtp
   attribute line may be defined (after [RFC3640]) as:

   ;
   ; SDP fmtp line definition
   ;

   fmtp = "a=fmtp:" token SP param-assign 0*(";" SP param-assign) CRLF

   where <token> codes the RTP payload type.  Note that white space MUST
   NOT appear between the "a=fmtp:" and the RTP payload type.

   We now define the syntax of the parameters defined in Appendix C.
   The definition takes the form of the incremental assembly of the
   <param-assign> token.  See [RFC3640] for the syntax of the
   mpeg4-generic parameters discussed in Section 6.2.

   ;
   ;
   ; top-level definition for all parameters
   ;
   ;

   ;
   ; Parameters defined in Appendix C.1

   param-assign =   ("cm_unused="  (([channel-list] command-type
                                     [f-list]) / sysex-data))

   param-assign =/  ("cm_used="    (([channel-list] command-type
                                     [f-list]) / sysex-data))

   ;
   ; Parameters defined in Appendix C.2

   param-assign =/  ("j_sec="      ("none" / "recj" / *ietf-extension))

   param-assign =/  ("j_update="   ("anchor" / "closed-loop" /
                                    "open-loop" / *ietf-extension))

   param-assign =/  ("ch_default=" (([channel-list] chapter-list
                                     [f-list]) / sysex-data))

   param-assign =/  ("ch_never="   (([channel-list] chapter-list
                                     [f-list]) / sysex-data))

   param-assign =/  ("ch_anchor="  (([channel-list] chapter-list
                                     [f-list]) / sysex-data))

   ;
   ; Parameters defined in Appendix C.3

   param-assign =/  ("tsmode="     ("comex" / "async" / "buffer"))

   param-assign =/  ("linerate="    nonzero-four-octet)

   param-assign =/  ("octpos="      ("first" / "last"))

   param-assign =/  ("mperiod="     nonzero-four-octet)

   ;
   ; Parameter defined in Appendix C.4

   param-assign =/  ("guardtime="     nonzero-four-octet)

   param-assign =/  ("rtp_ptime="     four-octet)

   param-assign =/  ("rtp_maxptime="  four-octet)

   ;
   ; Parameters defined in Appendix C.5

   param-assign =/  ("musicport="     four-octet)

   ;
   ; Parameters defined in Appendix C.6

   param-assign =/  ("chanmask="     ( 1*( 16( "0" / "1" ) )))

   param-assign =/  ("cid="          double-quote cid-block
                                     double-quote)

   param-assign =/  ("inline="       double-quote base-64-block
                                     double-quote)

   param-assign =/  ("multimode="    ("all" / "one"))

   param-assign =/  ("render="       ("synthetic" / "api" / "null" /
                                      "unknown" / *extension))

   param-assign =/  ("rinit="        mime-type "/" mime-subtype)

   param-assign =/  ("smf_cid="      double-quote cid-block
                                     double-quote)

   param-assign =/  ("smf_info="     ("ignore" / "identity" /
                                     "sdp_start" / *extension))

   param-assign =/  ("smf_inline="   double-quote base-64-block
                                     double-quote)

   param-assign =/  ("smf_url="      double-quote uri-element
                                     double-quote)

   param-assign =/  ("subrender="    ("default" / *extension))

   param-assign =/  ("url="          double-quote uri-element
                                     double-quote)

   ;
   ; list definitions for the cm_ command-type
   ;

   command-type    = command-part1 command-part2 command-part3

   command-part1   = (*1"A") (*1"B") (*1"C") (*1"F") (*1"G") (*1"H")

   command-part2   = (*1"J") (*1"K") (*1"M") (*1"N") (*1"P") (*1"Q")

   command-part3   = (*1"T") (*1"V") (*1"W") (*1"X") (*1"Y") (*1"Z")

   ;
   ; list definitions for the ch_ chapter-list
   ;

   chapter-list  =  ch-part1 ch-part2 ch-part3

   ch-part1  = (*1"A") (*1"B") (*1"C") (*1"D") (*1"E") (*1"F") (*1"G")

   ch-part2  = (*1"H") (*1"J") (*1"K") (*1"M") (*1"N") (*1"P") (*1"Q")

   ch-part3  = (*1"T") (*1"V") (*1"W") (*1"X") (*1"Y") (*1"Z")

   ;
   ; list definitions for the ch_ channel-list
   ;

   channel-list       = midi-chan-element *("." midi-chan-element)

   midi-chan-element  = midi-chan / midi-chan-range

   midi-chan-range    = midi-chan "-" midi-chan

                      ; decimal value of left midi-chan
                      ; MUST be strictly less than decimal
                      ; value of right midi-chan

   midi-chan          = %d0-15

   ;
   ; list definitions for the ch_ field list (f-list)
   ;

   f-list             = midi-field-element *("." midi-field-element)

   midi-field-element = midi-field / midi-field-range

   midi-field-range   = midi-field "-" midi-field
                      ;
                      ; decimal value of left midi-field
                      ; MUST be strictly less than decimal
                      ; value of right midi-field

   midi-field         = four-octet
                      ;
                      ; large range accommodates Chapter M
                      ; RPN (0-16383) and NRPN (16384-32767)
                      ; parameters, and Chapter X octet sizes.

   ;
   ; definitions for ch_ sysex-data
   ;

   sysex-data         = "__"  h-list *("_" h-list) "__"

   h-list             = hex-field-element *("." hex-field-element)

   hex-field-element  = hex-octet / hex-field-range

   hex-field-range    = hex-octet "-" hex-octet
                      ;
                      ; hexadecimal value of left hex-octet
                      ; MUST be strictly less than hexadecimal
                      ; value of right hex-octet

   hex-octet          = 2("0" / "1" / "2"/ "3" / "4" /
                          "5" / "6" / "7" / "8" / "9" /
                          "A" / "B" / "C" / "D" / "E" / "F")
                      ;
                      ; rewritten version of hex-octet in [RFC2045]
                      ; (page 23).
                      ; note that a-f are not permitted, only A-F.
                      ; hex-octet values MUST NOT exceed 7F.

   ;
   ; definitions for rinit parameter
   ;

   mime-type          = "audio" / "application"

   mime-subtype       = token
                      ;
                      ; See Appendix C.6.2 for registration
                      ; requirements for rinit type/subtypes.

   ;
   ; definitions for base64 encoding
   ; copied from [RFC4566]

   base-64-block      = *base64-unit [base64-pad]

   base64-unit        =  4base64-char

   base64-pad         =  2base64-char "==" / 3base64-char "="

   base64-char        =  %x41-5A / %x61-7A / %x30-39 / "+" / "/"
                      ;  A-Z, a-z, 0-9, "+" and "/"

   ;
   ; generic rules
   ;

   ietf-extension     = token
                      ;
                      ; ietf-extension may only be defined in
                      ; standards-track RFCs.

   extension          = token
                      ;
                      ; extension may be defined by filing
                      ; a registration with IANA.

   four-octet         = %d0-4294967295
                      ; unsigned encoding of 32-bits

   nonzero-four-octet = %d1-4294967295
                      ; unsigned encoding of 32-bits, ex-zero

   uri-element        = URI-reference
                      ; as defined in [RFC3986]

   double-quote       = %x22

                      ; the double-quote (") character

   token              =  1*token-char
                      ; copied from [RFC4566]

   token-char         =  %x21 / %x23-27 / %x2A-2B / %x2D-2E /
                         %x30-39 / %x41-5A / %x5E-7E
                      ; copied from [RFC4566]

   cid-block          = 1*cid-char

   cid-char           =  token-char
   cid-char           =/  "@"
   cid-char           =/  ","
   cid-char           =/  ";"
   cid-char           =/  ":"
   cid-char           =/  "\"
   cid-char           =/  "/"
   cid-char           =/  "["
   cid-char           =/  "]"
   cid-char           =/  "?"
   cid-char           =/  "="

                      ;
                      ; add back in the tspecials [RFC2045], except for
                      ; double-quote and the non-email safe () <>
                      ; note that "cid" defined above ensures that
                      ; cid-block is enclosed with double-quotes

   ; external references
   ; URI-reference: from [RFC3986]

   ;
   ; End of ABNF

   The mpeg4-generic RTP payload [RFC3640] defines a "mode" parameter
   that signals the type of MPEG stream in use.  We add a new mode
   value, "rtp-midi", using the ABNF rule below:

   ;
   ; mpeg4-generic mode parameter extension
   ;

   mode              =/ "rtp-midi"
                     ; as described in Section 6.2 of this memo

E.  A MIDI Overview for Networking Specialists

   This appendix presents an overview of the MIDI standard, for the
   benefit of networking specialists new to musical applications.
   Implementors should consult [MIDI] for a normative description of
   MIDI.

   Musicians make music by performing a controlled sequence of physical
   movements.  For example, a pianist plays by coordinating a series of
   key presses, key releases, and pedal actions.  MIDI represents a
   musical performance by encoding these physical gestures as a sequence
   of MIDI commands.  This high-level musical representation is compact
   but fragile: one lost command may be catastrophic to the performance.

   MIDI commands have much in common with the machine instructions of a
   microprocessor.  MIDI commands are defined as binary elements.
   Bitfields within a MIDI command have a regular structure and a
   specialized purpose.  For example, the upper nibble of the first
   command octet (the opcode field) codes the command type.  MIDI
   commands may consist of an arbitrary number of complete octets, but
   most MIDI commands are 1, 2, or 3 octets in length.

       ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
       |     Channel Voice Messages     |      Bitfield Pattern      |
       ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
       | NoteOff (end a note)           | 1000cccc 0nnnnnnn 0vvvvvvv |
       |-------------------------------------------------------------|
       | NoteOn (start a note)          | 1001cccc 0nnnnnnn 0vvvvvvv |
       |-------------------------------------------------------------|
       | PTouch (Polyphonic Aftertouch) | 1010cccc 0nnnnnnn 0aaaaaaa |
       |-------------------------------------------------------------|
       | CControl (Controller Change)   | 1011cccc 0xxxxxxx 0yyyyyyy |
       |-------------------------------------------------------------|
       | PChange (Program Change)       | 1100cccc 0ppppppp          |
       |-------------------------------------------------------------|
       | CTouch (Channel Aftertouch)    | 1101cccc 0aaaaaaa          |
       |-------------------------------------------------------------|
       | PWheel (Pitch Wheel)           | 1110cccc 0xxxxxxx 0yyyyyyy |
        -------------------------------------------------------------

                 Figure E.1 -- MIDI Channel Messages

       ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
       |      System Common Messages    |     Bitfield Pattern       |
       ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
       | System Exclusive               | 11110000, followed by a    |
       |                                | list of 0xxxxxx octets,    |
       |                                | followed by 11110111       |
       |-------------------------------------------------------------|
       | MIDI Time Code Quarter Frame   | 11110001 0xxxxxxx          |
       |-------------------------------------------------------------|
       | Song Position Pointer          | 11110010 0xxxxxxx 0yyyyyyy |
       |-------------------------------------------------------------|
       | Song Select                    | 11110011 0xxxxxxx          |
       |-------------------------------------------------------------|
       | Undefined                      | 11110100                   |
       |-------------------------------------------------------------|
       | Undefined                      | 11110101                   |
       |-------------------------------------------------------------|
       | Tune Request                   | 11110110                   |
       |-------------------------------------------------------------|
       | System Exclusive End Marker    | 11110111                   |
        -------------------------------------------------------------

       ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
       |    System Realtime Messages    |     Bitfield Pattern       |
       ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
       | Clock                          | 11111000                   |
       |-------------------------------------------------------------|
       | Undefined                      | 11111001                   |
       |-------------------------------------------------------------|
       | Start                          | 11111010                   |
       |-------------------------------------------------------------|
       | Continue                       | 11111011                   |
       |-------------------------------------------------------------|
       | Stop                           | 11111100                   |
       |-------------------------------------------------------------|
       | Undefined                      | 11111101                   |
       |-------------------------------------------------------------|
       | Active Sense                   | 11111110                   |
       |-------------------------------------------------------------|
       | System Reset                   | 11111111                   |
        -------------------------------------------------------------

                      Figure E.2 -- MIDI System Messages

   Figure E.1 and E.2 show the MIDI command family.  There are three
   major classes of commands: voice commands (opcode field values in the
   range 0x8 through 0xE), system common commands (opcode field 0xF,
   commands 0xF0 through 0xF7), and system real-time commands (opcode
   field 0xF, commands 0xF8 through 0xFF).  Voice commands code the
   musical gestures for each timbre in a composition.  Systems commands
   perform functions that usually affect all voice channels, such as
   System Reset (0xFF).

E.1.  Commands Types

   Voice commands execute on one of 16 MIDI channels, as coded by its
   4-bit channel field (field cccc in Figure E.1).  In most
   applications, notes for different timbres are assigned to different
   channels.  To support applications that require more than 16
   channels, MIDI systems use several MIDI command streams in parallel,
   to yield 32, 48, or 64 MIDI channels.

   As an example of a voice command, consider a NoteOn command (opcode
   0x9), with binary encoding 1001cccc 0nnnnnnn 0aaaaaaa.  This command
   signals the start of a musical note on MIDI channel cccc.  The note
   has a pitch coded by the note number nnnnnnn, and an onset amplitude
   coded by note velocity aaaaaaa.

   Other voice commands signal the end of notes (NoteOff, opcode 0x8),
   map a specific timbre to a MIDI channel (PChange, opcode 0xC), or set
   the value of parameters that modulate the timbral quality (all other
   voice commands).  The exact meaning of most voice channel commands
   depends on the rendering algorithms the MIDI receiver uses to
   generate sound.  In most applications, a MIDI sender has a model (in
   some sense) of the rendering method used by the receiver.

   System commands perform a variety of global tasks in the stream,
   including "sequencer" playback control of pre-recorded MIDI commands
   (the Song Position Pointer, Song Select, Clock, Start, Continue, and
   Stop messages), SMPTE time code (the MIDI Time Code Quarter Frame
   command), and the communication of device-specific data (the System
   Exclusive messages).

E.2.  Running Status

   All MIDI command bitfields share a special structure: the leading bit
   of the first octet is set to 1, and the leading bit of all subsequent
   octets is set to 0.  This structure supports a data compression
   system, called running status [MIDI], that improves the coding
   efficiency of MIDI.

   In running status coding, the first octet of a MIDI voice command may
   be dropped if it is identical to the first octet of the previous MIDI
   voice command.  This rule, in combination with a convention to
   consider NoteOn commands with a null third octet as NoteOff commands,
   supports the coding of note sequences using two octets per command.

   Running status coding is only used for voice commands.  The presence
   of a system common message in the stream cancels running status mode
   for the next voice command.  However, system real-time messages do
   not cancel running status mode.

E.3.  Command Timing

   The bitfield formats in Figures E.1 and E.2 do not encode the
   execution time for a command.  Timing information is not a part of
   the MIDI command syntax itself; different applications of the MIDI
   command language use different methods to encode timing.

   For example, the MIDI command set acts as the transport layer for
   MIDI 1.0 DIN cables [MIDI].  MIDI cables are short asynchronous
   serial lines that facilitate the remote operation of musical
   instruments and audio equipment.  Timestamps are not sent over a MIDI
   1.0 DIN cable.  Instead, the standard uses an implicit "time of
   arrival" code.  Receivers execute MIDI commands at the moment of
   arrival.

   In contrast, Standard MIDI Files (SMFs, [MIDI]), a file format for
   representing complete musical performances, add an explicit timestamp
   to each MIDI command, using a delta encoding scheme that is optimized
   for statistics of musical performance.  SMF timestamps usually code
   timing using the metric notation of a musical score.  SMF meta-events
   are used to add a tempo map to the file, so that score beats may be
   accurately converted into units of seconds during rendering.

E.4.  AudioSpecificConfig Templates for MMA Renderers

   In Section 6.2 and Appendix C.6.5, we describe how session
   descriptions include an AudioSpecificConfig data block to specify a
   MIDI rendering algorithm for mpeg4-generic RTP MIDI streams.

   The bitfield format of AudioSpecificConfig is defined in [MPEGAUDIO].
   StructuredAudioSpecificConfig, a key data structure coded in
   AudioSpecificConfig, is defined in [MPEGSA].

   For implementors wishing to specify Structured Audio renderers, a
   full understanding of [MPEGSA] and [MPEGAUDIO] is essential.
   However, many implementors will limit their rendering options to the
   two MIDI Manufacturers Association renderers that may be specified in

   AudioSpecificConfig: General MIDI (GM, [MIDI]) and Downloadable
   Sounds 2 (DLS 2, [DLS2]).

   To aid these implementors, we reproduce the AudioSpecificConfig
   bitfield formats for a GM renderer and a DLS 2 renderer below.  We
   have checked these bitfields carefully and believe they are correct.
   However, we stress that the material below is informative, and that
   [MPEGAUDIO] and [MPEGSA] are the normative definitions for
   AudioSpecificConfig.

   As described in Section 6.2, a minimal mpeg4-generic session
   description encodes the AudioSpecificConfig binary bitfield as a
   hexadecimal string (whose format is defined in [RFC3640]) that is
   assigned to the "config" parameter.  As described in Appendix C.6.3,
   a session description that uses the render parameter encodes the
   AudioSpecificConfig binary bitfield as a Base64-encoded string
   assigned to the "inline" parameter, or in the body of an HTTP URL
   assigned to the "url" parameter.

   Below, we show a simplified binary AudioSpecificConfig bitfield
   format, suitable for sending and receiving GM and DLS 2 data:

       0                   1                   2                   3
       0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
      | AOTYPE  |FREQIDX|CHANNEL|SACNK|  FILE_BLK 1 (required) ...    |
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
      |1|SACNK|              FILE_BLK 2 (optional) ...                |
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
      |  ...  |1|SACNK| FILE_BLK N (optional) ...                     |
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
      |0|0|        (first "0" bit terminates FILE_BLK list)
      +-+-+

                  Figure E.3 -- Simplified AudioSpecificConfig

   The 5-bit AOTYPE field specifies the Audio Object Type as an unsigned
   integer.  The legal values for use with mpeg4-generic RTP MIDI
   streams are "15" (General MIDI), "14" (DLS 2), and "13" (Structured
   Audio).  Thus, receivers that do not support all three mpeg4-generic
   renderers may parse the first 5 bits of an AudioSpecificConfig coded
   in a session description and reject sessions that specify unsupported
   renderers.

   The 4-bit FREQIDX field specifies the sampling rate of the renderer.
   We show the mapping of FREQIDX values to sampling rates in Figure
   E.4.  Senders MUST specify a sampling frequency that matches the RTP
   clock rate, if possible; if not, senders MUST specify the escape

   value.  Receivers MUST consult the RTP clock parameter for the true
   sampling rate if the escape value is specified.

                       FREQIDX    Sampling Frequency

                         0x0            96000
                         0x1            88200
                         0x2            64000
                         0x3            48000
                         0x4            44100
                         0x5            32000
                         0x6            24000
                         0x7            22050
                         0x8            16000
                         0x9            12000
                         0xa            11025
                         0xb             8000
                         0xc          reserved
                         0xd          reserved
                         0xe          reserved
                         0xf         escape value

                     Figure E.4 -- FreqIdx encoding

   The 4-bit CHANNEL field specifies the number of audio channels for
   the renderer.  The values 0x1 to 0x5 specify 1 to 5 audio channels;
   the value 0x6 specifies 5+1 surround sound, and the value 0x7
   specifies 7+1 surround sound.  If the rtpmap line in the session
   description specifies one of these formats, CHANNEL MUST be set to
   the corresponding value.  Otherwise, CHANNEL MUST be set to 0x0.

   The CHANNEL field is followed by a list of one or more binary file
   data blocks.  The 3-bit SACNK field (the chunk_type field in class
   StructuredAudioSpecificConfig, defined in [MPEGSA]) specifies the
   type of each data block.

   For General MIDI, only Standard MIDI Files may appear in the list
   (SACNK field value 2).  For DLS 2, only Standard MIDI Files and DLS 2
   RIFF files (SACNK field value 4) may appear.  For both of these file
   types, the FILE_BLK field has the format shown in Figure E.5: a 32-
   bit unsigned integer value (FILE_LEN) coding the number of bytes in
   the SMF or RIFF file, followed by FILE_LEN bytes coding the file
   data.

       0                   1                   2                   3
       0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
      |     FILE_LEN (32-bit, a byte count SMF file or RIFF file)     |
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
      |  FILE_DATA (file contents, a list of FILE_LEN bytes) ...      |
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

                  Figure E.5 -- The FILE_BLK field format

   Note that several files may follow CHANNEL field.  The "1" constant
   fields in Figure E.3 code the presence of another file; the "0"
   constant field codes the end of the list.  The final "0" bit in
   Figure E.3 codes the absence of special coding tools (see [MPEGAUDIO]
   for details).  Senders not using these tools MUST append this "0"
   bit; receivers that do not understand these coding tools MUST ignore
   all data following a "1" in this position.

   The StructuredAudioSpecificConfig bitfield structure requires the
   presence of one FILE_BLK.  For mpeg4-generic RTP MIDI use of DLS 2,
   FILE_BLKs MUST code RIFF files or SMF files.  For mpeg4-generic RTP
   MIDI use of General MIDI, FILE_BLKs MUST code SMF files.  By default,
   this SMF will be ignored (Appendix C.6.4.1).  In this default case, a
   GM StructuredAudioSpecificConfig bitfield SHOULD code a FILE_BLK
   whose FILE_LEN is 0, and whose FILE_DATA is empty.

   To complete this appendix, we derive the
   StructuredAudioSpecificConfig that we use in the General MIDI session
   examples in this memo.  Referring to Figure E.3, we note that for GM,
   AOTYPE = 15.  Our examples use a 44,100 Hz sample rate (FREQIDX = 4)
   and are in mono (CHANNEL = 1).  For GM, a single SMF is encoded
   (SACNK = 2), using the SMF shown in Figure E.6 (a 26 byte file).

               --------------------------------------------
              |  MIDI File = <Header Chunk> <Track Chunk>  |
               --------------------------------------------

   <Header Chunk> = <chunk type> <length>     <format> <ntrks> <divsn>
                    4D 54 68 64  00 00 00 06  00 00    00 01   00 60

   <Track Chunk> = <chunk type>  <length>     <delta-time> <end-event>
                   4D 54 72 6B   00 00 00 04  00           FF 2F 00

            Figure E.6 -- SMF file encoded in the example

   Placing these constants in binary format into the data structure
   shown in Figure E.3 yields the constant shown in Figure E.7.

       0                   1                   2                   3
       0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
      |0 1 1 1 1|0 1 0 0|0 0 0 1|0 1 0|0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0|
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
      |0 0 0 0 0 0 0 0 0 0 0 1 1 0 1 0|0 1 0 0|1 1 0 1|0 1 0 1|0 1 0 0|
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
      |0 1 1 0|1 0 0 0|0 1 1 0|0 1 0 0|0 0 0 0|0 0 0 0|0 0 0 0|0 0 0 0|
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
      |0 0 0 0|0 0 0 0|0 0 0 0|0 1 1 0|0 0 0 0|0 0 0 0|0 0 0 0|0 0 0 0|
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
      |0 0 0 0|0 0 0 0|0 0 0 0|0 0 0 1|0 0 0 0|0 0 0 0|0 1 1 0|0 0 0 0|
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
      |0 1 0 0|1 1 0 1|0 1 0 1|0 1 0 0|0 1 1 1|0 0 1 0|0 1 1 0|1 0 1 1|
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
      |0 0 0 0|0 0 0 0|0 0 0 0|0 0 0 0|0 0 0 0|0 0 0 0|0 0 0 0|0 1 1 0|
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
      |0 0 0 0|0 0 0 0|1 1 1 1|1 1 1 1|0 0 1 0|1 1 1 1|0 0 0 0|0 0 0 0|
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
      |0|0|
      +-+-+

            Figure E.7 -- AudioSpecificConfig used in GM examples

   Expressing this bitfield as an ASCII hexadecimal string yields:

      7A0A0000001A4D546864000000060000000100604D54726B0000000600FF2F000

   This string is assigned to the "config" parameter in the minimal
   mpeg4-generic General MIDI examples in this memo (such as the example
   in Section 6.2).  Expressing this string in Base64 [RFC2045] yields:

      egoAAAAaTVRoZAAAAAYAAAABAGBNVHJrAAAABgD/LwAA

   This string is assigned to the "inline" parameter in the General MIDI
   example shown in Appendix C.6.5.

References

Normative References

   [MIDI]      MIDI Manufacturers Association.  "The Complete MIDI 1.0
               Detailed Specification", 1996.

   [RFC3550]   Schulzrinne, H., Casner, S., Frederick, R., and V.
               Jacobson, "RTP: A Transport Protocol for Real-Time
               Applications", STD 64, RFC 3550, July 2003.

   [RFC3551]   Schulzrinne, H. and S. Casner, "RTP Profile for Audio and
               Video Conferences with Minimal Control", STD 65, RFC
               3551, July 2003.

   [RFC3640]   van der Meer, J., Mackie, D., Swaminathan, V., Singer,
               D., and P. Gentric, "RTP Payload Format for Transport of
               MPEG-4 Elementary Streams", RFC 3640, November 2003.

   [MPEGSA]    International Standards Organization.  "ISO/IEC 14496
               MPEG-4", Part 3 (Audio), Subpart 5 (Structured Audio),
               2001.

   [RFC4566]   Handley, M., Jacobson, V., and C. Perkins, "SDP: Session
               Description Protocol", RFC 4566, July 2006.

   [MPEGAUDIO] International Standards Organization.  "ISO 14496 MPEG-
               4", Part 3 (Audio), 2001.

   [RFC2045]   Freed, N. and N. Borenstein, "Multipurpose Internet Mail
               Extensions (MIME) Part One: Format of Internet Message
               Bodies", RFC 2045, November 1996.

   [DLS2]      MIDI Manufacturers Association.  "The MIDI Downloadable
               Sounds Specification", v98.2, 1998.

   [RFC4234]   Crocker, D. and P. Overell, "Augmented BNF for Syntax
               Specifications: ABNF", RFC 4234, October 2005.

   [RFC2119]   Bradner, S., "Key words for use in RFCs to Indicate
               Requirement Levels", BCP 14, RFC 2119, March 1997.

   [RFC3711]   Baugher, M., McGrew, D., Naslund, M., Carrara, E., and K.
               Norrman, "The Secure Real-time Transport Protocol
               (SRTP)", RFC 3711, March 2004.

   [RFC3264]   Rosenberg, J. and H. Schulzrinne, "An Offer/Answer Model
               with Session Description Protocol (SDP)", RFC 3264, June
               2002.

   [RFC3986]   Berners-Lee, T., Fielding, R., and L. Masinter, "Uniform
               Resource Identifier (URI): Generic Syntax", STD 66, RFC
               3986, January 2005.

   [RFC2616]   Fielding, R., Gettys, J., Mogul, J., Frystyk, H.,
               Masinter, L., Leach, P., and T. Berners-Lee, "Hypertext
               Transfer Protocol -- HTTP/1.1", RFC 2616, June 1999.

   [RFC3388]   Camarillo, G., Eriksson, G., Holler, J., and H.
               Schulzrinne, "Grouping of Media Lines in the Session
               Description Protocol (SDP)", RFC 3388, December 2002.

   [RP015]     MIDI Manufacturers Association.  "Recommended Practice
               015 (RP-015): Response to Reset All Controllers", 11/98.

   [RFC4288]   Freed, N. and J. Klensin, "Media Type Specifications and
               Registration Procedures", BCP 13, RFC 4288, December
               2005.

   [RFC3555]   Casner, S. and P. Hoschka, "MIME Type Registration of RTP
               Payload Formats", RFC 3555, July 2003.

Informative References

   [NMP]       Lazzaro, J. and J. Wawrzynek.  "A Case for Network
               Musical Performance", 11th International Workshop on
               Network and Operating Systems Support for Digital Audio
               and Video (NOSSDAV 2001) June 25-26, 2001, Port
               Jefferson, New York.

   [GRAME]     Fober, D., Orlarey, Y. and S. Letz.  "Real Time Musical
               Events Streaming over Internet", Proceedings of the
               International Conference on WEB Delivering of Music 2001,
               pages 147-154.

   [RFC3261]   Rosenberg, J., Schulzrinne, H., Camarillo, G., Johnston,
               A., Peterson, J., Sparks, R., Handley, M., and E.
               Schooler, "SIP: Session Initiation Protocol", RFC 3261,
               June 2002.

   [RFC2326]   Schulzrinne, H., Rao, A., and R. Lanphier, "Real Time
               Streaming Protocol (RTSP)", RFC 2326, April 1998.

   [ALF]       Clark, D. D. and D. L. Tennenhouse. "Architectural
               considerations for a new generation of protocols",
               SIGCOMM Symposium on Communications Architectures and
               Protocols , (Philadelphia, Pennsylvania), pp. 200--208,
               IEEE, Sept. 1990.

   [RFC4696]   Lazzaro, J. and J. Wawrzynek, "An Implementation Guide
               for RTP MIDI", RFC 4696, November 2006.

   [RFC2205]   Braden, R., Zhang, L., Berson, S., Herzog, S., and S.
               Jamin, "Resource ReSerVation Protocol (RSVP) -- Version 1
               Functional Specification", RFC 2205, September 1997.

   [RFC4288]   Freed, N. and J. Klensin, "Media Type Specifications and
               Registration Procedures", BCP 13, RFC 4288, December
               2005.

   [RFC4289]   Freed, N. and J. Klensin, "Multipurpose Internet Mail
               Extensions (MIME) Part Four: Registration Procedures",
               BCP 13, RFC 4289, December 2005.

   [RFC4571]   Lazzaro, J. "Framing Real-time Transport Protocol (RTP)
               and RTP Control Protocol (RTCP) Packets over Connection-
               Oriented Transport", RFC 4571, July 2006.

   [RFC2818]   Rescorla, E., "HTTP Over TLS", RFC 2818, May 2000.

   [SPMIDI]    MIDI Manufacturers Association.  "Scalable Polyphony
               MIDI, Specification and Device Profiles", Document
               Version 1.0a, 2002.

   [LCP]       Apple Computer. "Logic 7 Dedicated Control Surface
               Support", Appendix B.  Product manual available from
               www.apple.com.

Authors' Addresses

   John Lazzaro (corresponding author)
   UC Berkeley
   CS Division
   315 Soda Hall
   Berkeley CA 94720-1776

   EMail: lazzaro@cs.berkeley.edu

   John Wawrzynek
   UC Berkeley
   CS Division
   631 Soda Hall
   Berkeley CA 94720-1776

   EMail: johnw@cs.berkeley.edu

Full Copyright Statement

   Copyright (C) The IETF Trust (2006).

   This document is subject to the rights, licenses and restrictions
   contained in BCP 78, and except as set forth therein, the authors
   retain all their rights.

   This document and the information contained herein are provided on an
   "AS IS" basis and THE CONTRIBUTOR, THE ORGANIZATION HE/SHE REPRESENTS
   OR IS SPONSORED BY (IF ANY), THE INTERNET SOCIETY, THE IETF TRUST,
   AND THE INTERNET ENGINEERING TASK FORCE DISCLAIM ALL WARRANTIES,
   EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO ANY WARRANTY THAT
   THE USE OF THE INFORMATION HEREIN WILL NOT INFRINGE ANY RIGHTS OR ANY
   IMPLIED WARRANTIES OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR
   PURPOSE.

Intellectual Property

   The IETF takes no position regarding the validity or scope of any
   Intellectual Property Rights or other rights that might be claimed to
   pertain to the implementation or use of the technology described in
   this document or the extent to which any license under such rights
   might or might not be available; nor does it represent that it has
   made any independent effort to identify any such rights.  Information
   on the procedures with respect to rights in RFC documents can be
   found in BCP 78 and BCP 79.

   Copies of IPR disclosures made to the IETF Secretariat and any
   assurances of licenses to be made available, or the result of an
   attempt made to obtain a general license or permission for the use of
   such proprietary rights by implementers or users of this
   specification can be obtained from the IETF on-line IPR repository at
   http://www.ietf.org/ipr.

   The IETF invites any interested party to bring to its attention any
   copyrights, patents or patent applications, or other proprietary
   rights that may cover technology that may be required to implement
   this standard.  Please address the information to the IETF at
   ietf-ipr@ietf.org.

Acknowledgement

   Funding for the RFC Editor function is currently provided by the
   Internet Society.

 

User Contributions:

Comment about this RFC, ask questions, or add new information about this topic: