Patent application title: DIRECT DIGITAL SYNTHESIZER FOR REFERENCE FREQUENCY GENERATION
Anthony L. Tsangaropoulos (San Carlos, CA, US)
David Francois Guillou (Pittsburgh, PA, US)
Cymatics Laboratories Corp.
IPC8 Class: AH03K2100FI
Class name: Frequency or repetition rate conversion or control of output rectangular waveform frequency division
Publication date: 2013-01-24
Patent application number: 20130021069
A direct digital frequency synthesizer includes a multi-modulus divider,
a numerically controlled oscillator and a programmable delay generator.
The divider receives an input clock having an input pulse frequency and
outputs some integer fraction of those pulses at an instantaneous
frequency that is some integer fraction (1/P) of the input frequency. The
divider selects between at least two ratios of P (1/P or 1/P+1) in
response to a signal from the oscillator. The oscillator receives a value
which is the accumulator increment (i.e. the number of divided pulse
edges) required before an overflow occurs that causes the divider to
change divider ratios in response to receiving an overflow signal. The
oscillator also outputs both the overflow signal and a delay signal to
the delay generator. The delay signal contains phase-dithering noise that
is induced by input into the accumulator of an increment generated from a
pseudo-random noise generator.
1. A method for providing an output signal with an adjusted frequency,
the method comprising: receiving a clock reference signal at a first
input frequency; dividing the input frequency to generate an intermediate
signal with an adjusted frequency, wherein the intermediate signal
comprises a plurality of pulses; providing a phase value from an
oscillator in response to receiving an accumulator increment value;
clocking a latch circuit using the intermediate signal with the adjusted
frequency; outputting, from the numerically controlled oscillator, a
delay value and an overflow value; selecting between at least a first
divider ratio and a second divider ratio based on the overflow value; and
in response to receiving at least the overflow value and the delay value,
calculating a delay time for application to each pulse in the
intermediate signal to provide an output signal with a final adjusted
CROSS-REFERENCE TO RELATED APPLICATIONS
 The present application is a continuation application of U.S. patent application Ser. No. 12/800,808, filed May 21, 2010 and scheduled to issue as U.S. Pat. No. 8,242,850 on Aug. 14, 2012, which is a continuation-in-part application of U.S. patent application Ser. No. 12/229,948 filed Aug. 28, 2008, which issued as U.S. Pat. No. 7,724,097 on May 25, 2010, the disclosures of which are incorporated by reference herein.
BACKGROUND OF THE INVENTION
 1. Field of the Invention
 The present invention provides frequency synthesizer circuit architectures that are particularly adapted for use in computation, control and communication applications.
 2. Description of Related Art
 Low frequency (32 KHz-300 MHz) clock or timing signals are employed in electronic devices for many different applications. A typical reference oscillator may use a quartz resonator or another resonator which may operate on a fundamental frequency (less than about 40 MHz) or an overtone mode of oscillation (about 30 to 300 MHz). However, certain electronic devices (e.g., mobile communication devices) require higher frequency (500 MHz-3 GHz) timing signals which either cannot be generated directly by quartz resonators or other electro-acoustic resonators, or are prohibitively expensive to generate using such resonators. Also, high-frequency oscillators that use non-acoustic resonator technology (e.g., an inductor/capacitor resonant tank) cannot achieve the low phase noise or low power consumption required by many applications. Furthermore, conventional oscillator solutions may be too costly or too bulky for certain product applications and/or fail to provide a sufficient variety of output frequencies with sufficiently low noise.
 An alternate approach to generating a reference signal is frequency synthesis, which can be performed either indirectly, by employing a phase lock loop ("PLL"), or directly, by employing a direct digital synthesizer ("DDS").
 In PLL frequency synthesis, a reference oscillator operating at a relatively low frequency (fREF) is employed to generate a higher output frequency (fOUT>fREF) with a desired accuracy. To accomplish this synthesis, the frequency of a voltage controlled oscillator ("VCO") is adjusted until the phase error between the reference oscillator and the VCO is minimized. The VCO is adjusted by a feedback loop that compares the frequency and phase of the VCO to that of the reference oscillator. When the loop settles, the VCO frequency closely tracks both the frequency and phase of the reference signal according to a predetermined harmonic relationship defined by the division ratios of the dividers used in the PLL circuit, e.g. fVCO/N=fREF/M. Non-harmonic scaling may be obtained by rapidly switching the divider between adjacent ratios P and P+1 with the aid of a controller often employing a delta-sigma modulator loop. The instantaneous output frequency alternates between fREF*P/M and fREF*(P+1)/M, and the average frequency equals fREF*(P+N)/M where N is a non-integer value between 0 and 1. Simultaneously with the divider modulus control, a phase correction is also applied to the divider output signal before it is compared with the reference signal to produce an error signal that is low-pass-filtered and applied to the VCO input. This implementation is also known as a fractional-N synthesizer.
 The output signal of a fractional-N PLL may be degraded by the presence of spurious signals and noise that result from the constant switching of the P divider. These undesired signals must be minimized to meet the requirements of practical applications, which results in increased power consumption and loop settling time.
 In typical DDS architectures, a higher reference frequency generator is used with a numerically controlled oscillator ("NCO") to produce an output signal having controlled frequency and phase. The DDS output frequency range and resolution is mainly determined by the reference frequency and the length of the NCO word. As a result, DDS circuits that deliver a higher synthesized output frequency tend to have higher power consumption.
 While existing phase interpolating DDS architectures may provide generation of a wide range of output frequencies from a single reference oscillator, improved DDS architectures are desired.
SUMMARY OF THE INVENTION
 In accordance with aspects of the present invention, DDS synthesizer architecture derive a signal with selectable output frequency ("fOUT") from an oscillator signal with higher fixed frequency ("fOSC") using a combination of a multi-modulus divider providing at least two possible modulii for reducing the frequency of an input signal to achieve the desired frequency for fOUT and a variable delay to achieve a desired instantaneous phase for fOUT. The resultant fOUT signal may then be used to clock or drive circuitry in a user device that requires a lower reference frequency.
 In one embodiment the frequency synthesizer has an input for receiving a clock reference signal at a first frequency fosc. The direct digital synthesizer (DDS) architecture has a multi-modulus divider for dividing the frequency of the input signal to provide an intermediate signal. Multi-modulus, as used herein, refers to the multiple programmable division ratios used to divide the frequency of the input signal into a lower frequency. The embodiment also has a numerically controlled oscillator (NCO) having an accumulator receiving an accumulator increment (also referred to as a phase increment) value. The NCO provides a phase value to a latch circuit. The latch circuit receives and is clocked by the intermediate signal wherein the numerically controlled oscillator outputs a delay value comprised of one or more bits and an overflow signal comprised of one or more bits wherein said overflow signal is provided to the multi-modulus divider to select between at least a first divider ratio and a second divider ratio.
 The DDS also has a programmable delay generator. In one embodiment, the delay generator receives the delay signal having one or more bits and the overflow signal containing one bit. In preferred embodiments the delay generator receives one or more reference clock signals (e.g. the clock input signal to the multi-modulus divider and or the signal with the intermediate adjusted frequency output from the multi-modulus divider. From the inputs, the delay generator calculates a delay time for application to each pulse in the signal with an intermediate adjusted frequency to provide an output signal which achieves improved phase noise and timing jitter performance (fout).
 In one embodiment the accumulator of the NCO receives "dithered" accumulator increment values from a Pseudo-Random Noise generator. These values introduce a noise-like phase dither to the output of the delay generator. This noise-like phase addition spreads spurious signals that would exist in the DDS output signal if dithering were not performed.
 In certain preferred embodiments discussed herein, the multi-modulus divider selects between first and second divider ratios, while in other preferred embodiments the multi-modulus divider selects between a first divider ratio, a second divider ratio and a third divider ratio. In some instances selected embodiments also have one or more programmable dividers. In these embodiments, the programmable divider operates on the signal input into the multi-modulus divider or on the intermediate signal output from the programmable divider. If the former, the programmable pre-multi-modulus divider reduces the overall power consumption by providing a lower frequency signal to the multi-modulus divider. If the latter, the programmable post-multi-modulus divider provides a wide range of division ratios by which the frequency of the signal input to the multi-modulus divider is divided.
BRIEF DESCRIPTION OF THE DRAWINGS
 The foregoing and other objects and advantages of the invention will be appreciated more fully from the following further description thereof, with reference to the accompanying drawings, wherein:
 FIG. 1 illustrates a functional block diagram of a DDS architecture in accordance with aspects of the invention.
 FIG. 2 illustrates a delay generator that can be employed with aspects of the present invention.
 FIG. 3 shows the waveform timing relationships for the circuit of FIG. 2.
 FIG. 4 is an exemplary timing diagram for the direct digital synthesis block diagram of FIG. 1.
 FIGS. 5a and 5b are tables illustrating the behavior of a 3 and 4-bit NCO DDS systems in accordance with aspects of the present invention.
 FIG. 6 illustrates an alternative embodiment of the invention.
 FIGS. 7a and 7b show details of an exemplary frequency plan of a preferred embodiment of the present invention.
 FIG. 8a shows the role of the fosc Offset setting of the reference oscillator in the frequency planning of a preferred embodiment.
 FIG. 8b illustrates the functionality of a MCpolarity setting in accordance with aspects of the present invention.
 FIG. 9 illustrates an alternate embodiment of the invention that employs a digitally generated randomization signal to create a controlled spread-spectrum reference frequency output.
 FIG. 10 illustrates yet another embodiment of the invention.
 FIG. 11 illustrates another alternate embodiment of the invention that utilizes a triple-modulus divider.
 FIGS. 12a-c illustrate the operation of one embodiment of FIG. 11.
 FIG. 13 illustrates the architecture of a further embodiment of the present invention.
 FIG. 14 illustrates a variation of the embodiment shown in FIG. 13.
 FIG. 15a illustrates an alternative embodiment of the invention in which a non-zero-mean pseudo-random noise generator provides a non-zero-mean dither signal to the accumulator.
 FIG. 15b illustrates a detail of the embodiment presented in FIG. 15b. where the K tuning word is split to provide input to the non-zero-mean pseudo-random noise generator.
 FIG. 16a illustrates an alternate embodiment of the invention that features a Controlled PRN Generator.
 FIG. 16b illustrates the Controlled PRN Generator employed in FIG. 16a.
 The present invention is described in terms of several embodiments. In describing these embodiments, including the drawings, specific terminology will be used for the sake of clarity. Also, in the discussion of certain mathematical relationships, certain variable values are discussed by way of illustration. However, the invention is not intended to be limited to the specific embodiments described below.
 FIG. 1 is a schematic illustration of a multi-modulus divider direct digital synthesizer ("DDS") architecture 10 in accordance with aspects of the present invention. This exemplary DDS architecture 10 includes a reference oscillator 12, a multi-modulus divider 14, an NCO 16 and a delay generator 18, which is also referred to herein as a "phase interpolator."
 Reference oscillator 12 produces a periodic waveform with a frequency fosc. As shown in the figure, fOSC is provided to the multi-modulus divider 14 and the delay generator 18. The reference oscillator 12 may, by way of example, utilize a bulk acoustic wave resonator (e.g., FBAR-type or SMR-type), a micromechanical or nanomechanical resonator, a dielectric resonator, an LC tank or a quartz resonator.
 The multi-modulus divider 14 is illustrated as a dual modulus divider which is operable to switch between division ratios P and P+1 to modify the received fOSC signal. As will be discussed in detail below, the division ratio is controlled by the overflow output of the NCO. This overflow output is also referred to as modulus control ("MC"). The multi-modulus divider 14 is preferably synchronous with the oscillator frequency fOSC.
 The NCO 16 includes an adder or accumulator 20 and a holding circuit/register 22. The holding circuit 22 may be, for example, a bank of delay flip-flops or latches. The NCO 16 may also include a look-up table and other phase control circuits (not shown).
 The delay generator 18 generally has a programmable delay circuit 24 whose function is to delay individual pulse edges based on a control word. The delay generator 18 also includes a delay control circuit 26 that receives various timing signals along with the NCO output word and provides the control word to the delay circuit 24. The delay generator 18 is used to delay the phase of a resultant output signal relative to the phase the signal input to the delay generator 18.
 As shown in FIG. 1, in response to the input signal fOSC, the multi-modulus programmable divider 14 provides a clocking signal VP to the NCO 16 and the phase interpolator 18. The frequency fVp of signal VP is related to fOSC as follows: fVp=fOSC/Pset where Pset can be P or P+1 as determined by the modulus control signal MC output from the NCO 16.
 The NCO 16, in particular holding circuit 22, is clocked by the output VP of the multi-modulus divider 14. As further shown in FIG. 1, a digital tuning word K is provided as an input to the accumulator 20 of the NCO 16. The holding circuit 22 of NCO 16 outputs an overflow value ("OVF") or MC as well as an output word NA. The input word K sets the value by which the accumulator output value ("A") of word length NA is incremented for each VP pulse generated by the multi-modulus programmable divider 14. The value D is the portion of the accumulator output value A that is output from the NCO 16. D is used to set the delay of the delay generator/phase interpolator 18. The word length of D is ND. As described in further detail ND≦NA (NA is the word length of value A).
 In particular, the NCO 16, clocked at fVp, computes the phase value of the synthesized signal. Tuning word K, provided to accumulator 20 of the NCO 16, specifies how many VP periods are counted by the accumulator 20 for each fOUT OUT period. The holding circuit 22 in the NCO 16 updates the phase value calculated by the accumulator 20 at intervals of 1/fVp. The delay value (as represented by D and which, in certain embodiments, is a truncated word of that output from the holding circuit) and the overflow signal OVF are provided to the delay generator 18 to generate a residual phase correction via delay control 26.
 The instantaneous frequency of signal VP oscillates between fOSC/P and fOSC/(P+1), where P is the multi-modulus programmable divider base value, depending on the value of the modulus control signal MC. It can be shown that for NA=ND, the average value of the output frequency fOUT is as follows:
where P is the multi-modulus programmable divider base value, K is the programmed NCO accumulator increment value as described above and NA is the NCO word length. The ratio K/2NA can range from 0 to 1-(2-NA). As a result, the instantaneous frequency deviation of signal VP at the nth pulse instant is:
ΔfVp(n)=fOUTideal-fVp(n)=fOSC*[1/(P+K/2N.- sup.A)-1/Pn]=fOSC/Pn*[Pn-(P+K/2NA)]/(P+K/2.s- up.NA). (3)
 It is easy to prove that the maximum error occurs for K=2.sup.(NA-1). An example of this is illustrated in FIG. 5B discussed below. For large NA values, this is -fOSC/(2P2+P) for Pn=P, or fOSC/(2P2+3P+1) for Pn=P+1.
 Holding circuit 22 in NCO 16 latches at least the NA bits of the accumulator output word A. It is advantageous if the holding circuit 22 also latches the overflow signal bit of the accumulator 20. The holding circuit 22 is latched by the signal VP. A portion of the accumulator output word, which is referred to herein as delay value D.sub.(n-1), is provided to the delay control circuit 26 of the delay generator 18 along with the overflow value OVF, with a delay of one VP pulse period. The NCO phase value at each overflow instant represents the residual phase portion of the accumulator output word A and is used to provide a phase correction to the delay control circuit 26 while the overflow pulse OVF (MC) controls the modulus of multi-modulus divider 14. It should be noted that the edges of the VP pulse train are delayed by the programmable delay generator 18 based on the phase value originating from the NCO 16.
 The delay generator 18 may be implemented in different ways. Examples of delay generator implementations are described in U.S. Pat. No. 6,188,261 entitled "Programmable Delay Generator and Application Circuits Having Said Delay Generator" which issued on Feb. 13, 2001 and is incorporated by reference herein. One exemplary implementation of such circuits is shown in FIG. 2. FIG. 2 shows a specific implementation of a delay generator employing an integrator 50 that starts a programmable-slope ramping voltage 52 VRamp (e.g. by charging a capacitor with a constant current) after it receives a start pulse (Pulse In). The ramp slope is programmable via a slope control word 54 that determines the delay control range. Signals VP and Reset are fed to an integrator 56 in addition to the slope control word 54. Signal VP starts the integrator ramp and signal reset forces it to assume its initial value.
 Before the input pulse is applied, a Delay-Set word (derived from delay word D) is input to D/A converter 58 which in turn provides a threshold voltage (VThreshold). The voltage ramp VRamp 52 is compared with the threshold voltage VThreshold by comparator 60. An output of the comparator 60 is provided to a one-shot multivibrator 62. The resultant output of one-shot multivibrator 62 (VD) gets triggered each time there is a crossover of VD below VThreshold. This crossover determines the delay (tD) of the output pulse referenced to the input edge. The corresponding waveforms are shown in FIG. 3.
 Returning to the embodiment of FIG. 1, for the DDS to operate as desired, the applied delay should be a fraction of the fOSC period. At the nth VP pulse, the applied delay should ideally be Dn-1/2NA*1/fOSC, where Dn-1 was the value of the NCO output word at the previous overflow (n-1th VP pulse). Depending on the actual implementation of the DDS, ND can equal NA or just be a fraction of it (i.e. NA≧ND). In any case, the words D and A are related as follows:
 Therefore, the correction delay added to the nth VP pulse edge is Tc(n)=D.sub.(n-1)/2ND*1/fOSC. If NA>ND, the residual error ΔTcorr=[D.sub.(n-1)/2ND-A.sub.(n-1)/2NA]*1- /fOSC results in deterministic quantization noise, and therefore deterministic frequency jitter on the DDS output signal. One skilled in the art would realize that the operation of the overall embodiment of FIG. 1 remains unchanged if the delay has one or more but generally less than P-2 fOSC periods added to all of the delay values.
 In addition to the selected portion of the accumulated phase value A and OVF, the delay control circuit also could receive signal fOSC from the reference oscillator 12 as this could enhance the delay accuracy. VP is also optionally provided to the delay generator circuit 18. Since the more timing information that is provided to the delay generator, the better it will perform, FIG. 1 illustrates the delay control with inputs of VP and fOSC. The delay control 26 uses the truncated accumulator output value D, the value OVF and optionally, VP and fOSC to perform delay correction on VP that essentially shifts the VP pulses by a fraction of the fOSC period according to the previous NCO output word. The resultant signal VD is a phase corrected signal.
 The output frequency is given by the following relationship: fOUT=fOSC/(P+K/2NA) where fOSC is the reference oscillator frequency, P is the dual modulus divider base ratio, K is the NCO accumulator increment, and NA is the NCO word length.
 The maximum frequency synthesis resolution (taking K=0) is given as:
 The average frequency deviation over M VD output pulses is defined as:
where finst(n) is the instantaneous frequency of the corrected pulses after the phase correction has been applied. This is given as:
fOUT=1/TD(n)=1/[TD(n+1)-tD(n)]=1/[TP(n)+TO- SC/2ND*(D.sub.(n)-D.sub.(n-1))] (7)
Where tD(n+1) and tD(n) are the delay-corrected output pulse edges at instances n+1 and n, corresponding to delay correction words D.sub.(n) and D.sub.(n-1) and TP(n) is the period of the P divider output at the nth instance.
 It can be seen in FIG. 1 that NCO 16 is clocked by the dual modulus divider output signal VP to provide a delay correction value to the delay generator circuit 18. As will be discussed in more detail below, an optional binary frequency divider 104 shown in FIG. 6 may provide a fixed binary frequency division while it will also correct the duty cycle of the output waveform (fOUT). As the reference oscillator signal at frequency fOSC in FIG. 1 is operated on by programmable divider 60, the frequency of VP is necessarily lower than fOSC.
 In fact, the higher the fOSC/fOUT ratio, the finer the fOUT frequency synthesis resolution for a given NCO word length as shown above. The advantage of a high fOSC/fOUT ratio is that the delay generator and the NCO now operate at a much lower rate than in conventional phase-interpolating DDS architectures and therefore the circuit architecture of FIG. 1 features reduced power consumption over these DDS implementations. Furthermore, in FIG. 1, interpolation by the delay generator circuit (phase interpolator) 18 occurs at lower speeds, which offers a significant improvement over the prior art as the lower clocking speed results in a lower power dissipation.
 The output pulse VD has a duty cycle that depends on the synthesized frequency fOUT. Optionally, a toggle Flip-Flop (T-FF) with an output of half the frequency (fOUT/2) or other latching mechanism can be used to produce a duty-cycle corrected version of the synthesized output frequency signal.
 FIG. 4 is a timing diagram showing an example of operation of the circuit architecture of FIG. 1. FIG. 5A is a table that illustrates the behavior of the system described in FIG. 4. A number of signals are shown in the plots illustrated in FIG. 4. Before addressing the various plots, it is important to understand that certain parameters associated with the architectures have been selected in this example for illustrative purposes. For instance, the divider modulus base value (P) has been set to 4. This means that unless there is an overflow, four fOSC pulses are typically passed before a VP pulse comes from the programmable divider 18. When an overflow occurs, the divider counts 5 fOSC pulses before a VP is produced. A larger value of P results in a lower output frequency. While P has been set to a value of 4 in this example, it should be understood that P may be any integer value. P may be set at manufacture to a specific value or range of values depending upon operating conditions. By way of example only, P may be any integer from 3 to 30. Said another way, the larger the value of P, the better control and finer resolution there is over the synthesized output signal VP.
 In the example of FIG. 4, the word length NA of the NCO accumulator 16 is selected to be 3 bits. This provides a maximum value of A to be 7. A larger word length NA will result in a finer fOUT resolution as described above. The resolution of the accumulator output value determines the resolution of the phase correction word. In other words, the precision of the ratio of the accumulator increment to the full-scale range determines the time-scale granularity of the fOUT pulse edges. Another value set in this example is the word length of the delay being applied, which here is shown to be equal to ND.
 As mentioned above, in general NA≧ND. The value of NA determines the granularity of the average frequency programming as shown above through the ratio K/2NA. As the instantaneous frequency of signal VP will vary between fOSC/P and fOSC/P+1, the value of ND determines the effectiveness of the phase interpolation which ultimately determines the phase jitter (or Phase Noise) of the synthesizer.
 Obviously, the larger the NCO word length the finer the output frequency synthesis resolution. For example, in a practical frequency synthesis application with a fOSC frequency in the neighborhood of 2 GHz, if one wants to generate any arbitrary frequency with an accuracy of 1PPM in the range of 400 to 500 MHz the P divider should be set up for division-by 4 or 5, while a word length NA greater than 18 bits should be used. In this particular application, given that the instantaneous value of the frequency fVp will be switching between 400 and 500 MHz and given the current state of the art of delay generator designs, and given that implementations of digital to delay converters with larger ND values become exponentially more complex and consume exponentially more power (i.e. the more bits, the more complicated, with, for example, 6 bits being more complicated than 4 bits and 9 bits more complicated than 6 bits, etc.), ND would be chosen considerably smaller than NA in this case. In particular, the value of ND will generally be set as small as possible while still achieving sufficient cycle-to-cycle jitter performance and sufficient high frequency phase noise performance for the target application.
 The rate of overflow is dependent on the accumulator increment value K. The rate of overflow affects how the delay value applied to the phase interpolator varies (note that it can only be greater than zero). In this example, the increment value, K, is set at 3. The ratio of K to 2NA determines the effective divide ratio of the synthesizer (i.e. fractional division). As K assumes values in the range of 1 to 2NA-1 the output frequency fOUT ranges from fOSC/(P+2-NA) to fOSC/(P+1-2-NA).
 In the timing diagram of FIG. 4, the topmost plot shows the waveform output by the reference oscillator. The frequency of the reference oscillator output is fOSC as shown.
 The second plot from the top shows the waveform of signal VP output by multi-modulus divider 14. As discussed above, VP is provided to the delay generator 18, as well as to D-FF 22 of the NCO 16. The timing of the pulses VP depends upon the multi-modulus divider 14 toggling between P and P+1. Signal VP sets the NCO operational frequency.
 In FIG. 4, the third plot from the top presents the value of the accumulator output word (D), which produces the delay control word to be applied to VP. A delayed version of this value is applied to the delay generator as described below.
 The fourth plot from the top shows the overflow signal OVF which issues from holding or latching circuit 22. The OVF signal controls the division ratio (modulus control) of multi-modulus divider 14. It is also input into delay control circuit 26 of the delay generator 18. As shown in FIG. 4, OVF lasts for one VP cycle every time it is triggered. The accumulator values that cause the overflow are set forth in the table in FIG. 7A. Note, in this example, the overflow value causes the P value to change from P to P+1 which causes the Tosc count to increase from 4 to 5. The increment of tP and the value of TP also increase in duration during overflow.
 The fifth plot from the top shows a signal representing the delay value (D.sub.(n)), which is output by the accumulator 20. In the present example, this signal may range from a value of zero to a value of 7. Other values may be selected depending upon a design choice for various operating conditions. For K=3, the leftmost delay value is zero, while the next delay value is 3, the third delay is 6, the fourth delay is 1, the fifth delay is 4, the sixth delay is 7, the seventh delay is 2, the eighth delay is 5 and the ninth delay is 0.
 The sixth plot from the top presents delay value (D.sub.(n-1)) applied to delay generator/phase interpolator 18 via signal D output from holding/latching circuit 22 of NCO 16. It can be easily observed that the D.sub.(n-1) are staggered by one VP cycle relative to D.sub.(n).
 Finally, the seventh plot from the top presents a pulse train showing the delay-compensated output pulse signal VD of FIG. 1, in relation to signal VP. Referring to FIG. 5A it is observed that the increments of tD are uniform, and the effect of the uniform values of tD results in a uniform value for fout. As noted from FIG. 3 tD is the time delay applied to the pulse VD relative to the pulse edge for VP. FIG. 5B illustrates the effect of a change in the value of NA and K from 3 (in FIG. 5A) to 4 and 7, respectively, in FIG. 5B. In addition to a change in the increments for an overflow to occur, FIG. 5B illustrates that TD and therefore fout. oscillates between two values. FIG. 5B illustrates a system operation where the input increment K≈2NA-1. In this example, the K increment (7) is approximately equal to 24-1 (i.e. 8). As stated below, it is preferred if K does not assume values at or near the limits of its range of values.
 As explained above with regard to FIGS. 1 and 4, the edge-corrected VD pulses can be duty-cycle corrected by passing them through a device such as a toggle flip-flop or a binary frequency divider, as will be described in more detail below with regard to FIG. 6.
 FIG. 6 is a schematic illustration of an alternative embodiment of the invention. It illustrates a multi-modulus divider direct digital synthesizer architecture 100 in accordance with aspects of the present invention. The architecture 100 includes many of the same components as the architecture of FIG. 1, which operates in the manner discussed above. For instance, the reference oscillator 12 produces a waveform with a frequency fOSC. In the present embodiment the reference oscillator 12 also includes a frequency offset control function fOSCOffset that causes the frequency fOSC of its output signal to be shifted by a given amount (e.g. 2%). The architecture 100 also includes the multi-modulus programmable divider 14, programmable NCO 16, and delay generator circuit 18.
 As shown in FIG. 6, the architecture 100 preferably also includes two programmable binary dividers 102 and 104, a programmable modulus control inverter 106, and a temperature sensing unit such as digital thermometer 108. The digital thermometer 108 may receive a signal Tmeas.enable as shown in the figure. This signal may enable/disable measurement by the thermometer 108.
 The reference oscillator frequency fOSC can be optionally divided using the binary divider 102 to a lower master clock frequency fCLK to enable lower power consumption but provides cruder delay resolution (i.e. introduces more jitter). The binary divider 102 receives signal fOSC from the reference oscillator 12 and outputs a signal of frequency fCLK, which is provided to the multi-modulus programmable divider 14 as well as the delay control circuit 26 of delay generator 18. The binary divider 102 divides down the oscillator frequency fOSC, e.g., by a selectable but fixed integer value (B1) to generate fCLK. This provides reduced power dissipation, as the DDS operates now at a reduced speed, while it also increases the total delay range that must be covered by the phase interpolator. This typically degrades the absolute jitter performance of the phase interpolator because digital-to-time converters, like digital-to-analog data converters, tend to have a fixed ratio between the magnitude of the absolute errors and the full scale range. Doubling the maximum delay which the phase interpolator must provide will, to first order, double the absolute value of the maximum errors in the delay. Note that, in this embodiment, the input into the multi-modulus divider 14 is fclk and not fosc (as in FIG. 1). In this embodiment fclk is optionally input into delay control 26.
 The multi-modulus divider 14 is preferably synchronous with the reference oscillator 12. As discussed above with regard to FIG. 1, the output of the multi-modulus divider 14 is signal VP. The parameters B1 of the binary divider 102 and Pset of the multi-modulus divider 14 are used to set a coarse frequency range for VP, which also is the operational frequency of the delay generator 18. The frequency of VP is related to fOSC as follows:
where Pset can be P or P+1 as determined by the modulus control signal MC. The setting of B1 effectively enables a trade-off between noise performance (i.e. jitter) and power consumption.
 As shown in FIG. 6, input signal K is provided to the accumulator 20 of NCO 16. K is composed of a KFset word along with an additional correction value KFC originating from the digital thermometer. As explained above, the resulting increment K is used to determine the rate of overflows and subsequently defines how many VP pulses are counted by the programmable divider 18. The output A of accumulator 20 is a delay value D.sub.(n) and an overflow signal. A latching circuit such as a bank of NA+1 data flip-flops ("D-FF") 22 in the NCO 16 updates the output value of the accumulator 20, shown as signal A at specified intervals. The accumulator output values A (i.e. the phase delay values) are optionally truncated into values D which are provided to delay control circuit 26 of the delay generator 18 along with an overflow signal OVF, the master clock signal fCLK and optionally the signal VP. As previously noted, it is advantageous, but not required, to provide VP and the signal input into the multi-modulus divider (fCLK in FIG. 6) to synchronize the delay generator to the VP pulse.
 As shown, VP is also provided to the delay generator circuit 18. The delay control 26 uses the accumulated value D and OVF to perform delay generation on VP and to output a resultant signal VD, which is a corrected VP signal. Delay control 26 uses optional inputs fclk, and VP to synchronize the delay generator with VP. VD is provided to the binary divider 104. As with the binary divider 102, an input to the binary divider 104, B2, may be a user-specified or predetermined input that sets the binary division ratio at the output of the delay generator 18. The binary division ratio is controlled through a B2 programming word supplied to the binary divider 104. A buffer 110 may be employed to isolate the operation of the DDS from a variety of external loads (not shown). The resultant output is a signal fOUT.
 As discussed above, coarse adjustments of the output frequency can be primarily determined through programming words to the binary frequency dividers 102 and 104. This range of frequencies is also governed by the range of values for Pset in the multi-modulus P/(P+1) divider 14. To those skilled in the art it is obvious that if the selected range of possible P/(P+1) values spans an octave, i.e., max(Pset)>2*min(Pset), in combination with a programmable binary divider such as B2, the system can yield a multi-octave output frequency (fOUT) coverage. While programmable dividers 102 (B1) and 104 (B2) are not required to be binary, the use of other integer dividers require more power consumption and do not provide for duty cycle correction.
 Additional frequency control can be had through the configuration of the divider modulus control polarity MCpolarity applied to the programmable modulus control inverter 106. In the previous embodiments (MCP=0) it has been assumed that an overflow OVF condition causes the dual modulus divider to divide by P+1. In this case, the output frequency in the architecture of FIG. 6 is given by the following relationship:
 If MCpolarity is set so that during regular NCO increments the multi-modulus divider 14 divides by P+1 and during overflows it divides by P (MCp=1) then the output frequency is:
 This configuration also requires that the delay compensation word is bit inverted (i.e. D.sub.MCp-1=1-D.sub.MCp-0) So that maximum delay settings correspond to pulses resulting from division by P+1. The frequency coverage dependence on the MCpolarity signal is illustrated in FIG. 8B.
 Fine adjustment of the output frequency is controlled by K as explained above. The increment word K of word length NK is input to the NCO accumulator 20 and determines how fast the accumulator output A advances for each clock pulse VP generated by the programmable multi-modulus divider 14. Depending on the application, the increment K can be fixed at KFset producing a fixed output frequency or can be dynamically adjusted producing phase or frequency deviations of the output signal.
 FIG. 6 shows the increment K as a sum of the programming words KFset and KFC with word lengths NFset and NFC, respectively. While KFset may be programmed during the manufacturing process to determine the nominal output frequency fOUT,nom, the frequency correction word KFC is updated periodically to correct the output frequency for possible drifts of fOSC due to temperature fluctuations as measured by the thermometer 108. The intermittently-operated thermometer 108 may include an accurate temperature sensor, an analog-to-digital converter and a numerical circuit that produces a supplementary correction value that modifies the effective synthesizer division ratio as follows:
 If fOSC drifts due to a difference in temperature by an amount ΔfOSC then the output frequency will also drift as follows:
fOUT+ΔfOUT=(fOSC+ΔfOSC)*2-(B1+B2)/(- P+[KFset+KFC]/2NA) (12)
 To eliminate the output drift the correction increment should be:
 A variety of temperature-to-increment converters and drift correction mechanisms can be implemented in this block.
 Alternatively, KFset, can be dynamically adjusted according to a phase modulation word (not shown) to produce an output signal featuring continuously variable phase and/or frequency. In this case, a separate phase modulation controller would be inserted at the KFset input of the adder before the NCO 16.
 FIG. 7A shows an exemplary frequency plot for a preferred embodiment with B1=1, B2=1, and P having values of 2, 3 and 4. Also fOSC=2000 MHz and NA=10. It can be easily seen that no gaps in fOUT coverage (a range of about 500 MHz to 1000 MHz) exist as P transitions through various values. It should also be understood that these values for B1, B2, P, fOSC and NA are merely illustrative and do not limit the scope of the invention.
 FIG. 7B shows another exemplary frequency plan for a preferred embodiment with B1=1, B2=1 and 2, and P=2, 3 and 4. And as with FIG. 7A, fOSC=2000 MHz and NA=10. It can be easily seen that the output coverage (a range of about 250 MHz to 1000 MHz) is expanded by an additional factor of 2 by setting the output divider to divide by 2 (i.e., when B2=2).
 FIG. 8a illustrates the use of the oscillator offset function fOSCOffset. This setting shifts the frequency of the reference oscillator fOSC by a small amount (e.g. less than ˜2%) and may be programmed during the manufacturing of the product. It can be easily seen that this option can enable alternative output frequency coverage around the boundary frequencies for different P divider settings.
 FIG. 8b illustrates the effects of the MCpolarity setting in the frequency selection. The output frequency is plotted against the normalized K value for both configurations of the MC polarity. It can be easily seen that one curve is the inverse of the other. As the two MC polarity settings configure both the programmable divider and the delay generator in complementary ways, they provide diversity in selecting the output frequency.
 FIG. 9 is a schematic illustration of an alternate embodiment of the invention. It illustrates a multi-modulus divider direct digital synthesizer architecture 200, which incorporates many of the elements shown and described with regard to FIGS. 1 and 8. FIG. 9 further includes a zero-mean Pseudo-Random Noise ("PRN") generator 202 that produces a dynamically variable (dithering) accumulator increment. This in turn results in a phase-dithered output delay word that applies a controlled noise-like phase correction to the VP signal. This noise-like phase addition "frequency-spreads" any discrete spurious signals and thus reduces their amplitude. The coherence of the fOUT instantaneous frequency fluctuations is reduced by this random jitter process that also features a zero-mean. The PRN generator is clocked by the VP signal and has a word length NPRN. One skilled in the art is aware of different approaches for the construction of PRN generators. One example is a linear code sequence generator that can be made up of any set of delay elements in conjunction with linear combining elements in a feedback path such that the number of states the generator can assume is a function of the length (in time) of the delay elements and the particular combination of feedback. These structures are referred to as Linear Feedback Shift Registers (LFSR). One skilled in the art can readily integrate such structures into a circuit. The resulting pseudo-random instantaneous output frequency jumps are as shown below:
fOUT(n)=fOSC*2-(B1+B2)/(P+[KFset+KFC+KPRN(- n)]/2NA) (14)
where fOUT(n) is the output frequency at the nth VP instant and KPRN(n) is the dithering PRN increment issued by the PRN generator 202.
 FIG. 10 is a schematic illustration of another alternate embodiment 300 of the invention. It illustrates a multi-modulus divider direct digital synthesizer architecture in accordance with aspects of the present invention also shown in FIGS. 1 and 6 and 9. The only difference between the embodiments of FIG. 9 and FIG. 10 is the point where the dithering from the PRN generator 202 is inserted. The PRN 202 phase increment is added at the output of the NCO 16. The adder 203 sums delay value Di and KPRN to supply the output DO along with the summation overflow bit OVFD to the delay control block 26. In this embodiment, the PRN word length NPRN is smaller than the accumulator input/NCO output word length ND and much smaller than accumulator output word length NA. It should be noted that after the PRN addition, the word length of D0 is truncated back to ND. This embodiment can be more suitable for the purposes of further reduction in spurious signals than the embodiment illustrated in FIG. 9.
 Other dithering schemes are well known to one skilled in the art. One such example is described in U.S. Pat. No. 4,410,954 entitled "Digital frequency synthesizer with random jittering for reducing discrete spectral spurs," which issued on Oct. 18, 1983 and is hereby incorporated by reference.
 Another application where the increment K is also modulated in a controlled fashion is in "spread-spectrum" clocks to enable the suppression of Electro-Magnetic Interference ("EMI") products in noise sensitive applications. The difference with the PRN increment dithering described above is that spread spectrum clock oscillators spread out the concentrated clock energy on their nominal output frequency to a broader bandwidth and controlled frequency range (e.g. ˜1% of the output frequency). The total energy remains the same but the peak energy is spread out to near-by frequencies. The embodiments illustrated in FIGS. 9 and 10 can also implement spread-spectrum PRN functions but it is advantageous if frequency deviations do not exceed the allowed range of values of the K/2NA ratio.
 Thus, it can be seen that the output frequency fOUT for architectures implementing aspects of the present invention is dependent upon a number of parameters, including the reference oscillator frequency, the two binary divider ratios B1 and B2, the dual modulus divider ratio P, the increment associated with the accumulator, and the settings for the P modulus control polarity configuration and the fOSCOffset setting.
 FIG. 11 is a schematic illustration of another alternate embodiment 400 of the invention. This embodiment addresses the use of the invention when temperature, spread spectrum frequency modulation, and external modulation inputs added together could cause the value of K/2 NA to move above its maximum value (1) or below its minimum value (0). In such applications a problem is created when the nominal value of K must be selected near its maximum or minimum value in order to set the output frequency to its desired value. To understand this problem, consider that the maximum output frequency for the architecture in FIG. 1 is fosc/Pmin (the oscillator frequency over the minimum value of P) and the highest frequency of the output is fosc/(P.sub.max+1) (the oscillator frequency over 1+ the maximum value of P). By selecting the appropriate value of P and K, any arbitrary frequency between these two limits can be selected. However, the value of P and K that give a frequency are unique since K can only represent a fraction (less than 1) of an oscillator period, Tosc. Therefore, some possible output frequencies will require the normalized value of K to be near 0 or near 1. If temperature correction terms, external modulation terms, or spread spectrum modulation terms are added to K, the ratio K/2NA would need to exceed the limits of 0 or 1 and the loop would fail to operate correctly. The discussion of FIG. 8a describes a method to keep the normalized value of K from being too close to either end of the (0,1) range by pulling the oscillator frequency when KFset falls too close to a limit moving the nominal value of K further from the limit.
 FIG. 11 illustrates an improved structure and method for achieving this result. Instead of requiring a change in the oscillator frequency, which may compromise the oscillator's performance as a frequency reference, FIG. 11 extends the allowed normalized range of K (i.e., K/2 NA) to be from 0-2 oscillator periods instead of the 0-1 oscillator period range provided by the architecture in FIG. 1.
 The embodiment in FIG. 11 operates in a way that is similar to the embodiment in FIG. 1 except that the division modulus of the multi-modulus divider 14' is now controlled by both the overflow signal OVF (MC1) (which is like the OVF from FIG. 1) and the MSB (MC0) of word D instead of just using the OVF pulse as in the embodiment illustrated in FIG. 1. Also, only the NA-1 lower bits from the NCO output word are used for feedback to the accumulator input, as well as for delay compensation (before truncation to ND bits). For example, for the embodiment in FIG. 1, when the phase accumulator 20 output is 16 bits, 1 bit (OVF) is fed to the multi-modulus divider 14 and retired and the remaining 15 bits are added to K in the accumulator 20 to form the next accumulator 20 output. In FIG. 1, K is also a 15 bit binary word (i.e., it ranges from 1 up to 32,767); therefore, the sum of K and the 15 bit fed back remainder is a 16 bit word that can take on all possible 16-bit values, 0-65,535. In the embodiment in FIG. 11, the range of allowed increments is increased 2-fold compared to the fed-back remainder (i.e. NK=NA). The sum of a 15 bit remainder and a 16 bit K is a 17 bit word that ranges from 0-98,303. By increasing to a triple modulus divider P, (e.g., P, P+1, P+2), the two-bit overflow signal (MC1 and MC0) pulses are used to select the modulus while the same 15 bit remainder is returned for accumulation in the next period. As in the embodiments described above, a portion of this word is also used for the (n+1)th instance delay compensation of the multi-modulus divider output pulse. It should be obvious that this concept can be extended to higher order modulus dividers (e.g., P, P+1, P+2, P+3 etc.) in a straightforward manner by simply using more of the MSBs of the accumulator output as part of the OVF signal to control the multi-modulus divider and returning only the remaining accumulator output bits to form the remainder to be accumulated in the next period.
 Operationally, the formula for the output frequency of the embodiment in FIG. 11 is:
where fOSC is the reference oscillator frequency, P is the base value of the P/P+1/P+2 multi-modulus programmable divider, K is the increment and NA=NK is the increment word length. Since K can range from 0 to 2NA-1, fOUT, can range from (fOSC/2)/P to (fOSC/2)/(P+2-1/2NA). At the nominal center frequency, the ratio 2K/2NA would be centered around 1 and set to range from 0.5 to 1.5 in order to select any desired frequency in the range around fosc/(P+1). Note, the advantage of the embodiment of FIG. 11 over that of FIG. 1 is the added redundancy in defining fout. There are now two values of P and K that can be used to create any specific frequency. The added degree of freedom is what makes it possible to avoid selecting a nominal value of K near either end of its allowed range. In the specific case of a triple modulus programmable divider DDS, having the normalized 2K/2NA range from 0.5-1.5 allows the selection of all possible frequencies with the appropriate selection of P. And, this allows any dynamic components (e.g., temperature compensation, external modulation, spread spectrum modulation, etc.) to vary the overall output period by up to +/-0.5 of a reference oscillator period before the minimum or maximum value of K is exceeded. If larger dynamic components need to be accommodated, a higher modulus programmable divider can be employed and the range of K can be further increased.
 FIG. 12a illustrates the operation of the triple modulus DDS with P=4, fOSC=2 GHz, NA=NK=4, K=11 and ND=3. FIG. 12b shows the Modulus control truth table of the multi-modulus divider for values of MC0 and MC1 and FIG. 12c shows the P division cycle count and the output pulse period for all possible increments of a 4 bit word.
 FIG. 13 shows an embodiment using the circuit architecture of FIG. 11 with elements of the embodiment of FIG. 9. For instance, it includes a P, P+1, P+2 multi-modulus divider 14' controlled by both the overflow OVF and the MSB as well as a spreading generator 202. The output frequency fOUT is given by the following relationship:
fOUT=(fOSC/2.sup.B1+B2)*(1/(P+[KFset+KFC+KSS(n)- ]/2NA)) (16)
where fOUT(n) is the output frequency at the nth instant and KSS(n) is the spreading fractional increment issued by the spreading generator (202). One skilled in the art is aware of several different approaches for the effective spreading sequence that is provided by the spread generator (202). Examples include a predetermined noise generation pattern as described above, a linear deterministic function or other deterministic function (e.g. a "Hershey Kiss" function).
 FIG. 14 shows a variation of the embodiments of FIGS. 11 and 13 in which the spreading word KSS is added at the output of the accumulator, in a fashion similar to the embodiment of FIG. 10.
 FIG. 15a depicts an alternate embodiment of the invention. It illustrates a multi-modulus divider direct digital synthesizer architecture 1500 that incorporates the elements shown and described with regard to FIGS. 1 and 9. The embodiment in FIG. 15a illustrates one mechanism for inducing the generation of the dynamically variable (dithering) accumulator increment.
 Dithering mitigates the spurious tones that can result from the discrete response of the NCO and the delay generator in the DDS. In an ideal DDS, the NCO would have infinite resolution and the delay generator would delay pulses Vp by the exact time delay required to synthesize the desired output signal frequency. However, in practice, the delay generator has a discrete response, i.e. it can delay pulses Vp only by certain values of time delay. Also, the NCO has a higher resolution than the delay generator. As a result, for most values of increment K, the instantaneous period of the DDS output signal toggles between two values, and the toggling action repeats with a fixed pattern. In the frequency domain, this fixed-pattern toggling action yields spurious tones that degrade the spectral purity of the DDS as a reference frequency generator. The fixed-pattern toggling action can be disrupted by arithmetically dithering the increment K, the accumulator output A, or the delay word D, or by directly dithering the response of the phase generator. Ideally, dithering should be performed by a random signal. In practice, it is sufficient to employ a pseudo-random signal that repeats with a long sequence length. By disrupting the fixed-pattern toggling action, dithering effectively spreads the spurious tone energy across a wide frequency range and therefore decreases the spurious tone peak amplitude in the frequency domain. One trade-off is that dithering, because it adds randomness to the phase increment K, increases the cycle-to-cycle jitter of the DDS output signal.
 Architecturally, in the embodiment illustrated in FIG. 15a, the dithering mechanism may be provided upstream of the accumulator 20 of the NCO 16. As illustrated, FIG. 15a includes a Pseudo-Random Noise ("PRN") generator 1502 that generates a pseudo-random sequence KPRN with a non-zero mean. That is, the mean of KPRN is some value (positive or negative) other than zero (although one skilled in the art knows that the mean value may be zero). Preferably the pseudo-random sequence is a sequence of binary words, although one skilled in the art understands that other numeral systems may be employed to represent the sequence of pseudo-random numbers. The pseudo-random binary sequence word width is NPRN bits. Typically, NPRN ranges from 2 to 5. However, one skilled in the art understands that NPRN may range from 1 (in which case KPRN is a single-bit sequence) to 10 or even higher.
 The non-zero-mean PRN generator 1502 is clocked by the VP signal and receives Klow as data input. It produces pseudo-random sequence KPRN with a mean value of Klow.
 Klow is a binary word composed of the NA-ND least-significant bits of the total tuning increment K=KFset+KFC. The ND most-significant bits of K form Khigh, such that the relationship of Khigh and Klow is:
KFset+KFC=K=Khigh*2.sup.(NA-ND.sup.)+K.sub- .low (17)
 FIG. 15b illustrates a detailed configuration of the arithmetic circuit 1504 that computes K, forms Khigh and Klow and computes the dithered accumulator increment K'. Circuit 1504 comprises an adder 1506, a split circuit 1508 and another adder 1510. Adder 1506 sums the KFset with KFC to produce the K word with NA=NK bit length, as described in FIG. 9. Then the K word is split by the split circuit 1508 into its ND most-significant bits (to form Khigh with length ND) and its NA-ND least-significant bits (to form Klow with length NA-ND). Note that, when word K is represented in binary on a bus, the split operation implemented by circuit 1508 trivially consists in selecting the signals off MSB bus wires to form Khigh and the signals off LSB bus wires to form Klow.
 Adder 1510 sums KPRN with Khigh to produce the K' word, which is the dithered accumulator increment K' to be applied to the NCO. Therefore, the accumulator produces a reduced word of ND bits. In contrast, the accumulator employed with the zero-mean PRN generator in FIG. 9 processes data with word length NA≧ND.
 The dithered accumulator increment K' induces the NCO to produce a phase-dithered D signal to set the delay of the delay generator 18. As stated above, a non-zero-mean PRN generator enables the accumulator to produce a shorter word length. Thus, a simpler and cheaper implementation of the accumulator and NCO may be realized with a non-zero-mean PRN than with a zero-mean PRN generator.
 The non-zero-mean PRN generator 1502 may be integrated with a DDS as a separate component, or as part of the NCO. The skilled person is aware that there are many potential configurations of the non-zero-mean PRN generator 1502, depending upon the design of the DDS with which it is integrated. For example, for many applications of the DDS, it is advantageous to reduce the phase noise of the DDS output signal in certain frequency ranges (such as below 20 MHz, in the case of a 100 MHz output frequency). To reduce spurious tones in those frequency ranges, the non-zero-mean PRN generator can be configured to shape the pseudo-random noise out of those frequency ranges. A non-zero-mean PRN generator implemented with a digital Delta-Sigma modulator (commonly noted ΔΣ modulator, and alternately referred to as digital Sigma-Delta modulator and noted ΣΔ modulator) may provide such noise shaping properties.
 A multi-stage noise shaping (MASH) ΔΣ modulator may be used for its relatively simple hardware implementation and guaranteed stability. Detailed implementations of MASH ΔΣ modulators are described in "Understanding Delta-Sigma Data Converters" by Schreier and Temes (Wiley, 2004), which is incorporated by reference herein in its entirety.
 Preferably, a ΔΣ modulator with long sequence length may be used. Various techniques may be adopted to generate a long (or maximum) sequence length, such as those described in the following publications: Hosseini, K. et al. "Maximum Sequence Length MASH Digital Delta-Sigma Modulators" IEEE Transactions on Circuits and Systems--I: Regular Papers, 54(12) pp. 2628-2638 (December 2007) and Hosseini, K. et al. "Architectures for Maximum-Sequence-Length Digital Delta-Sigma Modulators" IEEE Transactions on Circuits and Systems--II: Express Briefs, 55(11), pp. 1104-1108, November 2008).
 The architectures illustrated in FIGS. 9 and 15a-b may be modified to selectively enable the use of PRN generator or disable its use. As such, the DDS may operate with dithering as illustrated in FIG. 9 or 15a-b or operate without dithering as illustrated in FIG. 1. This enables a DDS to selectively suppress dithering for those values of accumulator increment K that do not induce undesired spurious tones.
 FIG. 16a depicts an alternate embodiment of the invention. It illustrates a multi-modulus divider direct digital synthesizer architecture 1600 that incorporates the elements shown and described with regard to FIGS. 1, 9 and 15a. The embodiment in FIG. 16a incorporates a Controlled PRN Generator 1602, which is illustrated in details in FIG. 16b. The Controlled PRN Generator 1602 may comprise one or more PRN generators, which may be zero-mean PRN or non-zero-mean PRN. It takes KFset and KFC as input signals, and outputs a tuning word K'' with NK word length to the NCO 16. Various types of control signals, such as "dither enable" and "dither select", may be provided to the Controlled PRN Generator to facilitate PRN enable/disable and selection functions.
 FIG. 16b illustrates a detailed architecture of the Controlled PRN Generator. It comprises a zero-mean PRN generator 202 as described in FIG. 9, and a non-zero-mean PRN generator 1502 as described in FIG. 15a. It also comprises the elements of arithmetic circuit 1504 as described in FIG. 15b. These elements include adders 1506 and 1510, and a split circuit 1508. Additional circuits, such as bit-shift multiplier 1606 and adder 1610, may also be used.
 In addition, the Controlled PRN Generator may include a 3-to-1 multiplexer 1608 that may be used to select from the signals of K, K2 and K3. Signal K is the accumulator increment to be applied to the NCO, which is obtained by K=KFset+KFC. Signal K2 is a dithered signal of K generated by the non-zero-mean PRN generator 202. Signal K3 is a dithered signal of K obtained by the zero-mean PRN generator 1502.
 As shown, the Controlled PRN Generator comprises a dither control circuit 1604 that enables or disables the PRN generators and controls the 3-to-1 multiplexer 1608. The dither control circuit takes KFset as input, and is controlled by a user-provided dither enable signal and a user-provided dither select signal. Control signals may be output from the dither control circuit to disable zero-mean PRN generator 202, and non-zero-mean PRN generator 1502, and to control the multiplexer 1608.
Patent applications by Anthony L. Tsangaropoulos, San Carlos, CA US
Patent applications by David Francois Guillou, Pittsburgh, PA US
Patent applications by Cymatics Laboratories Corp.
Patent applications in class Frequency division
Patent applications in all subclasses Frequency division