Direct digital synthesizer with selectably randomized accumulator.

BACKGROUND OF THE INVENTION

1. Field of the Invention

0001The invention relates generally to digital frequency synthesizer apparatus, and more particularly to such apparatus employing an accumulator for synthesizing a selected frequency of a number of preselected frequen­cies in conjunction with a randomizer which adds random or pseudorandom values to a selectable few of the least significant bits of the accumulator.

2. Description of Related Art

0002In the design of radio communications systems, a continuing goal has been the design of a simple, low-cost frequency synthesizer that can provide a large library of discrete frequencies over a very wide frequency spectrum. Design efforts toward this goal have introduced an associ­ated spectrum of spurious outputs, the minimization of which has not been entirely successful or satisfactory.

0003Prior art approaches have typically been too complex for commercial feasibility. Most such prior-art solutions employ digital logic and have required the use of hybrid devices such as digital-to-analog (D/A) converters, as well as table look-up devices such as sine/cosine Read Only Memories (ROM's). A primary purpose of these additional elements has been to reduce the poor output spectral quality. Such associated poor output spectral quality is mainly evidenced by the presence of a large number of unwanted or spurious line spectral components referred to commonly as "spurs."

0004Direct Digital Frequency Synthesis is a digital technique of frequency synthesis that is often used in phase and frequency modulation and frequency hopping schemes in spread spectrum communications. This technique is advantageous because of the speed at which frequency can be changed, the small frequency step size possible, and the digital control of phase or frequency available. An apparatus employing this technique is referred to as a Direct Digital Synthesizer or "DDS."

0005The output frequency of a DDS is determined by an input digital frequency control word. This frequency control word can be generated by a digital processor. Many modulation schemes can be implemented, for example, FSK, FM, PM, DPSK, etc. Thus, for example, the output frequency can be frequency modulated according to the information digitized by the processor (FSK modulation). In addition to the FSK modulation, the window of modu­lated frequencies can hop, thereby moving the window of modulated frequencies in a coded scheme to avoid jamming or decoding by unauthorized transceivers. It is the agility and the frequency resolution of the DDS that makes it superior to other synthesizers in such an application and gives an unauthorized transceiver less time to lock on the correct frequency and then jam or decode it. Naturally, the authorized receiver must hop to the correct frequency to receive the modulated informa­tion. The receiver's digital processor generates the correct digital control word for the receiver's DDS and subsequent down-conversion, demodulation, and decoding.

0006The heart of the Direct Digital Synthesizer is the accumulator. The accumulator is a digital integrator consisting of a latch and full adder with the output sum and the input digital control word as inputs to the adder. The frequency word determines the step size by which the accumulated sum is incremented. Each step represents a step or increment in phase; thus, larger frequency words result in larger steps in phase. In addition, the rate of overflow of the accumulator determines the output frequency. The output frequency is a fraction of the clock frequency; thus its spurious performance has similarities to a fractional divider. However, some frequencies are not submultiples of the clock frequency, and the accumulator overflow occurs with varying remainders left in the accumulator. The residue left in the accumulator varies the output periods such that a sequence of different output periods result. The average of this sequence of periods is the desired period and, thus, determines the output frequency. However, since this sequence of periods is also different from an ideal repetition of the exact period, spurious line frequencies or "spurs" unique to each output frequency are created. Such spurs represent departure of the output waveform from a precisely periodic signal.

0007If the overflow of the accumulator is used to generate a binary output, the spurious levels will be significant. The conventional method used to suppress these spurious frequencies utilizes a sine approximation technique. This technique essentially consists of the accumulator, a sine look-up table in Read Only Memory (ROM), a digital analog converter (DAC), and an output filter. The instantaneous error from an ideal sinusoidal frequency of constant period is reduced; therefore, the spurious levels are suppressed.

0008An alternative to such solutions is disclosed in <u style="single">Wheatley</u> U.S. Patent No. 4,410,954. That patent discloses an apparatus for feeding a value corresponding to a selected phase increment at a clock rate to an adder, which is in mutual cooperation with an accumu­lator. The resultant periodic overflow of the accumu­lator corresponds to the selected frequency of interest. Additional cooperating apparatus is provided for generating a random sequence of values. The difference between any one of these values and the selected phase increment is less than the selected phase increment. Jitter logic is responsive to the overflow output of the accumulator to add or subtract a sequential one of the random values to or from the phase increment value as an input to the accumulator for at least one clock interval occurring within the duration between successive over­flows of the accumulator. The Wheatley dithered DDS has a binary output; therefore, it has the advantage that no ROM or DAC is needed.

0009The Wheatley method randomly varies or dithers the accumulator at each overflow or alternate overflows such that the edges or just one edge of the binary output are dithered. The Wheatley circuit thus reduces spurs by summing a random value with a uniform distribution up to the selected phase increment or "input vector." However, the Wheatley operation introduces flat phase noise, although, in many applications, the introduced flat phase noise is within the phase noise performance margin and hence does not limit system performance, whereas the spurious performance attributable to spurs in the absence of spur reduction would limit system performance.

0010One disadvantage of the Wheatley circuit is that the frequency deviation density decreases with decreasing frequency offset. Hence, the deviation density is inadequate to reduce close-in spurs, i.e., those spurs near the desired frequency output.

0011A second disadvantage of the Wheatley circuit is the lack of flexibility in selecting between the amount of spur reduction and phase noise. Since the dither is uniform up to the input vector, a trade-off between spur reduction and phase noise is established. With Wheatley's circuit, selectability between these two performance criteria can be achieved only through selection of the clock rate, a relatively inflexible approach.

SUMMARY OF THE INVENTION

0012It is therefore an object of the invention to improve direct digital synthesizer (DDS) circuitry.

0013It is another object of the invention to improve DDS circuits employing an accumulator as a frequency generator.

0014It is yet another object of the invention to provide a DDS employing an accumulator frequency genera­tor in which close-in spurs are effectively reduced.

0015According to the invention, an accumulator applicable in a DDS is employed in conjunction with a randomizer that adds random or pseudorandom values to a selectable few of the least significant bits of the accumulator to introduce flat or nearly flat frequency deviation density in the vicinity of significant spurs for spurious frequency suppression, instead of dithering enough bits of the accumulator to introduce flat phase noise as in prior art. The frequency deviation density at the spur's offset in frequency from the carrier effectively spreads the energy of the spur; therefore, by this technique all in-band spurs are equally reduced.

0016The invention particularly departs significantly from the approach taught by Wheatley. The random value is uniformly distributed over a range which is not up to and just less than the input vector as taught by Wheatley, but is rather much less than the input vector for a given set of system specifications. In addition, the random value added to the accumulator is not neces­sarily subtracted out and can be allowed to circulate in the accumulator. Permitting the random value to circu­late in the accumulator results in a known offset in output frequency. This offset can be subtracted from the input digital frequency control word (input vector) in applications where it proves objectionable. An addi­tional advantage is that, if the bits dithered are below the lowest input vector for a given output bandwidth, the dithering no longer needs to be sensitive to the input vector. This significantly reduces the complexity of the hardware since the comparator and memory (stack) in the prior art Wheatley circuit are no longer required. Addi­tionally, in a sine DDS design, if the significance of the low bits are truncated in the sine approximation algorithm, the accumulator can be sectioned into addi­tional pipelined stages, and the dither can be imple­mented without decreasing throughput. In fact, the accumulator can be lengthened while maintaining the same number of LSB's dithered. As a result, the spurs are spread into as low a level as before, but the phase noise introduced by dithering is reduced.

0017For a direct digital synthesizer employing the invention, the levels of all in-band sampling spurs are reduced, including the spurs close-in to the fundamental, which the prior art is ineffective in reducing. The dither introduces a frequency deviation of the signal which is less than the frequency offset of the closest spur to the fundamental. Consequently, the frequency deviation density at the offset frequencies of the spurs is constant, and all in-band spurs are reduced by the same amount including the close-in spurs.

0018In addition, since the number of bits dithered is selectable and the accumulator can be any length, a wide range of spur levels and phase noise trade-offs are possible. The number of bits dithered may be selected by hardwiring or by processor or other digital control.

BRIEF DESCRIPTION OF THE DRAWINGS

0019The just-summarized invention will now be described in conjunction with the drawings of which: <ul id="ul0001" list-style="none"><li>FIG. 1 is a block diagram of accumulator circuitry according to the preferred embodiment of the invention;</li><li>FIG. 2 is a circuit block diagram of a DDS circuit employing the preferred embodiment of the invention;</li><li>FIG. 3 illustrates spurs present about the fundamental in the output of an undithered accumulator; and</li><li>FIG. 4 illustrates suppression of spurs according to the preferred embodiment.</li></ul>

DESCRIPTION OF THE PREFERRED EMBODIMENTS

0020A dithered accumulator circuit according to the preferred embodiment is shown in FIG. 1. This circuit employs a subtractor 11 which receives an M-bit frequency word or "input vector" on a first input and subtracts from it a D-bit offset frequency word supplied on a second input. The output of the subtractor 11 is supplied as one input to a first full adder 13, which receives a second input from the output of a latch 19. The first full adder 13 outputs an M-bit word to one input of a second full adder 15 and an overflow bit to an enable gate 17. The second full adder 15 receives a second D-bit input from the enable gate 17 and outputs an M-bit output to the latch 19. The accumulator output comprises K bits where K ≦ M. A pseudorandom number (PRN) generator 21 provides a D-bit pseudorandom number to the enable gate 17. The pseudorandom number genera­tor 21 and the latch 19 receive a clock input on a clock line 22. The accumulator may be viewed as an integrator which integrates phase increments, Δ0̸, at its input, represented by the frequency control word, to yield an output K representing phase 0̸.

0021In typical prior art DDS circuits, the accumu­lator consists of an adder and a latch, analogous to the first full adder 13 and the latch 19. The embodiment of the invention includes a second full adder 15, enable gate 17 and pseudorandom number generator 21. As will be appreciated, the second full adder 15 adds the pseudo­random number from the PRN generator 21 to the output of the first full adder 13. The pseudorandom number is added only when an overflow occurs from the first full adder 13. Such an overflow is signalled over a line 14 from the first full adder to the enable gate 17, which then gates the pseudorandom number to the second full adder 15. The pseudorandom number added to the output of the first full adder 13 is selected to satisfy the following equation:<maths id="math0001" num=""><img file="EP0338742A2_D0001.tif" /></maths> Where f_m′ is the frequency offset in the spur closest to the fundamental, f_CLK is the clock frequency on line 22, which updates the latch 19 and clocks the number generator 21, D is the number of bits in the PRN generator 21 and in the offset frequency word, and M is the number of bits in the input vector. To satisfy this equation, the number of bits "D" in general must be very small compared to "M." If "D" becomes too large, the right-hand side of the equation becomes much greater than the offset in the frequency spectrum of the first spur, f_m′ and effectiveness in reducing close-in spurs is decreased.

0022When the pseudorandom number is added in by the second full adder 15, a slight frequency offset occurs. Depending on system specifications and the amount of dither added in, it may be unnecessary to compensate for this offset. In such case, the subtractor 11 is not required. For example, if only one or two bits are dithered, i.e., D«M, it will cause a very small offset in the output frequency. It may be noted that the second full adder 15 can be placed elsewhere in the circuit, for example in the place of the subtractor 11.

0023Selection of the number of bits "D" to be very small satisfies the relationship of equation (1) and therefore results in the desired substantially flat frequency deviation density for suppression of close-in spurs. It is also possible to achieve such flat fre­quency deviation density by arranging the circuitry to decode certain conditions where worst case spurs occur and respond to such conditions to randomize the accumulator.

0024FIG. 2 illustrates the incorporation of the preferred embodiment into a pipelined DDS circuit. The circuitry within the dashed line 66 comprises the dithering circuitry 66 according to the invention, while the remaining circuitry comprises a typical sine look-up DDS. This embodiment employs an accumulator sectioned into stages in order to maintain throughput of the pipelined system.

0025The dithering circuitry 66 receives a frequency word which is shifted serially into a shift register 53 clocked by a clock on a line 68. In the illustration shown, a twenty-bit frequency word is used as an example. The twenty-bit frequency word is input in parallel to a latch 51. The latch 51 is activated by a control line when a strobe (LOAD), e.g., from a digital processor, is supplied to a clocked timing circuit 57, which synchronizes the edge of the load strobe with the DDS system clock (CLK). The four least significant bits of the output of the latch 51 are transmitted to one input of a four bit adder 49, while the other sixteen bits contained in the latch 51 are transmitted to one input of a sixteen bit adder 50. Six carry generate and propagate outputs of the sixteen bit adder 50 are out­putted to a look ahead carry generator 37, which supplies an overflow bit on a line 38 to one input of a dither generator 43. The other input of the dither generator 43 is connected to receive the output 42 of the four bit adder 49 and has a four bit output 46 connected to a latch 41. The four bit output of the latch 41 is connected as the input to the four bit adder 49.

0026In the embodiment of the dither circuit 66 shown in FIG. 2, the four bit adder 49, sixteen bit adder 50 and look ahead carry generator 37 comprise the first full adder 13 of FIG. 1. Accordingly, the carry-out line 44 of the four bit adder 49 is connected to the carry-in input of the sixteen bit adder 50 and to the carry generator 37. The dither generator 43 includes the pseudorandom number generator 21, the enable gate 17 and the second full adder 15 illustrated in FIG. 1. Thus, the first full adder 13 of FIG. 1 is sectioned into two parts 49, 50, and the dithered accumulated value out­putted by the dither generator 43 is latched into the latch 41 and then added into the four bit adder 49 for accumulation of the lower section of the accumulator represented by the part of the M bits from latch 19 which lead back to the input of the first full adder 13 of FIG. 1. The sixteen bit adder 50 outputs to a sixteen bit latch 31. This latch 31 and the latch 41 perform the function of the latch 19 in FIG. 1.

0027Those skilled in the art will recognize the circuitry of FIG. 2 outside of dashed line 66 as typical DDS circuitry employing a sine approximation technique and quadrant decoding method to generate addresses for a sine look-up table stored in read only memory (ROM) circuits 45, 47. The circuitry for configuring the addresses includes a buffer 33, exclusive-or gates 35 for performing the one's complement, and a latch 39. The twelve most significant bits of the latch 31 are supplied to the first input of the one's complement circuit 35. The output of the latch 31 is fed back to the second input of the sixteen bit adder 50. The two least significant bits are supplied as an input to the buffer 33, which provides a large fanout to thirteen bits. The twelve significant bits of the output of the buffer 33 form the second input into the comparator 35. The one's complement circuit 35 comprises an exclusive-or (EX-OR) of respective pairs of the twelve incoming bits on first and second input lines from the latch 31 and the buffer 33. The upper four bits of the output of the one's complement circuit 35 are supplied to the first four bit positions of a twenty bit latch 39 and to the second four bit positions of the latch 39. The second four bits output by the one's complement circuit 35 are supplied to the next four bit positions of the latch 39 and to the next four bit positions of the latch 39. The last four bits of the twenty bit latch 39 are filled by the last four bits output by the comparator circuit 35.

0028According to the quadrant decoding method, the twenty output bits of the latch 39 are connected to address the respective ROM's 45, 47. The two read-only memory circuits 45, 47, supply a fifteen bit output to the latch 55. The latch 55 further receives the most significant bit from the latch 39 and transfers that most significant bit, which is the sign bit, to a buffer 59. The outputs of the table look-up operation, respectively, eleven bits from ROM 45 and four bits from ROM 47, are added by an eleven-bit adder 60 which outputs nine bits to a one's complement circuit 61. The one's complement circuit 61 receives a nine bit input from the buffer 59, and outputs nine bits to a latch 63. Latch 63 also receives the sign bit from the buffer 59. The latch 63 supplies a ten bit digital output to a digital to analog converter 67. As those skilled in the art will recog­nize, the one's complement circuit 61 supplies the appro­priate sign to the output since only one quadrant of the sinewave is stored in the read-only memory circuits 45, 47. A buffer tree timing generator 65 is also supplied to provide appropriate clock signals to the circuitry.

0029The desired output of the DDS is a sinusoidal signal; therefore, the closer the output approximates the sinewave the better the output spectral purity will be. The digital approach of FIG. 2 is to use a sine look-up table in ROM which is addressed by the output of the accumulator 66. The ROM output is then input to a DAC for an approximation to the desired output sinewave. The accuracy of the approximated output will depend on the phase information contained in the accumulator, trunca­tion in the sine algorithm and DAC, and other DAC errors.

0030The sine approximation technique used in the embodiment of FIG. 2 is unmodified from the DDS LSI design of Sunderland et al. Sunderland's DDS sine algo­rithm utilizes a well-known technique of ROM reduction by storing only one quadrant of the symmetrical sinewave in memory. In addition, his DDS design utilizes a unique trigonometric reduction to substantially reduce the memory requirements further. The argument of the sine function is separated into three terms, and the following trigonometric identity is applied: sin (A + B + C) = sin (A + B) cos C + cos A cos B sin C - sin A sin B sin C      (2) Approximations are made, and the equation becomes sin (A + B + C) ≅ sin (A + B) + cos A sin C      (3) Compensation is added to the terms B and C to reduce the errors introduced by the approximation. The reader is referred to Sunderland et al., "CMOS/SOS Frequency Synthe­sizer LSI Circuit for Spread Spectrum Communications," IEEE Journal of Solid-State Circuits, Vol. SC-19, No. 4, August 1984, for further details of this approximation. The two terms on the Right Hand Side of equation (3) represent the separation of the ROM into the two parts shown in FIG. 2: a coarse ROM 45 (of size 256x11) and a fine ROM 47 (of size 256x4). The two terms are added together by the eleven bit adder 60 and combined with the sign bit for a twelve bit offset binary output to the DAC 67, which is truncated to ten bits in the embodiment shown. The one's complement circuit 35 preceding the ROM's 45, 47 decodes the quadrant, while the one's complement circuit 61 following the eleven bit adder 60 decodes the sign. The phase information propagates through the DDS from the accumulator latch 31, to the ROM latch 39, to the ROM latch 55, to the DAC latch 63 with each step, from latch to latch, representing a pipelined stage. Each stage must settle within a clock period, and the throughput is determined by the clock period, the number of pipelined stages, and the settling time of the DAC 67.

0031The twenty bit accumulator 66 of FIG. 2 is sectioned into a sixteen bit high accumulator 50 and four bit low accumulator 49. The carry of the low accumu­lator 49 into the LSB of the high accumulator 50 is truncated in the sine approximation algorithm and DAC 67: the twelve bits retained in the sine algorithm represents the truncation of the least significant four bits of the sixteen bit high accumulator 50. A DAC 67 with ten bits of resolution represents additional truncation. Therefore, the carry from the low accumu­lator 49 can lag behind the high accumulator 50 by one clock cycle without affecting the significance of the bits in the sine algorithm. Similarly, additional accumulator sections can be added to the accumulator.

0032Since the low accumulator 49 is parallel to the high accumulator 50 and requires only a four bit adder, the low accumulator 49 is an advantageous place to inject dither without decreasing the DDS throughput. A pseudo­random value can be added to the low accumulator 49 to inject dither at each overflow of the high accumu­lator 50. But the previous random number which Wheatley subtracted from the accumulator may require a borrow from the high accumulator. If the previous random number is not subtracted out, the average of the random value will offset the output frequency just as though this average value had been added to the frequency control word. Since the pseudorandom number generator has a known average value, it can simply be subtracted from the frequency control word. If the processor is not capable of this, a twenty bit subtractor is required. To avoid the twenty bit subtractor, the carry which is added to the high accumulator 50 can be effectively subtracted out later by removing a "1" from the LSB of the high accumu­lator register when such a condition occurs. Care must be taken not to subtract this bit out too soon, or the random phase values may not accumulate to the extent necessary to generate the carry into the significant bits which address the sine ROM's 45, 47.

0033The random value is added to the accumulator sum before the accumulator latch 41, but it can be added anywhere in the accumulator loop or at the input vector. It is assumed that the overflow from the low accumula­tor 49 would not affect more than the least significant bit of the high accumulator 50. If the input vector to the low accumulator 49 is not zero, the possibility of carries into the two least significant bits of the high accumulator 50 could be required in one clock period. One way of avoiding this problem is to extend the accumu­lator, such that the lowest section of the accumula­tor 49, which is dithered, is lower in significance than the LSB of the input vector. The condition for the greatest spurs occurs when only a few of the MSBs are high and the LSBs of the accumulator are zero.

0034In order to determine the requirements for the random number generator 21, the DDS implementation must be considered. At exactly half the clock frequency, the alias will coincide with the output frequency. The alias occurs at the frequency determined by subtracting the output frequency from the clock frequency. The output frequency must be less than the alias by a margin necessary for the transition region of the output alias filter (a low pass filter to filter out the alias). Therefore, the random number generator 21 must have a propagation delay which is at most slightly less than twice the clock period at the highest possible DDS output frequency. A suitable random number generator is a design by Rader et al. disclosed in the article "A Fast Method of Generating Digital Random Numbers," The Bell System Technical Journal, November 1970. This design has a delay equal to the settling time of a flip-flop and the propagation delay of an exclusive-or gate. Rader lists the sequence period for register lengths from L=1 to 25. A longer sequence period represents a better pseudorandom number approximation to a true random number. Rader shows a period of 159783 for L=13, which requires a reasonable amount of hardware, and it has a good auto­correlation function of noise generation. The distribu­tion is uniform, and any number of bits required (up to thirteen) for dithering can be selected from it to dither the accumulator of the DDS.

0035The DAC 67 is the final stage in the disclosed DDS. Following the DAC 67 is a filter for smoothing the output sinewave and reducing the alias amplitude (not shown in FIG. 2). The performance of the DAC 67 is of concern since the spurious performance of the DDS can be affected. The circuitry disclosed in FIG. 2 is all digital and can be implemented in Very High Speed Inte­grated circuit (VHSIC) technology with clock rates in the gigahertz. The DAC 67, however, must settle to within a specified accuracy to the desired level within one clock period. In addition, the spectral performance of the DDS is degraded by DAC errors.

0036At the state-of-the-art, Hewlett-Packard Labora­tories has developed a twelve bit GaAs DAC capable of conversion rates of 1 gigaword/s. The packaging required for linearity performance at full speed requires microwave techniques. Such a design demonstrates the advancements being made in DAC technology and will certainly have application in advanced DDS designs. However, conversion rates of typical off-the-shelf ten bit DACs can range up to 100 megawords/s, which has proved satisfactory in simulations of the circuit of FIG. 2.

0037The operation of the preferred embodiment in suppressing spurs may be graphically illustrated using FIGS. 3 and 4. FIG. 3 shows spurs 101, 103, 105, typi­cally present about the fundamental 107 in the output of a DDS with an undithered accumulator. The preferred embodiment functions to suppress the amplitude of such spurs 101, 103, 105 as shown in FIG. 4. Suppression of the close-in spurs 101, 103 is a particular advantage of the invention over the prior art. The spectrum of FIG. 4 exhibits decreasing phase noise with frequency offset from the output frequency and therefore has a substantially flat frequency deviation density and reduces all spurs including those close-in.

0038The foregoing discussion has illustrated a dithered accumulator circuit and its application to reduce spurs in the output of a DDS circuit. Those skilled in the art will appreciate that many adaptations and variations of the disclosed embodiments may be made without departing from the scope and spirit of the inven­tion. Therefore, it is to be understood that, within the scope of the appended claims, the invention may be practiced other than as specifically described herein.