Decimation Filter Having a Selectable Decimation Ratio

This application is related to patent application Ser. No. 08/265,343, entitled "Decimation Filter Using a Zero-Fill Circuit For Providing a Selectable Decimation Ratio," filed concurrently with the present application, assigned to the assignee of the present invention and herein incorporated by reference.

This invention relates to decimation filters, and more particularly, to decimation filters such as for use with sigma-delta modulators.

Several conflicting objectives are present in the field of high performance data acquisition for electrical power measurement, metering, and management. A first objective is high data resolution. The availability of high resolution analog-to-digital conversion methods, such as with sigma-delta modulators, combined with the power of state of the art digital signal processors provides the capability to achieve significant levels of accuracy. Thus, complex signal processing may be performed on precise data using conventional signal processing architectures. However, a second objective is increased signal processing bandwidth. An inherent tension exists between these two objectives in that increasing bandwidth typically occurs at the expense of data resolution. Yet, a third objective is compactly integrated or electronic circuitry that requires less power to operate. A need thus exists for a decimation filter having the capability to balance high data resolution demands against large bandwidth demands, as desired for power measurement, metering, and management systems, while having the size and power requirements of a conventional electronic or integrated circuit component.

It is further desirable to provide a decimation filter with improved normalization or scaling characteristics in order to improve data access speed and minimize filter circuitry complexity. U.S. patent application Ser. No. 08/025,456 by J. E. Krisciunas et al, filed Mar. 3, 1993, assigned to the assignee of the present application and herein incorporated by reference, describes a technique which, although effective in providing suitable normalization for a desired decimation ratio, employs relatively complex synchronous conversion on the filter output signals. The technique described therein uses a parallel-to-serial (PISO) converter and/or tapped delays which, in general, are not suitable for asynchronously reading out filter output signals for any additional signal processing. The present invention advantageously provides a coefficient generator having the capability of providing coefficient signals with variable scaling. In accordance with another objective of the present invention, an overflow detector is employed in the decimation filter to detect and comet any overflow condition which can occur under predetermined conditions. As described in U.S. patent application Ser. No. 08/025,456 by J. E. Krisciunas et al, one way to avoid the overflow condition is to modify the ideal response of the filter. In general, this modification results in a slight alteration in the magnitude response of the decimation filter realization which can introduce substantial distortion at relatively low decimation ratios. Thus, there is a need to provide a decimation filter in which the magnitude response is impervious to any selected decimation ratio.

Generally speaking, the present invention fulfills the foregoing needs by providing a decimation filter for filtering at least one externally derived stream of quantized electrical signals having a predetermined rate. The filter comprises a coefficient generator responsive to a set of externally derived decimation-ratio select signals to provide a separate normalized coefficient signal at each respective one of a plurality of output ports. An accumulator is coupled to the coefficient generator to receive each normalized coefficient signal generated therein. The accumulator receives the stream of quantized electrical signals so as to produce, upon masking with respective ones of the received normalized coefficient signals, a plurality of accumulator output signals. An overflow detector is coupled to the accumulator to detect any overflow condition arising in the accumulator.

A method for decimation filtering at least one stream of externally derived binary signals in accordance with the invention comprises the steps of: generating a predetermined sequence of normalized coefficient signals in response to a set of externally derived decimation-rate select signals; receiving the externally derived stream of binary signals; masking the received externally derived stream of binary signals with the predetermined sequence of normalized coefficient signals; and accumulating the masked signals to provide a filtered signal. The step of generating the predetermined sequence of normalized coefficient signals may comprise the steps of: generating N counter signals wherein N is a predetermined integer chosen so that 2.sup.N constitutes an upper decimation ratio bound of the decimation filter; and generating N+S-1 normalized signals upon predeterminedly shifting and zero-filling respective ones of the generated N counter output signals, wherein S is a predetermined integer less than N corresponding to a number of selectable decimation ratios provided by the decimation filter and chosen so that 2.sup.N-(S-1) constitutes a lower decimation ratio bound of the decimation filter.

The subject matter regarded as the invention is particularly pointed out and distinctly claimed in the concluding portion of the specification. The invention, however, both as to organization and method of operation, together with further objects and advantages thereof, may best be understood by reference to the following detailed description when read with the accompanying drawings in which:

FIG. 1 is a block diagram of one embodiment of a sigma-delta modulator such as may be incorporated in an analog-to-digital converter having a decimation filter which has a selectable decimation ratio capability in accordance with the invention;

FIG. 2 illustrates an embodiment of a sigma-delta modulator comprising electrical or electronic components, such as may be incorporated in an analog-to-digital converter having a decimation filter in accordance with the invention;

FIG. 3 shows spectral shaping of quantization noise such as may be accomplished by the use of a sigma-delta modulator;

FIG. 4 is a schematic block diagram of a decimation filter which uses a coefficient generator with variable scaling to provide normalized coefficients in accordance with the invention;

FIG. 5 shows further details in block diagram form of the coefficient generator shown in FIG. 4 including its normalizer circuit;

FIGS. 6A-6E collectively illustrate the normalizer circuit of FIG. 5;

FIG. 6 is a block diagram showing how FIGS. 6A-6E are assembled with respect to each other;

FIG. 7 shows the frequency response for a rectangular "window" decimation filter and a triangular "window" decimation filter;

FIG. 8 is a plot of filter coefficients such as may be realized by an embodiment of a decimation filter in accordance with the invention;

FIGS. 9A-9C illustrate respective exemplary multi-channel stages for one embodiment of the accumulator of FIG. 4 in accordance with the invention;

FIG. 9 is a block diagram showing how FIGS. 9A-9C are assembled with respect to each other;

FIG. 10 is a timing diagram showing waveforms associated with the accumulator of FIGS. 9A-9C, plotted as a function of time; and

FIG. 11 is a block diagram showing details for one exemplary embodiment of the overflow detector of FIG. 4.

Conventional analog-to-digital conversion frequently involves use of oversampled analog-to-digital converters. One example of such an analog-to-digital converter that employs a sigma-delta modulator is illustrated in FIG. 1. Systems in which analog-to-digital converters and sigma-delta modulators may prove useful are described in U.S. Pat. No. 5,181,033 entitled "Digital Filter for Filtering and Decimating Delta Sigma Modulator Output Signals," by Yassa et al., issued Jan. 19, 1993, U.S. Pat. No. 5,126,961, entitled "Plural-Channel Decimator Filter, as for Sigma-Delta Analog-to-Digital Converters," by Garverick, issued Jun. 30, 1992, U.S. Pat. No. 5,134,578, entitled "Digital Signal Processor for Selectively Performing Cordic, Division or Square-Rooting Procedures," by Garverick et al., issued Jul. 28, 1992, U.S. Pat. No. 4,951,052, entitled "Correction of Systematic Error in the Oversampled Analog-to-Digital Converters," by Jacob et al., issued Aug. 21, 1990, and U.S. Pat. No. 4,896,156, entitled "Switched-Capacitance Coupling Network for Differential-Input Amplifiers Not Requiring Balanced Input Signals," by Garverick, issued Jan. 23, 1990, all of the foregoing patents assigned to the assignee of the present invention and herein incorporated by reference, and described in "A Programmable Mixed Signal ASIC for Power Metering," by S. L. Garverick, K. Fujino, D. T. McGrath, and R. D. Baertsch, IEEE Journal of Solid State Circuits, Vol. 26, No. 12, December 1991, pp. 2008-16, and "A Programmable Mixed Signal ASIC for Power Management," by D. T. McGrath, P. Jacobs, and H. Sailer, IEEE 1992 Custom Integrated Circuits Conference, pp. 19.4.1-19.4.2, both of which are herein incorporated by reference.

The technique of oversampling is frequently used in performing analog-to-digital conversion. An example of an analog-to-digital converter employing this technique is illustrated in FIG. 1. In this illustration, a converter 700 comprises a modulator 710 and a decimator 720. Modulator 710 produces coarse, or low-resolution, samples of its input signal, at a rate much greater than that required by the Nyquist theorem. These coarse samples are subsequently low-pass filtered by decimator 720 to produce high-resolution samples at or above the Nyquist rate of the input signal of the modulator.

As illustrated, modulator 710 includes negative feedback to node 810, which is coupled to the input of an integrator 820, the output of which is coupled to the input of a Q-bit analog-to-digital converter 760. The output of analog-to-digital converter 760 is coupled to a Q-bit digital-to-analog converter 800, the output of which is coupled to node 810. An analog input signal applied to terminal 790 is sampled by a sampling device 780, illustrated as a single-pole single throw switch, at a rate f.sub.ds much higher than the Nyquist rate. These samples are compared to a Q-bit estimate of the input signal provided by digital-to-analog converter 800, using the difference signal at node 810. The difference, or error signal, produced at node 810 is integrated by integrator 820, then quantized by Q-bit analog-to-digital converter 760 to produce the Q-bit estimate of the input signal at terminal 795 which is supplied to the input port of digital-to-analog converter 800 and a decimator 720. The combination of feedback and error-signal integration serves to shape the spectrum of the quantization noise such that it may be more completely removed by the low-pass filtering. The effectiveness of the noise shaping is improved when the number of integrators in the feedback loop (i.e., the order of the modulator) increases, but serious stability problems are encountered for modulators of order greater than two.

As illustrated in FIG. 1, the decimator comprises a digital low-pass filter 730 followed by signal sampling device 740, shown as a single-pole single throw switch, to provide a discrete output signal at a node or terminal 750. This filtering and decimation results in a large fraction of the quantization noise being removed to thereby provide a high resolution output signal. Nonetheless, it will be appreciated that the resolution is gained by having a throughput much lower than the initial sampling rate, f.sub.ds. The ratio of the initial sampling rate to the modulator conversion rate is typically referred to as the oversampling ratio R of the modulator. As described in "A Comparison of Modulation Networks for High-Order Oversampled S D Analog-to-Digital Converters," IEEE Transactions On Circuits and Systems, Vol. 38, pp. 145-159, by D. B. Ribner, published in February 1991, and herein incorporated by reference, the resolution (bits) of such an analog-to-digital conversion is governed by the number of quantization bits Q, the oversampling ratio R, and the order of the modulator, L, in accordance with the following expression: ##EQU1## This expression may be derived from a linear approximation to the modulator and by assuming an ideal lowpass filter. Equation [1] shows, in essence, that each time the oversampling ratio R is doubled, resolution is improved by L+1/2 bits.

Equation [1] may be evaluated for a first order modulator for various oversampling ratios. It will be appreciated that in accordance with equation [1] above, the higher the oversampling ratio the greater the corresponding number of bits, given a predetermined order for the sigma-delta modulator. For some applications, use of a one-bit quantizer may be desirable. One advantage of such a quantizer is that it may avoid harmonic distortion or other nonlinearities typically associated with multiple bit quantizers. In contrast with multiple bit quantizers, a one-bit quantizer is inherently linear because its output signal takes on one of two distinct values and, thus, determines a straight line. A sigma-delta modulator using a one-bit quantizer has been analyzed in "A Use of Limit Cycle Oscillations to Obtain Robust Analog-to-Digital Converters," IEEE Transactions on Communications, COM-22(3), pp. 298-305, by J. C. Candy, published in March, 1974, and herein incorporated by reference. Equation [2] below provides q(n.tau.), the nth quantized signal, where .tau. represents the cycle period: ##EQU2## The quantization represented by the addition of an error, .epsilon., is assumed to be uncorrelated to the input signal, x. Because q(t) is a sampled function, its integral may be expressed as a summation, which may be rearranged and divided by R, the number of cycles in a sample time, providing equation [3] below. ##EQU3## Equation [3] indicates that the average quantization error will be R times smaller than the "coarse" quantizer error. Thus, high resolution may be obtained by repeated feedback with a sufficiently large R.

FIG. 2 illustrates one embodiment of the first order sigma-delta modulator for incorporation in an analog-to-digital converter having a decimation filter in accordance with the invention. As illustrated, the input signal, such as an electrical voltage, is provided at a terminal 980 and scaled or attenuated by a resistance 960. An amplifier 940 performs integration due to the presence of a capacitor 950 in a negative feedback configuration. Likewise, digital quantization is accomplished by coupling the output signal of amplifier 940 to the positive input terminal of an analog comparator 930. As illustrated in FIG. 2, digital-to-analog conversion takes place because the output signal of comparator 930 is latched by a flip-flop 910 and provided to amplifier 940, as feedback, through a resistance 970. More specifically, one bit analog-to-digital conversion takes place due to the feedback provided through resistance 970 to integrating amplifier 940. This feedback through resistor 970 is equivalent to the feedback through digital-to-analog converter 800 in FIG. 1. The output signal at a node 990 of the modulator will have a spectral output corresponding to the shape illustrated in FIG. 3. Similar approaches have been employed, such as described in "A Seven-Channel Mixed Analog/Digital Signal Acquisition and Processing Architecture," a masters thesis for Rensselaer Polytechnic Institute, prepared by P. L. Jacob in December 1988, and herein incorporated by reference.

As previously discussed, an oversampling modulator shapes the spectrum of the quantization noise such that most of the noise energy is allocated to high frequencies. This is illustrated in FIG. 3, which is a plot of the simulated quantization noise spectral density of a third-order, one-bit sigma-delta modulator; a first-order modulator is qualitatively similar. The highest frequency illustrated, bin number 16000, corresponds to f.sub.ds /2. Since the decimator filters quantization noise energy between f.sub.ds /2R and f.sub.ds /2, then reduces the sampling rate to f.sub.ds /R, a trade-off exists between resolution and sampling rate which may be controlled by adjusting parameter R.

One embodiment of a decimation filter is a rectangular window filter. This conventional type of lowpass filter offers simplicity. Rectangular window filtering performed in the time domain has the frequency response of a sinc function. For a window of duration or length t, the frequency response is sinc (ft) with the first zero occurring at f=1/t, where in the context of the invention sinc(x) refers to [sin (.pi.x)]/.pi.x. The window length or duration t is selected such that the first zero occurs at the conversion rate f.sub.ds /R. Thus, t=R/f.sub.ds. This filter has the desirable characteristic of linear phase. The various channels of a multi-channel system pass through the same linear phase decimation filter and thus realize substantially the same time delay for the relevant range of input frequencies.

Despite the simplicity of a rectangular window decimation filter, a "sharper" cutoff in the frequency domain may be achieved by using higher order filtering. One such embodiment is the so-called double interpolation filter in which the cutoff is sharper and stopband attenuation is increased in comparison with the rectangular window. One advantage of such a filter is that it reduces the amount of noise "leaking" through the stopband and thus aliased back into the baseband following decimation. One embodiment of a double interpolation filter is the triangular window, having a frequency response of sinc.sup.2, such as described in the aforementioned Candy paper and in "Using Triangularly Weighted Interpolation to Get 13-Bit PCM from a Sigma-Delta Modulator," IEEE Communications, Com. 24, pp. 1268-1275, by J. C. Candy, Y. C. Ching, and D. S. Alexander, published in November, 1976 and herein incorporated by reference. Thus, for a filter of length t, the frequency response is sinc.sup.2 (f.multidot.t/2) with the first zero at f=2/t. It will now be appreciated that to maintain the first zero at the decimation frequency, the filter length for this particular embodiment becomes doubled. Thus, t=2R/f.sub.ds places zero frequency response at the conversion rate f.sub.ds /R. FIG. 7 illustrates the frequency response of a first order rectangular filter of length t=16 and a second order triangular filter of length 2t=32, demonstrating that a sharper cutoff and improved stopband attenuation are achieved by the second order filter.

FIG. 4 illustrates one embodiment of a decimation filter having a selectable decimation ratio in accordance with the present invention. As used herein the expression "decimation ratio" refers to the reduction or down-conversion factor in sampling rate provided by the decimation filter. In particular, coefficient generator 10 is designed to provide a separate normalized coefficient signal at each of a plurality of output ports in response to a set M of externally derived decimation-ratio select signals 11. A multi-bit accumulator 100 is coupled to generator 10 to receive the normalized coefficient signals 12. Accumulator 100 also receives a stream of quantized or binary signals 14 from modulator 710, for example (FIG. 1). Normalization is desirable because depending on the decimation ratio selected, the output signals of the filter will be provided in binary multiples for the same modulator input signals, i.e., without suitable normalization for any given decimation ratio selected, a user will not be able to properly interpret the correct value for the filtered signals produced by the decimation filter. U.S. patent application Ser. No. 08/025,456 by J. E. Krisciunas et al, filed Mar. 3, 1993, describes a technique which, although effective in providing suitable normalization for a desired decimation ratio, employs a somewhat complex synchronous conversion on accumulator output signals. Briefly, the technique described therein uses at least a parallel-to-serial (PISO) converter and/or suitable tapped delays. The aforementioned technique, in general, is not applicable to asynchronous readout of the accumulator output signals for any additional signal processing. As will be explained shortly hereafter, the present invention advantageously provides in coefficient generator 10 the capability of providing coefficient signals with variable scaling. Since the coefficient signals supplied to the accumulator are suitably normalized, the accumulator output signals are ready to be utilized for further processing without having to employ the aforementioned synchronous conversion.

In accordance with another advantage of the present invention, an overflow detector 200 is coupled to accumulator 100 to detect and correct any overflow condition which may occur in the accumulator under predetermined conditions. The overflow condition arises whenever the modulator output signal is at full scale, that is, during a predetermined triangular "window" the modulator supplies a stream of quantized signals, each having a binary value of "ONE", and consequently the summing or accumulating capacity of the accumulator is exceeded when the modulator output signal is at full scale. As described in U.S. patent application Ser. No. 08/025,456 by J. E. Krisciunas et al, one way to avoid the overflow condition is to always reduce the triangular "window" count by one, for example, by dropping the next to last coefficient of the triangular "window". This modification results in a slight alteration in the magnitude response of the decimation filter realization which can introduce substantial distortion at relatively low decimation ratios. Because of overflow detector 200, the decimation filter is no longer subject to the above-described modification to its triangular window and consequently, regardless of the selected decimation ratio, the magnitude response remains unaffected.

Coefficient generator 10, as shown in FIG. 5, comprises, for example, an N-bit "up" counter 20 responsive to a suitable counter clock or control signal CTRL which substantially corresponds to a predetermined multiple of the rate of arrival of the stream of quantized electrical signals. Counter 20 includes N counter output ports, each designed to provide a separate counter output signal. Letter N represents a predetermined integer chosen so that 2.sup.N constitutes an upper decimation ratio bound of the decimation filter.

As illustrated in FIG. 5, triangular "window" weights or coefficients are generated after selected counter output signals are inverted by an inverter circuit 30 in response to a suitable inverter control signal UPDNB (waveform shown in FIG. 10). The inversion of the counter output signals allows for generating the down-ramp of the triangular "window" while the uninverted counter output signals in conjunction with a suitable carry-in signal (not shown) allow for generating the up-ramp of the triangular window. This feature of the coefficient generator takes advantage of the fact that the down-ramp of a triangular filter is simply the inverse of the up-ramp, as illustrated in FIG. 8. Other alternative implementations may conveniently provide such triangular "window" coefficients, for example, suitable "up/down" counters can be used in lieu of the "up" counter/inverter arrangement to generate the triangular "window".

The inverter output signals are coupled to a normalizer circuit 400 which, in response to set M of decimation-ratio select signals, provides the separate normalized coefficient signal at each respective one of N+S-1 normalizer output ports. Letter S represents a predetermined integer less than N corresponding to the number of selectable decimation ratios provided by the decimation filter and chosen so that 2.sup.N-(S-1) constitutes a lower decimation ratio bound of the decimation filter. It will be appreciated that the N+S-1 normalizer output ports comprise the plurality of output ports of coefficient generator 10. A suitable roll-over reset circuit 50 is also responsive to set M of decimation control signals so as to provide a predetermined reset signal to counter 20 upon completion of a desired count sequence. Provision is also made for providing an externally derived master reset signal (not shown) which allows for suitably resetting the counter and additional circuitry of the decimation filter upon the occurrence of predetermined events such as start of operations.

FIG. 6 (i.e., FIGS. 6A-6E, considered as an assemblage) shows an exemplary embodiment of a normalizer circuit employing a plurality of multiplexors 401-419, each respectively coupled to receive a respective set of inverter output signals represented by italic letters A-L, that is, each one of italics letters A-L represents one of N (e.g., N-12) inverter output signals (conversely, roman letters within a respective multiplexer represent pin designation for any given multiplexer). For example, the respective set of inverter signals received by multiplexer 401 is limited to the inverter output signal designated by the italic letter A. Similarly, the respective set of inverter output signals received by multiplexer 406 comprises inverter output signals B, D and F.

Each respective one of multiplexers 401-419 is responsive to predetermined combinations of set M of externally derived decimation-ratio select signals so as to produce a respective one of the normalized coefficient signals being supplied at each respective one of the N+S-1 output ports. Conventional Boolean logic gates such as AND, OR, and inverter gates 420-430 conveniently cooperate to provide such predetermined combinations. As will be explained below, in the exemplary embodiment of FIG. 6, set M comprises a number of three decimation-ratio select signals M.sub.0 -M.sub.2. For example, the predetermined combination of decimation-ratio select signals received by multiplexers 401 and 402 is conveniently provided by AND gate 420.

To facilitate understanding of the operation of normalizer circuit 400, Table 1 is provided with reference to a decimation filter having an upper decimation ratio bound chosen so that 2.sup.N =4096 (i.e, N=12); and having a lower decimation ratio chosen so that 2.sup.N-(S-1) =32 (i.e., S=8).

It should be appreciated that these respective bounds for the decimation filter are chosen in the manner indicated in Table 1 only by way of example and not of limitation. As previously explained, since S corresponds to the number of different decimation ratios that the decimation filter provides, it can be shown that the number of decimation-ratio control signals in set M is chosen in general so that the following property is satisfied

so that in the case of S=8, the number of control signals is three, previously designated as M.sub.0, M.sub.1 and M.sub.2.

In Table 1 and in FIG. 6, B.sub.0 -B.sub.18 represent the N+S-1 normalizer circuit output ports. As previously stated, the inverter output signals are represented by the italic letters A-L which in this example are 12 (since N=12). Those skilled in the art will appreciate that for a decimation ratio of 2.sup.N (here 2.sup.N =4096) the required scaling factor is equal to unity. It can be shown that multiplexers 401-419 in FIG. 6 will supply at their respective output ports B.sub.0 -B.sub.18, in response to predetermined combinations of signals M.sub.0 -M.sub.2, normalized signals as designated in the first row of Table 1, which correspond to a scaling factor of unity. For example, multiplexer 401 will supply the A signal at its output port B.sub.0 whenever the decimation-ratio select signals are in a state or condition such that each has a value corresponding to logic ONE, which correspond to a decimation ratio of 4096. For each remaining condition for signals M.sub.0 -M.sub.2, multiplexer 401 will supply a zero at its output port B.sub.0 in accordance with the B.sub.0 column in Table 1; that is, multiplexer 401 conveniently provides a zero-fill function as illustrated in Table 1. Similarly, multiplexer 402 will supply the B signal at its output port B.sub.2 whenever the decimation-ratio select signals are in the state or condition which corresponds to a decimation ratio of 4096. It can be shown that each remaining multiplexer will provide at its respective output port normalized signals in accordance with Table 1. It will be appreciated that output ports B.sub.12 -B.sub.19 are filled with zeros since, for a decimation ratio of 4096, only output ports B.sub.0 -B.sub.11 are used to provide the N output signals from the inverter with a scaling factor of unity.

For a decimation ratio of 2.sup.N-1 =2048, which is the next available decimation-ratio selection in Table 1, those skilled in the art will appreciate that to achieve proper normalization the normalizer circuit must provide a scaling factor of four, which is equivalent to shifting two places to the left the least significant bit of a binary representation. Inspection of FIG. 6 reveals that multiplexers 401-419 will supply at their respective output ports B.sub.0 -B.sub.18 normalized signals as designated in the second row of Table 1. For example, the A signal, which in the case of a decimation ratio having a value of 4096 was supplied at the B.sub.0 output port, is now supplied at the B.sub.2 output port. This is because whenever each of decimation-ratio select signals M.sub.0 -M.sub.2 has a respective state represented by 110, then, for instance, multiplexer 403 will supply signal A at its output port B.sub.2. From inspection of Table 1, it will be apparent that the normalizer circuit conveniently provides both a predetermined shift and zero-fill to the received inverter output signals A-L in order to supply suitably normalized coefficient signals at the N+S-1 output ports. It will be appreciated that the combinational logic illustrated in FIG. 6 conveniently operates asynchronously once the inverter signals are provided to normalizer circuit 400.

FIG. 9 (i.e., FIGS. 9A-9C, considered as an assemblage) illustrates an exemplary embodiment for accumulator 100 (FIG. 4). The embodiment of FIG. 9 shows, for the purpose of simplicity of explanation, only three stages 100.sub.0 -100.sub.2 of a 2N-bit accumulator comprising a number of 2N stages, each adapted to generate a respective one of the 2N accumulator bits. FIG. 9 shows that each stage is a multi-channel stage that includes a respective signal-stream multiplexer 102 having two input ports.

FIG. 9 can be conveniently used in conjunction with FIG. 10 to explain in more detail operational aspects of the accumulator. For example, during respective cycles of multiplexer control signal CHS (waveform shown in FIG. 10), multiplexer 102 supplies to AND gate 104 a respective stream of quantization signals, such as stream of signals DSV and the additional stream of signals DSI, each of which constitutes a respective modulator signal output. In accordance with this embodiment of the invention, decimation filter signal outputs may be produced by multiplying or masking each modulator signal output, such as signals DSV and DSI which may represent voltage and current measurements, respectively, with the normalized coefficient, or normalizer circuit signal outputs. FIG. 8 illustrates the time domain representation of the coefficients employed in the processing performed by the programmable decimator filter, that is, a decimation filter which is capable of being programmed to provide different decimation ratios as discussed in the context of Table 1 and FIG. 6.

Multiplication of the quantized electrical signals by the coefficient is conveniently performed at each respective stage by AND gate 104, by masking the output signals from coefficient generator 10 (FIG. 4) with each sigma-delta modulator output signal. Each AND gate 104 of the accumulator has two input ports and an output port. A respective one of the two input ports is coupled to receive the output signal from multiplexer 102. The other AND gate 104 input port is coupled to receive a respective normalized coefficient signal (designated COEF[0], COEF[1], and COEF[2] in FIGS. 9A, 9B and 9C, respectively) from the normalizer circuit. It will be appreciated that the coefficient signal received by the last N-(S+1) (here 5) respective stages of the 2N (here 24) accumulator stages is simply set to zero since the normalizer circuit is designed to supply only N+(S+1) (here 19) coefficient signals.

The output signal of AND gate 104 is supplied to one summand input port of a suitable full adder (FA) 106 having two summand input ports and a stun output port for supplying respective cumulative masked output signals. Adder 106 is coupled to receive masked signals from AND gate 104 at a respective one of the summand input ports. Adder 106 has a carry-in input port for receiving a carry-in signal, if any, and a carry-out output port coupled to a subsequent of the 2N stages for passing a carry-out signal to the subsequent stage, except the last stage wherein the carry-out output port is coupled to overflow detector 200 (FIG. 4). It will be appreciated by those skilled in the art that depending on the decimation ratio selected, not every carry-in input port in the accumulator has to be enabled at each stage; thus, to reduce computational burden, a predetermined enabling signal CARRYEN, suitably combined in logical gates 107.sub.1 and 107.sub.2, conveniently enables or disables predetermined ones of the carry-in input ports in the accumulator, depending on the decimation ratio being selected by the user.

In each stage, delay means, such as serially connected delay units 108.sub.1 -108.sub.4, impart a respective delay to the cumulative masked output signals from adder 106. Each respective delay is imparted at predetermined cycles of a suitable delay unit clocking signal (not shown) being received at a suitable clock port of each delay unit and predeterminedly synchronized with the rate of arrival of the stream of quantized signals. It will be appreciated that the delay units cooperate to provide in a time multiplexed manner of operation respective filter output signals. For example, after a complete number of accumulator cycles during which a triangular "window" has been formed, the content of predetermined delay units is latched into a respective set of latching units 110.sub.1 and 110.sub.2 in response to a LATCH signal that pulses to ZERO from its normal ONE state (waveform being shown in FIG. 10). It should now be appreciated that for the embodiment of FIG. 9, such time-multiplexed operation, conveniently allows for consecutive "windows" to overlap, (as shown in FIG. 8) resulting in the first zero of the triangular "window" filter occurring at the selected decimation frequency. A multiplexer 112 is coupled to respective delay means (e.g., 108.sub.3 and 108.sub.4) for providing two-separate feedback paths which are in turn selectively coupled to the other of the two summand input ports of adder 106 in response to a PIPE signal (waveform shown in FIG. 10) applied to multiplexer 112. It can be shown that this arrangement conveniently allows for computation of the suggested overlapping consecutive "windows". As shown in the embodiment of FIG. 9, first delay delay unit 108.sub.1, is coupled to directly receive the cumulative output signal from adder 106 while third and fourth delay units 108.sub.3 and 108.sub.4, respectively, am selectively coupled to the other summand input port of adder 106 through a respective one of the two feedback paths provided by multiplexer 112 whenever the PIPE signals reaches one of two predetermined levels. Decimation filter output samples or signals may be latched by the embodiment illustrated in FIG. 9 at the times indicated by the arrows in FIG. 8.

As will be appreciated by those skilled in the art, a required decimation ratio may be determined based upon the desired bandwidth, assuming that a triangular "window" decimation filter embodiment is desirable. Although this particular embodiment of the invention implements a triangular filter, it will be appreciated that a rectangular filter may likewise be implemented. Upon completion of a triangular "window" accumulation, a suitable RESET signal (waveform shown in FIG. 10) is employed for suitably resetting summer 116 for computation of the next triangular "window". For example, AND gate 114 provides one exemplary implementation for supplying the RESET signal to summer 106. Signals ISEL and VSEL can be provided by a suitable microprocessor (not shown) in order to read out a respective filtered output signal from the latch set 110.sub.1 -110.sub.2.

FIG. 11 shows an exemplary embodiment of overflow detector 200 comprising suitable detector stages, such as detector stages 200.sub.1 and 200.sub.2, for detecting a respective overflow condition in accumulator 100 (FIG. 4 and FIG. 9) due to one of the externally derived quantizing signals being supplied at full scale. Each detector stage includes a respective AND gate 201 coupled to supply a signal indicative of the origin of a respective overflow condition. In this embodiment, AND gate 201 in detector stage 200.sub.1 is coupled to combine the carry-out signal from the last accumulator stage with signal CHS while AND gate 201 in detector stage 200.sub.2 is coupled to combine the same carry-out signal with the complement of signal CHS. Such arrangement for AND gates 201 conveniently allows for determining in a multi-channel architecture which is the specific channel of accumulator 100 experiencing a respective overflow condition. This follows since signal CHS and its complement, when coupled to AND gates 201 as shown, allow for determining which modulator signal output is being accumulated within any giving accumulating cycle. Each detector stage further includes suitable latching units, such as flip-flops 204.sub.1 and 204.sub.2, which cooperate to latch or grab any signal or bit indicative of a respective overflow condition. In each detector stage an OR gate 202 is coupled to combine the output signal from a respective AND gate 201 with the signal indicative of the present state of a respective flip-flop 204.sub.1 clocked by a suitable clock signal CLK1 substantially synchronized with the clocking signal for the delay units discussed in the context of FIG. 9. Similarly, flip-flop 204.sub.2 is clocked by the LATCH signal (waveform shown in FIG. 10) so that upon completion of a triangular "window" each respective flip-flop 204.sub.2 can separately reach a respective present state which enables signal OVFV and OVFI, respectively, and which is indicative of a respective overflow condition. It will be appreciated that the RESET signal (waveform shown in FIG. 10) applied at a respective flip-flop CLEAR port allows each flip-flop 204.sub.1 to be suitably reset upon commencement of a subsequent triangular "window". Each respective overflow indication signal in turn is supplied to suitable combinational logical gates (not shown) in order to correct any overflow condition; for example, an "ORing" technique can be used for setting each accumulator output signal to ONE. In this manner, the overflow detector advantageously allows the decimation filter to advantageously operate at decimation ratios as low as 32, for example, without any distortion to its magnitude response.

Decimation filtering at least one stream of externally derived binary signals in accordance with the present invention may be accomplished by the following method. As exemplified in the embodiment for the programmable decimation filter illustrated in FIG. 4, a predetermined sequence of normalized coefficient signals is generated, such as by coefficient generator 10, in response to a set of externally derived decimation-rate select signals. An externally derived stream of binary signals is received, such as from a delta-sigma modulator. For instance, the externally derived stream may comprise a single bit signal stream from a one-bit sigma delta modulator. The received externally derived stream of binary signals is masked or multiplied, such as with an AND gate having two input ports and an output port, with the predetermined sequence of normalized coefficient signals and the masked signals are then suitably accumulated, such as by multibit accumulator 100, to provide a filtered signal. The step of generating the predetermined sequence of normalized coefficient signals may comprise the steps of: generating N counter signals wherein N is a predetermined integer chosen so that 2.sup.N constitutes an upper decimation ratio bound of the decimation filter; and generating N+S-1 normalized signals upon predeterminedly shifting and zero-filling respective ones of the generated N counter output signals (as described in the context of Table 1 and FIG. 6), wherein S is a predetermined integer less than N corresponding to a number of selectable decimation ratios provided by the decimation filter and chosen so that 2.sup.N-(S-1) constitutes a lower decimation ratio bound of the decimation filter.

While only certain features of the invention have been illustrated and described herein, many modifications, substitutions, changes, and equivalents will now occur to those skilled in the art. It is, therefore, to be understood that the appended claims are intended to cover all such modifications and changes as fall within the true spirit of the invention.