CA1176317A - Digital filter bank - Google Patents

Digital filter bank

Info

Publication number
CA1176317A
CA1176317A CA000397894A CA397894A CA1176317A CA 1176317 A CA1176317 A CA 1176317A CA 000397894 A CA000397894 A CA 000397894A CA 397894 A CA397894 A CA 397894A CA 1176317 A CA1176317 A CA 1176317A
Authority
CA
Canada
Prior art keywords
filter
output
sampling period
input
filters
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Expired
Application number
CA000397894A
Other languages
French (fr)
Inventor
Thomas G. Marshall, Jr.
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
AT&T Corp
Original Assignee
Western Electric Co Inc
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Western Electric Co Inc filed Critical Western Electric Co Inc
Application granted granted Critical
Publication of CA1176317A publication Critical patent/CA1176317A/en
Expired legal-status Critical Current

Links

Classifications

    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04JMULTIPLEX COMMUNICATION
    • H04J4/00Combined time-division and frequency-division multiplex systems
    • H04J4/005Transmultiplexing

Abstract

DIGITAL FILTER BANK

Abstract of the Disclosure A digital filter bank which has a discrete cosine transform section and a polyphase network section has the polyphase network section configured with a plurality of single input-double output digital filters. The filters have one of their outputs independent of their present input signal. This permits a single multiplier-accumulator to use a "memory swapping" procedure to accumulate all the requisite information from a complementary pair of the filters and to perform the necessary operations on it to compute an output value, while clearing the accumulator only once each sampling period.

Description

7631~

DIGITAL FILTER BANK

Technical Field The present invention relates generally to digital filter banks and relates more particularly to such filter banks which include a transform section in combination with a polyphase network section of digital filters.
Background of the Invention .
Digital filter banks are useful for performing certain selective operations on a broadband signal, such as for spectral analysis and for telephone communications~ In telephone communications, for example, it is frequently necessary to provide signal conversion between a TDM
(time-division multiplexed) signal format and an FDM
(frequency-division multiplexed) signal format. This can be accomplished in each direction in known ways by a signal translator incorporating a digital filter bank of appropriate design for effecting the conversion.
A digital filter bank can be constructed of a transform section connected to a polyphase network of digital filters. A polyphase network allows all the operations to be done at the input bit rate and only then multiplexed to a higher rate. Both such sections can be provided by one or more appropriately adapted large scale integrated circuits known as "digital signal processors"
which each contain on a single chip a large number of memory elements and an arithmetic unit in the form of a multiplier-accumulator for operating on values from the memory elements in a selected mode.
Since each digital signal processor generally has only a single multiplier-accumulator available for operating on the values from the memory, it is important to make eEficient use of it. ~dditionally, it is highly desirable to minimize the number of times that the contents of the accumulator of the multiplier-accumulator must be ~ ~76317 cleared and transferred to the memory each sampling period, because each such operation tends to introduce round-off noise in the signal.
Summary of the Invention .
In accordance with an aspect of the invention there is provided a digital filter bank of the type including a discrete transform means connected to a polyphase network comprising at least one complementary pair of first and second digital filters, characterized in that each of said filters comprises a single input which provides a filter input signal during the course of a present sampling period; at least a first output which during the present sampling period gives a first output signal value independent of the present sampling period filter input signal and dependent on the filter input signal of a previous sample period; a second output which during the present sampling period gives an output signal value dependent in general on the value of the present filter input signal, and a shift register which for the present sampling period receives a register input signal dependent on the present filter input signal; said shift register including a first set of stages comprising a first stage and every other one of a plurality of successive stages and including a second set of stages comprising a second stage and every other one of said plurality of successive stages, said first filter output being accumulated from weighted values of said first set of stages and said second filter output being accumulated from the weighted value of said register input signal and ~0 weighted values of said second set of stages.
Brief Description of the Drawing FIG. 1 is a schematic signal flow path block diagram of a digital filter bank in accordance with a preferred embodiment of the present invention, including ~ ~76317 - 2a -a polyphase network filter made up of a plurality of SIDO
(single input, double output~ digital filters;
FIG. 2 is a schematic signal flow path block diagram of one of the SIDO digital filters of the polyphase network of FIG. l; and FIG. 3 is a schematic signal flow path schematic block diagram of an alternative structure for the digital filter of FIG. 2 in accordance with another preferred embodiment of the invention.
Description of the Preferred Embodiment The schematic signal flow diagram of FIG. 1 shows a digital filter bank 10 in accordance with the present invention which is designed to serve as a 14-channel TDM/FDM signal translater. The diagram of FIG. 1 is ~ ~6317 configured to represent the various operations of the filter bank 10 as occurring simultaneously and there~ore includes a number of signal delay devices~ In actual practice these delays are in large part provided in known ways by appropriate time frame ordering of the operations performed on the signal.
The digital filter bank 10 includes a discrete cosine transform section 12 and a polyphase network section 1~. The polyphase network section 14 is constructed with 14 digital filters 16 shown enclosed by broken lines. The structure and general mode of operation of the cosine transform section 12 is well known to those of ordinary skill in the art of digital filter transforms and is not discussed in detail here. The general structure and function of the polyphase network section 14 and the nature of its interaction with the cosine transform section 12 is also well known to those of ordinary skill in the art of digital filter bank design and is not discussed in detail here, except as regards features which pertain to the present invention.
The designing of digital filter banks and their description generally involves numerous mathematical relationships which are best expressed in mathematical notation. In the interest of conciseness, such notation is used in the following description of the filter bank 10 and is present in the drawings. The symbols used are ones customarily used in the art of digital filtering~ and as such the description is believed to be in a form readily understandable to a person of ordinary skill in the art of digital filtering. Various mathematical expressions which are set out from the text are numbered for reference elsewhere.
The digital filter bank in accordance with the present invention can be effectively implemented by a person of ordinary skill in the art of digital filter banks by appropriately adapting one or more digital signal processors. Such processors are presently in use and may ~ 1176317 include a memory unit, an arithmetic unit, a control unit, an input-output unit and a machine language storage unit in a single V~SI circuit or may alternately be a com-bination of a number of separate VLSI circuits interconnected to provide the needed functions.
One device which is suitable for implementing the digital filter bank of the present invention is a DSP
tDigital Signal Processor~ manufactured by the Western Electric Co., a corporation of New York, U.S.A. and described, for example, in the following:
Copending Canadian Patent application of J.R.
Boddie, R.N. Gadenz and J.S. Thompson, Serial No. 370,509 fil-ed February 10, 1981 and Copending Canadian Patent application of J.R.
Boddie and J.S. Thompson, Serial No. 370,508 filed February 10, 1981.
Another suitable device is a Signal Processing Interface NEC ~ PD7720, manufactured by the Nippon Electric Co. of Japan and marketed by Microcomputers, Inc., Wellesley, Mass., U.S.A. A suitable combination of devices can be assembled by combining memory and controller circuits with a MAC-16 (TDC 10 10 J) multiplier-accumulator manufactured and marketed by the TRW Co. of California, U.S.A. The programming of a particular VLSI digital signal processor circuit configuration can be readily performed by a person of ordinary skill in the art of digital signal processor implementations.
The following items, some of which are referred to in the discussion below, are cited as examples of publications describing various aspects of digital filtering. An understanding of these aspects may be helpful for gaining an understanding of the digital filter bank 10:
U.S. Patents 3,891,803 3,971,922 6 3 ~ 7 ~ 4,237,551 Technical Publications M. G. Bellanger and J. L. Daguet, "TDM-FDM
Transmultiplexer: Digital Polyphase and FFT", IEEE Trans. Comm., vol. COM-22, pp. 1199-1205, Sept. 1974.
G. Bonnerot, M. Coudreuse and M. G. Bellanger, "Digital Processing Techniques in the 60 Channel Transmultiplexer", IEEE Trans. Comm., Vol. COM-26, pp. 698-706, May 1978.
S. L. Freeny, R. B. Kieburtz, K. V. Mina and S. K. Tewksbury, "Design of Digital Filters for an All Digital Frequency Division Multiple~-Time Division Multiplex Translator", IEEE Trans. Circuit Theory, Vol. CT-18, pp. 702-711, November 1971~
"Systems Analysis of a TDM-FDM Translator/Digital A-Type Channel Bank", IEEE Trans. Comm., Vol. COM-l9, pp. 1950-1959, December 1971.
M. J. Narasimha and A. M. Peterson, "Design of a 24-Channel Transmultiplexer", IEEE Trans. Acoust., Speech, Signal Processing, Vol. ASSP-27, pp. 752-761, December 1979.
John Makhoul, "A Fast Cosine Transform in One and Two Dimensions~, IEEE Trans. Acoust., Speech, Signal Processing, Vol. ASSP-28, pp. 27-34, February 1980.
P. Vary and U. Heute, "A Short-Time Spectrum Analyzer with Polyphase-Network and DFT", Signal Processing, Vol. 2, pp~ 55-65, January 1980.
P. Vary, "On the Design of Digital Filter Banks Based on A Modified Principle of Polyphase", AEU, Band 33, Heft 7-8, 1979, pp. 293-300.
While the polyphase network section 14 of the filter bank 10 is often designed with 14 or 64 digital filters 16, the following discussion of the filter structure and the interconnections of the polyphase network section 14 will relate to a more general case of a design .

.

, . . ' ~ ~76317 suited for any desired number N of digital filters 16, where N is an even integer. In this regard, it should be recognized that while the filter bank 10 includes a discrete cosine transform section 12, other types of fast Fourier transform sections can also be used and may even be preferable for a different number of input channels. The labeling of the inputs 18 is from X0 to XN_l, instead of from Xl ~o XN~ as a convenience in making calculations for the design parameters of the filter bank 10.
The schematic diagram of FIG. 1 shows the signal flow of the filter bank 10. The signals in their various states in the filter bank 10 are represented throughout by the notations X, W, and Y, as abbreviations for their respective z-transform notations, which would be X(zN), W(zN), and Y(zN). The signals themselves are sequences of values, such as the input sequences usually obtained by sampling analog signals at a given sampling rate, the inverse of this sampling rate being a sampling period.
A number N of input signal channels 18 supply N
input signals X0 to XN_l to the discrete transform section 12, which here performs a discrete cosine transform. Typically, N=14, so that there would be fourteen such input signals X in the form of sequences of values coming from a demultiplexer (not shown) receiving a single TDM input signal. In this instance, the first X0 input channel 18 and the last XN_l input channel 18 function as guard bands and have no values which may be regarded as zero values for all sampling periods. The even-number channel signals are often voice signals sampled at an 8 kHz rate, whereas the odd-numbered channels are often similarly sampled voice signals modified by multiplying every other sample by minus one to facilitate the filtering or selecting of the lower sideband of all signals by the filter bank. This is described, for example, in the Freeny et al publication cited aboveO
The output of the discrete transform section 12 is a set of N transformed signals 20 labeled from W0 to ~ 176317 WN 1~ which then pass to the pol~phase network section 14 having N input branches 22. The term "transformed signals" refers to signals derived from a discrete transform and does not refer to the use of z-transform notation, which applies to all signals. Each input branch 22 is associated with a corresponding one of the trans-formed signals 20, and will therefore be individually identified herein by reference to the W transformed signal 20 associated with it. The transformed signals 20 are processed by the SIDQ's 16 to eventually result in a corresponding number of signals 24, labeled Y0 to YN 1 which pass to a multiplexer 26 having a single Y(z) FDM
output 28. The notation ,,zN,, as the argument corresponds to a sampling period N T which is N times that of a signal with the argument "z" which has sampling period T.
The signals 24, of sampling period N-T have been delayed from 1 to N-l periods T by delays 42 are combined sequentially by multiplexer 26 to give signal Y(z) of sampling period T.
With the exception of the filter 16 of the W0 branch 22, the filters 16 are SIDO (single input, double output) filters with identical 1 recursive portions 30 and two non-recursive portions 32 labeled A and -A, and also include a delay register 34, labeled z N, which delays one of the output signals by N output periods for a total delay of N-T. Since the ~An non-recursive functions has a factor of z N for all n ~ N, it is ~onvenient to show this delay explicitly as the z N
delay register 34 in FIG. 1.
In the W0 branch 22, the signal passes through a conventional single input, single output digital filter 15 which includes a recursive portion 30 and a single non-recursive portion 32 leading directly to the multiplexer 26. Each of the SIDO filters 20 of the remaining branches 22 has a single input W, a first ~ 17~317 output 36, and a second output 38.
It can be seen from FIGo 1 that from the SIDO
filters 16 there is a set of first outputs 36 denoted by Y2N-l,N-l to YN+l,l which are independent of the transformed signals 20 for the present sampling period but dependent on these signals for a previous sampling periodO
For this notation ~yl~ as well as for other notatlons herein which identify a signal or a component associated with two different paths of transformed signals 20, there are provided a pair of subscripts which indicate the nature of the association~ There is also a set of second outputs 38 from Yl 1 to YN_l N-l which are dependent in general on the transformed signals 20 for the present sampling period. Between the non-recursive filter portions -A and the first outputs 36, the signals are delayed by the delay registers 34. The signals Y2~_1 N-l to YN+l 1 from each of the first output 36 of the SID0 filters 16 are accumulated with those Gf the second output 38 of a complementary one of the SIDO filters 16 by adders 40. The pair of subscripts of the symbol Y of each of the two outputs 36, 38 includes a first one which identifies the input branch 22 of the non-recursive filter portion A, -A from which the signal is coming and a second one which identifies the input branch 22 of the complementary SID0 filter 16 to which it is being sent for output accumulation. This notation therefore describes the network interconnections of the filters 16 without showing each of the actual paths, which would result in needless crowding of detail in FIG. 1. The interconnection of the WN/2 input branch 22 SIDO 16 is also shown in particular in FIG. 1 because its SIDO filter 16 is its own complement.
The Y3N/2 NJ2 first output 36 of its SID0 filter 16 is accumulated with its own ~N/2 N/2 second output 3B, rather than with the second output 38 of one of the other SIDO
filters 16. From the adders 40 the signals proceed to delay registers ~2, so that they sequentially enter the multiplexer 26 to yield the single FDM output signal at the ~ 1 7 6 3 1 7 g output 28.
The operation of the SIDO filters 16 is shown in more detail in the flow path diagram of FIG. 2. Signal flow symbols in FIG. 2 are provided with subscripts which indicate that the filter 16 is for the nth input branch 22, where n is a variable integer ranging from 1 to N-l. It can therefore be the SIDO filter 16 for any one of the input branches 22 from the Wl branch 22 to the WN_ branch 22.
The nth SIDO filter 16 has as its input the Wn transformed signal 20. It has a first Y2N_n N-n output 36 and a second Yn n output 38. A central feature of the SIDO
filter 16 is a shift register 44 (outlined by broken lines) having a set of N stages 46, labeled z-N, with an access tap 48 below each stage 46. The Wn transformed signal 20 passes through a multiplier 50 with coefficient ~n and a first one of a first set of adders 52 to become a present input signal To n for the shift register 44. From there it enters the first one of the z-N register stages 46 and is transferred to the second and successive stages 46 at the rate of one shift each sampling period.
The taps 48 of the first stage 46 and of every other successive stage 46 are respectively multiplied by a corresponding set of multipliers 54 with coefficients being ~1 n to ~-2L-l n~ and accumulated by the second set of adders 56 to yield a signal Y2N-n,N-n for the f output 36. It is noted that the Y2N_n N-n first output 36 is independent of the present register input signal To n~
and can be co~puted either before or after To n is computed during substantially the entire sampling period. The Y2N_n N-n first output 36 is, however, dependent in general on the register input signal of a previous sampling period because the coefficients n of the multipliers 5~ are in general non-zero for making the best use of the filter bank 10. In specific instances, however, certain of the multipliers 54 could have zero coefficients ~n' thereby removing the dependence in a technical sense. The , , . ~ . .

~ ~76317 /
independence of Y2N_n,N_n from To,n allows computed immediately following the computations of YN_n N-nt so that the two signals can be added with adder 40 leading to signal YN_n; this can be done e~en if N-n is less than n because of the stated independence from To n Therefore, all the signals Yn can be computed in numerical order even though their summands, YN+n n~ are derived from N-n SIDO's whose indices change in reverse order.
The taps 4~ of the z-N second register stage 46 and of every other z-N successive stage 46 are connected through one of a respective set of multipliers 60 from ~2 n to ~2L-2 n to a set of adders 62. These taps 48 are also connected through another respective set of multipliers 64 from -~2 to -~2 L to the set of adders 52.
The accumulated value from the set of adders 52 is connected so that it contributes to input To n~ thus resulting in a feedback, or recursive path. The signal To n does not within its present sampling period depend upon the accumulated values of either set of adders 56 or 64, so that the paths incorporating these adders is feed-forward, or non-recursive. The accumulated value from the set of adders 62 is accumulated with the register input si~nal To n via a multiplier 68 with coefficient ~o n by the adder 70 to generate a sum signal Yn n. It is noted that the signal Yn n of the second output 38 depends on the present register input signal To n.
An alternate form of the SID0 filter 16 of FIG. 2 is the SIDO filter 74 shown in FIG. 3 which can realize the same functional relations between Yn n and Wn and between Y2N_n N-n and Wn as that of FIG. 2. The SID0 filter 74 is constructed of a parallel combination of a number r (where r is an integer) of smaller, component SIDO filters 76 from Sl to Sr which each include a shift register with fewer stages than the filter 16 of FIG. 2 for the same functional relationship, typically 4. The sum of the number of stages in FIG. 2 and FIG. 3 is the same for the same functional a ~76317 relationship. I~ is seen that all the first outputs from Vl to Vr of the parallel component SIDO filters Sl to Sr accumulate via a set of adders 78 to provide a signal Y2N_n N-n~ for the first output 36 while all the second outputs Ul to Vr of the parallel component SIDO filters Sl to Sr accumulate via a set of adders 80 to provide a signal Yn n for the second output 38. The SIDO filter 7~ has the advantage that the multiplier coeEficients ~ and ~ present in the functions of the Sl to Sr component SIDOs can be represented with fewer bits than those of the SIDO
filter 16 for comparable performance~ It is well Isnown in the art of digital filter design that a single-input, single-output filter consisting of a parallel combination of smaller filters, obtained by a well-known partial fraction expansion design procedure, can be constructed using fewer bits for the coefficients than a single-input, single-output filter of the direct form, i. e., having a simple but larger shift register. The SIDO 74 of FIG. 3 employs the same coefficientc as would two single-input,single-output filters each designed by the partial fraction expansion procedure to serve the same function, and therefore shares this property.
DIGITAL FILTER BANK FUNCTIONS
The operational relationships of the filter bank 10 will now be discussed. The number N is assumed to be 14, as would commonly be the case. The xk(m) are signals which are sampled at 8 kHz. They are voice signals for k-even and are t-l)~ times voice signals for k-odd.
The input signals to the filter bank, xk(m), can be represented in transform notation as Xk(zN)- The complete filter bank which is used in a TDM to FDM
translator is described by N-l X ( N
k-0 k k (1) - . . ~, .

~ li76317 în which, because of the use of guard bands, it is often true that XO(zN) and XN_l(zN) are zero. Hk(z) is the transfer function of the desired filtering operation for channel k of the filter bank. It is derived from the transfer function of a prototype filter having transfer function H(z). The prototype's impulse response is [h(r)] = [hot hl, ..., hrt~ hM, ....], (2) with z transform H(z) = ho + hlz + ... + hrz + hMz M + ....... . (3) The ratio M = 142kHzHZ = 28 (4) is the ratio of the sampling rate of the FDM signal to the channel bandwidth. It is convenient to express H(z) in the decimated form ' I ~6317 H(z) = (ho + hMZ + h2Mz 2M + .--) + z-l(h ~ h z-M + h z-2M +

+ z-2(h + h z-M + h z-2M +

+ + z-(M-l)(hM 1 + h2M_lZ + h3M-lZ

O ) z Pl(z ) + z 2p2(zM) + + -(M-l) M

~E n( n=0 The filters Pn(zM) resulting from the decimation are unique to the polyphase approach which is described in further detail in the above-cited publication of Bellanger and Daguet. The translation of H(z) to Hk(z) will be now discussed.
The translation of H(~), with ~ representing the digital frequency expressed in radians/sample and the center frequency of the kth channel by ~k~ which is given by the expression, ~ = (2k 1)~ (6) is summarized by Table 1.

1l ~7~317 TABLE 1. Lowpass to Bandpass Transformation H(A) -~ Hk(~) = 1/2 H(~ ~ ~k) ~ 1/2 H(~ ~ ~k) z ~ ze~ ~k M M (2k+1)~
z ~ -z since ~k = M

H(z) ~ ~k(Z) = -- + h ¦(ze j k) + ~ze ~k)--r = ... + hrZ cos~kr ~ ...

Hk ( z ) ~o ( c o s ~ kn ) z nPn ( z It is important to observe that the polyphase network filters, Pn(-zM), of the translated filter, Hk(z), are independent of k since z~ translates to _zM independent of k. The expression for Hk(z) given in Table 1 can be introduced into Equation (1), land, in view of the independence of the Pn(-zM) with respect to the index k, the order of summation can be reversed to yield, ~(z) = ~ z Pn( z )Wn(z ), (7) in which n k=0 k k (8) .

- ' , 1 ~7~317 or, reverting to the time domain, wn(m) = ~ coS~kn Xk(m) Equation (9) will be referred to as the discrete cosine transform (DCT~, although a constant often included in other contexts is not present here.
The change in order of summation in going from (l) to (7) changes the structure radically from that employing 12 separate high rate bandpass filters to that of a discrete cosine transform and a polyphase network of 28 low rate polyphase network filters, shown in Figure l. All filtering and discrete transform operations in Figure l occur at the input rate, typically of 8 kHz, so one discrete cosine transform and 28 filtering operations, each with the new input wn(m), need to be performed at this rate. The computation rate is reduced by about a factor of 8 by using the structure of Figure l as opposed to the use of separate filters, Hk(z), shown in Equation (l).
The efficient solution of the DCT equations of Eq. (9) was considered in the above-cited Narasimha and Peterson publication. The solution methods for the DCT are adequately described in the published literature and will not be described further here.
THE POLYPHAS~ NETWORK SECTION l~ STRUCTURE AND
ALGORITHM
A. A Single-Input-Double-output Filter Structure The realization of a TDM/FDM translator with 28 polyphase network filters, Pn(-zM), as in the prior art and as suggested by Equation (7), requires about one/third more filtering operations than necessary. The reason for these .:

., , '~
, ~ ~7~3~7 extra operations is that each filter resulting from the polyphase network breakdown procedure is of the form p ( zM) = n( (10) in which the denominator is the same for all paths, as illustrated in Figure 1. In view of this, it is seen that the recursive iltering operations defined by B(-zM) need only be done on Wn(zN) for 0 < n<l3, as shown in FIG~ 1.
In order to avoid the use of extra memory, the memory required for the recursive operations upon the Wn can be combined with that for the operations of An and A28_n shown in FIG. 1 for 1 < n< 13. The single-input-double-output (SIDO) filter in FIG. 2 results. The degrees of the numerator and denominator of the prototype filter have both been assumed to be L, and would be expected to be 8 or 9 for this application.
Double-input-double-output structures were disclosed in the above-cited publication by Bonnerot et al, and achieve similar reductions in memory requirements to those achieved by the SID0 16 of FIG. 2. In the Bonnerot et al publication it is assumed that the numerator of H(z) has the symmetry of a linear-phase FIR filter which permits a reduction in the required num~er of multiplications. It is for this additional reason that the structures there described have two inputs as well as two outputs. This assumption and a similar structure for FIR
filters was employed in the Narasimha and Peterson publication.
For the filter bank 10, the linear-phase assumption has not been made, since the resulting structure does not reduce the number of additions, as would be desirable when a digital signal processor with an arithmetic unit consisting of only a multiplier-accumulator is employed, and the linear-phase restriction increases the . ~. ' .
' : ~, ' ' ' ~ ~76317 order of the non-recursive filters. It is seen here, that economies in mernory and recursive operations can be obtained even though the linear-phase assumption is not made. It will also be seen that the SIDO filter 20 lends itself to an efficient realization with the minimal number of round off noise sources.
B. Memory Swapping Algorithm for SIDO Filter The SIDO filter 16 structure appears, at first glance, to require the use of three separate accumulators for accumulation of the recursive products and the two sets of non-recursive products. This would be the case if one starts at the last tap ~8 of the shift register 44 and proceeds to the first, forming and accumulating all products and shifting the register's contents, in order, in the conventional manner. However, the sets of alternate locations of the register 44 form two groups whose functions may be swapped each sampling period. Only one of the groups need be shifted, thereby eliminating half of the shifting, as shown in the following algorithm. It is assumed in this algorithm that ~o n is a power of 2, so that multiplication by this constant can be done by a scaling operation. This assumption is a convenience, but not a necessity, and a modified algorithm which has one more multiplication will accommodate values of ~o n which are not powers of 2.
ALGORITHM: MEMORY SWAPPING
I. For each sample m A. Compute Y0 for the conventional single-input-single-output filter.
B. For each value of n in the range 1 < n < 13, 1. accumulate T2i,n~2i for i L to 1 while shifting T2i n to register 2i + 2 (except that T2L n can be discarded rather than shifted),
2. accumulate ynWn with the former accumulator contents to obtain To n~ shift To n to register (2,n) and scale accumulator by ~o n~
3- accumulate ~2i~nT2i+2,n for i - (L-l) to 1 .
.
. . .

~ 1!76317 with the former accumulator contents, ~0 nTo n~
to obtain Yn n and, for the complementary SIDO N-n,
4. accumulate 2i-1,N-nT2i-1,N-n for i to 1 with the former accumulative contents, Yn,n, to obtain Yn.
C. Swap the functions of the groups of even and odd register memory locations for all filters, i. e., P s to T2i,n and T2i_l,n are interchanged so that the odd nu~bered register stages in this sampling period will serve the role of even numbered stages in the next sampling period.
It can be seen in the above algorithm that each Yn is co~puted with no need for clearing the accumulator in the computation of Yn during sampling period m, except at the outset. Since the accumulator width, i.e., the number of bit~, in most designs exceeds the width of the product, only two significant round-off noise sources then occur, that due to the storing of To n in memory which normally has shorter word length than that of the accumulator and that due to the storing of Yn in the output register. This is the least possible introduction of round-off noise.
There is also a savings in the number of computations required for computational overhead, since intermediate accumulator contents need never be stored and retrieved~
The multiplier-accumulator is seen to be utilized in every operation and therefore at maximum efficiency.
THE FIG. 3 SIDO 74 REALIZED BY T~E PARAI,LEL CONNECTION
OF r SMALLER, COMPONENT SIDO'S
It is sometimes desirable to realize a digital filter as an interconnection of smaller filters. A known method is the parallel connection which is obtained by expanding the desired transfer function into partial fractions. This approach has been disclosed for the particular case of digital filters employed in polyphase networks in the second-cited P. Vary publication cited above. The parallel realization can be used, in - , - - . .

- .: . , :
~, - ~ , :
-: : ` . : ,, ~ .-' ~ ~76317 particular, for the SI~O networks introduced here. A
factorization of B(-zM) into r factors, usually linear or quadratic in the variable z-M, such that B(--ZM) = Bl(--zM) B2(--Z ) ^-Br( Z ) (11) is assumed. Using the conventional approach, as described for example in the P. vary publication, the Equation (10) is written or Pn(-zM) and P~(-zM) with m = 2N-n, p ( zM) = An1( Z ) An2(~Z ) + Anr(-z ) and M Am1( Z ) + Am2( Z ) + + Amr( z ) (13) The same factorization as in Equation (11) of B(-zM) must be employed for both Pn(-zM) and Pm(-zM), as is shown in Equations (12) and (13) used for de~signing SIDO n.
The component SIDO's, denoted Sl, S2, ..., Sr, are 2a each obtained by employing like terms in (12) and (13), for example, output Ui of Si (where "i" is a variable integer between 1 and r) i5 determined by An i (~Z
i( ) Wn (14) .

~ ~76317 and output Vi of Si by i B (-ZM~ Wn (15) The factor z N in the above equation (15) corresponds to the delay of the delay register 34 in Fig. 3 and so must be considered in the definition of Vi in the equation (15).
Since the denominators are the same and the numerators are each of lower degree in the variable z-M than the denominator, a well-known property of partial fraction expansions, Si has the same structure as the SID0 in FIG. 2. The integer L in FIG. 2 will be replaced by the degree of Bi(-zM), which will usually be l or 2. The parallel connection of the r filters, Si to Srl yields the realization for SIDO filter 74 shown in FIG. 3.
The realization in FIG. 3 has the previously mentioned advantage of requiring fewer bits to represent ti~e multiplier coefficients. However, the structure in FIG. 3 will require the accumulator to be cleared r-l times while computing the Ui for each of the first r-l SIDO's, since these are independent functions described by the independent ter~s of Equation (12). The rth SIDO's output Ur may then be computed, accumulated with the previously com~uted r-l Ui to yield Yn n~ and then with all the Vi oE
SID0 N-n to yield Yn without further clearing of the accumulator. Therefore, the coefficient accuracy advantage of the realization in Figure 3 is offset by the disadvantage of additional round-off noise. However, the value of r can be varied to obtain the best balance between coefficient accuracy and round-off noise. It should be noticed that the least value of the number of factors of s(z-M) is 1, which corresponds to the SID0 of Figure 2, and that the maximum is normally the smallest integer greater than L/2 for conventional filtering functions. Therefore, the use of the structure in Figure 3 gives added flexibility in balancing the effects of coefficient quantization and round-off noise in designing filter banks.

Claims (5)

Claims:
1. A digital filter bank of the type including a discrete transform means connected to a polyphase network comprising at least one complementary pair of first and second digital filters, characterized in that each of said filters comprises:
a single input which provides a filter input signal during the course of a present sampling period;
at least a first output which during the present sampling period gives a first output signal value independent of the present sampling period filter input signal and dependent on the filter input signal of a previous sample period;
a second output which during the present sampling period gives an output signal value dependent in general on the value of the present filter input signal, and a shift register which for the present sampling period receives a register input signal dependent on the present filter input signal, said shift register including a first set of stages comprising a first stage and every other one of a plurality of successive stages and including a second set of stages comprising a second stage and every other one of said plurality of successive stages, said first filter output being accumulated from weighted values of said first set of stages and said second filter output being accumulated from the weighted value of said register input signal and weighted values of said second set of stages.
2. The filter bank according to claim 1, wherein said first stage receives said register input signal, said second stage receives a signal from said first stage delayed by one sampling period and said second set of stages are connected as active elements in feedback for said sampling period.
3. The filter bank according to claim 2, said second filter also having a single filter input and at least first and second outputs, the second filter first output being indepenndent of the second filter present input and dependent on the second filter input of a previous sampling period, said second filter second output being dependent in general upon the second filter present input, said filter bank further comprising means for accumulating in each sampling period the value of said first filter first output and the value of said second filter second output and for accumulating the value of said first filter second output and said second filter first output.
4. The filter bank according to claim 3 and comprising a plurality of complementary pairs of said first and second filters in said polyphase network.
5. The filter bank according to claim 4 wherein at least one of said first and second digital filters comprises:
a parallel network of component digital filters each having a single input and first and second outputs, and means for accumulating all the first outputs of said component digital filters to generate a first output of said digital filter and for accumulating all the second outputs of said component digital filters to generate a second output of said digital filter.
CA000397894A 1981-03-19 1982-03-09 Digital filter bank Expired CA1176317A (en)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
US06/245,194 US4393456A (en) 1981-03-19 1981-03-19 Digital filter bank
US245,194 1981-03-19

Publications (1)

Publication Number Publication Date
CA1176317A true CA1176317A (en) 1984-10-16

Family

ID=22925689

Family Applications (1)

Application Number Title Priority Date Filing Date
CA000397894A Expired CA1176317A (en) 1981-03-19 1982-03-09 Digital filter bank

Country Status (6)

Country Link
US (1) US4393456A (en)
JP (1) JPS57168517A (en)
CA (1) CA1176317A (en)
DE (1) DE3209450A1 (en)
GB (1) GB2095515B (en)
NL (1) NL8201131A (en)

Families Citing this family (56)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPS59185421A (en) * 1983-04-06 1984-10-22 Nec Corp Signal processor
US4691292A (en) * 1983-04-13 1987-09-01 Rca Corporation System for digital multiband filtering
DE3484701D1 (en) * 1983-10-05 1991-07-18 Nec Corp DIGITAL SIGNAL PROCESSING DEVICE WITH A DIGITAL FILTER.
US4794556A (en) * 1984-09-19 1988-12-27 Massachusetts Institute Of Technology Method and apparatus for sampling in-phase and quadrature components
JPS61154214A (en) * 1984-12-26 1986-07-12 Nec Corp Digital filter
US4745567A (en) * 1985-02-01 1988-05-17 American Telephone And Telegraph Company At&T Bell Laboratories Moment generator
US4709344A (en) * 1985-10-02 1987-11-24 Motorola, Inc. Programmable multifrequency digital tone receiver
US4825396A (en) * 1986-02-14 1989-04-25 Siemens Aktiengesellschaft Digital circuit for sampling rate variation and signal filtering and method for constructing the circuit
DE3610195A1 (en) * 1986-03-26 1987-10-01 Ant Nachrichtentech DIGITAL FILTER BANK
DE3732085A1 (en) * 1986-03-26 1989-04-06 Ant Nachrichtentech DIGITAL FILTER TREE
DE3627676A1 (en) * 1986-08-14 1988-02-25 Blaupunkt Werke Gmbh FILTER ARRANGEMENT
US4843581A (en) * 1987-05-11 1989-06-27 American Telephone And Telegraph Company Digital signal processor architecture
JPH0770964B2 (en) * 1988-06-10 1995-07-31 日本電気株式会社 Chirp filter
NO165980C (en) * 1988-11-07 1991-05-08 Frobe Radio As PROCEDURE FOR TRANSFORMING RESPECTIVE PROCESSING OF ELECTRONIC MULTIPLE SIGNALS, AND DEVICE FOR PROCESSING SUCH SIGNALS.
US5034907A (en) * 1989-09-12 1991-07-23 North American Philips Corporation Dynamically configurable signal processor and processor arrangement
US5283780A (en) * 1990-10-18 1994-02-01 Stanford Telecommunications, Inc. Digital audio broadcasting system
CA2066540C (en) * 1991-06-13 1998-01-20 Edwin A. Kelley Multiple user digital receiving apparatus and method with time division multiplexing
FR2685593B1 (en) * 1991-12-20 1994-02-11 France Telecom FREQUENCY DEMULTIPLEXING DEVICE WITH DIGITAL FILTERS.
US5528522A (en) * 1992-10-30 1996-06-18 United States Of America Portable radar or beacon emulator
WO1994019791A1 (en) * 1993-02-18 1994-09-01 Aware, Inc. Improved filter for use in audio compression and decompression systems
US5420891A (en) * 1993-03-18 1995-05-30 New Jersey Institute Of Technology Multiplierless 2-band perfect reconstruction quadrature mirror filter (PR-QMF) banks
US5535240A (en) * 1993-10-29 1996-07-09 Airnet Communications Corporation Transceiver apparatus employing wideband FFT channelizer and inverse FFT combiner for multichannel communication network
US5606575A (en) * 1993-10-29 1997-02-25 Airnet Communications Corporation FFT-based channelizer and combiner employing residue-adder-implemented phase advance
GB9414664D0 (en) * 1994-07-20 1994-09-07 British Aerospace Digital signal processing apparatus
JP3302229B2 (en) 1994-09-20 2002-07-15 株式会社リコー Encoding method, encoding / decoding method and decoding method
US5748786A (en) * 1994-09-21 1998-05-05 Ricoh Company, Ltd. Apparatus for compression using reversible embedded wavelets
US5881176A (en) * 1994-09-21 1999-03-09 Ricoh Corporation Compression and decompression with wavelet style and binary style including quantization by device-dependent parser
US6549666B1 (en) 1994-09-21 2003-04-15 Ricoh Company, Ltd Reversible embedded wavelet system implementation
US6195465B1 (en) 1994-09-21 2001-02-27 Ricoh Company, Ltd. Method and apparatus for compression using reversible wavelet transforms and an embedded codestream
US5966465A (en) * 1994-09-21 1999-10-12 Ricoh Corporation Compression/decompression using reversible embedded wavelets
US6229927B1 (en) 1994-09-21 2001-05-08 Ricoh Company, Ltd. Reversible embedded wavelet system implementation
US6873734B1 (en) * 1994-09-21 2005-03-29 Ricoh Company Ltd Method and apparatus for compression using reversible wavelet transforms and an embedded codestream
US5710763A (en) * 1995-07-31 1998-01-20 Motorola, Inc. Filtered fast Fourier transmultiplexer and method
US5999656A (en) * 1997-01-17 1999-12-07 Ricoh Co., Ltd. Overlapped reversible transforms for unified lossless/lossy compression
US6023718A (en) * 1997-05-09 2000-02-08 Matsushita Electric Industrial Co., Ltd. High speed interpolation filter and a method thereof
US5867479A (en) * 1997-06-27 1999-02-02 Lockheed Martin Corporation Digital multi-channel demultiplexer/multiplex (MCD/M architecture)
US6044172A (en) * 1997-12-22 2000-03-28 Ricoh Company Ltd. Method and apparatus for reversible color conversion
US6411647B1 (en) * 1998-10-30 2002-06-25 Broadcom Corporation Fully integrated ethernet transmitter architecture with interpolating filtering
US6584102B1 (en) 1998-12-21 2003-06-24 At&T Corp. Communication network apparatus and method
US6314452B1 (en) 1999-08-31 2001-11-06 Rtimage, Ltd. System and method for transmitting a digital image over a communication network
US7095781B1 (en) 1999-11-24 2006-08-22 Northrop Grumman Corporation Polyphase filter with stack shift capability
US20010047516A1 (en) * 2000-02-01 2001-11-29 Compaq Computer Corporation System for time shifting live streamed video-audio distributed via the internet
WO2001080561A1 (en) * 2000-04-18 2001-10-25 Rtimage Inc. System and method for the lossless progressive streaming of images over a communication network
WO2002048966A1 (en) * 2000-12-14 2002-06-20 Rtimage Inc. Three-dimensional image streaming system and method for medical images
US6898323B2 (en) * 2001-02-15 2005-05-24 Ricoh Company, Ltd. Memory usage scheme for performing wavelet processing
US6895120B2 (en) * 2001-03-30 2005-05-17 Ricoh Co., Ltd. 5,3 wavelet filter having three high pair and low pair filter elements with two pairs of cascaded delays
US7006697B1 (en) 2001-03-30 2006-02-28 Ricoh Co., Ltd. Parallel block MQ arithmetic image compression of wavelet transform coefficients
US6859563B2 (en) 2001-03-30 2005-02-22 Ricoh Co., Ltd. Method and apparatus for decoding information using late contexts
US6950558B2 (en) 2001-03-30 2005-09-27 Ricoh Co., Ltd. Method and apparatus for block sequential processing
US7062101B2 (en) 2001-03-30 2006-06-13 Ricoh Co., Ltd. Method and apparatus for storing bitplanes of coefficients in a reduced size memory
US7581027B2 (en) * 2001-06-27 2009-08-25 Ricoh Co., Ltd. JPEG 2000 for efficent imaging in a client/server environment
US7280252B1 (en) 2001-12-19 2007-10-09 Ricoh Co., Ltd. Error diffusion of multiresolutional representations
US7095907B1 (en) 2002-01-10 2006-08-22 Ricoh Co., Ltd. Content and display device dependent creation of smaller representation of images
US7120305B2 (en) * 2002-04-16 2006-10-10 Ricoh, Co., Ltd. Adaptive nonlinear image enlargement using wavelet transform coefficients
US7450659B2 (en) * 2004-03-29 2008-11-11 Agilent Technologies, Inc. Digital modulator employing a polyphase up-converter structure
JP6913099B2 (en) * 2016-07-21 2021-08-04 マイクロチップ テクノロジー ジャーマニー ゲーエムベーハー Digital filter with trusted input

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
FR2188920A5 (en) * 1972-06-15 1974-01-18 Trt Telecom Radio Electr
US3971922A (en) * 1974-11-29 1976-07-27 Telecommunications Radioelectriques Et Telephoniques T.R.T. Circuit arrangement for digitally processing a given number of channel signals
US4237551A (en) * 1978-12-22 1980-12-02 Granger Associates Transmultiplexer

Also Published As

Publication number Publication date
NL8201131A (en) 1982-10-18
US4393456A (en) 1983-07-12
GB2095515A (en) 1982-09-29
DE3209450A1 (en) 1982-10-07
GB2095515B (en) 1985-04-17
JPS57168517A (en) 1982-10-16

Similar Documents

Publication Publication Date Title
CA1176317A (en) Digital filter bank
US4131766A (en) Digital filter bank
Lim et al. Frequency-response masking approach for digital filter design: Complexity reduction via masking filter factorization
Valenzuela et al. Digital signal processing schemes for efficient interpolation and decimation
US3665171A (en) Nonrecursive digital filter apparatus employing delayedadd configuration
WO1979000271A1 (en) Fdm/tdm transmultiplexer
US4843581A (en) Digital signal processor architecture
Gustafsson et al. On the use of multiple constant multiplication in polyphase FIR filters and filter banks
Ambede et al. A modified coefficient decimation method to realize low complexity FIR filters with enhanced frequency response flexibility and passband resolution
Lian Complexity reduction for FRM-based FIR filters using the prefilter-equalizer technique
US4612625A (en) Decimator employing finite impulse response digital filters
US4192008A (en) Wave digital filter with multiplexed arithmetic hardware
Lehto et al. Synthesis of narrowband linear-phase FIR filters with a piecewise-polynomial impulse response
Gustafsson et al. Single filter frequency-response masking FIR filters
Pandel et al. Design of bireciprocal wave digital filters for high sampling rate applications
US10972318B2 (en) Data stream processing device with reconfigurable data stream processing resources and data stream processing method
Taxen Polyphase filter banks using wave digital filters
Constantinides et al. An efficient and modular transmultiplexer design
Ramstad Fractional rate decimator and interpolator design
KR100270165B1 (en) Paralled fir filter system with a digital interpolation function
Vaithiyanathan et al. Comparative Study of Single MAC FIR Filter Architectures with Different Multiplication Techniques
Rossiter et al. A modular transmultiplexer system using custom LSI devices
Johansson et al. Two-channel FIR filter banks based on the frequency-response masking approach
Haghparast et al. A computationally efficient coefficient update technique for Lagrange fractional delay filters
DeBrunner et al. Using entropy to build efficient FIR digital filters

Legal Events

Date Code Title Description
MKEC Expiry (correction)
MKEX Expiry