CN103956990B - The digital filter of parallel processing - Google Patents

The digital filter of parallel processing Download PDF

Info

Publication number
CN103956990B
CN103956990B CN201410099339.1A CN201410099339A CN103956990B CN 103956990 B CN103956990 B CN 103956990B CN 201410099339 A CN201410099339 A CN 201410099339A CN 103956990 B CN103956990 B CN 103956990B
Authority
CN
China
Prior art keywords
signal
module
adder
chronotron
modular construction
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 - Fee Related
Application number
CN201410099339.1A
Other languages
Chinese (zh)
Other versions
CN103956990A (en
Inventor
黄朝耿
周志光
林剑
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.)
Zhejiang University of Finance and Economics
Original Assignee
Zhejiang University of Finance and Economics
Filing date
Publication date
Application filed by Zhejiang University of Finance and Economics filed Critical Zhejiang University of Finance and Economics
Priority to CN201410099339.1A priority Critical patent/CN103956990B/en
Publication of CN103956990A publication Critical patent/CN103956990A/en
Application granted granted Critical
Publication of CN103956990B publication Critical patent/CN103956990B/en
Expired - Fee Related legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Abstract

Parallel processing digital wave filter, has N rank (N is even number), including: N/2 modular construction A: comprise 4 multipliers, two adders;N/2 modular construction B: comprise 4 multipliers, two adders;N+1 tap coefficient, is N+1 multiplier;Modular construction A as the first module only comprises 2 multipliers and 1 adder, and whole device only comprises 1 multiplier of 5N and 1 adder of 3N;The denominator of described modular construction A and modular construction B alternately cascade shaping filter transmission function, the parameter in described a series of tap coefficients and all modules is collectively forming the molecule of filter transfer function.

Description

The digital filter of parallel processing
Technical field
The invention belongs to digital processing field, be specifically related to the numeral filter of a kind of parallel processing Ripple device.
Technical background
Digital filter is one of most important ingredient in Digital Signal Processing, by extensively It is applied to voice, image procossing, digital communication, analysis of spectrum, pattern recognition, automatically controls Field.Reality is applied, does not require nothing more than digital filter configuration simple, limited wordlength is imitated Should possess the strongest robustness, because a digital filter designed needs by special Digital device realizes, and the digital device of these limited wordlengths is often substantially reduced wave filter Performance, it can be caused what is more to realize unsuccessfully.Therefore, study and design has been highly resistant to The most succinct digital filter of limit word length effect is always the study hotspot of scholars.
Since the forties, parallel processing technique is as application problem research and the product of industrial development Thing, has obtained development at full speed, and its application includes that the area of computer aided of VLSI circuit sets Meter, artificial intelligence, Aero-Space and defense strategy research etc..Parallel processing technique is mainly to calculate Method is core, and software and hardware is as a kind of structure connecting each other and mutually restricting of implementation tool Technology.Described parallel processing thought is that the efficiently realization of digital filter provides possibility.
For a N rank digital filter, normalized lattice type digital filter (A.H.Gray, Jr. and J.D Markel,“A normalized filter structure,”IEEE Trans.on Acoust., Speech, Signal Processing, vol.ASSP-23, pp.268-277, Jun., 1975.) containing only There is 5N+1 multiplier, and there is the strongest anti-finite word length effect ability.Described normalizing Although changing lattice digital filter is to have modular structure, but when processing in real time module it Between do not accomplish Parallel Implementation, this limit to a certain extent its application and development.
Summary of the invention
The present invention is the disadvantages mentioned above overcoming prior art, it is provided that a kind of have modular structure also Row processes digital filter.
Parallel processing digital wave filter of the present invention, it is characterised in that: there is N rank (N For even number), including:
N/2 modular construction A: comprise 4 multipliers, two adders;Often come one Pulse clock, the value in chronotron is as signal fk+1Enter rear stage module;Previous stage comes Signal fkThe signal b come with rear stagek+1Respectively with parameter-sin φkWith cos φkIt is multiplied defeated Entering to first adder A1, the value obtained is deposited in described chronotron, for subsequent time Calculating is prepared;Meanwhile, the signal f that described previous stage comeskThe letter come with described rear stage Number bk+1Respectively with parameter cos φkWith sin φkIt is multiplied and is input to second adder A2, formed Signal bkPrepare for previous stage module;
N/2 modular construction B: comprise 4 multipliers, two adders;Often come one Pulse clock, the signal f that previous stage comeskThe signal value characterized with chronotron respectively with parameter -sinφkWith cos φkIt is multiplied and is input to first adder B1, the signal f of formationk+1For rear one Level module is prepared;Meanwhile, the signal f that described previous stage comeskThe letter characterized with chronotron Number value respectively with parameter cos φkWith sin φkIt is multiplied and is input to second adder B2, formation Signal bkPrepare for previous stage module;And the signal b that rear stage comesk+1To in chronotron Value be updated, the calculating for subsequent time is prepared;
N+1 tap coefficient, is N+1 multiplier;
Modular construction A as first module only comprises 2 multipliers and 1 addition Device, whole device only comprises 5N-1 multiplier and 3N-1 adder;
The alternately cascade shaping filter transmission of described modular construction A and modular construction B The denominator of function, the parameter in described a series of tap coefficients and all modules is collectively forming filter The molecule of ripple device transmission function.The parallel principle of described wave filter shows:
During assuming that wave filter calculates, the time needed for a sub-addition and multiplication of doing is respectively TaAnd Tm.Alternately cascade in view of described modules A and described module B, it is to avoid signal edge Forward direction branch road (or backward branch road) from first module directly to last module, simultaneously, Described alternately cascade defines the parallel computation of wave filter.It is characterized in that following two walks Rapid: 1) the chronotron output of the chronotron output of first module and second module respectively with The most adjacent parameter is multiplied, i.e.-sin φ2、cosφ2With cos φ2、sinφ2, and respectively Enter two adders, be output as step 2 accordingly) to prepare, this calculating process takies Time be Ta+Tm.In this time, 2k+1 module and the 2nd (k+1) individual module Calculation as described first module and second module (k=1,2 ..., N/2-1), And complete to calculate accordingly.2) according to the result of calculation of step 1), to prolonging in N number of module Time device carry out numerical value renewal: for first module, input signal u (n) and-sin φ1It is multiplied Result adds backward output signal and the cos φ of second module1The result being multiplied, the knot obtained Fruit updates chronotron;For second module, the output of forward signal and cos φ3The knot being multiplied Fruit updates chronotron plus the result of the backward output signal of the 3rd module, the result obtained; The chronotron of 2k+1 module updates similar with first module;2nd (k+1) individual module Chronotron updates similar with second module.The renewal of described N number of chronotron is also to be parallel Process, all time-consuming T of the calculating between each two templatea+Tm
Beneficial effects of the present invention is mainly manifested in:
1. the two generic module structures of the present invention alternate cascade, effectively cut off letter Number from input a step calculate output defect (compared to normalized lattice type filter Device) so that complete the operand of equivalent in clock pulses, it is achieved that locate parallel every time The purpose of reason, greatly increases the handling capacity of wave filter.
2. two generic module structures described in entirely define dividing of transmission function by cascade Mother, and meet the condition of input Balancing relization, i.e. reach controlled Lindsey Graham square Battle array is the purpose of unit battle array, greatly reduces the probability that filter status overflows.
3. two generic module structures described in are prone to VLSI and realize and reduce cost.
Accompanying drawing explanation
Fig. 1 is the structural representation of modules A of the present invention.
Fig. 2 is the structural representation of module B of the present invention.
Fig. 3 is the structural representation of the present invention 4 rank wave filter.
Fig. 4 is the structural representation of N rank of the present invention wave filter.
Detailed description of the invention
Referring to the drawings, the present invention is further illustrated:
As it is shown in figure 1, the modules A of the present invention comprises four multipliers, two adders and one Individual chronotron.Often carrying out a pulse clock, the value in chronotron is as signal fk+1After entrance one Level module;The signal f that previous stage comeskThe signal b come with rear stagek+1Respectively with parameter -sinφkWith cos φkBe multiplied and be input to first adder A1, the value obtained be deposited into described in prolong Time device in, the calculating for subsequent time is prepared;Meanwhile, the signal f that described previous stage comesk The signal b come with described rear stagek+1Respectively with parameter cos φkWith sin φkIt is multiplied and is input to Second adder A2, the signal b of formationkPrepare for previous stage module.
As in figure 2 it is shown, module B of the present invention comprises four multipliers, two adders and one Individual chronotron.Often carry out a pulse clock, the signal f that previous stage comeskCharacterized with chronotron Signal value respectively with parameter-sin φkWith cos φkIt is multiplied and is input to first adder B1, shape The signal f becomek+1Prepare for rear stage module;Meanwhile, the signal f that described previous stage comesk The signal value characterized with chronotron respectively with parameter cos φkWith sin φkIt is multiplied and is input to second Adder B2, the signal b of formationkPrepare for previous stage module;And the letter that rear stage comes Number bk+1Being updated the value in chronotron, the calculating for subsequent time is prepared.
Based on two generic module structures shown in Fig. 1 and Fig. 2, Fig. 3 gives the present invention 4 rank The structural representation of wave filter.The parallel process of of the present invention wave filter is described in detail below. During assuming that wave filter calculates, the time needed for doing a sub-addition and multiplication is respectively TaWith Tm.Alternately cascade in view of described modules A and described module B, it is to avoid signal is along forward direction Branch road (or backward branch road) from first module directly to last module, simultaneously, described Alternately cascade define the parallel computation of wave filter.It is characterized in that following two step: 1) First module chronotron output with the chronotron of second module export respectively with respective phase Adjacent parameter is multiplied, i.e.-sin φ2、cosφ2With cos φ2、sinφ2, and respectively enter two Individual adder, is output as step 2 accordingly) prepare, the time that this calculating process takies For Ta+Tm.Simultaneously, the calculating of the 3rd module and the 4th module with described first Module is the same with second module.2) according to the result of calculation of step 1), to four modules In chronotron carry out numerical value renewal: for first module, input signal u (n) and-sin φ1 The result being multiplied adds backward output signal and the cos φ of second module1The result being multiplied, The result arrived updates chronotron;For second module, the output of forward signal and cos φ3 The result being multiplied updates plus the result of the backward output signal of the 3rd module, the result obtained Chronotron;The chronotron of the 3rd module updates similar with first module;4th module is prolonged Time device value directly updated by the output valve of its forward signal.The renewal of described four chronotron is Parallel processing, need time-consuming Ta+Tm
Fig. 4 is that to characterize is the structural representation of N rank wave filter, the principle of its parallel processing Ibid.
The computational methods that the present invention relates to digital filter configuration parameter are given below.
As shown in Figure 4, the structural parameters of described digital filter include { sin φk,cosφkAnd {ξm}.Wherein, described { sin φk,cosφkCalculating can be found in document " A canonical representation of input-balanced realizations for discrete-time systems,” (S.P.Zhu,Y.Wang,G.Li and C.R.Wan,in Proc.of the2010IEEE International on Green Circuits and Systems-ICGCS2010,Jun.2010.)。 For convenience, the digital filter that document above is corresponding is defined as R0={A0,B0,C0, d}, Digital filter corresponding to the present invention is defined as R={A, B, C, d}.Theoretical according to system structure Understand, { ξm, it is the element that vector [C d] is corresponding, can be by similarity transformation by R0Obtain. According to Fig. 4, as known { sin φk,cosφkOn the premise of }, arranged side by side by analyzing state variable Go out corresponding relation formula and i.e. can get A, B, then similitude transformation matrix T can be given by:
T=W0W-1
Wherein, W0A is had respectively with W0,B0With A, B determine, meet
W 0 = B 0 A 0 B 0 · · · A 0 N - 1 B 0 W = B AB · · · A N - 1 B
Therefore, it can be easy for obtaining C=C0T。
Content described in this specification embodiment is only the row of the way of realization to inventive concept Lifting, protection scope of the present invention is not construed as being only limitted to the concrete shape that embodiment is stated Formula, protection scope of the present invention also and is caned according to present inventive concept in those skilled in the art The equivalent technologies means expected.

Claims (1)

1. parallel processing digital wave filter, it is characterised in that: there are N rank (N is even number), Including:
N/2 modular construction A: comprise 4 multipliers, two adders;Often come one Pulse clock, the value in chronotron is as signal fk+1Enter rear stage module;Previous stage comes Signal fkThe signal b come with rear stagek+1Respectively with parameter-sin φkWith cos φkIt is multiplied defeated Entering to first adder A1, the value obtained is deposited in described chronotron, for the meter of subsequent time It is considered preparation;Meanwhile, the signal f that described previous stage comeskThe signal come with described rear stage bk+1Respectively with parameter cos φkWith sin φkIt is multiplied and is input to second adder A2, the letter of formation Number bkPrepare for previous stage module;
N/2 modular construction B: comprise 4 multipliers, two adders;Often come one Pulse clock, the signal f that previous stage comeskThe signal value characterized with chronotron respectively with parameter -sinφkWith cos φkIt is multiplied and is input to first adder B1, the signal f of formationk+1For rear stage Module is prepared;Meanwhile, the signal f that described previous stage comeskThe signal characterized with chronotron Value respectively with parameter cos φkWith sin φkIt is multiplied and is input to second adder B2, the signal of formation bkPrepare for previous stage module;And the signal b that rear stage comesk+1Value in chronotron is entered Row updates, and the calculating for subsequent time is prepared;
N+1 tap coefficient, is N+1 multiplier;
Modular construction A as first module only comprises 2 multipliers and 1 adder, Whole device only comprises 5N-1 multiplier and 3N-1 adder;
Described modular construction A and modular construction B alternately cascade shaping filter transmission letter The denominator of number, the parameter in described a series of tap coefficients and all modules is collectively forming filtering The molecule of device transmission function;
Corresponding coefficient ζ is multiplied by the output of every one-levelk(k=1,2 ..., N) it is added afterwards, add ζ0With The product of u (n) obtains exporting y (n).
CN201410099339.1A 2014-03-18 The digital filter of parallel processing Expired - Fee Related CN103956990B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201410099339.1A CN103956990B (en) 2014-03-18 The digital filter of parallel processing

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201410099339.1A CN103956990B (en) 2014-03-18 The digital filter of parallel processing

Publications (2)

Publication Number Publication Date
CN103956990A CN103956990A (en) 2014-07-30
CN103956990B true CN103956990B (en) 2016-11-30

Family

ID=

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2003179466A (en) * 2001-12-10 2003-06-27 Sanyo Electric Co Ltd Digital signal processing apparatus and sound reproducing apparatus
CN1830142A (en) * 2003-07-29 2006-09-06 松下电器产业株式会社 Surface acoustic wave device
CN101763629A (en) * 2010-01-19 2010-06-30 武汉科技大学 Construction method of integrative filter bank with cellular structure

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2003179466A (en) * 2001-12-10 2003-06-27 Sanyo Electric Co Ltd Digital signal processing apparatus and sound reproducing apparatus
CN1830142A (en) * 2003-07-29 2006-09-06 松下电器产业株式会社 Surface acoustic wave device
CN101763629A (en) * 2010-01-19 2010-06-30 武汉科技大学 Construction method of integrative filter bank with cellular structure

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
高鲁棒性低复杂度数字滤波器结构设计的研究;黄朝耿;《中国博士学位论文全文数据库 信息科技辑》;20140315(第03期);全文 *

Similar Documents

Publication Publication Date Title
US11645529B2 (en) Sparsifying neural network models
Brown The massless higher-loop two-point function
JP3636361B2 (en) Digital filter with high accuracy and high efficiency
CN103888104B (en) Method and system for designing FIR digital filter
Magnus Riccati acceleration of Jacobi continued fractions and Laguerre-Hahn orthogonal polynomials
CN110807522B (en) General calculation circuit of neural network accelerator
CN102135951B (en) FPGA (Field Programmable Gate Array) implementation method based on LS-SVM (Least Squares-Support Vector Machine) algorithm restructured at runtime
CN107241081B (en) Design method of sparse FIR prototype filter of cosine modulation filter bank
Abdesselam The ground state energy of the massless spin-boson model
CN105243241B (en) The two passage biorthogonal figure filter set designing methods based on lift structure
CN103956990B (en) The digital filter of parallel processing
CN106982045A (en) A kind of design method based on the SOCP EIR CMFB structures optimized
CN104504205A (en) Parallelizing two-dimensional division method of symmetrical FIR (Finite Impulse Response) algorithm and hardware structure of parallelizing two-dimensional division method
CN107565932A (en) A kind of FIR ptototype filter design methods based on linear neural network
CN109687843A (en) A kind of algorithm for design of the sparse two-dimentional FIR notch filter based on linear neural network
Chernyshev Time-dependent Schrödinger equation: statistics of the distribution of Gaussian packets on a metric graph
CN104198584A (en) Method for obtaining sound transfer rate and sound transmission loss of circular opening
Aggarwal A comparative study of X-tree, pyramid and related machines
CN106160702A (en) The method for designing of approximation Perfect Reconstruction monoarch DFT modulated filter bank
CN102064796B (en) Simplified weighted repeat pseudo-median filtering method with negative coefficients
CN103973254B (en) A kind of across resistance type integrated band pass filter method for designing
CN105589833A (en) Storage method of LSQR algorithm-based frequency domain waveform inversion
Dimakis et al. Bidifferential graded algebras and integrable systems
CN103956990A (en) Parallel processing digital filter
Bilski et al. Parallel realisation of the recurrent Elman neural network learning

Legal Events

Date Code Title Description
PB01 Publication
SE01 Entry into force of request for substantive examination
GR01 Patent grant
CF01 Termination of patent right due to non-payment of annual fee

Granted publication date: 20161130

Termination date: 20170318