CN107241081B - Design method of sparse FIR prototype filter of cosine modulation filter bank - Google Patents
Design method of sparse FIR prototype filter of cosine modulation filter bank Download PDFInfo
- Publication number
- CN107241081B CN107241081B CN201710429939.3A CN201710429939A CN107241081B CN 107241081 B CN107241081 B CN 107241081B CN 201710429939 A CN201710429939 A CN 201710429939A CN 107241081 B CN107241081 B CN 107241081B
- Authority
- CN
- China
- Prior art keywords
- filter
- vector
- linear phase
- prototype filter
- filter bank
- 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
Links
Images
Classifications
-
- H—ELECTRICITY
- H03—ELECTRONIC CIRCUITRY
- H03H—IMPEDANCE NETWORKS, e.g. RESONANT CIRCUITS; RESONATORS
- H03H17/00—Networks using digital techniques
-
- H—ELECTRICITY
- H03—ELECTRONIC CIRCUITRY
- H03H—IMPEDANCE NETWORKS, e.g. RESONANT CIRCUITS; RESONATORS
- H03H17/00—Networks using digital techniques
- H03H2017/0072—Theoretical filter design
- H03H2017/0081—Theoretical filter design of FIR filters
Landscapes
- Mobile Radio Communication Systems (AREA)
- Filters That Use Time-Delay Elements (AREA)
Abstract
The invention discloses a design method of a sparse linear phase FIR prototype filter of a cosine modulation filter bank. The method comprises the following specific steps: 1, initializing design parameters of a sparse linear phase FIR prototype filter of a cosine modulation filter bank; and 2, iteratively calculating the sparse linear phase FIR prototype filter of the cosine modulation filter bank meeting the complete reconstruction condition, wherein the sparse linear phase FIR prototype filter comprises the determination of the number, the position and the value of the coefficient of the nonzero tap of the unit impulse response. The invention can design a prototype filter with low nonzero tap number, and the sparsity of the filter can reduce the number of the adder multipliers used for realizing the filter, thereby improving the operation speed, reducing the operation error and the energy consumption and further reducing the production cost.
Description
Technical Field
The invention belongs to the technical field of digital signal processing, and provides a design method of a linear phase FIR (finite impulse response) prototype filter of a sparse and efficient cosine modulation filter bank.
Background
The theory and design of multi-rate filter banks is of great interest because of their wide application in the fields of communications, speech and image coding/compression, system recognition, fast computing, etc. And the filter bank can be generally divided into a DFT filter bank and a cosine modulation filter bank. The cosine modulation filter bank is obtained by optimally designing a low-pass prototype filter and by fast Discrete Cosine Transform (DCT), and has wide application in the fields of signal processing, communication, biomedical engineering and the like because of the advantages of low computational complexity, simple design process and the like. A cosine modulated filter bank with a sparse linear phase FIR (finite impulse response) prototype filter is a filter bank in which the filter coefficients of each channel have a sparse characteristic (the number of non-zero tap coefficients is less than the filter order). The number of addition and multiplier used for realizing the sparse filter is far less than that of the similar filter with the equivalent filtering effect, so that the sparse filter has the advantages of high operation speed, small operation error, low energy consumption and the like.
The design method of the cosine modulation filter bank is mainly divided into the steps of respectively designing an analysis filter bank and a comprehensive filter bank in the filter bank and independently designing a low-pass prototype filter and then carrying out cosine modulation to obtain the filter bank, and in the proposed design method of the cosine filter bank, the design methods of P.P.Vaidyanathan and R.D.Koilpilai are classic, and the filter bank which meets the minimum standard of amplitude distortion and aliasing distortion is estimated through an analysis method.
Disclosure of Invention
The invention aims to design a linear phase FIR prototype filter of a cosine modulation filter bank for realizing small ripples, low tap number, low amplitude distortion and aliasing distortion, and provides a brand new design method, namely a method for designing the linear phase FIR prototype filter of the cosine modulation filter bank with sparseness and high efficiency.
The design method of the sparse linear phase FIR prototype filter of the cosine modulation filter bank provided by the invention comprises the following specific steps:
1, initializing design parameters of a sparse linear phase FIR prototype filter of a cosine modulation filter bank;
and 2, iteratively calculating the sparse linear phase FIR prototype filter of the cosine modulation filter bank meeting the complete reconstruction condition, wherein the sparse linear phase FIR prototype filter comprises the determination of the number, the position and the value of the coefficient of the nonzero tap of the unit impulse response.
(type II linear phase FIR filter is exemplified below):
constructing initial parameters according to design requirements:
according to the invention, the sampling numbers L respectively corresponding to the pass band, the transition band and the stop band are selected according to the channel number M of the cosine modulation filter bankp,Lt,LsSum ripple valuep,t,sDetermining an initial order N of the linear phase FIR prototype filter, wherein a tap coefficient of the linear phase FIR prototype filter is expressed by a vector h as:
h=2[h1,h2,…,hm…,hN/2]T(1)
wherein h ism(1 m is less than or equal to N/2) represents the m-th tap coefficient of the FIR prototype filter; the design problem of a sparse linear phase FIR prototype filter of a cosine modulation filter bank is converted into the following mathematical optimization problem:
s.t.|Bh-d|≤e (2b)
wherein h does not calculation0Represents 0-norm operation, i.e. represents the number of nonzero taps in the tap coefficient vector; the combined equations (2a) - (2c) of "min" and "s.t." represent solving for | | h | | computational complexity that satisfies the requirements of (2b) and (2c)0Minimum value of (d); the sampling matrix B is expressed as B ═ Bp;Bt;Bs]In which B isp、BtAnd BsThe sampling matrices, representing pass band, transition band and stop band, respectively, are represented as
WhereinTo representDimension of row vector; (L)p+Lt+Ls) The x 1-dimensional vector d is the discretized ideal frequency response and is expressed as:
where f (ω) is the frequency response function to which the transition band is to be approximated, expressed as:
wherein ω is0α pi/2M (0 ≦ α ≦ 1), equation (7) satisfies the complete reconstruction condition,frequency sampling points representing a transition band; the error vector e is (L)p+Lt+Ls) A x 1-dimensional column vector, expressed as:
e=[p…p t…t s…s]T, (8)
(II) setting (L)p+Lt+Ls) The initial value of the x 1-dimensional weight vector is w(1)=[1,1,…,1]TIn the invention, in the kth iteration (k is more than or equal to 1 and less than or equal to N/2), the column vector of the matrix B is normalized:
s.t.||h(k)||0≤k (10b)
and calculating (L)p+Lt+Ls) X 1-dimensional residual vector r(k)Expressed as:
r(k)=Φ(k)s(k)-d (11)
where k x 1 dimensions of s(k)As a result of the operation of equation (10),representing OMP algorithm from B(k)Set of column vectors, set Λ selected from(k)={n1,n2,…,nkDenotes an index set of non-zero tap coefficients.
(III) the invention utilizes the index set Lambda of the obtained nonzero tap coefficient(k)Solving the following linear programming problem:
s.t.|B(k)h(k)-d|≤e+μ·1L×1(11b)
judging whether mu is smaller than zero, if so, updating the weight vector w, wherein the updating formula is as follows:
wherein w(k+1)(l) Representing a new weight vector w(k+1)Value of (1), rl (k)Representing a residual vector r(k)A value of (1); new weight vector w(k+1)Carrying out cyclic calculation in the step 2; stopping iterative operation if mu is less than or equal to zero, and calculatingThe final sparse linear phase FIR prototype filter is obtained.
The invention has the following beneficial effects:
1. the invention provides a design method of a linear phase FIR prototype filter of a sparse and efficient cosine modulation filter bank for the first time.
2. The invention can design a prototype filter with low nonzero tap number, and the sparsity of the filter can reduce the number of the adder multipliers used for realizing the filter, thereby improving the operation speed, reducing the operation error and the energy consumption and further reducing the production cost.
3. Simulation results show that under the requirement of the same design index, the number of the nonzero tap coefficients of the filter is less than that of the optimal similar filter at home and abroad by more than 35%.
Drawings
FIG. 1 is a flow chart of a sparse linear phase FIR prototype filter design method implementing a cosine modulated filter bank of the present invention;
FIG. 2 according to function eam(omega) calculating to obtain an amplitude distortion figure of an approximately completely reconstructed cosine modulation filter bank;
FIG. 3 is a graph according to function ea(omega) calculating an aliasing distortion map of the approximately completely reconstructed cosine modulation filter bank;
fig. 4 is a frequency domain response plot of the sparse linear phase FIR prototype filter of the cosine modulated filter bank of table-2.
Detailed Description
Example 1:
the method for designing the sparse linear phase FIR prototype filter of the cosine modulation filter bank comprises the following steps:
1, initializing design parameters of a sparse linear phase FIR prototype filter of a cosine modulation filter bank;
and 2, iteratively calculating the sparse linear phase FIR prototype filter of the cosine modulation filter bank meeting the complete reconstruction condition, wherein the sparse linear phase FIR prototype filter comprises the determination of the number, the position and the value of the coefficient of the nonzero tap of the unit impulse response.
In order to verify the effectiveness of the filter bank design method, computer simulation was performed on the method.
The design requirement is as follows: the use of the literature: (F.Tan, et al.: optical design of cosine modulated filter banks using quality-corrected discrete optimization algorithm, "4 th International construction on Image and Signal Processing, vol.5, pp.2280-2284,2011.) (F.Tan, et al.: Quantum particle swarm optimization algorithm-based cosine modulated Filter Bank optimization design," fourth International Image and Signal Processing conference, prototype.5, pp.2280-2284,2011.), wherein the number of filter Bank channels M is 16, the number of Filter initial coefficients N is 256, and the sampling numbers L corresponding to the pass band, transition band and stop band are designedp=4,Lt=7,Ls94, ripple valuep=t=s=1·10-8Carry-over into the computation. The invention designs a prototype filter of a cosine filter bank by using an IROMP algorithm, and selects a column vector B corresponding to a nonzero coefficient position in an obtained variable set B according to the calculation of a weight valuejAnd carrying the filter coefficients obtained by iterative computation into an IROMP algorithm.
The method comprises the following steps: according to the design parameter requirement of a sparse linear phase FIR prototype filter of a cosine modulation filter bank, substituting each design parameter into an initialization condition to obtain a problem to be solved:
s.t.|Bh-d|≤e (2b)
the designed pass band, transition band and stop band are respectively corresponding to the sampling number Lp=4,Lt=7,LsSubstituting equation (6) with 94 yields the discretized ideal frequency response d in dimension (4+7+94) × 1, expressed as:
wherein the values of f (ω) are shown in Table-1:
TABLE-1
Will design the ripple valuep=t=s=1·10-8Substituting into the formula (7), a (4+7+94) × 1-dimensional error column vector e is obtained, each value in the vector being 1 · 10-8Setting the initial value of the (4+7+94) × 1-dimensional weight vector as w(1)=[1,1,…,1]T;
Step two: in the k (k is more than or equal to 1 and less than or equal to N/2) iteration, the column vector of the matrix B is normalized:
the following problem is then solved using the OMP algorithm in appendix 1:
s.t.||h(k)||0≤k (10b)
index set Lambda of non-zero tap coefficient is obtained through calculation(k);
Step three: index set Lambda of nonzero tap coefficient obtained by using step two for solving(k)The method is carried into the solution of the following linear programming problem:
s.t.|B(k)h(k)-d|≤e+μ·1L×1(11b)
further obtaining the tap coefficient of the sparse linear phase FIR prototype filter of the cosine modulation filter bankThe values are given in Table-2.
TABLE-2
Since the impulse response of a type II FIR filter is symmetrical, the filter tap coefficients found by the present inventionHalf of the tap coefficients of the prototype filter to be solved are symmetrical and the other half is equal, namely the total tap coefficients of the sparse FIR prototype filter to be solved are expressed as:
And selecting the variable matrix B by using the algorithm. For a 256-order filter, the algorithm selects all cases that result in half of the tap coefficients (the other half of the coefficients are symmetric with respect to the tap coefficients), i.e., the algorithm selects the tap coefficients for half of the tap coefficients (the other half of the coefficients are symmetric with respect to the tap coefficients), i.e., the filter selects the tap coefficientsWill be provided withThe total tap coefficient of the prototype filter is obtained by substituting the total tap coefficient expression of the filterThe finally obtained sparse linear phase FIR prototype filter of the cosine modulation filter bank is a filter with the nonzero coefficient of 166 orders, and compared with the particle swarm optimization algorithm, the sparse linear phase FIR prototype filter saves 35.6 percent.
Calculating and analyzing filter bank h by using the tap coefficient of the prototype filter obtained in the step threem(n) and a synthesis filter bank gm(n):
Wherein M is more than or equal to 1 and less than or equal to M, and calculating the amplitude distortion value e of the cosine modulation filter bankamAnd an aliasing distortion value eaThe calculation formula is expressed as:
eam(ω)=1-|A0(ejω)
wherein A is0(ejω) And Al(ejω) Expressed as:
Hk(ejω) For analyzing the frequency domain response of the filter bank, Gk(ejω) The frequency domain response of the synthesis filter bank.
In table-3, the order of the FIR notch filter, the number of non-zero tap weights, the amplitude distortion and aliasing distortion of the filter bank, which are obtained by the algorithm of the present invention and the particle swarm optimization algorithm, are respectively compared, and as shown in table-3, the amplitude distortion value e is obtainedamAnd an aliasing distortion value eaSimilarly, the prototype filter order of the invention is significantly less than the particle swarm optimization algorithm.
TABLE-3
In FIG. 2, according to function eam(omega) calculating to obtain an amplitude distortion figure of an approximately completely reconstructed cosine modulation filter bank, wherein the maximum value of the amplitude distortion figure is equal to the amplitude distortion value e obtained by the algorithm of the invention in the table-3amThe correspondence is equal; in FIG. 3, according to function ea(omega) calculating to obtain an aliasing distortion figure of the approximately completely reconstructed cosine modulation filter bank, wherein the maximum value of the aliasing distortion figure is equal to an aliasing distortion value e obtained by the algorithm of the invention in the table-3aThe correspondence is equal; fig. 4 is a plot of the frequency domain response of the sparse linear phase FIR prototype filter of the cosine modulated filter bank of table-2.
Formula (13) OMP Algorithm calculation Process
Using OMP algorithm calculation formula (10), M × N matrix B is the sensing matrix of OMP algorithm, N × 1 d is the observed value, riRepresenting the residual, t represents the number of iterations,represents the empty set, ΛtSet of indices, λ, representing t iterationstDenotes the index found in the t-th iteration, ajJ-column, B, representing matrix BETAtIndicating by indexΛtSelected set of columns, θ, of matrix BtIs a column vector of t × 1, the symbol @ represents a union operation,<rt-1,aj>the method is shown in the step of calculating the j-th column vector inner product of the residual error and the matrix BETA before the t-th iteration updating. The method comprises the following concrete steps:
1. initializing the residual to be equal to
r0=d; (1)
2. By the formula
Calculating to obtain an index lambdat;
3. To atAnd BtUnion operation, order
4. Finding new observed value d ═ BtθtLeast squares solution of (c):
6. if t is smaller than a preset value, returning to the step (2), otherwise, stopping iteration and entering the step 7;
Claims (3)
1. A method for designing a sparse linear phase FIR prototype filter of a cosine modulated filter bank, characterized in that the method comprises the following steps:
1, selecting sampling numbers L respectively corresponding to a pass band, a transition band and a stop band according to the number M of channels of a cosine modulation filter bankp,Lt,LsSum ripple valuep,t,sDetermining an initial order N of the linear phase FIR prototype filter, wherein a tap coefficient of the linear phase FIR prototype filter is expressed by a vector h as:
h=2[h1,h2,…,hm…,hN/2]T(1)
wherein h ism(1 m is less than or equal to N/2) represents the m-th tap coefficient of the FIR prototype filter; the design problem of a sparse linear phase FIR prototype filter of a cosine modulation filter bank is converted into the following mathematical optimization problem:
s.t.|Bh-d|≤e (2b)
wherein h does not calculation0Represents 0-norm operation, i.e. represents the number of nonzero taps in the tap coefficient vector; the combined equations (2a) - (2c) of "min" and "s.t." represent solving for | | h | | computational complexity that satisfies the requirements of (2b) and (2c)0Minimum value of (d); the sampling matrix B is expressed as B ═ Bp;Bt;Bs]In which B isp、BtAnd BsThe sampling matrices, representing pass band, transition band and stop band, respectively, are represented as
WhereinTo representDimension of row vector; wherein ω is0=απ/2M(0≤α≤1),(Lp+Lt+Ls) The x 1-dimensional vector d is the discretized ideal frequency response and is expressed as:
where f (ω) is the frequency response function to which the transition band is to be approximated,frequency sampling points representing a transition band; the error vector e is (L)p+Lt+Ls) A x 1-dimensional column vector, expressed as:
e=[p…p t…t s…s]T, (7)
2 nd, setting (L)p+Lt+Ls) The initial value of the x 1-dimensional weight vector is w(1)=[1,1,…,1]TIn the k (k is more than or equal to 1 and less than or equal to N/2) iteration, the column vector of the matrix B is normalized:
s.t.||h(k)||0≤k (9b)
and calculating (L)p+Lt+Ls) X 1-dimensional residual vector r(k)Expressed as:
r(k)=Φ(k)s(k)-d (10)
where k x 1 dimensions of s(k)As a result of the joint operation of equations (9a) and (9b),representing OMP algorithm from B(k)Set of column vectors, set Λ selected from(k)={n1,n2,…,nkDenotes an index set of non-zero tap coefficients;
3, index set Lambda of obtained non-zero tap coefficient(k)Solving the following linear programming problem:
s.t.|B(k)h(k)-d|≤e+μ·1L×1(11b)
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710429939.3A CN107241081B (en) | 2017-06-09 | 2017-06-09 | Design method of sparse FIR prototype filter of cosine modulation filter bank |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710429939.3A CN107241081B (en) | 2017-06-09 | 2017-06-09 | Design method of sparse FIR prototype filter of cosine modulation filter bank |
Publications (2)
Publication Number | Publication Date |
---|---|
CN107241081A CN107241081A (en) | 2017-10-10 |
CN107241081B true CN107241081B (en) | 2020-10-27 |
Family
ID=59987825
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710429939.3A Expired - Fee Related CN107241081B (en) | 2017-06-09 | 2017-06-09 | Design method of sparse FIR prototype filter of cosine modulation filter bank |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107241081B (en) |
Families Citing this family (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107947760B (en) * | 2017-12-18 | 2021-05-04 | 天津工业大学 | Design method of sparse FIR trap |
CN109510609B (en) * | 2018-12-11 | 2022-10-25 | 天津工业大学 | Design method of low-complexity sparse FIR low-pass filter |
CN109818890B (en) * | 2019-03-12 | 2020-04-10 | 北京科技大学 | Prototype filter determining method and device |
CN110365314B (en) * | 2019-06-04 | 2020-04-07 | 杭州电子科技大学 | Design method of separable two-dimensional FIR filter with sparse coefficient |
CN112528582B (en) * | 2021-02-07 | 2021-06-29 | 北京理工大学 | Prototype filter generation method and device, channelization method and satellite-borne communication device |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101972170A (en) * | 2010-10-22 | 2011-02-16 | 广东工业大学 | Self-adapting filter for least square support vector machine and filtering method thereof |
CN102882491A (en) * | 2012-10-23 | 2013-01-16 | 南开大学 | Design method of sparse frequency-deviation-free linear phase FIR (finite impulse response) notch filter |
EP3018906A1 (en) * | 2014-11-05 | 2016-05-11 | Dolby Laboratories Licensing Corp. | Systems and methods for rectifying image artifacts |
-
2017
- 2017-06-09 CN CN201710429939.3A patent/CN107241081B/en not_active Expired - Fee Related
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101972170A (en) * | 2010-10-22 | 2011-02-16 | 广东工业大学 | Self-adapting filter for least square support vector machine and filtering method thereof |
CN102882491A (en) * | 2012-10-23 | 2013-01-16 | 南开大学 | Design method of sparse frequency-deviation-free linear phase FIR (finite impulse response) notch filter |
EP3018906A1 (en) * | 2014-11-05 | 2016-05-11 | Dolby Laboratories Licensing Corp. | Systems and methods for rectifying image artifacts |
Non-Patent Citations (3)
Title |
---|
TENCON 2003. Conference on Convergent Technologies for Asia-Pacific Region;G. Abhilash,G.V. Anand;《TENCON 2003. Conference on Convergent Technologies for Asia-Pacific Region》;20040315;全文 * |
一种非均匀信道化滤波方法;杨君 等;《现代雷达》;20100930;第32卷(第9期);I135-329 * |
高效数字FIR滤波器及滤波器组的设计;黄绍广;《中国优秀硕士学位论文全文数据库信息科技辑》;20160215;第2016年卷(第2期);59-62、66 * |
Also Published As
Publication number | Publication date |
---|---|
CN107241081A (en) | 2017-10-10 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107241081B (en) | Design method of sparse FIR prototype filter of cosine modulation filter bank | |
Liu et al. | Filter design for autoregressive moving average graph filters | |
Vetterli | Wavelets, approximation, and compression | |
CN107256537B (en) | Design method for designing two-channel orthogonal diagram filter bank | |
Rafi et al. | An improved particle swarm optimization method for multirate filter bank design | |
Sakiyama et al. | Spectral graph wavelets and filter banks with low approximation error | |
CN109064412A (en) | A kind of denoising method of low-rank image | |
CN110365312B (en) | Multiphase structure of two-channel orthogonal mirror image filter bank and coefficient design method thereof | |
CN108092644B (en) | Design method of sparse two-dimensional FIR (finite impulse response) notch filter with accurately adjustable notch frequency | |
CN109510609B (en) | Design method of low-complexity sparse FIR low-pass filter | |
CN110365314B (en) | Design method of separable two-dimensional FIR filter with sparse coefficient | |
CN106651974B (en) | Utilize the compression of images sensing reconstructing system and method for weighting structures group Sparse rules | |
CN108305297A (en) | A kind of image processing method based on multidimensional tensor dictionary learning algorithm | |
Tseng et al. | Designs of fractional derivative constrained 1-D and 2-D FIR filters in the complex domain | |
CN109408765A (en) | Intelligent Matching based on quasi-Newton method tracks sparse reconstruction method | |
CN105787204B (en) | The design method of the complete over-sampling DFT modulated filter group of the double prototypes of bidimensional | |
Emirov et al. | Polynomial graph filters of multiple shifts and distributed implementation of inverse filtering | |
CN107133921B (en) | Image super-resolution reconstruction method and system based on multilevel neighborhood embedding | |
CN110335196A (en) | A kind of super-resolution image reconstruction method and system based on fractal decoding | |
CN107294512A (en) | A kind of non-homogeneous wave filter group filtering method based on tree | |
Sommariva et al. | Approximate Fekete points for weighted polynomial interpolation | |
CN110086452B (en) | Design method of low-complexity sparse FIR notch filter | |
Yazdanpanah et al. | Low-complexity feature stochastic gradient algorithm for block-lowpass systems | |
Bailey et al. | Nonuniform sampling and recovery of multidimensional bandlimited functions by Gaussian radial-basis functions | |
CN115861723A (en) | Monocular depth estimation method based on convolution sparse coding |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20201027 |
|
CF01 | Termination of patent right due to non-payment of annual fee |