Elsevier

Signal Processing

Volume 85, Issue 7, July 2005, Pages 1335-1351
Signal Processing

Designing N-band biorthogonal FIR systems given several of the analysis and synthesis filters

https://doi.org/10.1016/j.sigpro.2004.12.013Get rights and content

Abstract

The problem of designing an N-band biorthogonal FIR analysis/synthesis system, given K of its analysis and synthesis filters, is addressed in this paper. A necessary and sufficient condition on the K filters, that guides their selection, is derived. The solution set is completely parameterized via ladder-type structures that guarantee the perfect reconstruction property under both coefficient quantization and roundoff errors. Both nonlinear- and linear-phase systems are considered. A design example is included to illustrate the theory.

Introduction

Designing an N-band analysis/synthesis perfect reconstruction (PR) FIR filter bank (FB) given K1 of the analysis filters constitutes a powerful means of constructing scalar- and vector-valued wavelet bases with desired properties in their analysis functions, and has also been shown to facilitate the design and realization of PRFBs [1], [2], [3], [4], [5], [6], [7], [8], [9], [10], [11], [12], [13], [14], [15], [16], [17], [18], [19]. A complete solution to the above so-called analysis (N,K)-problem has been reported in [20] both for the paraunitary and the general biorthogonal linear-phase (LP) and non-LP cases.1 Related works include [22], [23], [24], [25]. However, such a design approach provides no direct control on the quality of the synthesis FB [15]. Of course, this is of no concern in the case of orthogonal analysis/synthesis systems where the synthesis filters are simply time-reversed versions of the analysis ones. For nonorthogonal systems, however, this becomes an important issue, especially in applications like wavelet design, where it is well-known that the regularity of the analysis and synthesis wavelets has to be separately enforced [26]. Furthermore, it has been pointed out that the regularity of the synthesis wavelets may be equally or even more important than that of the analysis ones [5], [27], [28], [29]. Moreover, the interest in considering biorthogonal multi-band wavelets was demonstrated in [26], among others.

Hence the question arises of whether one can build an N-band biorthogonal analysis/synthesis system with 1K<N filters being selected a priori in both the analysis and synthesis filter banks. This problem was posed in [26] in the context of wavelet design, though no solution was provided. A complete solution to the above problem (to be referred to hereafter as the analysis/synthesis (A/S) (N,K)-problem) is presented in this paper. Based on existing results on the analysis (N,K)-problem, a procedure of determining a particular set of the filters that complete the given analysis and synthesis filters to a PR A/S system is developed. The set of all possible completions is also parameterized. A ladder-type realization scheme for the system is suggested by the solution process, which generalizes the well-known lifting scheme [30], to multiple channels. The design of the K filters in the two banks as well as the optimization of the frequency characteristics of the N-K complementary filters are also studied.

The rest of this paper is organized as follows. In Section 2 the analysis (N,K)-problem is recalled along with results on its solution. These are employed to arrive at the solution of the corresponding A/S problem in Section 3, for both the general (non-LP) and LP cases. The problem of designing the K given analysis and synthesis filters is also addressed. The optimization of the complementary filters is discussed in Section 4, where a design example is presented in detail. Section 5 concludes the paper.

Section snippets

The analysis (N,K)-problem

Let {Hi(z)}i=0N-1 and {Gi(z)}i=0N-1 denote the analysis and synthesis FBs, respectively, in an N-band system, as shown in Fig. 1. Only real FIR FBs will be considered. A(n) (essentially) sufficient and necessary condition for this system to be PR is that [31]GpT(z)Hp(z)=z-kIN,ka nonnegative integer,whereHp(z)=H0,0(z)H0,1(z)H0,N-1(z)H1,0(z)H1,1(z)H1,N-1(z)HN-1,0(z)HN-1,1(z)HN-1,N-1(z)and similarly Gp(z)=[Gi,j(z)] denote the matrices of the (type-I) analysis and (type-II) synthesis

The analysis/synthesis (N,K)-problem

Now let the first K filters of the synthesis bank be also given a priori, and let g0T(z) denote the corresponding K×N matrix of their (type-II) polyphase components, defined in the same way as h0T(z) in (2). The problem then is to complete h0T(z) and g0T(z) up to N×N polynomial matrices Hp(z) and Gp(z), respectively, such that the PR condition (1) is satisfied. Clearly, the given filters must satisfyh0T(z)g0(z)=z-kIKWe look then for the choice of the parameters E(z) and U(z) in (6) that will

Optimization of the free parameters

As we already noted, the structure of Fig. 3 realizes all the PR A/S systems with the given responses for their first K analysis and synthesis filters, provided that U(z) is realized so as to be unimodular (l-ext for the LP case). Hence, unconstrained optimization techniques (MATLAB fminu was employed in our design examples) can be used to optimize the responses of the complementary N-K analysis and synthesis filters. The performance criterion that we employ here is the minimization of the

Conclusions

The problem of designing N-band biorthogonal FIR systems given KN-1 filters in both the analysis and synthesis FBs was studied. A general solution procedure was developed and a complete parameterization of the solution was provided. The latter implies a ladder-type realization of the system, that structurally guarantees PR, being robust to both coefficient quantization and multiplication roundoff errors. This feature greatly facilitates the optimization of the N-K complementary analysis and

Acknowledgements

The author wishes to thank Profs. Sergios Theodoridis and Nicholas Kalouptsidis, Department of Informatics and Telecommunications, University of Athens, Greece, and Prof. Phillip A. Regalia, Département Communications, Image et Traitement de l’Information, Institut National des Télécommunications, Evry, France, for their support in the course of this research. Fruitful discussions with Dr. Felix C.A. Fernandes, Texas Instruments DSPS R&D Cr., Dallas, TX, USA, are also gratefully acknowledged.

References (91)

  • A. Cohen et al.

    Biorthogonal bases of compactly supported wavelets

    Comm. Pure Appl. Math.

    (1992)
  • M. Vetterli et al.

    Wavelets and filter bankstheory and design

    IEEE Trans. Signal Process.

    (September 1992)
  • P. Steffen et al.

    Theory of regular M-band wavelet bases

    IEEE Trans. Signal Process.

    (December 1993)
  • K. Kurosawa et al.

    A simple design method of perfect reconstruction QMF banks

    IEEE Trans. Circuits Syst.—IIAnalog Digital Signal Process.

    (March 1994)
  • H. Cagˇlar, O. Alkin, B. Sankur, E. Anarim, Design of M-channel wavelet transform bases with maximum regularity,...
  • P.N. Heller

    Rank M wavelets with N vanishing moments

    SIAM J. Matrix Anal. Appl.

    (April 1995)
  • S.-M. Phoong et al.

    A new class of two-channel biorthogonal filter banks and wavelet bases

    IEEE Trans. Signal Process.

    (March 1995)
  • P. Saghizadeh, A.N. Willson Jr., A generic approach to the design of M-channel uniform-band perfect-reconstruction...
  • X.-G. Xia et al.

    FIR paraunitary filter banks given several analysis filtersfactorizations and constructions

    IEEE Trans. Signal Process.

    (March 1996)
  • T. Nagai et al.

    Design of linear phase M-channel perfect reconstruction FIR filter banks

    IEEE Trans. Signal Process.

    (September 1997)
  • C.W. Kok, T.Q. Nguyen, LU factorization of perfect reconstruction filter bank, Proceedings of SPIE’97, San Diego, CA,...
  • A. Muthuvel et al.

    Design of two-channel linear-phase FIR PR filter banks with even length filters using convolution matrices

    IEEE Trans. Circuits Syst.—IIAnalog Digital Signal Process.

    (December 2000)
  • Y.-J. Chen et al.

    M-channel lifting factorization of perfect reconstruction filter banks and reversible M-band wavelet transforms

    IEEE Trans. Circuits Syst.—IIAnalog Digital Signal Process.

    (December 2003)
  • E. Kofidis et al.

    On the perfect reconstruction problem in N-band multirate maximally decimated FIR filter banks

    IEEE Trans. Signal Process.

    (October 1996)
  • S. Basu, H.M. Choi, Hermite-like reduction method for linear phase perfect reconstruction filter banks design,...
  • S. Basu et al.

    Hermite reduction methods for generation of a complete class of linear-phase perfect reconstruction filter banks—Part Itheory

    IEEE Trans. Circuits Syst.—IIAnalog Digital Signal Process.

    (April 1999)
  • Y.-J. Chen, K.S. Amaratunga, On completion of M-channel perfect reconstruction filter banks with prescribed admissible...
  • Y.-J. Chen, K.S. Amaratunga, How to complete paraunitary filter banks and simultaneously preserve linear phase?,...
  • P.N. Heller, T.Q. Nguyen, H. Singh, W.K. Carey, Linear-phase M-band wavelets with application to image coding,...
  • M. Antonini et al.

    Image coding using wavelet transform

    IEEE Trans. Image Process.

    (April 1992)
  • J. Tian, R.Q. Wells Jr., A remark on vanishing moments, Proceedings of 30th Asilomar Conference of Signals, Systems,...
  • S.O. Aase, T.A. Ramstad, Parallel FIR filter banks for robust subband image coding, Proceedings of ICASSP’93, vol. 5,...
  • I. Daubechies et al.

    Factoring wavelet transforms into lifting steps

    J. Fourier Anal. Appl.

    (1998)
  • P.P. Vaidyanathan

    Multirate Systems and Filter Banks

    (1993)
  • T. Kailath

    Linear Systems

    (1980)
  • M. Vidyasagar

    Control System SynthesisA Factorization Approach

    (1985)
  • A.G. Dewey

    Computation of the Smith canonical form of a polynomial matrix

    Electron. Lett.

    (March 1967)
  • T.Q. Nguyen et al.

    Structures for M-channel perfect reconstruction FIR QMF banks which yield linear-phase analysis filters

    IEEE Trans. Signal Process.

    (March 1990)
  • T.D. Tran et al.

    On M-channel linear phase FIR filter banks and application in image compression

    IEEE Trans. Signal Process.

    (September 1997)
  • W. Dahmen et al.

    Biorthogonal multiwavelets on the intervalcubic Hermite splines

    Constr. Approx.

    (2000)
  • P.P. Vaidyanathan

    A note on general parallel QMF banks

    IEEE Trans. Circuits Syst.

    (January 1987)
  • F.A.M.L. Bruekers et al.

    New networks for perfect inversion and perfect reconstruction

    IEEE J. Selected Areas Commun.

    (January 1992)
  • T.A.C.M. Kalker, I.A. Shah, Ladder structures for multidimensional linear phase perfect reconstruction filter banks and...
  • L. Tolhuizen et al.

    On the realizability of biorthogonal m-dimensional two-band filter banks

    IEEE Trans. Signal Process.

    (March 1995)
  • R.E. Van Dyck et al.

    Wavelet video coding with ladder structures and entropy-constrained quantization

    IEEE Trans. Circuits Syst. Video Technol.

    (October 1996)
  • Cited by (1)

    View full text