Designing N-band biorthogonal FIR systems given several of the analysis and synthesis filters
Introduction
Designing an N-band analysis/synthesis perfect reconstruction (PR) FIR filter bank (FB) given 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 -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 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) ()-problem) is presented in this paper. Based on existing results on the analysis -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 complementary filters are also studied.
The rest of this paper is organized as follows. In Section 2 the analysis -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 -problem
Let and 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]whereand similarly denote the matrices of the (type-I) analysis and (type-II) synthesis
The analysis/synthesis -problem
Now let the first K filters of the synthesis bank be also given a priori, and let denote the corresponding matrix of their (type-II) polyphase components, defined in the same way as in (2). The problem then is to complete and up to polynomial matrices and , respectively, such that the PR condition (1) is satisfied. Clearly, the given filters must satisfyWe look then for the choice of the parameters and 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 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 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 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 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)
Biorthogonal wavelets
- et al.
Efficient state-space approach for FIR filter bank completion
Signal Processing
(2003) - et al.
Mirror-image symmetric perfect reconstruction FIR filter banksparameterization and design
Signal Processing
(1998) - et al.
Local decomposition of refinable spaces and wavelets
Appl. Comput. Harmon. Anal.
(1996) - et al.
Biorthogonal spline-wavelets on the interval—stability and moment conditions
Appl. Comput. Harmon. Anal.
(1999) - et al.
Construction of semi-recursive PR-filter banks via generalised lifting
Signal Processing
(2004) - et al.
Improved technique for design of perfect reconstruction FIR QMF banks with lossless polyphase matrices
IEEE Trans. Acoust. Speech Signal Process.
(July 1989) - C. Herley, M. Vetterli, Linear phase wavelets: theory and design, Proceedings of ICASSP’91, Toronto, Ontario, Canada,...
- C.W. Kim, R. Ansari, A.E. Cetin, A class of linear-phase regular biorthogonal wavelets, Proceedings of ICASSP’92, San...
- H. Kiya, M. Yae, M. Iwahashi, A linear phase two-channel filter bank allowing perfect reconstruction, Proceedings of...
Biorthogonal bases of compactly supported wavelets
Comm. Pure Appl. Math.
Wavelets and filter bankstheory and design
IEEE Trans. Signal Process.
Theory of regular M-band wavelet bases
IEEE Trans. Signal Process.
A simple design method of perfect reconstruction QMF banks
IEEE Trans. Circuits Syst.—IIAnalog Digital Signal Process.
Rank M wavelets with N vanishing moments
SIAM J. Matrix Anal. Appl.
A new class of two-channel biorthogonal filter banks and wavelet bases
IEEE Trans. Signal Process.
FIR paraunitary filter banks given several analysis filtersfactorizations and constructions
IEEE Trans. Signal Process.
Design of linear phase M-channel perfect reconstruction FIR filter banks
IEEE Trans. Signal Process.
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.
M-channel lifting factorization of perfect reconstruction filter banks and reversible M-band wavelet transforms
IEEE Trans. Circuits Syst.—IIAnalog Digital Signal Process.
On the perfect reconstruction problem in N-band multirate maximally decimated FIR filter banks
IEEE Trans. Signal Process.
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.
Image coding using wavelet transform
IEEE Trans. Image Process.
Factoring wavelet transforms into lifting steps
J. Fourier Anal. Appl.
Multirate Systems and Filter Banks
Linear Systems
Control System SynthesisA Factorization Approach
Computation of the Smith canonical form of a polynomial matrix
Electron. Lett.
Structures for M-channel perfect reconstruction FIR QMF banks which yield linear-phase analysis filters
IEEE Trans. Signal Process.
On M-channel linear phase FIR filter banks and application in image compression
IEEE Trans. Signal Process.
Biorthogonal multiwavelets on the intervalcubic Hermite splines
Constr. Approx.
A note on general parallel QMF banks
IEEE Trans. Circuits Syst.
New networks for perfect inversion and perfect reconstruction
IEEE J. Selected Areas Commun.
On the realizability of biorthogonal m-dimensional two-band filter banks
IEEE Trans. Signal Process.
Wavelet video coding with ladder structures and entropy-constrained quantization
IEEE Trans. Circuits Syst. Video Technol.
Cited by (1)
Design of three-channel filter banks for lossless image compression
2009, Proceedings - International Conference on Image Processing, ICIP