Design Methods Of Subband Filter Banks And Its Applications

Filter banks have many important applications in communications,transmultiplexing,subband coding of speech and images,radar,adaptive filter,denoising,system identification and many other fields. Over the past decade,there have been considerable advances in subband (or multirate) signal processing,with much of the attention being given to subband decompositions and wavelet transforms. The research on theory and design of filter banks has become popular. One reason for their popularity is that they can easily extract the spectral components of a signal while providing very efficient implementation. Second reason is that they offer many advantages,such as reduced computational complexity for a given task,transmission rate and storage requirement. This dissertation is mainly about the design methods for subband filter banks and its applications in denoising for signal detection and interference suppression for Direct Sequence Spread system (DSSS).The dissertation mainly includes the following three aspects: 1. The designs of structurally perfect reconstruction (PR) FIR filter bands,including two-channel and M-channel filter banks. 2. The design of nonuniform PR filter banks. 3. The applications of filter banks,including denoising and interference suppression.The first aspect is contributed to the design of structurally PR filter bands. In this part,an introduction and analysis to previous work on theory and design of filter banks are firstly given. The drawbacks existing in previous approaches are that the filter banks are in general pseudo PR. In many applications of filter banks,a common requirement is that reconstructed output signal should be as close as possible to input signal. Therefore we focused on the problem about designing a structurally PR filter band and its implementation. Secondly,the Euclidean algorithm about polynomials and the PR condition of two-channel filter banks are analyzed. The PR condition can be factorized into a cascade of matrixes with determinant I by Euclidean algorithm. A proposition about the factorization was presented and proven. Based on the cascade of matrixes,an algorithm for design two-channel filter bands with linear phase and low delay was proposed. The algorithm is also suitable for designing the two-channel filter banks with unequal length of two filters. Thirdly,the proposed method is extended to the design of M-band cosine-modulated filter banks (CMFB) with length N = 2mM and arbitrary-length. A problem for designing transmultiplexer with arbitrary-length CMFB is also addressed. Many design examples with good frequency are given,but there is a drawback,which the coefficients of matrixes in large dynamical range in these examples.UIvSo,a heuristic-searching algorithm based on tree-structure is proposed to design filter banks with coefficients in low dynamical range fourthly. A class of filter banks with sum of power of two (SOPOT) coefficients is also designed by evolutionary strategies (ES) algorithm with given seed.Nonuniform filter banks with perfect-reconstruction (PR) property are desirable in applications such as audio coding. The second aspect is to study the design of nonuniform filter banks. The design of nonuniform cosine-modulated filter banks based on a two-stage structure can be formulated as a problem with two objective functions to be optimized. A new hybrid optimization method for designing perfect-reconstruction nonuniform filter banks is presented,which can be used to solve a class of problems with two objective functions. This method is of good convergence and mezzo calculation cost. Satisfactory results free of aliasing in analysis and synthesis filters can be obtained by the proposed method.The last aspect is devoted to applications of filter banks. In this part,two applications are addressed. The first application is interference suppression in DSSS. Narrow-band interference often exists in DSSS. In this application,a threshold method of subband transform is proposed for interference suppression. Com...