The Design Of Sub-band Filter Banks Based On Cosine-Modulation

Sub-band filter banks are widely used in various fields, such as voice signal processing, image processing and mobile communications ever since sub-band signal processing was proposed in the mid-1970's. Sub-band signal processing applying technology can effectively reduce the complexity of signal processing, lower data transmission rate and reduce data storage memory. The design aims of the filter banks are to make the reconstructed output signal and the original input signal the same. From a practical application point of view, the distortion control in a certain range of conditions, a design method that has the characteristics of few limits, high efficiency, and simple to achieve is more valuable.This paper focus on the work of the filter banks design theory. The design theory of M-channel cosine-modulated filter banks is fully studied. An indirect method based on the linear iterative optimization algorithm for designing cosine-modulated filter banks is presented. In accordance with this method, there is only a one-dimensional linear optimization objective function to be disposed. A prototype filter of good characteristic and near perfect-reconstruction cosine-modulated filter banks with lower reconstruction error can be gained. Also, the configuration of parameters has design freedom, it can be acquired according to the need of practical application. The computer simulation shows that the filter banks designed by this method have the characteristics of higher stopband attenuation and small system distortion.In fields of sub-band coding, adaptive sub-band filtering and speech signal processing and others practical applications, the analysis of the signals distributing in different frequency bands often request different resolving power, so non-uniform filter banks are more popular.In the traditional merging algorithm, it is necessary to design two uniform filter banks of different channels respectively, the whole design process is complex, and the whole merging process is limited to the parameters of each uniform filter banks. According to the improved iterative linear algorithm, the cosine-modulated filter banks can be gained and then the non-uniform filter banks will be achieved by directly merging the adjacent channel coefficients of the uniform filter banks. Computer simulation shows that there are no ups and downs in the frequency bands of the non-uniform filter banks after the merging resistance.