Genzao Zhang
Published: 1992
Total Pages: 378
Get eBook
New design techniques of digital filter banks are presented in this thesis. The research has been focused on systematic design methods for computationally efficient filter banks with arbitrary center frequencies, which could be easily implemented and are useful in acoustic, speech signal processing, and communication systems. Frequency interpolation filter banks (FIFB) are first derived based on an adaptive filtering structure. FIFB banks show good frequency responses, interesting sensitivities and negligible parameter quantization effects and roundoff noise when they are implemented using modern DSP processors with fixed-point arithmetic. For maximally flat FIFB filter banks, the adjusted pole selection strategy by an optimization procedure is applied. Then, the resonator-based arbitrarily spaced center frequency filter banks (RFB) are proposed. They show better behaviour than general FIFB banks from the point of view of frequency responses and the implementation. In order to suppress the non-negligible sidelobes and further improve the passband performance of the non-uniform band filter banks, frequency domain windowing techniques are studied and linear programming is used to produce optimal windows for different banks. Thus the design methods of non-uniform band filter banks with windowing processing are presented and significant improvements are obtained on the filter bank performance. Finally, the FIFB filter bank techniques are successfully extended to the field of allpass and perfect reconstruction analysis/synthesis systems. This widens the application area of the FIFB filter banks. In an allpass system, a synthesis FIFB filter bank is derived based on a general analysis FIFB bank and the allpass requirement. In the perfect reconstruction system, multirate techniques are applied and the efficient general FIR synthesis filter bank and IIR FIFB filter bank are designed to construct the system.