| Summary: | 碩士 === 國立臺灣大學 === 電信工程學研究所 === 89 === In this thesis, we introduce the integer-modulated filter banks based on the concept of losslessness in digital filter structures. Given the filter coefficient word length, many IMFB’s satisfying the perfect reconstruction conditions can be obtained. Owing to the ith dyadic symmetry feature of modulation vectors that makes vectors similar to cosine function, we proposed a systematic method to find all solutions satisfying the perfect reconstruction constraints, we also minimize the searching region as small as possible and do not lose any good solutions. The simulation results prove that the approach make the number of solution sets decrease much.
In chapter 4, we present a new method to add constraints on four cases of FIR filters that are constructed by some sub-filters. There are three major kinds of constraints: zero, flatness and Mth-band constraints. The implementation is to formula a linear constraint matrix directly derived from the property of constraints and filter coefficients. Moreover, synthesizing various constraints in the same design formulation can be done. We compute the optimal solution that minimizes the least-square error with orthogonality principle. Simulation results show that the magnitude with zero constraint decay speedily at the specific frequency and magnitude with flatness constraint is more flat than other filters without any constraints. Adopting this method, the FIR filters’ implementation in VLSI will be more flexible.
|
|---|
