A Combined Optimization Method of Finite Wordlength FIR Filters

FIR filters are very important in digital system design. The fix-point hardware implementation of FIR filters has the advantages of low cost and high performance. But this kind of hardware implementation introduces the unavoidable quantization error. In this paper, we present a combined method of GA and the local-search to optimize the coefficients of finite wordlength FIR filters. The search ability of GA and the efficiency of the local-search method are combined, and the genetic operators are selected carefully. The experiment result shows that the quantization error of the finite wordlength FIR filters is suppressed and within limited iterations, the method is evidently better than the simple GA method in terms of the average convergence performance