Word-length optimization of LMS adaptive FIR filters

Abstract

The paper deals with the word-length optimization of LMS adaptive FIR filters using fixed-point arithmetic. As the optimization criterion, a weighted sum of the mean-square error /spl xi/, area and delay of the adaptive filter considering target implementation technology is used. Because the optimization is performed by a computer and there is no need to have a simple expression for manual reviewing, a more accurate expression for the mean-square error has been derived considering an improved quantization model. Word-length optimization was performed by a genetic algorithm because classical optimization methods were not suitable for such optimization. Simulation results are given for an adaptive echo canceller.