Quantization noise compensation by analysis filter bank optimisation

Filter banks proved to be effective in many signal processing contexts. A method to improve and to enlarge the set of filter banks applications consists in overcoming the effects of developing the perfect reconstruction condition without considering distortions inducted, inevitable, by quantization and transmission over additive noise disturbed channels. In this paper is presented an analysis filter design algorithm that, based on the knowledge of the input- AR signal model, optimally compensates the quantization noise effects. Simulation results, worked out on speech signal, prove significant performance improvements related to those obtained by conventional approach.