Index assignment for predictive wideband LSF quantization

Abstract

In this paper we summarize some results derived earlier for the mean-square channel distortion of an autoregressive moving average (ARMA) vector quantizer with a maximum entropy encoder when the channel is assumed binary symmetric and memoryless. We discuss the required assumptions and their practical consequences in index assignment of ARMA vector quantizers. The discussion relates also to channel optimization of these quantizers. Furthermore, we compare noisy channel performance of memoryless, moving average, and autoregressive two-stage vector quantizers in line spectrum frequency quantization applied to wideband speech coding.