Abstract
An adaptive non-uniform quantizer is used at the transmitter for improving the SNR. Without using any extra channel for transmitting quantization step sizes, they are predicted at the receiver or corrected otherwise, thus increasing the channel capacity. The prediction algorithm has been used to optimize the system by maximizing SNR. Three different methods for step-size correction at the receiver have been discussed.
Using random samples with first-order Gauss-Markovian sequence (which symbolises speech and picture signals) as the input source, it is shown that the proposed schemes yield better SNR (only for large values of C) than most of other existing quantization schemes.