Effects of mismatch and low rate for coding of autoregressive data by tree-searched DPCM

An experimental evaluation is undertaken of the performance of differential PCM (DPCM) codes over a range of all-pole, autoregressive sources whose correlations are typical of speech and images. Such codes employ code searching in order to improve SNR. Test results, for the cases of rates 1, 2, and 7, with Gaussian and Laplacian variates as excitation, indicate that the gain from a fixed moderate degree of code searching increases as the rate drops and grows consistently with the reciprocal of the Berger-Gray critical distortion parameter (which is inversely related to correlation). It is established that code searching will be most effective for low rate coding of highly correlated sources.