Analysis of the performance of predictive SNR scalable coders

In this paper, we present an analysis of the performance of three predictive fine granular SNR-scalable coders and compare them with their non-scalable version. In our study, we assume an exponential model for the quantization noise and the use of linear prediction. Coders efficiency is assessed through the signal-to-noise ratio as a function of rate (SNR(R)) and the mean SNR. Validity of our analysis is tested by comparing theoretical results with simulations of the encoding of realizations of first order autoregressive processes. Results show that the use of coders which tolerate some prediction drift provides better results than other conventional scalable schemes.