In this paper, we propose a novel feed-forward adaptive coding scheme called SACS (sample adaptive coding scheme) for the lossy compression of a discrete-time memoryless stationary source. SACS is based on a concept of adaptive quantization to the varying samples of the source and is very different from traditional adaptation techniques for non-stationary sources. SACS quantizes each solirce sample using a sequence of quantizers. Even when using scalar quantization in SACS, we can ach.ieve performance comparablse to vector quantization (with the complexity still of the order of scalar quantization). We also show tihat important lattice based vector quantizers can be constructed using scalar quantization in SACS. We mathematically analyze SACS and propose a simple algorithm to implemeint it. We numerically study SACS for independent and identically distributed Gaussian sources. Through our numerical study, we :find that SACS using scalar quantizers achieves typical gains of 1-2 dB signal to noise ratio over the non-adaptive scheme based on the Lloyd-Max quantizer. We also show that SACS can be used in conjuncfion with vector quantizers to further improve the gains.
Date of this Version