ANALYSIS AND FURTHER RESULTS ON ADAPTIVE ENTROPY-CODED QUANTIZATION

Low-distortion coding of memoryless sources continues to be an important research problem. All coding schemes for correlated sources attempt in one way or another to transform the source data to a memoryless or uncorrected sequence. This sequence, be it LPC or DCT coefficients or even vector-quantizer indices, can be treated as a new memoryless source and should be efficiently coded to make best use of the available channel capacity. Entropy-coded quantization of memoryless sources is known to be an efficient source-coding technique in a rate-distortion (R-D) sense. In recent work, an adaptive entropy-coded quantizer (AECQ), which used buffer-state feedback to control the quantizer characteristics, was investigated. Simulation results demonstrated that the buffer underflow/overflow problems that are normally associated with entropy coding could be reduced with minimal rate or distortion penalty. In the present paper, buffer underflow and overflow problems are completely eliminated by effectively imposing reflecting walls at the buffer endpoints. Synchronous operation of the AECQ encoder and decoder is examined in detail, and it is shown that synchronous operation is easily achieved without side information. A method is then developed to explicitly solve for the buffer-state probability distribution and the resulting average distortion when memoryless buffer-state feedback is used as well as when the source is stationary and memoryless. This method is then used as a tool in the design of low-distortion AECQ systems, with particular attention being given to the development of source scale-invariant distortion performance. We show that the introduction of reflecting buffer walls in a properly designed AECQ system results in a very small rate-distortion performance penalty and that the resulting AECQ system can be an extremely simple and effective solution to the stationary memoryless source coding problem for a wide range of source types. Operation with nonstationary sources is also examined. © 1990 IEEE