Universal Source Coding of Finite Alphabet Sources

Abstract

The application of Shannon's rate distortion theory to real data compression practice has been limited by a lack of effective distortion measures for most real users and incomplete statistical knowledge of real sources. Universal source coding techniques attempt to overcome the latter limitation by developing codes that are effective (in the sense of approaching the theoretical rate distortion limit) for a large class of sources. Here we examine one such technique for sources with finite alphabets where source sequences are divided into composition classes and a universal code is formed from codes designed for each composition class.