Entropy coding using equiprobable partitioning

We present a simple source coding algorithm for independent and identically distributed (i.i.d.) sources that gives coding efficiency performance close to that of arithmetic coding, but with much lower computational complexity and much higher robustness to mismatches between the assumed and actual symbol probabilities. The method is based on the principle that the probability of occurrence of a symbol sequence is determined by the total number of occurrences of each member of the symbol alphabet, but not by the order of occurrence. Thus, the coding of a string of symbols can be accomplished in three steps. First, the sequence length M is encoded using an exp-Golomb code. Second, the symbol occurrence frequencies are coded using exp-Golomb codes. Third, a set of fixed length codes are used to select among the equiprobable candidate sequences. In contrast with arithmetic coding, which involves significant computation during the process of encoding and decoding, in the method described here the actual encoding and decoding are extremely simple. Furthermore, the proposed algorithm is robust to mismatches between the assumed and actual symbol probabilities.