Citation

Abstract

In this article, the authors consider the following question about Huffman coding, which is an important technique for compressing data from a discrete source. If p is the smallest source probability, how long, in terms of p, can the longest Huffman codeword be? It is shown that if p is in the range 0 < p < 1/2, and if K is the number, then the longest Huffman codeword for a source whose least probability is p is at most K, and no better bound is possible. Asymptotically, this implies the surprising fact that for small values of p, a Huffman code’s longest codeword can be as much as 44 percent larger than that of the corresponding Shannon code.

Details

Volume
42-110
Published
August 15, 1992
Pages
188–193
File Size
246.4 KB