Fast Canonical Huffman Decoder

Abstract

We propose a novel method to decode canonical Huffman codes specifically designed to decode multiple symbols in one decoding cycle. The encoding method used is the classic canonical Huffman encoding. Our decoding method is specifically designed for speed and provides big throughput of decoding data; within our tests, we have achieved a decompression output of more than 2.1GiB/s when decompressing highly redundant data. Our method holds minimal storage requirements for decoding tables while still operating at very high speeds. We will also show that for limited-length canonical Huffman codewords that extend up to 12 bits, the memory size and construction times of the decoding tables are negligible.