Bit-wise arithmetic coding for data compression

Consider the problem of compressing a uniformly quantized IID source. A traditional approach is to assign variable length codewords to the quantizer output symbols or groups of symbols (e.g., Huffman coding). We propose an alternative solution: assign a fixed length binary codeword to each output symbol in such a way that a zero is more likely than a one in every codeword bit position. This redundancy is then exploited using a block-adaptive binary arithmetic encoder to compress the data. This technique is simple, has low overhead, and can be used as a progressive transmission system.