Lossless data compression with low complexity

Abstract

An adaptive linear predictor is used to predict samples, and residuals from such predictions are encoded using Golomb-Rice encoding. Linear prediction of samples of a signal which represents digitized sound tends to produce relatively low residuals and those residuals tend to be distributed exponentially. Accordingly, linear prediction combined with Golomb-Rice encoding produces particularly good compression rates with very efficient and simple implementation. A code length used in Golomb-Rice, which is typically referred to as the parameter k, is adapted for each sample in a predictable and repeatable manner to further reduce the size of a Golomb-Rice encoding for each sample. An infinite incident response filter of processed residuals automatically reduces influences of previously processed residuals upon such adaptation as additional samples are processed. The efficiency of Golomb-Rice encoding is improved by limiting the predicted samples to an efficient range. The maximum of the efficient range is the maximum valid value of a sample less the maximum positive value of the fixed-length, binary portion of an encoded residual. The minimum of the efficient range is the minimum valid value of a sample plus the minimum negative value of the fixed-length, binary portion of an encoded residual. Such reduces the number of bits required to represent a variable-length, unary portion of the encoded residual by improving the efficiency with which the fixed-length, binary portion can represent a particular residual.