Integer Coefficient Predictive Technique And Its Application In Lossless Audio Compression

Lossless audio compression is a compression technology of using the compressed data to construct new data that is identical to the originals. In the process, there have no data losing. A variety of utilities such as PkZip are designed to be very effective for text data compression. Unfortunately, they do not succeed in compressing audio data very well. What is needed are algorithms specifically designed for digital audio data compression.In this thesis, the current state-of-the-art lossless audio codec is analyzed and a study of some compression algorithms is made, the compression ratio of experimental audio files compressed by some lossless compression utilities is also given. The study suggests that lossless compression algorithms rarely obtain a compression ratio larger than 3:1; so, the design of a lossless coder should be focused on reducing algorithm complexity.After a survey of the coder AudioPak, this thesis brings up an improved integer coefficient predictor. The improved predictive algorithm atones for the shortage of AudioPak on frequency responses and reduces the range of prediction error. With the improvement on predicting and entropy coding, performance indexes are raised on minimum, maximum, average value and variance per frame, together with first-order entropy, spectrum of the prediction error and the influence of frame length on prediction error. Experimental studies indicate that the improved predictive algorithm present significantly better performance than the AudioPak.Based on the study of design of framing, predictor and entropy coding, this thesis expounds how to design, realize and analyze lossless audio coder under MATLAB. Finally, a great deal of experimental data is obtained to show the coder performance.