Degree Name

Master of Engineering Research


School of Electrical, Computer and Telecommunications Engineering - Faculty of Informatics


This thesis studies the techniques and feasibility of embedding a perceptual audio coder within a lossless compression scheme. The goal is to provide for two step scalability in the resulting bitstream, where both a perceptual version of the audio signal and a lossless version of the same signal are provided in the one bitstream. The focus of this thesis is the selection of the perceptual coder to be used as the perceptual base layer and the techniques to be used to compress the lossless layer by using backward linear prediction followed by entropy coding. The perceptual base layer used is MPEG-4 AAC, chosen based on entropy measurements of the residual signal. Results of the work in this thesis show that the embedded lossless coding scheme could achieve an average compression ratio of only 6% larger compared to lossless only coding. Performing decorrelation on the AAC residual signal by means of backward linear predictive coding and measuring the entropy of the resulting LPC residual signal of various orders revealed that an 8% decrease in coding rate is achievable using 15th order prediction. Furthermore, this thesis also investigates an entropy coding technique known as cascade coding which is originally designed to compress hydroacoustic image data and is modified to compress audio data. Cascade coding is an entropy coding technique that uses multiple cascaded stages where each stage codes a specific range of integers and is used to perform entropy coding of the backward linear prediction residual signal. The cascade coding technique explored in this thesis includes using a frame based approach and trained codebooks.

02Whole.pdf (569 kB)