A DSP-based decompressor unit for high-fidelity MPEG-Audio over TCP/IP networks

Copyright 1997 by Björn Wesén at Axis Communications AB, Sweden

[ Home | Contents | Component and tool archive ]

Previous chapter: Contents, next chapter: Compression of digital audio

Introduction

Multimedia as an invention is old by now, but still normal home-computers are astonishingly bad at both audio and video processing. This is due to the immense amount of data resulting from the digitizing of audio and video - the computers cannot neither store nor transmit audio and video in their raw digital form. A stream of digitized broadcast-quality video requires a transmission speed in the order of 100 Mbit/s, and a stream of CD-quality stereo audio requires 1.4 Mbit/s. Four minutes of music (a normal song) thus require around 40 Mbyte of storage space. It is apparent that some means of compressing both video and audio is necessary, and the compression needs to be specialised and more efficient than conventional compressors commonly used for compressing computer programs, due to the major difference in content.

The MPEG organisation has released several compression algorithms for both audio and video. Fairly recently, in the autumn of 1996, Fraunhofer (a member of MPEG) released their new audio-compression algorithm called MPEG-1 Audio layer-3. The good news was that it could compress audio signals like music by a factor of 10 to 12, with almost no audible loss. This breakthrough meant that a song that previously needed 40 Mbyte of storage now could be stored in 3-4 Mbyte instead - suddenly it was feasible to keep music on your harddisk and download it from the Internet without spending a fortnight doing it. Programs that could decode and play these music files were made for both PC, Macintosh and Unix systems, and the standard became popular very fast.

The idea came up about using this or a similar compression technique to create a small, stand-alone decoder, that would attach itself directly to a network and play music or any other audio it had fetched or got pushed from the net. It could be built into a loudspeaker or anything. This Master's thesis is about the design of such a device - from the analysis of which compression algorithm to use, to the implementation of the algorithm in a Digital Signal Processor, to the design of the hardware and circuit board containing the DSP and interfacing it to the network. The project name of the thesis is Whitney, because I like Whitney Houston.

Specificially, the goal of the thesis was to:

• Find an audio compression algorithm that could compress over 10 times with as little perceptible loss as possible
• Determine if there are any existing Digital Signal Processor or other hardware implementations of the chosen algorithm
• Choose a suitable Digital Signal Processor chip that would be powerful enough to perform audio decoding
• Develop hardware around the chosen DSP to interface it to an Axis ETRAX design (providing network access) and to a Digital-to-Analog converter to provide audio output
• Develop the decoding software implementation of the chosen algorithm on the DSP
• Make the final unit play compressed audio streams over a TCP/IP network such as the Internet

The thesis was carried out at Axis Communications AB, a swedish company strongly profiled in the design and manufacture of networkable computer components.

Previous chapter: Contents, next chapter: Compression of digital audio

This document may be freely distributed for educational purposes. See the copyright notice for additional information.