Degree Name

Doctor of Philosophy


Department of Electrical and Computer Engineering


Packet switching is used extensively in Local Area Networks (LANs) for data communications, and is becoming increasingly important in the trend towards integrated services Wide Area Networks (WANs). As most existing speech codecs were designed with circuit switched connections in mind, they are vulnerable to packet loss, and are unable to fully exploit the variable capacity of packet switched connections. Consideration is therefore given in this thesis to the design and implementation of a speech codec specifically intended for use with packet switched networks.

The thesis starts with a discussion of the general characteristics of local and wide area networks. Then a model for a packet voice terminal, consisting of a speech codec, network voice protocol, and access controller, is described. The way in which the network and the components of the packet voice terminal can affect the quality of speech communications is then discussed, and this leads to a detailed set of requirements for the codec itself.

The codec design makes use of an "embedded" coding scheme, which allows rapid flow control of voice traffic to be performed, and enables the variable activity of the signal to be exploited for bandwidth compression purposes. The fundamental coding technique used is Adaptive Differential Pulse Code Modulation (ADPCM), and particular attention is given to the design of the adaptive quantizer in this algorithm.

A new structure for this device is developed, and the result is called the Generalized Hybrid Adaptive Quantizer (GHAQ). The GHAQ is easily optimized to the statistics of a particular signal by means of an iterative procedure, and is shown to yield improved signal-to-noise ratio over other well-known adaptive quantizers in Adaptive Delta Modulators and 2-bit ADPCM coders.

The codec is implemented on an IBM PC expansion card using a programmable digital signal processor. Associated interface hardware, designed to allow the packetization of coded speech with minimal processing overhead, is also included. This hardware/software system represents an economical means of adding voice traffic to an existing data LAN, and is a flexible vehicle for further research into packet voice communications.