Abstract
Packet communications have increasingly become popular since the launch of the first ever packet-switched network, the ARPANET. Evolution of such networks resulted in creation of the Internet, the global packet-switched network. Amongst the well-known advantages of packet-switching are service transparency, cost-effectiveness, and high system capacity. A natural pathway will incorporate tetherless mobility features to packet services. The key component to realisation of wireless packet communication is the multiple access technology (the subject of this thesis), capable of optimum sharing of scarce radio resources amongst a diverse set of packetized information sources. Based on core features of packet multiple access protocols, a classification is presented by which most proposed protocols could be mapped into. For mixed services, two equivalent offered loads, namely ''access load" and "traffic load", are derived. The former can be used for design, optimisation, and stabilisation of the access mechanism, whilst the latter is to evaluate the overall system performance. The proposed normalised offered loads are found accurate and instrumental in predicting and designing packet access protocols for mixed services. "Statistical upper-bounds" for packet access protocols are derived (both for random access and general access mechanisms), based on which an "efficiency factor" is defined to evaluate and compare merits of protocols against each other. Conventionally, analysis of resource-sharing systems with mixed services (such as multiple access protocols) involve multidimensional Markov chains which become quite intractable for more than two service types. In order to avoid multidimensional Markov analysis, an analytical method called "Aggregate Source Modelling" is proposed by means of which various service types are aggregated into a representative service type. The proposed method is deployed to analyse various queueing and multiple access scenarios. Finally, a series of multiple access protocols are presented for adapting real-time and non-real-time packet services to existing cellular and satellite systems. For evaluation purposes, the proposed statistical models of packet services (such as FTP, E- mail, WWW, packet voice, and so forth) are used throughout the thesis. These models have components of various statistical distributions, namely, Poisson (Exponential), lognormal, Pareto, Cauchy, Gaussian, Geometric, and Hyper-Exponential. Extensive discrete-time simulation models have been developed for evaluation purposes.