This paper introduces a method of designing optimized MAC layer algorithms using Genetic Programming By evolving entire algorithmic behaviour rather than optimizing a set of values to tune a parameterized design, a much wider space of behaviour can be explored automatically. This technique is illustrated using the variation of contention window size that is part of the Distributed Coordination Function of 802.11. When applied to the example of a variable sized network under saturated load this approach produces expressions that comfortably outperform the standard 802. JIb behaviour. Also, despite being automatically generated, these solutions achieve the throughput performance of the best enhancements to this aspect of the protocol.

-As the potential data-rates of wireless local area networks (WLANs) continue to rise, the ability of such systems to support a rich set of applications increases. The centralized control functions in the IEEE802.11 family of standards have been developed to enable both data-oriented (browsing, email, etc.) traffic and quality of service (QoS) sensitive traffic to coexist. Balancing the demands of the two types of traffic has, to date, been achieved by algorithms based on experimental, heuristic data. In this paper we present a non-linear optimization theory-based approach for deriving optimum configurations with the IEEE802.11/e centralized control functions in mind. The optimization algorithm itself (the "Barrier Method") is well-known; the challenge in problems such as this is in the formation of the utility function and its constraints, so these are explored in detail. Finally, we show the advantages of this approach over discrete look-up based approaches.

No abstract