Dynamic channel sharing strategies through game-theoretic reinforcement learning

Abstract

—Random access protocols are used by multiple nodes in wireless networks to effectively share a wireless channel for data transmission. While competing for the channel, the nodes seek to achieve an individual or group objective. Game theory, can thus be applied to analyze and model individual or group behavior of nodes in random access networks. It can also be used as an ‘engineering’ application and subsequently re–engineer the system. In this paper, the current CSMA/CA mechanism was analyzed using game theory. Based on the analysis, the strategy space available to individual nodes was increased so that the optimal strategies for different situations learnt using reinforcement learning. From the analysis it was determined that the Nash equilibrium was not Pareto optimal. Simulation experiments yielded better results for the modified algorithm pointing to moving the Nash equilibrium towards being fair and Pareto optimal.