Multiple access with multi-dimensional learning for cognitive radio in open spectrum

Abstract

In this paper, we present an Multi-Dimensional Learning multiple access control scheme for cognitive radio users in unlicensed spectrum. The proposed scheme is developed by applying reinforcement learning to the perspectives of both channel assignment and retransmission policy based upon a multichannel p-persistent CSMA algorithm. In a distributed open spectrum sharing situation, decisions regarding channel assignment and the p-persistent probability level are made by each individual cognitive radio user considering historical transmission experiences. Through the learning process, cognitive radio users are expected to avoid each other, and maintain an optimum transmission rather than contending for a channel. In the situation where two groups of users have different levels of offered traffic, simulation results have shown that our proposed scheme significantly increases the throughput, and optimizes the transmission delay of both cognitive user groups, compared with the schemes that only apply reinforcement learning to channel assignment.