Analysis and Implementation of Reinforcement Learning on a GNU Radio Cognitive Radio Platform

Spectrum today is regulated based on fixed licensees. In the past radio operators have been allocated a frequency band for exclusive use. This has become problem for new users and the modern explosion in wireless services that, having arrived late find there is a scarcity in the remaining available spectrum. Cognitive radio (CR) presents a solution. CRs combine intelligence, spectrum sensing and software reconfigurable radio capabilities. This allows them to opportunistically transmit among several licensed bands for seamless communications, switching to another channel when a licensee is sensed in the original band without causing interference. Enabling this is an intelligent dynamic channel selection strategy capable of finding the best quality channel to transmit on that suffers from the least licensee interruption. This thesis evaluates a Q-learning channel selection scheme using an experimental approach. A cognitive radio deploying the scheme is implemented on GNU Radio and its performance is measured among channels with different utilizations in terms of its packet transmission success rate, goodput and interference caused. We derive similar analytical expressions in the general case of large-scale networks. Our results show that using the Q-learning scheme for channel selection significantly improves the goodput and packet transmission success rate of the system