Transfer learning and cooperation management: balancing the quality of service and information exchange overhead in cognitive radio networks

This paper introduces and studies a novel solution of transfer learning applied to spectrum management in cognitive radio networks in order to improve the Quality of Service (QoS) and convergence performance of conventional full distributed learning. Cooperation management has been investigated to enhance transfer learning as a novel way of reducing the need for control information exchange between distributed cognitive agents while providing the same effective QoS as that achieved in a fully coordinated network. A structured approach is taken to transfer learning, including a source agent selection function defining how the agents exchange learning information, and a target agent training function reinforcing the knowledgebase. It is demonstrated in simulation and analysis that transfer learning achieves a significantly higher QoS and throughput than distributed reinforcement learning. The cooperation management algorithm is shown to effectively reduce the need for information exchange by 90 per cent whilst still providing adequate QoS compared with a fully coordinated network. Copyright © 2014 John Wiley & Sons, Ltd.