Secondary User Experience-oriented Resource Allocation in AI-empowered Cognitive Radio Networks Using Deep Neuroevolution

Abstract

Secondary user (SU)-experience-oriented resource allocation (RA) will become increasingly important in cognitive radio networks (CRNs) in future wireless networks. For efficient real-time processes, cognitive radios (CRs) are usually combined with artificial intelligence (AI) to improve better adaptation and intelligent RA. However, deep learning (DL), which is a key AI strategy with remarkable capabilities towards advancing this vision, has several built-in limitations. Firstly, the most successful DL applications require training with large amounts of data; secondly, they assume that the data samples to be independent, while in CRNs one typically encounters sequences of highly correlated states. To circumvent this issue, this paper introduces a deep neuroevolution (DNE) technique for dynamic RA. Using this technique, a stable learning framework was achieved by introducing the phenotypic plasticity of transmission rates and delay constraints inside a multi-layer perceptron (MLP). The stability of SU satisfaction as they increased in number was achieved at 36 SUs, which is a 13.3% decrease from when they were only 6 SUs in the CRN for all learning mechanisms.