LSTM-enabled prediction-based channel switching scheduling for multi-channel cognitive radio networks

Abstract

To protect primary users (PUs) from harmful interference and to maintain seamless communication, secondary users (SUs) need to trigger channel switching in cognitive radio networks (CRNs). However, due to frequent channel switching, SUs lose considerable energy. In order to perform channel switching efficiently in multi-channel CRNs (MC-CRNs), an optimal channel switching schedule is desirable, which can assist SUs in minimizing the number of switching and the switching energy overhead. In this work, we propose a machine learning based prediction-driven channel-switching scheduling scheme for MC-CRNs. The proposed scheme optimizes the utility of SUs by balancing the trade-off between spectral efficiency and channel switching overhead. Initially, a Long-Short Term Memory (LSTM) network-based channel prediction scheme is presented using which individual SUs locally predict PU occupancy in future time slots based on their spectrum sensing history. Thereafter, a cooperative channel prediction framework is introduced using which SUs perform cooperation of local prediction results to enhance the prediction results. Using the cooperative prediction results, a channel-switching scheduling problem is formulated, which captures the trade-off between spectral efficiency and channel-switching overhead. A proof is carried out to establish that the formulated scheduling problem is NP-Hard. To solve the formulated problem in polynomial time, a greedy-based heuristic solution is proposed for scheduling the SUs into different slots of channels to achieve a sub-optimal solution. A simulation-based analysis has been carried out to demonstrate our proposed scheme’s efficacy compared to a conventional exhaustive search and fixed channel switching schemes.