Reinforcement Learning for Spectrum Prediction and EE Maximization in D2D Communication

This paper proposes a reinforcement learning (RL) based Q-learning to address the issues of joint spectrum prediction (SP) and device-to-device (D2D) data communication in cognitive radio (CR) framework. An optimization problem is formulated that addresses energy efficiency (EE) maximization of D2D communications under the constraints of its total transmit power and a certain data transmission rate while meeting an interference threshold and cooperation rate in primary user (PU) transmission. The high accuracy in SP offers reward as an improvement on EE while a compulsion of meeting an interference threshold and a penalty on PU data transmission are made based on the relative degree of wrong prediction. A large set of simulation results shows that the proposed method offers 30% gain in EE while 20% reduction in data collision with PU over the existing works.