Deep Reinforcement Learning based Adaptive Discontinuous Reception Method for Cognitive Radios

Discontinuous reception (DRX) is one of the important power-saving mechanisms for 5G/B5G cognitive radio systems. Conventional DRX methods improve the power-saving performance at the cost of data transmission delay. To address this issue, a deep reinforcement learning based DRX method is proposed. An adaptive time-slot DRX structure is propsed and the DRX mechanism is modeled as a Markov decision process. The state set, action space and reward expressions are jointly designed for the DRX mechanism. A DRX decision strategy is obtained by deep reinforcement learning, which adaptively adjusts the window length of wake-up period. Simulation results are provided to demonstrate that the proposed DRX method improves the power-saving performance while the average transmission delay of data packages closes to the lower bound.