Deep Reinforcement Learning Optimal Transmission Policy for Communication Systems With Energy Harvesting and Adaptive MQAM

In this paper, we study an optimal transmission problem in a point-to-point wireless communication system with energy harvesting and limited battery at its transmitter. Considering the non-availability of prior information about distribution on energy arrival process and channel coefficient, we propose a deep reinforcement learning (DRL) based optimal policy to allocate transmission power and adaptively adjust multi-ary modulation level according to the obtained causal information on harvested energy, battery state, and channel gain to achieve maximum throughput of the system. This optimization problem is formulated as a Markov decision process with unknown state transition probability. Applying the principle of the DRL, we use a deep Q-network to find the optimal solution in continuous state space, which provides rapid convergence since there is no additional memory required. Simulation results show that the proposed policy is effective and valid and it can improve the throughput of the system compared with Q-learning, greedy, random, and constant modulation level transmission policies.