Energy Efficient Resource Allocation for D2D Communications using Reinforcement Learning

The millimeter-wave (mmWave) device-to-device (D2D) communication is already being employed to satisfy the high datarate demand of the internet-of-things nodes. Using mmWave signals has its own challenges as it suffers from high penetration losses. Therefore, presence of dynamic obstacles further complicates the already hard problem of allocation of channel resources to the demanding nodes. In this work, we have proposed a reinforcement learning (RL) based framework to jointly allocate the frequency channels as well as assign the transmit-powers to the demanding D2D pairs in order to maximize the energy-efficiency in presence of dynamic obstacles. We justify our choice of reward function through a formal proof and also ensure the convergence of the algorithm. Through extensive simulations, we show that our proposed RL framework not only converges, but also outperforms an existing approach.