Energy Efficient Multi-Pair Massive MIMO Two-Way AF Relaying: A Deep Learning Approach

We consider two-way amplify and forward half-duplex massive multiple-input multiple-output (MIMO) relaying, where multiple user-pairs exchange information via a shared relay. Most of the existing work on massive MIMO relaying solves the weighted sum energy efficiency (WSEE) maximization problem using iterative optimization algorithms, which are not suitable for real-time implementation due to high computational complexity. We develop a deep neural network (DNN) based power allocation to maximize the WSEE by learning a unknown function which maps the input (i.e. channel fading coefficients, system total transmit power and relay antennas) and the output optimal power vector. Once the DNN learned the unknown map, DNN provides a non-iterative closed form expression to solve the WSEE maximization problem in real-time with much lower computational complexity. We numerically demonstrate the performance of the proposed approach achieves optimal performance as the existing iterative optimization methods.