Power Allocation for NOMA-Based Two-Way Full-Duplex Relaying Systems

Abstract

In this paper, the design of power allocation is studied to maximize the spectral efficiency of a non-orthogonal multiple access-based two-way full-duplex relaying system. Due to the non-convexity of the original problem, a polyblock outer approximation (POA)-based offline mechanism is first proposed to approach the global optimum in polynomial time. To fulfill the low-latency requirement in 5G and beyond, a deep neural network (DNN)-based online mechanism is then proposed to directly build up the mapping from the channel state information to the power allocation policy by using the solutions of the POA-based offline mechanism. In case the output power allocation of the trained DNN is infeasible, orthogonal projection onto the feasible region is further invoked. Based on the simulation results, we show that the POA-based mechanism outperforms the popular successive convex approximation-based benchmark, and the DNN-based mechanism substantially cuts down the online computational time at the cost of only minor performance loss when compared with the POA-based mechanism.