DRL-Based Sum-Rate Maximization in D2D Communication Underlaid Uplink Cellular Networks

Device-to-device (D2D) communication affords the benefits of improved network spectral efficiency, throughput, energy efficiency, and delay performance. However, D2D communication underlaid cellular uplinks are degraded by interference between D2D user equipment (DUE) and cellular user equipment (CUE) that use the same frequency band. In this context, researchers have proposed the optimization of the transmit power of the devices to mitigate interference. This power allocation problem is generally modeled as a NP-hard combinatorial optimization problem with linear constraints; therefore, traditional optimization methods are ineffective in addressing the problem. In this paper, we apply a deep reinforcement learning (DRL) algorithm to optimize the transmit power for both DUEs and CUEs in the context of D2D communication underlaid uplink cellular networks, which can solve the optimization problem via optimal decision-making together with efficient deep network training. Our simulation results show that the proposed algorithm affords a near-global-optimum solution with lower computation complexity than exhaustive search, and outperforms the fixed-power, weighted minimum mean square error (WMMSE), gradient descent method, and Newton-Raphson approaches.