Learning-based joint optimization of mode selection and transmit power control for D2D communication underlaid cellular networks

Abstract

In device-to-device (D2D) communication, each transceiver pair can choose to operate in either direct D2D mode or device-infrastructure-device (DID) mode. This technique is called mode selection, which is important for establishing a reliable and efficient communication link that can significantly improve the network performance. However, the coexistence of direct D2D and DID communications in the same frequency band creates interference and degrades the performance of the system. Therefore, in this study, we aim to jointly optimize the mode selection and transmit power allocation of D2D pairs in D2D communication underlaid cellular networks. For this purpose, we first formulate a joint optimization model considering both mode selection and transmit power control of D2D pairs, which is an NP-hard combinatorial optimization problem with linear and nonlinear constraints. To solve this optimization problem, we design a deep neural network (DNN) structure. The proposed DNN algorithm is trained by minimizing the loss function which is obtained from Lagrange duality function. The weighting factor in the loss function is designed to decrease (increase) the rate of the receiver with strong (weak) interference compared to the mean channel gain. Simulation results show that the proposed DNN algorithm achieves a near-global optimal solution with lower computational complexity compared with the exhaustive search (ES) approach, and outperforms the solution obtained by using sub-optimal methods for Lagrange duality function.