Deep-Learning-Based Resource Allocation for Time-Sensitive Device-to-Device Networks

Ultra-reliable and low-latency communication (URLLC) is considered as one of the major use cases in 5G networks to support the emerging mission-critical applications. One of the possible tools to achieve URLLC is the device-to-device (D2D) network. Due to the physical proximity of communicating devices, D2D networks can significantly improve the latency and reliability performance of wireless communication. However, the resource management of D2D networks is usually a non-convex combinatorial problem that is difficult to solve. Traditional methods usually optimize the resource allocation in an iterative way, which leads to high computational complexity. In this paper, we investigate the resource allocation problem in the time-sensitive D2D network where the latency and reliability performance is modeled by the achievable rate in the short blocklength regime. We first design a game theory-based algorithm as the baseline. Then, we propose a deep learning (DL)-based resource management framework using deep neural network (DNN). The simulation results show that the proposed DL-based method achieves almost the same performance as the baseline algorithm, while it is more time-efficient due to the end-to-end structure.