Deep Learning Based Power Allocation in 6G URLLC for Jointly Optimizing Latency and Reliability

Abstract

Supporting ultra-reliable and low-latency communications (URLLC) is one of the primary goals for the sixth-generation (6G) cellular networks. Minimizing latency while maintaining high reliability is the central concept of URLLC. In this paper, we propose a deep learning (DL) based power allocation mechanism for jointly optimizing latency and reliability in 6G URLLC. Existing iterative algorithm-based solutions provide valuable insights, but there are significant challenges in implementing them in a real world system. The central theme of our paper revolves around merging theoretical network models and channel information in analyzing latency and reliability and training deep neural networks (DNNs) for satisfying the requirements of URLLC. This paper demonstrates a distinct approach on how to apply data-driven supervised DL in URLLC. The performance of the proposed system is evaluated through extensive simulations. Scrupulous comparison of results with those of the weighted MMSE (WMMSE) based systems validate that the proposed DNN models reduce latency drastically and simultaneously ensure service reliability.