Joint User Grouping and Resource Allocation for Network Massive MIMO Communications

In this paper, we investigate multi-cell resource management for massive MIMO communications, where users within different cells reuse the same time-frequency resource further aggravating the inter-user interference. Based on re-inforcement learning, which has been a useful technique for resource management, we propose a joint user grouping and time-frequency resource allocation algorithm (JUGRA). JUGRA contains user grouping network (UGNet) and time-frequency resource allocation network (TFRANet). TFRANet is used to select time-frequency resource allocation scheme, which maximizes system throughput with fixed user grouping scheme. UGNet is built to select the multi-cell user grouping scheme and calls TFRANet to evaluate the system performance, realizing joint optimization of user grouping and resource allocation. Simulation results demonstrate that the performance of JUGRA can be improved by more than 30% over the greedy algorithm with lower computational complexity. The proposed time-frequency resource allocation algorithm based on TFRANet can achieve 95 % of exhaustive search performance.