Dynamic Interference Management by Using the Enhanced Clustering and Deep Reinforcement Learning in VLC-Enabled UAV Communication

Unmanned aerial vehicle (UAV) equipped with visible light communication (VLC) emerges as a promising technology for nighttime outdoor consumer applications. Nevertheless, the interference from the overlapped coverage of multiple UAVs will degrade the signal quality thus constrain the overall system performance. In this paper, an efficient two-stage dynamic interference management approach is investigated to maximize the total sum-rate by jointly optimizing the user association, UAV trajectory, frequency band assignment and power allocation while satisfying the specific requirements of data rate and target illumination in VLC-enabled multi-UAV system. Firstly, the total flying time is split into multiple overlapping short time periods, and the enhanced K-means clustering algorithm is employed to establish the user association indirectly for each time window. Then, the multi-agent deep reinforcement learning framework is invoked to determine the optimal UAV trajectory and frequency-power resource blocks accordingly. After that, the sub-problems are sequentially solved over the time windows and the sub-optimal solution is obtained through several iterations and updates. Simulation results reveal that the proposed scheme exhibits excellent convergence performance, and it can also achieve at least 21.6% average throughput improvements as compared with the conventional schemes in complex interference patterns.