Abstract
We investigate the problem of beam tracking in a cell-free millimeter wave (mmWave) massive MIMO scenario with multiple access points (APs) and multiple mobile users (UEs). The difficulty in handling multiple APs, UEs and the unknown of UEs’ movement, such as their directions and speeds which is crucial for the beam prediction in the next time slot, poses overhead and performance challenges for beam tracking. To address these challenges, we propose a novel adaptive distributed training and interaction framework and develop a distributed deep Q-network (DDQN) beam tracking algorithm based on the beam image stacking (BIS) method. To enhance beam tracking performance, we make significant improvements and optimizations to the action design of DDQN. Simulation results demonstrate significant performance improvements compared to existing methods. Notably, our algorithm is able to dynamically adapt to changes in the environment, leading to more efficient and accurate beam tracking.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call