Power allocation in a spatial multiplexing free-space optical system with reinforcement learning

Abstract

The multiple-input multiple-output (MIMO) technique for free-space optical (FSO) system was initially designed for combating fading events in the diversity mode. However, as people demand for higher throughput, extra freedom can be obtained from the multiple apertures in the spatial multiplexing mode, where the system transmits independent parallel data streams over multiple apertures to increase data rate. In this paper, we study a MIMO FSO system in the multiplexing mode. By maximizing long-term benefits on the average capacity within limited time slots, we propose a power allocation algorithm based on the reinforcement learning (RL) method. Our RL algorithm utilizes an actor–critic structure, where both action space and state space are continuous. We also add the constraints on the peak power and total power. A novel reward function is designed with a punishment item for remaining power. The proposed RL algorithm can achieve a better performance than the existing benchmarks.