Reinforcement learning-based adaptive SLIPT for free-space optical communication under Málaga turbulence

This paper investigates a reinforcement learning (RL)-based adaptive simultaneous lightwave information and power transfer (SLIPT) scheme for free-space optical (FSO) communication systems operating over M’alaga turbulence channels with pointing errors. The proposed framework dynamically selects the SLIPT operating mode to balance information decoding reliability and harvested energy under varying channel conditions. Unified closed-form analytical expressions for harvested energy, bit error rate (BER), and outage probability are derived for time-switching (TS), power-splitting (PS), and hybrid time-switching power-splitting (TSPS) schemes, incorporating a nonlinear energy harvesting model. A deep Q -network (DQN) agent adaptively selects the optimal SLIPT mode and associated parameters based on prevailing channel state information. The reward function enforces a priority hierarchy: outage probability is given the highest priority, followed by BER, while harvested energy is treated as a secondary objective. Simulation results demonstrate that the proposed framework significantly outperforms fixed SLIPT strategies in terms of energy harvesting efficiency and communication reliability across different turbulence regimes and weather-dependent attenuation conditions. Furthermore, the RL agent learns a non-trivial mode selection policy that certifies the suboptimality of PS under the nonlinear harvesting model, providing a data-driven design insight with practical implications for adaptive SLIPT receiver hardware.