Channel correlation-aware DQN-based adaptive power transmission in an FSO network

In a free-space optical (FSO) communication network, the optical signal suffers from turbulence-induced fading, which leads to serious performance distortion. To address this challenge, we propose a channel correlation-aware deep Q-network (CCA-DQN)-based adaptive transmission algorithm. This technique constructs a joint state space by leveraging the Pearson correlation coefficient related to the zenith angle and the downlink received signal strength, enabling the agent to learn transmit power decisions without requiring explicit uplink state feedback. The optimization objective is formulated as minimizing the average transmit power under the constraint of long-term average bit error rate (BER). Recognizing that average BER merely characterizes transmission quality in statistical significance, the outage probability is further introduced as a complementary reliability metric to evaluate the link survivability under deep fading scenarios. Simulation results demonstrate that, compared to baseline methods of max transmit power, random transmit power, truncated channel inversion, feedback-based moving-average power control, and DQN without CCA, the proposed approach reduces the average transmit power by approximately 60% while maintaining the average BER at the order of 10^-5, and its outage probability performance approaches that of max transmit power.