Analysis of Mode Switching Metrics for Deep Reinforcement Learning-Based Adaptive Modulation

Aiming at the characteristics of random fading and strong noise caused by the multipath effect of the underwater acoustic channel, this paper proposes an improved adaptive underwater acoustic communication scheme. The scheme uses signal-to-noise ratio (SNR) and correlation coefficient(ρ) as the two-dimensional channel quality evaluation criteria in this paper, and uses the dual-channel-based quality evaluation method and deep Q-network learning (DDQN) algorithm for adaptive selection of modulation methods. Simulation experiments verify that the proposed method can improve system throughput while maintaining bit error rate constraints. Compared with traditional schemes, the new algorithm has the ability to extract channel state information, learn parameter expressions, and complex underwater acoustic communication environments. The experimental results prove that the scheme significantly improves the system performance in the complex underwater acoustic communication environment, and provides an innovative solution for realizing high-reliability and high-throughput underwater acoustic communication. The algorithm proposed in this paper has faster convergence speed and lower outage probability.