A state-of-the-art survey and benchmarking of Adaptive Modulation and Coding for Underwater Acoustic Communications

Adaptive Modulation and Coding (AMC) is a critical technique for optimising data transmission in the highly variable and challenging domain of Underwater Acoustic Communication (UWAC). This survey reviews AMC methodologies across three paradigms: traditional (non-ML), supervised machine learning (ML), and reinforcement learning (RL). The evolution is traced from simple rule-based heuristics to data-driven polices that address noise, long propagation delays, multi-path, Doppler, and outdated channel state information (CSI). We then implement four representative classes of approach - SNR thresholding baselines, boosted tree classifiers, model-free multi-armed bandits and a model-based LSTM DQN. These approaches are comprehensively evaluated on an open-source experimental dataset. While traditional and supervised methods perform strongly in stable regimes, model-based deep RL shows superior adaptability, achieving the best throughput-reliability balance in time-varying channels. Finally, we discuss key open research challenges including data scarcity and model generalisation and highlight future directions for developing more robust, intelligent, and efficient AMC systems for UWAC.