Adaptive Modulation and Coding With Feedback Scheduling for an Underwater Acoustic Link

Underwater acoustic channels exhibit significant temporal and spatial variability, making it challenging to design a single communication scheme that works well everywhere and at all times. Adaptive modulation and coding (AMC) techniques offer a solution by dynamically selecting the optimal modulation and coding scheme (MCS) for specific channel conditions but require an accurate model to predict communication performance. We propose a bit error rate (BER) estimation model that fuses domain knowledge to aid the evaluation of MCSs. In complex sea conditions, we enhance the reliability of AMC by extending our BER prediction model from a point prediction to an interval predictor. This extension involves incorporating Gaussian process regression to address the uncertainty in BER. Predictions from such an algorithm are used to drive AMC to maximize communication throughput reliably. For AMC, regular feedback from the receiver to the transmitter is necessary to gather channel state information (CSI). On the one hand, obtaining feedback too often reduces the communication throughput in channels with long propagation delays, but on the other hand, insufficient feedback leads to suboptimal AMC decisions and hence poor throughput. We propose an algorithm that integrates tree search and deep Q-network (DQN) for feedback scheduling to automatically find the right balance and optimize communication performance. We demonstrate the advantages of our algorithm through experiments in a test tank and at sea in Singapore. Furthermore, our algorithm also exhibited reliability and achieved optimal throughput in various underwater environments in simulation.