Machine Learning Based Link Adaptation on an Underwater Communication Network

Underwater acoustic communication (UAC) networks face significant challenges due to rapid spatial and temporal variations in channel conditions influenced by environmental factors. To enhance system efficiency, adaptive modulation and coding (AMC) can be employed to adjust transmission parameters in response to these channel variations. A thorough analysis of a measured sea trial dataset is essential to determine an appropriate link adaptation strategy tailored to channel quality. This analysis employs a rule-based approach that includes three-dimensional evaluation, modulation-specific analysis, and fixed-SNR strategy. However, non-reversible nature of the rule-based strategy presents limitations, prompting an extension of the study to incorporate machine learning (ML) algorithms for classifying modulation levels and coding rate levels based on channel characteristics. Among four ML algorithms evaluated, the boosted regression tree (BRT) algorithm demonstrated exceptional accuracy, achieving a 99.97% success rate in MCS level classification. Further ML analysis involved various parameter combinations to classify coding rates, assessing both individual and overall classification accuracies. Additionally, the rule-based strategy was utilized to determine optimal threshold values for coding rate labels. This was achieved through the analysis of Gaussian distribution plots and multi-dimensional scatter plots, providing a comprehensive approach to link adaptation in UAC networks.