Link Adaptation Strategy for Underwater Acoustic Sensor Networks: A Machine Learning Approach

Abstract

Abstract Due to growing concerns regarding their use in fields including oceanography, commercial marine operations, and military surveillance, demand in the exploration of underwater sensor networks for marine studies has developed. Network channels for underwater sensor network (USN) rapidly change (spatially and temporally) depending on the surroundings. To increase system efficiency by adjusting transmission parameters to channel fluctuations, it is alluring to utilize adaptive modulation and coding (AMC) for USNs. In order to determine the best link adaptation method based on the channel quality, this article focuses on evaluating a measured sea trial dataset utilizing a rule-based approach (i.e., three-dimensional evaluation, modulation-wise analysis, and a fixed-SNR strategy). To determine the optimum AMC combinations in terms of channel adaptively, we draw a situation of the measured USN data rate versus Bit Error Rate (BER) and Signal to Noise Ratio (SNR). The work further extends to apply machine learning (ML) methods to identify the MCS levels by looking into the channel characteristics due to the non-reversibility limitation of the rule-based strategy. One of the ML methods we used for the investigation, gradient boosted regression tree (GBRT), exhibits impressive accuracy of 99.988% in classifying MCS levels. The MCS levels are related to channel statistics and signal characteristics, particularly those that are susceptible to SNR and BER limitations, using an ensemble of trees that learns from the buoy and base station’s uplink data.