Channel Quality Prediction for Adaptive Underwater Acoustic Communication

In this paper, the communication quality of an underwater acoustic link between two nodes is quantified by the predicted channel gain and delay spread using a stochastic and reinforcement learning model. The stochastic model generates an ensemble of time-varying channel characteristics by capturing the effect of known environmental changes including changes in sound speed profile, tides and bathymetry. Along with the stochastic model to capture the impact of unknown environmental parameters on channel quality a hidden Markov model is utilized to complement sparse channel measurements and predict the channel characteristics over a long time period spanning multiple days. In this work, the nodes are bottom mounted in a shallow turbulent water environment, with known tide cycles, physical oceanography conditions and channel geometry. As such, the channel characteristics can be estimated using a simulation software model at the remote nodes. While the simulation model is used to estimate the initial channel condition between the nodes in short-term deployment, as will be shown, the hidden Markov model provides an accurate channel characteristics prediction for long term deployment, which can be utilized by software-defined acoustic nodes such that they can adapt to the time varying acoustic channel.