Pre-setting of channel types for long range underwater acoustic communications

Long range underwater acoustic communication (LR-UWAC) is essential to applications including manned and unmanned operations such as control an unmanned underwater vehicle (UUV) over long term surveying, communications for submarines, and under-the-ice operations. While underwater communication over short range of a few km has been relatively established, this is not the case for LR-UWAC over distances of tens of km. This is partly because of complex propagation loss makes it hard to obtain data, but mostly due to uncertainty in the channel. Specifically, different than for short range UWAC where ray tracing models can be used, the channel for LR-UWAC is highly complex and relies greatly on the sound speed profile and the bathymetry. Further, feedback from the receiver is not available for LR-UWAC. In that context, the communication type should be chosen by the expected channel instead of actual channel, but choosing the appropriate modulation scheme blindly is challenging. Considering this challenge, in this paper, we propose a method to pre-set the modulation type according to an evaluation of the channel type. We based our scheme on a machine learning application aimed to classify the expected channels from a numerical parabolic equation (PE) model set by some (possibly mismatched) environmental knowledge sampled only at the transmitter. The classifier labels the expected channels into four types, and the modulation scheme is chosen as the one than is expected to perform best for the selected channel type. Our numerical simulations show the average classification accuracy for matching the channels into the correct types is 86.7%, which means the proposed method is a promising method of pre-setting modulation scheme for LR-UWAC.