Machine learning based delay tolerant protocols for underwater acoustic wireless sensor networks

[Truncated abstract] In recent years, the developments of underwater acoustic wireless sensor networks (UA-WSNs) have attracted considerable research interest due to their capabilities to support underwater missions. Underwater acoustic communication channels are featured with large attenuation, long propagation delay and constrained bandwidth, which limit communications between sensors and make the system intermittent. As a result, delay tolerance is one of the major design concerns for supporting UA-WSNs to carry out tasks in harsh subsea environments. Although a number of delay tolerant schemes have been proposed for terrestrial wireless sensor networks, the fundamental differences between underwater acoustic channels and radio frequency channels make those schemes perform poorly in subsea environments. Therefore, it is desirable to develop a feasible, reliable and robust protocol for high-speed underwater acoustic wireless communications. In this dissertation, we present a family of delay tolerant protocols for UA-WSNs, which employ reinforcement learning algorithms.