Distributed learning based message dissemination approach for underwater surveillance in OUSN

Opportunistic Underwater Sensor Network (OUSN) is deployed for various underwater surveillance. The nodes in OUSN always keep moving, and the movement laws are extremely complex, due to many factors such as the movement intentions and living habits of their carriers. Besides, the message dissemination in OUSN is implemented in an opportunistic manner, because the communication links between nodes are intermittent, making the stable communication paths hard to be formed. The link prediction technique can be applied to predict the future links that may appear in the network topologies, and these links indicate the probabilities of future encounters between nodes, which helps to improve the performance of message dissemination. Especially, because the communication links between nodes are varied over time, a centralized link prediction is not feasible, and the nodes should predict the future links and make dissemination decisions locally. In this paper, we propose a Distributed Learning based Message Dissemination Approach (DLMDA) for each node to disseminate the stored data messages distributedly. Specifically, the link prediction results are expressed by some adjacency matrices, based on which each node disseminates the stored data messages to several selected neighbours. By adopting a distributed learning framework, both the communication overhead and processing delay of DLMDA are significantly reduced by avoiding the uploads of historical links to a server. Simulation results demonstrate the superior performance of DLMDA, i.e., through predicting the future links with the distributed learning framework, DLMDA enhances the delivery ratio of data messages, and reduces the delivery delay of data messages and the communication overhead effectively.