Publication

Abstract : Oceannet is a heterogeneous wireless network of fishing vessels at sea to provide internet connectivity at sea. Cellular network coverage is available only up to few kilometers and there is a lack of low cost communication mechanism beyond 20 km from shore. The maritime communication network is heterogeneous in terms of the connectivity ranges, directionality, resources, and mobility to extend coverage up to deep sea. The dynamic and unpredictable nature of sea environment contributes to a time-varying link structure and hence the selection of neighbors for packet forwarding is a critical factor to provide better connectivity. In this paper, we propose a connectivity aware reinforcement learning algorithm to rank the neighborhood of each node for different traffic types. The system is modeled as a Markov Decision Process and each node updates the reward of link selection based on connectivity quality, duration of link availability, the direction of node movement and packet delivery status. Nodes perform collaborative learning and rank the neighbors for each traffic type in order to optimize the network utilization and throughput. Simulations and testbed experiments are conducted to demonstrate the effectiveness of the algorithm and the result shows improvement in throughput of the network.