Event Localization in Underwater Wireless Sensor Networks using Monitoring Courses

Abstract

In this thesis we consider different methods to localize events in a multi-hop wireless sensor network operating underwater using acoustic modems. The network consists of surface gateway nodes and relay nodes. Localization of surface gateways can be achieved through GPS, but we cannot rely on this technology for localizing underwater nodes. Surface Gateway nodes can distribute their locations through the network using the incoming signals by the acoustic modems from the relay nodes. Relay nodes are deployed to remain static but due to water currents, floating, and the untethered nature of the nodes, they often suffer from frequent drifting which can result in a deployed network suffering link failures. In this work, we developed a novel concept of an underwater alarming system, which adapts a cyclic graph model. In the event of link failure, a series of alarm packets are broadcasted in the network. These alarms are then captured through a novel concept of underwater Monitoring Courses (M-Courses), which can also be used to assure network connectivity and identify node faults. M-Courses also allow the network to localize events and identify network issues at a local level before forwarding any results upwards to a Surface Gateway nodes. This reduces the amount of communication overhead needed and allowing for distributed management of nodes in a network which may be constantly moving. We show that the proposed algorithms can reduce the number of send operations needed for an event to be localized in a network. We have found that M-Course routing reduces the number of sends required to report an event to a Surface Gateway by up to 80% in some cases when compared to a naive routing implementation. But this is achieved by increasing the time for an event to reach a Surface Gateway. These effects are both due to the buffering effect of M-Course routing, which allows us to efficiently deal with multiple events in an local area and we find that the performance of M-Course routing is not affected by these types of events while the naive case can experience problems. We also find that generating an M-Course solution even over large networks (up to 800 nodes) is relatively inexpensive and can typically be achieved in under a second. However we also find that the viability of M-Course solutions is heavily dependant on the size of the network used, where smaller networks often lead to more efficient solutions. So we also propose a partitioning of the network based on Surface Gateway locations to reduce computational complexity and improve the viability of M-Course solutions.