Deep Net Localization - eavesdropping in mobile acoustic underwater sensor networks

Abstract

Navigation is a challenging task for underwater devices, as there is no global navigation system available. This often restricted Autonomous Underwater Vehicles (AUV) in their operations, either spatially because they depend on reference markers that need to be deployed on the mission side, or by dive time as they need to surface regularly to geo-reference themselves via satellite. For most applications the position and attitude information of an AUV is crucial for a meaningful interpretation of the collected data. In this paper, we present Deep Net Localization (DNL), a novel localization approach for stationary and mobile nodes in an acoustic underwater sensor network. For that purpose a communication protocol was developed that enables network nodes to self-localize their position and attitude in the network by eavesdropping on the communication of other nodes. This is achieved by combining modern acoustic USBL modems and pressure sensors with a Bayes estimator. Without the need to actively transmit acoustic signals, DNL saves energy and reduces the overall workload of the communication channel. Furthermore this minimizes interference effects with other acoustic sensors, like multi-beam sonar, which are essential for the measurement purpose of a sensor node. In order to assess the approach we first tested the general functioning in a Monte-Carlo-Simulation. Subsequently the method was evaluated with real data that was collected during a sea trial in the Middle Atlantic Ocean.