Fully Decentralized Sequential Detection in an Underwater Setting

In this technical report we present a model for a detector network of acoustic sensor nodes to be used in the military underwater setting. The model accounts for well-known properties of acoustic sensing and communication equipment, and to this end we employ the traditional decentralized detection framework combined with ideas from the field of team decision theory, thus arriving at a network model which operates without a centralized fusion centre. The model instead gives the decision to sound the alarm to individual detector nodes, which classify the presence of a target using the classic Bayes´detector framework, applied for timedependent sequential hypothesis testing. The detector network employs an internal message-passing scheme, thus reducing outside communication to a minimum. Reduction of internal network communication is achieved through a sensor censoring scheme where uninformative messages are withheld, and the detector nodes adjust for these missing data. Indicative simulation results are presented.