MULTISENSOR TRACK TERMINATION FOR TARGETSWITH FLUCTUATING SNR Wayne Blanding, Peter Willett, Yaakov Bar-Shalom University of Connecticut Department of Elec. and Comp. Engineering Storrs, Connecticut 06269 Stefano Coraluppi NATO Undersea Research Center 19138 La Spezia, Italy ABSTRACT In active sonar tracking applications, targets frequently un- dergo fading detection performance in which the target’s de- tection probability can shift suddenly between high and low values. Using a multistatic active sonar problem, we examine the performance of sequential track termination tests where target detections are based on an underlying Hidden Markov Model (HMM) with high and low detection states. We show that the Page test is not optimal in this problem and that a K/N track termination rule yields better performance. Further we show that a Bayesian sequential test (the Shiryaev test) yields dramatic performance improvements over both the K/N rule and the Page test. Index Terms— Target tracking, multistatic sonar, track termination, Page test, Shiryaev test 1. INTRODUCTION In undersea surveillance of large areas, multistatic sonar net- works show promise in the ability to use many sensors to cover a large area with overlapping detection coverage, the achievement of higher data rates from use of multiple sen- sors (receivers) to process a single active transmission from a source, and in the geometric diversity that can be achieved by selecting receiver locations. However, it has been observed from at-sea testing that sensor detection performance varies significantly over the sensor network and for a single sensor over time. In particular, due to geometric, environmental, and geographic effects a target may be detected by a given sen- sor with high probability over a number of scans and then suddenly fade from view as the sensor detection probability decreases drastically. A key tracking issue therefore becomes how to adapt the tracking system to account for this fading detection performance in a multistatic active sonar problem. In this paper we address the track termination aspect of this problem. We consider a centralized track management model that processes time ordered measurements from all sen- sors and include sensor origin information. The sensor detec- tion performance is modeled as a two-state Markov chain with ICASSP 2007, Honolulu, HI, April 2007. This work was sponsored by the Office of Naval Research. high and low detection states. Target-originated measure- ments (binary detection events) can therefore be described us- ing a Hidden Markov Model (HMM) structure. There is a well-developed base of literature covering track termination for sensors with a fixed probability of detection (P d ) of the target on a single scan (see e.g., [3]). Examples include K/N tests (a track is terminated if K or fewer detec- tions are received in the last N scans) and track score tests. Track score tests may include the SPRT or Bayesian sequen- tial tests. If the track score (related to the probability that the detection sequence is the result of a true track) falls below a certain value, the track is terminated. However research per- taining to track termination for sensors with P d based on a Markov model has only recently been considered [6]. In this paper we analyze the performance of K/N-based and sequential track termination tests when target-originated measurements are described by a HMM with high and low P d Markov states. Using only the binary detection events, it is shown that the K/N test outperforms the Page test over a portion of its operating characteristic region. This result is surprising considering the fact that the Page test is proven to be the optimal sequential test for quickest detection of a change in measurement distribution and we show how when the HMM-based detection statistics are used, a key assump- tion in the optimality proof for the Page test is no longer sat- isfied. It is next shown that by using a Bayesian version of a sequential test (the Shiryaev test), significant performance im- provement is obtained compared to the K/N test. This work presents significantly updated and improved results from that originally presented in [4]. 2. PROBLEM DESCRIPTION In the multistatic sonar problem one or more transmitters trans- mit active sonar pings. Receivers are positioned to receive and process acoustic energy in order to detect reflected energy from the target(s). Because each transmitter-receiver combi- nation is unique and provides a set of measurements, a sensor is identified as a specific transmitter-receiver combination. A scan of data is defined as the measurement data set from one sensor resulting from one sonar transmission. Although II  1097 1424407281/07/$20.00 ©2007 IEEE ICASSP 2007