On Passive Emitter Tracking in Sensor Networks
Regina Kaune, Darko Mušicki, Wolfgang Koch · Sciyo eBooks · 2010
Passive emitter tracking in sensor networks is in general superior to emitter tracking using single sensors. Even a pair of sensors improves the performance strongly. The techniques of solving the underlying tracking problem are the same as in the single sensor case. The first step should be the investigation of the CRLB to know the optimal achievable estimation accuracy using the available measurement set. It reveals characteristic features of localization and gives an insight into the parametric dependencies of the passive emitter tracking problem under consideration. It shows that the estimation accuracy is often strongly dependent on the geometry. Secondly, a powerful estimation algorithm is needed to solve the localization problem. In passive emitter tracking, states and measurements are not linearly related. Therefore, only methods that appropriately deal with nonlinearities can be used. This chapter provides a review of different nonlinear estimation methods. Depending on the type of measurement and on different requirements in various scenarios, different estimation algorithms can be the methods of choice. E.g., to obtain good results near the CRLB the MLE is an appropriate method. Here, the computational effort is higher compared to alternatives such as Kalman filters. Tracking from the first measurement is possible using the UT in the TDOA/AOA case or using the GM filter or the GMM-ITS filter. They overcome the initialization difficulties of single Kalman Filters. The UT transform the measurement into the Cartesian state space and the GM filter and GMM-ITS filter approximate the first measurement by a Gaussian Mixture, a weighted sum of Gaussian densities. The first measurement is transformed into the Cartesian space and converted into a Gaussian sum. The tracking with the GM filter and GMM-ITS filter shows good performance and results near the CRLB. For passive emitter tracking in sensor networks different measurement types can be gained by exploiting the signal coming from the target. Some of them can be taken by single sensors: e. g. bearing measurements. Others are only gained in the sensor network, a minimum of two sensors is needed. The combination of different measurements leads to a significant gain in estimation accuracy.