From Decision Fusion to Localization in Radar Sensor Networks: A Game Theoretical View

Abstract

This paper considers the localization problem in a radar sensor network (RSN), where the estimation is made based on fusing the received signals from multiple radar sensors. For practical radar receivers, the moving target indication (MTI) technique is often adopted to suppress the clutter in the relatively small Doppler frequency shift regime, although it may filter out the desired target signal as well. As a result, when multiple radar sensors are deployed and the target is moving along one direction, it is likely that only a subset of the radar receivers can observe the target, which we call an observation pattern. In this paper, we explore how to utilize the information of all possible observation patterns to derive the Cramer-Rao lower bound (CRLB) for the localization problem, which is shown to hinge heavily on radars and the prior statistic information of the observation patterns. Next, we generalize the localization problem to the case for an area $\mathcal{S}$, and investigate the localization games between the RSN and the intrude target. We propose a two-stage Stakcelberg game framework to model the interactions between the RSN and the target, for cases that the target can adopt mixed and pure strategies, respectively. Finally, numerical results demonstrate that the proposed scheme can significantly improve the localization performance.