A Two-step Signal Registration Algorithm for Distributed Radar with Censored Observations

In a distributed radar system, a signal fusion-based target detection algorithm can yield better detection performance compared to traditional decision fusion-based algorithm, thus has drawn considerable attention. In practice, however, signal fusion-based detection algorithm for distributed radar encounters two primary challenges: one is excessive communication cost and the other is difficult to fuse locally unregistered signals which is usually ignored by existing researches. To resolve these challenges, this paper presents a two-step signal registration algorithm with locally censored observations. Specifically, only local signals exceeding an initial threshold are sent to a fusion center for signal registration. At the first step of this algorithm, Mahalanobis Distance is employed as a metric to identify physically admissible observations, referred to as candidate observations. At the second step, the maximum likelihood estimation criterion is applied to eliminate potential false alarms among these candidate observations, thus determining the final set of registered observations for signal fusion. Numerical results indicate that the algorithm presented herein is capable of efficiently tackling the signal registration problem as well as alleviating communication burden. Furthermore, it achieves nearly lossless detection performance in comparison with its unregistered counterpart. This algorithm can be applied to practical engineering applications, since the simulation parameters are aligned with real radar scenarios.