Distributed Detection With Generalized Locally Most Powerful Fusion of Compressed Local Multi-Frame Test Statistics

Abstract

Distributed radar multi-frame detection (DR-MFD) can detect dim targets through a longer fusion period of multi-radar observations. However, communication cost in the DR-MFD is usually huge, causing a heavy hardware burden in the real application. This paper investigates the dim target detection problem with the DR-MFD system and aims to realize the DR-MFD with a good detection performance and a low communication cost. To compress local data, local test statistics formed at each frame are quantized by the Fisher Information maximization-based quantization (FIMQ), and the quantized outputs are transmitted to a fusion center (FC). A generalized locally most powerful (GLMP) detector is derived to fuse the received multi-frame quantized outputs at the FC. It is proved that the FIMQ is asymptotically optimal for the GLMP detector in the sense of detection performance. A closed-form solution is provided to find the local quantization thresholds of the FIMQ. Simulation results indicate that the GLMP detector outperforms the generalized likelihood ratio detector in most situations. The asymptotic optimality of the FIMQ is also corroborated. In addition, for the GLMP detector with 3-bit FIMQ, the signal-to-noise ratio loss caused by data quantization can reach a 0.11 dB level.