Improved probability hypothesis density filter-based time difference of arrival passive localization and tracking method for UAV swarms

Abstract

This paper proposes a Peak-Width-Weighted Unscented Gaussian Mixture Probability Hypothesis Density (PWW-UGMPHD) algorithm to address the critical challenges of multi-target tracking divergence and cardinality estimation distortion in traditional PHD filtering. These issues typically arise from low Signal-to-Noise Ratios (SNRs) and multipath effects on Time Difference of Arrival (TDOA) measurements in complex electromagnetic environments. The proposed method extracts the Full Width at Half Maximum (FWHM) of the TDOA main lobe as a signal quality feature to dynamically construct an adaptive measurement noise covariance matrix. By seamlessly embedding this matrix into the likelihood update step of the Unscented Transform (UT), the algorithm imposes a "soft-limiting" penalty on inferior-quality measurements. Simulations demonstrate that, even under harsh conditions characterized by dense clutter and a high proportion of anomalous measurements, the PWW-UGM-PHD algorithm accurately maintains target cardinality estimation and significantly reduces the Optimal Subpattern Assignment (OSPA) error. The proposed approach substantially outperforms traditional methods in both tracking accuracy and track continuity, providing a highly robust, real-time data association baseline for practical anti-drone systems.