Robust Drone Tracking System Using Distributed FMCW Radars and State-Aware Adaptive IMM-EKF

Abstract

Small multi-rotor drones have become a significant security concern, necessitating robust and continuous tracking systems. Conventional tracking filters, primarily designed for fixed-wing aircraft, rely on standard kinematic models that are inadequate for capturing the highly non-linear and evasive maneuvers of quadcopters. Furthermore, implementing distributed radar networks with commercial off-the-shelf sensors introduces practical challenges, including severe path-loss-induced signal-to-noise ratio (SNR) variations and asymmetric communication latency among distributed nodes. To address these limitations, this paper proposes a delay-compensated adaptive extension of the interacting multiple model extended Kalman filter (IMM-EKF) framework. First, we introduce a drone-specific 6D motion model set—Hovering (HV), Constant velocity (CV), and a Singer-based high maneuver (HM) model—to precisely capture time-correlated agile dynamics. Second, we propose an adaptive measurement fusion strategy that dynamically scales the noise covariance based on the predicted target-to-radar distance, intrinsically prioritizing high-SNR observations against geometric dilution of precision (GDOP). Finally, to neutralize asymmetric communication delays, we propose a two-phase delay compensation architecture: a baseline independent forward prediction (Scheme I) to prevent spurious state contamination, and an advanced likelihood-steered forward prediction (Scheme II) that actively exploits real-time observations from the primary radar to seamlessly steer the mode probabilities. Extensive Monte Carlo simulations and real-world experiments demonstrate that the proposed framework significantly reduces position errors, enables rapid maneuver identification, and maintains robust track continuity even under severe network latencies and unmodeled real-world dynamics.