An Adaptive Extended Kalman Filter Incorporating Noise Adaptation and Maneuver Awareness

This paper proposes a novel adaptive filter, termed the Range/Bearing-Rescaled, Anisotropic-Q EKF (R2-AEKF), which enhances the conventional constant-acceleration-based Extended Kalman Filter (EKF-CA). The proposed method integrates channel-wise measurement noise rescaling based on innovation statistics with maneuver-aware anisotropic process-noise shaping aligned with the velocity direction of the target. This design allows the filter to retain the computational simplicity and efficiency of EKF-CA under nominal conditions, while significantly improving robustness in scenarios involving abrupt maneuvers or time-varying noise statistics. Simulation studies are conducted using two representative scenarios: a zigzag trajectory with axial bursts and a complex maneuver including S-curves, slow-and-go decelerations, and jerks. The results demonstrate that the R2-AEKF consistently reduces position errors and tracking delays, prevents track loss, and shortens recovery time compared with the EKF-CA. Furthermore, the evolution of the adaptation scale parameters confirms that the internal logic mechanism operates as intended, thus substantiating the observed performance gains. These findings verify that the R2-AEKF provides reliable and stable tracking across diverse maneuvering environments.