Skywave OTHR Tracking With Measurement Residual Learning Based Unscented Kalman Filter

Skywave over-the-horizon radar (OTHR) enables long-range sensing via ionospheric reflection, yet propagation variability and sensor miscalibration often induce geometrydependent measurement mismatch, making the effective measurement mapping only partially known. We propose a referenceaided measurement residual learning approach that captures the unknown component as a state-dependent residual added to a nominal physics-based mapping, and learns this residual without assuming an explicit low-dimensional bias parameterization. The learned correction is incorporated into a residual-corrected measurement mapping and applied in the unscented Kalman filter (UKF) update, providing mismatch compensation while keeping the state dimension unchanged. Monte Carlo simulations with biased skywave OTHR measurements and Automatic Dependent Surveillance-Broadcast (ADS-B) reference uncertainty show reduced measurement residuals and improved tracking accuracy relative to the Nominal-UKF and Joint-UKF baselines.