Hyperparameter Free RKHS-Based Non-Linear Parameter Estimation for Radar Sensors
Uday Kumar Singh, Rangeet Mitra, Thipparaju Rama Rao, K. Venkateswaran, Amit Kumar Mishra, Michał Lupa · IEEE Access · 2026
Estimation of the target’s location and velocity using radar sensors is a long-standing problem, and improving the accuracy of these estimates is still an open challenge. In fact, the viability of the radar systems is heavily based on the fact that how reliably it is in estimating the target related parameters. Traditional methods typically depend on the maximum likelihood (ML) estimation of frequencies, which correspond to the target’s range, velocity, and angle. However, frequency estimation is inherently a non-linear problem, and ML-based approach fails to provide a close form solution and hence often yields suboptimal results. To enhance the accuracy of the estimates using radar sensors while addressing the nonlinear nature of the radar measurements, various adaptive algorithms based on Reproducing Kernel Hilbert Space (RKHS) have emerged recently. However, the performance of RKHS-based adaptive algorithms is highly dependent on the selection of an appropriate kernel width. In this work, we introduce a kernel-width assignment based on stochastic sampling for the extensively used Gaussian kernel in the context of radar parameter estimation. The proposed approach is found to deliver improved performance in terms of mean-squared error convergence and computational complexity, while overpowering the results corresponding to manually tuned kernel width. The effectiveness and generalization of the proposed approach is validated through analytical results and computer simulations considering two types of practical radar system models.