Hyperparameter Free MEEF Based Adaptive Estimator for MIMO Radar

Abstract

In the context of parameter estimation for nextgeneration nonlinear radar system models impaired by nonGaussian clutter, reproducing kernel Hilbert space (RKHS)-based signal processing algorithms and information-theoretic learning (ITL) have emerged as promising. Notably, the RKHS and ITLbased approaches are found to outperform conventional maximum likelihood (ML) based methods. However, these RKHS/ITLbased approaches, together with neural network/Bayesian-based approaches, are known to depend on hyperparameters for parameter/criterion estimation. In the context of next-generation radar, this paper introduces a minimum error entropy with fiducial points (MEEF)-based parameter estimation for multi-input-multiple-output (MIMO) radar to estimate target position and velocity amid non-Gaussian additive noise processes, such as clutter. Furthermore, we utilize a kernel width sampling method to make the proposed MEEF estimator hyperparameter-free. The proposed hyperparameter-free MEEF-based RKHS estimator with kernel width sampling (MEEF-KWS) is found to outperform minimum mean squared error (MMSE) and other ITL-based adaptive estimation techniques with fixed best kernel width. Computer simulations are presented assuming practical MIMO radar scenarios, which indicate that the proposed hyperparameter-free MEEF-KWS delivers improved estimation accuracy and/or lower computations compared to the existing RKHS-based MMSE and other common ITL criteria with fixed kernel width. © 2025 IEEE.