Near-Field Localization With Physics-Compliant Electromagnetic Model: Algorithms and Model Mismatch Analysis

Accurate signal localization is critical for Internet of Things applications, but precise propagation models are often unavailable due to uncontrollable factors. Simplified models, such as planar and spherical wavefront approximations, are widely used but can cause model mismatches that reduce accuracy. To address this, we propose an expected likelihood framework for model mismatch analysis and online (on-the-fly) model selection. The expected likelihood idea is that under the Gaussian assumption of independent samples, the likelihood ratio of the actual covariance matrix of the received signal is described by a distribution that depends only on the number of samples and receive antennas, not the true covariance itself. This scenario independence allows us to avoid the true model need-to-know requirement for model mismatch analysis. In this paper, we formulate an expected likelihood approach for online model and estimation parameters selection, and demonstrate its applicability and efficiency using the analytical electromagnetic model for data generation and considering different simplified models for source localization in both direct and reconfigurable intelligent surface assisted diverse IoT environments.