Generalized cablibration of model errors for uniform linear arrays

Abstract

A generalizedcalibration technique for model errors is proposed for uniform linear arrays. Being different from the existing calibration methods, this technique constructs covariance matrices for two adjacent sampling periods, from a reference source with a known location, and then estimates the model errors from the difference between these matrices and an equation formed by using the MUSIC null space. The proposed technique can not only calibrate different types of model errors (such as gain-phase errors, geometry errors and mutual coupling) in different impinging source scenarios, but also has great robustness to different types of background noise. Meanwhile, it is computationally efficient and does not require an accurate clock synchronization between the reference source and the array receiving system. However, it requires that the statistical properties of both signal and noise must be constant in the two adjacent sampling periods. Numerical examples validate the superiority and effectiveness of the proposed technique.

Vehicle Positioning With Deep-Learning-Based Direction-of-Arrival Estimation of Incoherently Distributed Sources

“…It has been demonstrated that such array perturbations can degrade the estimator's performance substantially [32]. Although there have been some array calibration methods available, such as the eigenstructure-based method [33], the active calibration method [34], and the covariance difference based method [35], most of them are based on the point source model, uniform linear array and subspace theory, and it is very difficult to extend them to deal with the scenario considered in this work. Fortunately, based on the characteristics of DL, this issue can be tackled by enhancing the constructed D1D-CNN in the following two aspects: i) Transfer learning (TL) is used to promote the generalization ability of already trained D1D-CNN with a small amount of perturbed array data, which aims to align the features across array perturbations and reduce the distribution divergence.…”

“…This strategy can yield an improved positioning accuracy and robustness on vehicle positioning. • When there are array perturbations, one can exploit some approach to estimate it first with the estimated 2-D DOAs, such as the approach in [35], and then obtain the received SNR after compensating for such array perturbations. • Estimation of the received SNR ( 17) is based on the approximated model, which only holds when the angular spreads are sufficiently small.…”