Mixed far-field and near-field source separation and localization based on FOC matrix differencing

Abstract

This paper investigates the classification and localization of the mixed far-field sources (FFSs) and near-field sources (NFSs) on a symmetrical uniform linear array and proposes a new mixed source localization algorithm. Firstly, we analyze the persymmetric structure property of the far-field (FF) steering vector in matrix differencing. Based on that, we implement the matrix auto-differencing operation on the fourth-order cumulant (FOC) matrix, which eliminates the contribution of the FF signals from the mixed signals. Based on the reserved near-field (NF) parts, an effective subspace-based is performed to estimate NF parameters. Secondly, to reconstruct the NF FOC matrix, we propose a new method to estimate the NF kurtosis from the non-Hermitian statistics matrix. Then, the related NF components can be removed from the mixed signals to avoid interfering with FF direction-of-arrival (DOA) estimations. Based on the subspace technique, the FF DOAs can be determined by the one-dimensional spectrum search. The proposed algorithm does not require additional classification operations, even when NFSs have the same DOAs as FFSs. Finally, several simulations demonstrate the better estimation performance of the proposed algorithm than existing localization algorithms.