Machine Learning-Based Near-Field Localization in Mixed LoS/NLoS Scenarios

TL;DR

This paper introduces a CNN-based method for 3D near-field source localization in mixed LoS/NLoS scenarios, offering improved efficiency over traditional algorithms.

Contribution

It proposes a novel machine learning approach that maps covariance matrix eigenvectors to source locations, reducing computational complexity.

Findings

The CNN accurately estimates 3D source positions.

The approach significantly reduces localization time.

Numerical simulations validate effectiveness and efficiency.

Abstract

The conventional MUltiple SIgnal Classification (MUSIC) algorithm is effective for angle-of-arrival estimation in the far-field and can be extended for full source localization in the near-field. However, it suffers from high computational complexity, which becomes especially prohibitive in near-field scenarios due to the need for exhaustive 3D grid searches. This paper presents a machine learning-based approach for 3D localization of near-field sources in mixed line-of-sight (LoS)/non-LoS scenarios. A convolutional neural network (CNN) learns the mapping between the eigenvectors of the received signal's covariance matrix at the anchor node and the sources' 3D locations. The detailed description of the proposed CNN model is provided. The effectiveness and time efficiency of the proposed CNN-based localization approach is corroborated via numerical simulations.