.png){kind=icon}
Return
.png){kind=icon}
.png){kind=icon}
.png){kind=icon}
.png){kind=icon}
.png){kind=icon}
Abstract
En 中文
A vector sensor array (VSA) can efficiently estimate the directions of arrival (DOAs) of radio frequency (RF) sources. Conventionally, constraints on the computational complexity of VSA processing have motivated the assumption of free space propagation conditions, even in multipath scenarios. The discrepancy between the considered and actual propagation conditions limits the VSA DOA estimation performance and, as a result, restricts its adoption to a broad spectrum of RF source localization applications. This work introduces a computationally efficient model-based neural network (MBNN) approach for VSA 2D DOA estimation that accounts for multipath propagation. The proposed framework employs a conventional fully connected neural network within a model-assisted VSA estimation pipeline. Physical propagation knowledge is introduced outside the neural network architecture through the analytical multipath steering model, covariance-subspace preprocessing, electromagnetic-field output representation, deterministic Poynting-vector DOA recon struction, and a Poynting-vector-based DOA-domain loss. It is demonstrated through simulations that the auxiliary physical information enables the proposed MBNN-based VSA DOA estimation to achieve the Cramér-Rao lower bound at a low signal to-noise ratio, with significantly fewer estimation parameters and a shorter processing time compared to the maximum likelihood estimator (MLE). It is also demonstrated that the proposed model-assisted multi-layer perceptron (MLP) estimator outper forms a conventional unconstrained regression MLP and the MLE in terms of mean-squared error and empirical cumulative distribution function criteria. The proposed MBNN-based VSA DOA estimation approach is expected to facilitate the adoption of RF VSA across a broad spectrum of applications for resource limited, compact platforms in scenarios where the dominant propagation mechanism can be approximated by structured multipath models.
Keywords:
Vector sensor array
DOA estimation
Model based neural networks
Multipath model
Journal
IF:
5.7
Papers:
779
Citations:
2.4W
Organization
Cited Papers
No cited papers available