Prior-Knowledge Guided Deep Learning Algorithm for Robust Array Diagnosis Under Array Errors

Yinghao Zhang

doi:10.1109/tap.2026.3735036

Prior-Knowledge Guided Deep Learning Algorithm for Robust Array Diagnosis Under Array Errors

2026-09-23

0

PRE

AI

DOI：10.1109/tap.2026.3735036

Original

Original request for help

Save

Abstract

En 中文

In this communication, a prior-knowledge guided deep learning algorithm is proposed for robust array diagnosis under array errors. By establishing a cascaded convolutional neural network (CCNN), the faulty elements in the array can be detected in real-time with high diagnostic accuracy from far-field radiation patterns, even in noisy environment. Specifically, the first CNN is served as the denoiser to eliminate noise in the measured far-field radiation patterns, while the second CNN is served as the regressor to detect the faulty elements based on the denoised patterns. Prior knowledge, including the frequency offset and array position errors in the antenna array, is purposely applied at the dataset generation stage to guide the training of neural network. It enables the CCNN to realize high diagnostic accuracy in the presence of array errors, thereby enhancing the robustness. Numerical simulations are carried out on 4 × 5 and 10 × 10 planar arrays to demonstrate the effectiveness of prior knowledge and the improved robustness. The performance of the CCNN is compared with the genetic, compressed sensing, least square and CNN algorithms to validate the superior performance, including the high diagnostic accuracy under array errors and the low computational complexity.

Keywords:

Array error

cascaded convolutional neural network (CCNN)

deep learning

robust array diagnosis

Prior-Knowledge Guided Deep Learning Algorithm for Robust Array Diagnosis Under Array Errors

Abstract

Journal

Organization

Cited Papers