Far-Field Phaseless Diagnosis for Impaired Arrays Based on Artificial Neural Networks and Compressed Sensing

In this article, a novel artificial neural network (ANN)-based method for diagnosing failures in impaired antenna arrays is introduced. Distinct from existing unidirectional ANN-based diagnosis approaches, an architecture featuring a decoder–encoder ANN framework is constructed. The decoder serves as a power pattern calculator, while the encoder serves as a diagnostic tool. The proposed approach significantly enhances the efficiency of the diagnostic procedure compared to the conventional ANN-based methods through a reduction in measurements within the far-field pattern under the compressed sensing (CS) framework. To enhance the sparsity of the recovered excitation, a smoothed
$l_{0}$
-norm (SL0) approximation method is introduced for the first time in the context of array diagnosis. This technique is seamlessly incorporated into the ANN framework as a means of regularization for its loss function. The establishment of this tailored ANN framework, coupled with a reduction in the required measurements, leads to improvements in both diagnostic accuracy and efficiency. Numerical examples and comparisons with prevailing phaseless diagnosis methods provide compelling evidence of the effectiveness and precision of the proposed technique in diagnosing both ideal and real antenna arrays.