Unsupervised SVD-Based Correction of Antenna Measurements in Nonanechoic Environments

Abstract

Experimental validation of antenna prototypes is normally performed in expensive laboratories. A cost-efficient alternative involves measurements in nonanechoic conditions followed by post-processing of the obtained far-field responses. In this work, a framework for unsupervised, singular value decomposition-based correction of antenna measurements performed in uncontrolled propagation environments is proposed. The method involves data de-noising based on reconstruction of far-field responses from signals that correspond to high-energy singular values. Automatic adaptation of the algorithm-specific settings to the test site is performed using unsupervised machine learning oriented towards maximization of singular values sparsity. The method is validated using over a dozen measurements of five antennas. The improvement of responses fidelity due to post-processing is up to 14 dB. Benchmark against the state-of-the-art methods is also performed.