An Unsupervised Learning Framework for Determining the Excitation Coefficients Using Near-Field Antenna Measurements

This article presents a novel unsupervised learning framework based on multiscale dynamic mode decomposition for determining the excitation coefficients of antennas using time-domain near-field measurements. The proposed framework integrates temporal multiscale analysis to extract a joint distribution of frequency, damping factors, and spatial modes, enabling precise extraction of excitation frequencies, rising/falling edges, and phases without labeled data. We validate the effectiveness of the proposed approach through two examples involving
on
–
off
keying modulation and a phase-shift dipole antenna. It is found that the proposed method performs well in handling nonstationary excitation signals and proves particularly advantageous for calibrating tunable antenna systems. Our work could advance the capabilities of antenna measurements and optimization.