A Compact Machine Learning Architecture for Wideband Amplitude-Only Direction Finding

• Published in: IEEE transactions on antennas and propagation Vol. 70; no. 7; pp. 5189 - 5198
• Main Authors: Friedrichs, Gaeron R., Elmansouri, Mohamed A., Filipovic, Dejan S.
• Format: Journal Article
• Language: English
• Published: New York IEEE 01.07.2022; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Amplitudes; Angle of arrival; Antenna arrays; Antennas; Artificial neural networks; Direction finding; direction finding (DF); Estimation; Gain; Machine learning; machine learning (ML); Multiple signal classification; Noise levels; Retrofitting; Signal to noise ratio; support vector regression (SVR); Ultrawideband; ultrawideband antennas
• ISSN: 0018-926X; 1558-2221
• DOI: 10.1109/TAP.2021.3137505

A generalized, reduced-size machine learning architecture for single-snapshot amplitude-only direction finding (AODF) is proposed for uniform circular arrays. A method for reusing angle of arrival (AoA) estimation models that are accurate over narrower fields of view is described. The efficacy of the proposed method is demonstrated using an ultrawideband circular array of miniaturized transverse electromagnetic (TEM) horns covering 1.5-5.5 GHz. Reasonable azimuth estimations performed on this retrofitted system are obtained over 2.6:1 bandwidth (1.5-4.0 GHz). Antenna performance features that impact the accuracy of AODF are also recognized. Root mean square error less than 5° is achieved in simulation above 15 dB signal-to-noise ratio (SNR) and above 20 dB SNR in measurement. The improved accuracy over the conventional correlation method of 52%-85% is demonstrated in an SNR domain of 10-40 dB. This performance improvement is obtained while maintaining a footprint reduction of 80%-95%, and an AoA estimation time speed-up of at least 85%.