Antenna radiation pattern predictions with machine learning

Research output: Chapter in Book/Report/Conference proceedingConference contributionScientificpeer-review

Abstract

A machine-learning based method to characterize integrated antennas is presented. The technique allows fast characterization with significantly reduced complexity compared to the previous antenna tests with separate scanned probe and receiver. The broadband reflection from a quasirandom target conveys the antenna characteristics in the reflection coefficient or S11-parameter measurement. A neural network is trained to retrieve the beam characteristics from the measured reflection coefficient S11. The antenna measurement setup is simulated as a reflection measurement with the antenna under test (AUT) facing quasirandom reflective mask. The reflection coefficient is calculated as the coupling coefficient between the AUT radiated field and the back-reflected field at 75-110 GHz, and it is fed to a fully-connected neural network and trained to the beam-steering angles and beamwidths. The predicted median beam direction error is 4.1° and beamwidth error is 2.2°. The technique is promising, as it allows for antenna characterization without scanned or rotated antennas, yet providing sufficient accuracy for antennas with moderate directivity.

Original languageEnglish
Title of host publication2021 IEEE Conference on Antenna Measurements and Applications, CAMA 2021
PublisherIEEE
Pages434-437
Number of pages4
ISBN (Electronic)9781728196978
DOIs
Publication statusPublished - 2021
MoE publication typeA4 Article in a conference publication
EventIEEE International Conference on Antenna Measurements and Applications - Antibes Juan-les-Pins, France
Duration: 15 Nov 202117 Nov 2021

Conference

ConferenceIEEE International Conference on Antenna Measurements and Applications
Abbreviated titleCAMA
Country/TerritoryFrance
CityAntibes Juan-les-Pins
Period15/11/202117/11/2021

Fingerprint

Dive into the research topics of 'Antenna radiation pattern predictions with machine learning'. Together they form a unique fingerprint.

Cite this