Detection of Faulty Elements in IC-Controlled Phased Arrays Using Sparse Far-Field Data: A Machine-Learning Approach

Abstract

Active phased array antennas, generally consisting of a large number of antenna elements and integrated-circuit (IC) components, are essential for different applications like 5G/6G communications, radar, aerospace, satellite communications and so on. The study of radiation pattern measurements and analysis of pattern data is a pivotal step in characterizing and understanding the use of antennas in any application. Unwanted changes can be observed in the radiation pattern due to randomly failed elements, ICs, or IC channels in phased arrays. Hence, there is a vital need for fast testing and diagnosis to identify possible failures of these large antenna systems, before taking any immediate action to compensate for these changes. Due to integrated-circuits (IC) non-linearities and mutual coupling challenges, the existing deterministic and optimization-based fault detection techniques fail to provide reliable results in practical systems. Besides, it is desired to achieve real-time array diagnosis during in-field operation, which makes the problem even more challenging. In this thesis, a novel machine-learning assisted fault-detection algorithm is proposed which is trained by using sparse far-field measurements from a practical IC-integrated millimeter-wave (mm-wave) phased array prototype and fixed multi-probe measurement system (antenna dome). A global optimization approach is used as a benchmark for comparison to the state-of-the-art. The incorporation of machine learning techniques greatly improves the fault detection time and accuracy in comparison to global optimizers. Furthermore, studies based on the number of sampling points used and the varying number of faults are also conducted which indicate similar conclusions.