Machine Learning Accelerated Passive Electronically Scanned Phased Array Design

This paper presents a passive approach to electronically scanned millimetre-wave phased arrays. To avoid the need for additional amplification within the phased array, low-loss phase shifters that exhibit minimal amplitude variation are required. To achieve this, specific low-resolution switched-line phase shifters have been designed to meet the phase requirements of individual elements, avoiding supporting unused phase states associated with high-resolution phase shifters. Furthermore, the phase shifters are designed for thin-film ceramic circuits using low-loss phase change material switches to select different delay lines. However, as the unselected delay lines are left disconnected, these naturally resonate at specific frequencies causing additional narrow-band attenuation. To mitigate this loss, these resonances can be adjusted by changing the delay line geometry such that they occur outside the passband. A machine learning model has been trained to predict the frequencies at which these resonances occur to accelerate the design process for different phase requirements.