Inverse design of a phase shift circuit using deep learning and particle swarm optimization for simultaneous amplitude and phase control

Abstract This work presents an inverse design method for a π -type phase-shift circuit using deep learning and particle swarm optimization (DL-PSO). The proposed framework enables simultaneous control of amplitude and phase in reconfigurable arrays, including reconfigurable intelligent surfaces. The DL-PSO method resolves the non-unique mapping problem and enables precise control of both amplitude and phase. An external PSO loop optimizes fixed structural parameters by evaluating amplitude performance over a full 360° phase range. A prototype operating at 2.8 GHz was fabricated and measured. It achieves full-range phase tuning with a maximum amplitude fluctuation of 0.6 dB and an average phase error of 1.92°. Accurate amplitude control is also achieved at multiple target levels across the 360° phase range, with average amplitude and phase errors of 0.16 dB and 1.76°, respectively. The proposed method is scalable for reconfigurable array applications such as beam shaping and null steering.