Design and Optimization of an Ultra-Wideband Millimeter-Wave Circularly Polarized Metalens Antenna With Deep Learning Method

Abstract

To meet the growing demands of higher data rates and larger bandwidth in the future wireless communications, several methods including the optimization of reference phase, multiple frequency matching, and dispersion engineering method have been investigated to design millimeter wave (mm-wave) antennas. However, the optimization process in these methods necessitate complex calculation and experience. In this study, we proposed a novel approach that optimizes reference phases and transmission magnitudes at multiple frequencies assisting with deep learning method to design an ultra-wideband mmwave circularly polarization antenna. The first neural network intelligently provides the required geometric parameters of the unit cells, while the second neural network validates the phase shifts and predicts the transmission magnitudes based on inputs from the first neural network. The particle swarm optimization (PSO) algorithm is also utilized to optimizes the reference phases at multiple frequencies to minimize the matching errors. To validate the design, a 3-D printed compact CP antenna is fabricated and the measured results show that this antenna achieves a 1-dB gain bandwidth of 31.4%, and the 3-dB axis ratio (AR) covers the full W-band from 75 GHz to 110 GHz. The results show that our proposed method is more intelligent and faster than alternative approaches.