A Wideband Pseudo-CPW-Based Grid Antenna: Novel Design and Optimization via Space-Mapped Cascaded DNN and PSO Algorithm

In this paper, we propose a novel wideband grid array antenna (GAA) based on a pseudo-coplanar waveguide (PCPW) configuration. The short side of grid, slots, and shorted parasitic patches are effectively excited by differential feeding, generating three closely-spaced resonant modes. Compared to conventional GAAs, such design achieves an approximate 30.5% improvement in impedance bandwidth (IBW) without compromising gain performance. Subsequently, to facilitate efficient scaling of the proposed antenna to larger arrays, we develop a cascaded deep neural network (CDNN)-based surrogate model using space-mapping technique, combined with particle swarm optimization (PSO) algorithm, for rapid optimization of PCPW-GAAs with various array sizes. Notably, this forward surrogate model requires only 2010 data samples, with a 1:1 data ratio between the smallest and other array sizes, effectively reducing the initial data acquisition cost compared to directly training a single network model. Experimental validation confirms the performance advantages of the proposed wideband GAA, achieving a measured IBW of 34.7%, a 3-dB gain bandwidth of 31.2%, and a peak gain (PG) of 10.9 dBi for the smallest array size. To validate the effectiveness of the surrogate-model-assisted optimization method, we apply it to the design of a 17-element PCPW-GAA as a case study. The final optimized structural parameters are also evaluated through both full-wave simulation and experiment. Given these performance benefits, we believe that the proposed wideband grid structure and efficient optimization method present a compelling solution for millimeter-wave antenna design and applications.