Design and Development of CNN-based Dimension Optimized Dual-Band Antenna

This work presents a dual-band operating antenna whose design dimensions are optimized using Convolutional Neural Networks (CNN). The proposed design features a circular patch antenna (CPA) with C-shaped split-ring resonators (C-SRR) incorporated in the ground plane. The CNN is trained on a dataset comprising various antenna designs and their corresponding performance metrics, enabling it to predict optimal design parameters for achieving the desired resonance frequency, return loss, and gain. The proposed dual-band CPA is designed after three stages. Three different versions of the antenna are studied in detail, and the analysis of their results is detailed. The initial CPA with C-SRR in the ground plane shows a single resonance between 4.83 GHz and 5.4 GHz with a bandwidth of 570 MHz. Further stages aim at enhancing signal strength and minimizing loss, and the last design incorporates a split-ring resonator with slots on either side, resulting in the formation of the CPA – Lower Upper SRR (CPA-LU-SRR) structure operating from 4.82 GHz to 5.15 GHz and 9.65 GHz to 9.893 GHz, attaining return losses of -18.06 dB and -17.67 dB, and bandwidths of 330 MHz and 243 MHz, respectively. The dimensions of the CPA-LU-SRR are optimized using the CNN model. The CNN is trained on a dataset of antenna designs and their respective performance values, allowing the prediction of the best dimensions for the desired resonant frequency, return loss, and gain.