Towards Bi-Directional Deep Learning Approach in Patch Antenna Design and Optimization

This paper presents a novel approach for the design and optimization of high-gain bidirectional antennas using deep learning, aimed at reducing the computational complexity and energy consumption associated with traditional iterative simulation methods. By leveraging a neural network model trained on six key antenna parameters, our approach enables bidirectional functionality. It predicts the S11 parameter and gain from given antenna geometrical parameters, and inversely, determines the optimal antenna geometrical parameters based on desired S11 and gain values within the frequency band of 3.25 to 4.25 GHz. This dual capability not only accelerates the design process but also minimizes the need for extensive simulations, thereby conserving computational resources. The results significantly reducing the energy and time required for antenna design and optimization. The antenna is designed on a Rogers 5880 substrate with a thickness of 62 mils, and both simulation and measured results validate the high performance and efficiency of the proposed bidirectional antenna