Computationally-Efficient and Reliable Antenna Design within the System-by-Design ML-Based Framework

The increasing complexity of modern electromagnetic (EM) devices requires advanced solution strategies to reduce the computational burden of the synthesis process. In this work, the machine learning (ML)-based System-by-Design (SbD) framework is exploited to address the design of antennas in a reliable and computationally-efficient fashion. The SbD scheme is based on a "divide-et-impera" paradigm which aims at decomposing the problem at hand into smaller sub-tasks that are addressed by specific interconnected functional blocks. Hence, the overall computational cost can be reduced by choosing and integrating the proper functional blocks, including global optimization techniques and learning-by-examples (LBE) approaches. The effectiveness of the SbD paradigm is shown with an illustrative example concerning the design of a mm-wave slotted substrate integrated wave-guide (SSIW) antenna for 77 GHz automotive radar systems.