Circuit-Informed Neural Network for Broadening the Bandwidth of SIW-Fed Slot Antennas

Abstract

A circuit-informed neural network (CINN) is proposed for broadening the bandwidth of substrate-integrated waveguide (SIW)-fed slot antennas. The proposed approach optimizes the structural parameters for matching multiple stub pairs (SPs) efficiently by combining circuit knowledge and a well-trained artificial neural network (ANN) for single SP. The CINN significantly reduced the computational costs of optimization, dataset construction, and training. Experimental results illustrated the effectiveness of the proposed CINN in achieving a wide impedance fractional bandwidth of 43%. This approach features strong generalization capabilities, making it widely applicable to various SIW antennas with diverse structures and varying numbers of SPs.