Physics-Informed Deep Neural Network for Low Sidelobe Time-Modulated Antenna Array Synthesis With Harmonic Suppression
Tarek Sallam, Qun Wang, Ahmed M. Attiya
Abstract
This paper introduces an innovative approach utilizing a deep neural network (DNN) to optimize the modulation scheme for time-modulated antenna array to verify specific side lobe and maximum harmonics levels. The proposed method involves training a DNN with a physics-informed loss function designed to reduce the discrepancy between the desired and actual beam patterns. This is accomplished by exclusively adjusting the periodic switching time sequence of each element within the TMLA. Specifically, the physics-informed deep neural network (PIDNN) is trained to optimize the switching-on times of for each antenna element. Simulation results demonstrate that the proposed technique achieves the desired beam patterns with significantly lower side lobe level and maximum harmonic levels compared to previously published methods. Additionally, the approach is compared to genetic algorithm (GA) which corresponds to a representative evolutionary optimization algorithm. Numerical results indicate that the PIDNN surpasses the GA in both computational efficiency and loss function evaluation.