Physics-Informed Deep Neural Network for Low Sidelobe Cosecant-Squared Pattern Time-Modulated Antenna Array with Harmonic Suppression

In this paper, a physics-informed deep neural network (PIDNN)-based approach with phase-only weighting is proposed for the synthesis of a cosecant-squared radiation pattern time-modulated linear array (TMLA). A deep neural network is trained using a physics-informed loss function to minimize deviation between the desired shaped beampattern and the actual one, thus eliminating the need for timeconsuming optimization techniques, such as evolutionary optimization algorithms. The sidelobe level (SLL) of the fundamental signal and the maximum sideband levels (SBLs) of the harmonic signals are simultaneously reduced by controlling both the phase and periodic switching time sequence of each element of the TMLA. The simulation results reveal that the proposed method achieved the desired patterns along with very low SLL and SBLs for 30- and 40-element TMLAs. For the 30-element TMLA, an SLL of −39.54 dB and a maximum SBL of −42.27 dB was achieved. For the 40-element TMLA, the obtained SLL and maximum SBL were −29.28 dB and −32.02 dB, respectively. Moreover, PIDNN demands considerably less computation time than evolutionary optimization algorithms—for a 30-element TMLA, the genetic algorithm took 47.13 minutes to complete the optimization process while PIDNN took only 2.87 minutes.