Deep-Learning-Assisted Optimization of Pseudorandom Time-Modulated Arrays for Signal Transmission Under Nonlinear Distortion Constraints

A time-modulated antenna array (TMA) employing pseudorandom modulation schemes distributes the radiation power over a continuous spectrum. This feature facilitates robust suppression of sideband radiations; however, it inevitably introduces nonlinear distortions for signal transmission. In this paper, a novel deep-learning-assisted approach is proposed for the optimal design of pseudorandom TMAs (PS-TMAs) under signal nonlinear distortion constraints. A surrogate model is developed based on the deep neural network to predict the error vector magnitudes (EVMs) under various PS-TMA design parameters. By integrating the deep-learning surrogate model into the differential evolution optimization framework, the proposed approach realizes real-time calculation of design parameters that satisfy specified EVM requirements. Consequently, the proposed framework significantly reduces the optimization time from a typical 1.21 hr in traditional optimization methods to 0.12 s by eliminating the need for time-consuming Monte Carlo simulations during the iterative optimization process. Moreover, it is found that nonlinear distortion could be further mitigated by introducing an additional distortion correction module at the receiving end, which provides an effective solution for application scenarios with stringent signal quality requirements. Numerical and measured results of a Ku-band PS-TMA prototype for orthogonal frequency-division multiplexing (OFDM) transmission clearly demonstrate the effectiveness of the proposed approach.