ANN-Assisted Quasi-Time-Invariant Beamforming for Retrodirective Frequency Diverse Array
Yong Qiang Hei,Xiao Yan Ju,Long Yuan Ma,Wen Tao Li,Xiao Wei Shi
DOI: https://doi.org/10.1109/tap.2024.3375968
IF: 5.7
2024-01-01
IEEE Transactions on Antennas and Propagation
Abstract:Due to the range-angle-dependent beampattern, frequency diverse array (FDA) has attracted extensive attention. However, the inherent time-varying and range-angle coupling problems of FDA are still troublesome. In this paper, a quasi-time-invariant focusing beamforming scheme for retrodirective frequency diverse array (RD-FDA) is proposed to alleviate above tough problems. Specifically, linear and semicircular RD-FDAs are explored to achieve automatic beam alignment of the desired target without phase weights. By establishing an artificial neural network (ANN) framework, frequency offsets can be optimized to match the radiation requirements. By minimizing the loss value of network, the desired focusing pattern can be realized. Then, optimized frequency offsets are periodically invert to limit the signal phase, which allows flexible control of spatial variation range of the mainlobe beam. Simulation results show the effectiveness and superiority of the proposed scheme to achieve quasi-time-invariant beamforming.
telecommunications,engineering, electrical & electronic