Adaptive Learning of Probability Density Taper for Large Planar Array Thinning
Li Gu, Yanwen Zhao, Zhipeng Zhang, Lifeng Wu, Qiang‐Ming Cai, Runren Zhang, Jun Hu
Abstract
This article presents a novel optimization algorithm to design a probability density taper for large array thinning. This novel algorithm is based on the iterative Fourier technique (IFT) integrated with an innovative adaptive learning mechanism and denoted as the probability learning IFT (PLIFT) here. With successive forward and backward Fourier transforms, the traditional IFT can be capable of thinning large arrays with high convergence rates, but easily get trapped in local optima. The proposed PLIFT utilizes the probability estimation to describe the elements with possible states of “ON” and “OFF” in a thinned array. A probability model is then adopted to determine the locations of elements in the thinned array featuring minimum sidelobe level. In this PLIFT, on the one hand, the probability model of density taper is learned in optimization process and gradually trained to approach the global optima. With the previous information being honored, the probability model taper proves to be reliable and effective in generating diverse competitive seeds, which, in turn, helps training an optimal probability distribution to best-fit the design goal. Several representative examples of large planar arrays thinning are provided to demonstrate the validity, highlighted global searching ability and increased robustness of the PLIFT.