AI Deep Learning Optimization for Compact Dual-Polarized High-Isolation Antenna Using Backpropagation Algorithm

An artificial intelligence deep learning algorithm is proposed to analyze a dual-polarized high-isolation antenna effectively. The method is a building model of multi-input target characteristics and multioutput dimensional variables based on a backpropagation algorithm (MIMO-BP). The inputs are defined as the desired targets of the two-port impedance bandwidths, average isolations, and maximum gains, and the outputs are described as the antenna's dimensional variables. A demonstrated antenna prototype verifies the method's effectiveness and the predicted antenna's performance. The experimental results show that the proposed MIMO-BP method has the advantage in terms of convergence speed (i.e., the total electromagnetic simulated number to obtain the desired design) and time costs, high isolation of better than 40 dB over the bandwidth of 3.47–3.58 GHz, and a maximum gain of 4.3 dBi for both ports, which was obtained in about 22.7 h. These features make it a competitive candidate for antenna optimization design.

I. Introduction

In Contrast to the traditional way of empirically optimizing antenna structures, a new optimization tool using intelligent search algorithms can significantly reduce the workload of antenna designs from repetitively adjusting dimensional parameters to desired objectives. The particle swarm algorithm [1] and bat algorithm [2] were used to optimize the dimensions of the antenna structures to obtain broad impedance bandwidth. In addition, a multiobjective evolutionary algorithm [3] was implemented as an automated design solution for a compact and highly isolated antenna. However, intelligent search algorithms suffer from slow convergence and tend to fall into local optimum solutions.