An Efficient Method for Complex Digitally Coded Antenna Design Based on Evolutionary Computation and Machine Learning Techniques

Digitally coded antennas (DCAs), also called pixelized or fragmented antennas, show high potential for improving performance and size via unconventional structures. However, the bottleneck is the resolution that can be handled. When the resolution is more than a few hundred pixels, optimization quality and efficiency become severe challenges. Therefore, a new method, called digitally coded antenna-oriented surrogate model-assisted evolutionary algorithm (DC-SADEA), is presented in this article. The key innovations include: 1) the introduction of an ensemble learning-based surrogate modeling method for mapping the DCA design variables to performances and 2) a bespoke surrogate model-assisted global optimization framework and genetic algorithm (GA) operators for DCAs. An ultrawideband antenna (about 1900 pixels) and the feeding part of a 5G outdoor base station antenna (about 1500 pixels) are used to demonstrate DC-SADEA. Measurement results demonstrate the effectiveness and efficiency of DC-SADEA.