A machine-learning-assisted MPA design and optimization for multi-band applications

Within the 5G communication system framework, multi-band antennas serve as a solution that combined engineering practicality and technical advantages. Traditional antenna designs were often constrained by the designer's experience and electromagnetic (EM) simulators. This work proposes a Genetic Backpropagation Network (GBN), a hybrid framework combining the Genetic Algorithm (GA) and Backpropagation Neural Network (BNN). This network model optimizes the weights and structure of the backpropagation neural network through the genetic algorithm, thereby improving the model's performance. We use the GBN method to achieve multi-objective-driven customized multi-band patch antenna (MPA) design. To demonstrate the effectiveness of the proposed model, a low-profile multi-band patch antenna has used as a design example. Finally, the antenna prototype is fabricated based on the antenna parameters obtained through the GBN inverse prediction. The measurements demonstrate that the simulated and measured S-parameters are in good agreement.