High Isolation Base Station Antenna Array Based on Fully Connected Neural Network Optimization

In this work, we present a novel approach for enhancing the efficiency of antenna array optimization by introducing a fully connected neural network (FNN) model. The proposed methodology is specifically applied to the decoupling design of base station antennas, aiming to augment the port isolation of the antenna system. The FNN model takes the structural parameters of the antenna array as input and yields the corresponding S parameters as output. In the antenna array, the five structural parameters as input affect the port matching and isolation of the antenna. The S parameters as output represent the port matching and isolation of the antenna. Following training with a limited set of simulation data, the model effectively encapsulates the simulation process of the base station antenna array in the simulation software. As additional and more intricate structural parameters are incorporated into the model, a substantial volume of generated data is procured. A total of 675 sets of simulation data are employed for the training of the neural network. Following the generation of the FNN model, a dataset comprising 10125 sets of data is acquired. To assess the reliability of the generated data, eight randomly selected datasets are compared with the simulated results, affirming the fidelity of the generated data set. This data generation mechanism significantly economizes simulation time. Employing a search algorithm, we ascertain the structural parameters of the antenna array with the highest isolation in the datasets. At the same time, the results of this antenna array are verified in the simulation software. This methodology facilitates rapid optimization of the antenna array’s port performance and has significant potential for mitigating coupling issues in base station antennas.