Shaped Beam Synthesis Using Bayesian Optimization-Based Convolutional Neural Networks

The shaped beam synthesis of array antenna is known as a computional-cost task. In this article, a bayesian optimization (BO)-based convolutional neural network (CNN) is proposed for shaped beam synthesis (SBS). By establishing BO-CNN framework, the shaped beam pattern is synthesized in real-time without the array geometry limits. The CNN is composed of convolutional layers and fully-connected layers. The former is served as an extractor to extract the features, while the latter predicts the array excitation using the extracted the features in real-time. However, during the training of CNN, many architecture hyper-parameters of CNN need to be tuned by experience. In the proposed algorithm, by introducing the BO into the CNN, the automatic design of the CNN can easily achieved and the CNN performance is further improved. The effectiveness of the BO-CNN is validated by two examples, and its performance is compared with that of the convex optimization algorithm.