Array Nulling Synthesis Based on Hyper-Parameter Optimized Self-Paced Learning Convolutional Neural Networks

Array nulling synthesis (ANS) problem is usually solved iteratively via the optimization algorithms, resulting the processing time-consuming. The communication proposes a fast ANS algorithm based on a self-paced learning convolutional neural network (SPL-CNN). The CNN includes the convolution layers and the fully connected (FC) layers. The convolution layers serve as an extractor to learn the features of the covariance matrix of the array factor (CMAF) in the nulling regions, while the FC layers serve as a regressor to predict the array excitation and the radiation pattern. However, the significant differences among the CMAF elements for different wide nulling regions make the convolutional neural network (CNN) difficult to quickly learn the samples. To overcome this shortcoming, SPL is adopted to make the CNN select samples from "easy" to "hard" in the training process by comparing the loss function and the hyper-parameter. Moreover, the quantum particle swarm optimization (QPSO) is utilized to tune the architecture hyper-parameter values of the SPL-CNN. After the SPL-CNN is well-trained, the proposed algorithm can calculate the array excitation without iteration. Numerical simulations for point/wide nullings are carried out to validate the superiority of the proposed algorithm by comparing it with the convex algorithms and the other existing data-driven algorithms.