I. Introduction
In recent years, ANNs have been widely adopted for circuits and components modeling in RF and microwave because of its generalization capability and convergence rate for complex multidimensional nonlinear problems. Various ANN structures have been explored for RF and microwave applications, including dynamic neural networks (DNNs), knowledge-based neural networks (KBNNs) and convolutional neural networks (CNNs) [1] - [3]. There have also been advances in neural network algorithms, including genetic algorithm (GA) and particle swarm optimization (PSO) [4], [5]. However, one of the main challenges of these models and algorithms is the requirement for a large amount of data, which leads to a rapidly growing cost both in time and complicated fast instruments.