Autogenerating Training Set for Model Extraction of the behavioral model in RF Simulation Using Machine Learning Algorithm

This paper presents a method to autogenerate training set in model extraction of the behavioral model in RF simulation based on machine learning algorithm. The input signal is cut into a large number of short time length segments which are clustering into several groups using K-means clustering algorithms. The training signals are generated based on the transition possibility table acquired from analyzing the input signal. Results show that this method is more capable than traditional method in adapting to the input signal and suppressing overfitting during model extraction, especially for the circuit with long time memory effects and strong signal distortion.