Role of Machine Learning in Rapid Modeling of RF Devices: VO2 RF Switch Modeling as a Case Study

Ahmad Khusro, Shuai Yang, Mohammad Vaseem, Mohammad Hashmi, Atif Shamim

2021

Closed access Metadata (CSL-JSON)
DOI 10.1109/cama49227.2021.9703575
Data confidence 70 medium
How this score is calculated

It is built from checkable properties of the record itself: whether the article has a DOI, how many independent sources confirmed its data, whether the required fields are filled in, and how recent the latest confirmation is. Disagreement between sources and a missing identifier lower it. A value of 80 or above means the record was published without further checking; 60 to 79 means it is published at reduced confidence, with a manual check advisable. A record verified by an editor by hand is never pushed below 90 by a recalculation. This scores our data about the article, not the article itself.

Abstract

The paper provides an extensive overview on the potential of machine learning in rapid modeling of RF/Microwave devices. To demonstrate the role of machine learning in rapid modeling of RF devices, printed VO2RF switch has been selected as an example. Of many supervised machine learning techniques, Artificial Neural Network (ANN) has been preferred to model the printed VO2RF switch for reconfigurable applications. The model uses cascade feed-forward architecture of ANN along with Bayesian regularization algorithm to train the multi-input behavioral model with varied set of operating conditions (geometric dimensions and temperature) over a frequency range of 0.01 GHz to 35 GHz. Subsequently, the trained model is tested for novel set of test inputs for both interpolation and extrapolation cases to evaluate the accuracy, scalability, extrapolation capability and generalization ability. A very good performance metrics is obtained with the mean square error $3\mathrm{x}10^{-4}$ and $4.5\mathrm{x}10^{-4}$ for the interpolation and extrapolation set and correlation coefficient of 99% is obtained for the broad frequency range between the measured S-parameters and modeled which validates the effectiveness of the proposed modeling approach. Eventually, the potential of machine learning based model development is further enhanced by demonstrating the seamless integration of the model in CAD environment for real time design, simulation and analysis of bandwidth reconfigurable filters involving multiple ANN based VO2RF switch equivalent model.