Parametric modeling of microwave passive components using combined neural networks and transfer functions in the time and frequency

A novel parametric modeling technique is proposed to develop combined neural network and transfer function models for both time and frequency (TF) domain applications of passive components, where the neural network is trained to map geometrical variables to the coefficients of transfer functions. Built on our previous work, a new order-changing module is developed to enforce stability of transfer functions and simultaneously guarantee continuity of coefficients. A constrained optimization strategy is introduced to enforce passivity of transfer functions through a neural network training process. A general equivalent circuit for two-port passive components is generated directly from coefficients of arbitrary-order transfer functions. Once trained, the parametric model can provide accurate and fast prediction of the electromagnetic behavior of passive components with geometrical parameters as variables. Compared to our previous work, the proposed method enables models to work well in the time domain providing good accuracy in challenging modeling applications. Two parametric modeling examples of spiral inductors and interdigital capacitors, and their application in both time and frequency domain simulations of a power amplifier are examined to demonstrate the validity of the proposed technique. (C) 2012 Wiley Periodicals, Inc. Int J RF and Microwave CAE , 2013.