A new formulation of dynamic neural network for modeling of nonlinear RF/microwave circuits

In this paper, we propose a new formulation of dynamic neural network (DNN) for modeling of nonlinear RF/microwave devices or circuits in continuous time domain. The proposed model can be trained directly from input-output large-signal data irrespective of internal details of the circuit. The proposed approach maintains the accuracy even in presence of measurement noise in training data. A circuit representation of the proposed model is introduced in order to incorporate it into circuit simulators for high-level design. Examples of dynamic modeling of FET amplifier operating at high frequencies are presented.