Nonlinear Device Modeling Based on Dynamic Neural Networks: A Review of Methods

This paper reviews the recent modeling methods of nonlinear devices including dynamic Neuro-space mapping (Neuro-SM), time delay neural network (TDNN) and Wiener-type dynamic neural network (Wiener-type DNN). The dynamic Neuro-SM combines dynamic neural networks with existing models to reflect the dynamic characteristics of nonlinear devices. The TDNN and the Wiener-type DNN are dynamic neural networks to train input-output relation of nonlinear devices. The formulations of dynamic Neuro-SM, TDNN and Wiener-type DNN are presented allowing the training with dc, small signal, and large signal data. The modeling example of GaAs MESFET device illustrates that the Wiener-type DNN model can fast and accurately model microwave nonlinear devices and has good convergence and robustness.