A new family of neural network-based bidirectional and dispersive behavioral models for nonlinear RF/microwave subsystems(invited paper)

The article introduces a new approach to the ANN-based behavioral modeling of nonlinear RF/microwave systems or subsystems. The proposed model accounts for both the linear and the nonlinear distortion of modulated signals. The former is introduced by the two-port frequency-dependent admittance matrix. The latter is described by the means of a couple of describing functions (which may also be frequency-dependent), relating the two-port linear and the nonlinear responses. The describing functions are approximated by artificial neural networks (ANN) trained by harmonic-balance results. The resulting model is fast to compute, can accurately handle broadband modulated signals, and is fully bilateral. This means that the model can be routinely used to analyze systems where bidirectional signal flow takes place, e.g., due to mismatches between the interconnected subsystems, or even due to the presence of filters operating in the stopband. © 2002 John Wiley & Sons, Inc. Int J RF and Microwave CAE 12: 51–70, 2002.