A Time Delay Neural Network Based Technique for Nonlinear Microwave Device Modeling.

Area of Science:

• Electrical Engineering
• Computational Electromagnetics
• Machine Learning Applications

Background:

• Accurate modeling of nonlinear microwave devices is crucial for circuit design and performance prediction.
• Static neural network models often struggle with the dynamic behavior and generalization required for complex devices.
• Existing modeling techniques may lack efficiency or require extensive, specific datasets.

Purpose of the Study:

• To present a new nonlinear microwave device modeling technique using time delay neural networks (TDNNs).
• To improve the accuracy and generalization capabilities of device models compared to static neural network methods.
• To develop an efficient training algorithm for the proposed TDNN model.

Main Methods:

• Development of a time delay neural network (TDNN) based modeling technique.
• Formulation of a training methodology enabling the TDNN model to learn from DC, small-signal, and large-signal data.
• Implementation and verification of the TDNN model using Gallium Arsenide Metal-Semiconductor-Field-Effect Transistors (GaAs MESFETs) and Gallium Arsenide High-Electron Mobility Transistors (GaAs HEMTs).

Main Results:

• The proposed TDNN modeling technique demonstrated superior accuracy in modeling nonlinear microwave devices compared to static neural network approaches.
• The training formulation allowed for effective learning from varied signal data, enhancing model generalization.
• Experimental validation with GaAs MESFET and GaAs HEMT devices confirmed the efficiency and validity of the TDNN method.

Conclusions:

• The time delay neural network (TDNN) presents an efficient and accurate approach for modeling nonlinear microwave devices.
• The proposed TDNN technique offers improved generalization, making it suitable for a wide range of device types.
• This method provides a robust framework for advancing microwave device characterization and simulation.