Antenna Impedance Matching Using Deep Learning

Area of Science:

• Electrical Engineering
• Antenna Theory
• Machine Learning

Background:

• Antenna impedance matching is crucial for efficient signal transfer.
• Traditional matching methods often require complex mathematical descriptions and knowledge of both impedance magnitude and phase.
• Existing methods may struggle with non-ideal or unimplementable input conditions.

Purpose of the Study:

• To propose a novel deep neural network (DNN) for determining antenna impedance matching circuit parameters.
• To enable impedance matching using only the magnitude of the input impedance, eliminating the need for phase information.
• To approximate feasible solutions for matching circuits, even for non-ideal or unimplementable inputs.

Main Methods:

• A deep neural network (DNN) was designed to predict matching circuit element values.
• The DNN was trained using data where the input was the magnitude of the antenna's input impedance (S11) and the output was the corresponding matching circuit element values.
• A gamma-matching circuit, comprising series and parallel capacitors, was applied to an inverted-F antenna for learning and validation.

Main Results:

• The DNN successfully determined matching circuit parameters using only impedance magnitude.
• Training demonstrated convergence of loss as epochs increased, indicating effective learning.
• The DNN accurately predicted matching values for previously unlearned impedance characteristics, such as square and triangular waves.

Conclusions:

• The proposed DNN offers a simplified approach to antenna impedance matching by relying solely on impedance magnitude.
• This method bypasses the need for detailed mathematical models of matching techniques.
• The DNN shows potential for approximating solutions even for challenging or unimplementable input scenarios, enhancing matching robustness.