Transmission line impedance matching with neural networks

Impedance matching between transmission lines and antennas is an important and fundamental concept in electromagnetic theory. This paper verifies that feedforward neural networks can be trained to design accurate and efficient matching configurations to accomplish this task. Impedance matching is frequently performed with Smith charts or relatively complex formulas, but mathematical methods can yield unanticipated results unless the user has a solid grasp of the underlying theory. Graphical instruments, e.g., the Smith chart, can be a very useful approach, often achieving an accuracy of three significant digits if the normalized impedance is not excessively close to the origin of the chart. Smith charts can be difficult to use because of optical effects generated by the gridlines. This paper proposes a neural verification method to match shorted stubs to lines, resulting in a voltage standing wave ratio of 1:1, i.e., no reflected waves. The user enters the normalized load impedance and the neural network yields the distance of a shorted stub from the toad and the length of the stub. Test results indicate that a matching network can be computed very successfully using the approach described in this paper. It can be useful in verifying results obtained from the Smith chart and can indicate sources of graphical errors. Other applications are expected in further efforts.