Antenna Impedance Matching Using Deep Learning.

We propose a deep neural network (DNN) to determine the matching circuit parameters for antenna impedance matching. The DNN determines the element values of the matching circuit without requiring a mathematical description of matching methods, and it approximates feasible solutions even for unimplementable inputs. For matching, the magnitude and phase of impedance should be known in general. In contrast, the element values of the matching circuit can be determined only using the impedance magnitude using the proposed DNN. A gamma-matching circuit consisting of a series capacitor and a parallel capacitor was applied to a conventional inverted-F antenna for impedance matching. For learning, the magnitude of input impedance of the antenna was extracted according to the element values of the matching circuit. A total of 377 training samples and 66 validation samples were obtained. The DNN was then constructed considering the magnitude of impedance as the input and the element values of the matching circuit as the output. During training, the loss converged as the number of epochs increased. In addition, the desired matching values for unlearned square and triangular waves were obtained during testing.