Machine Learning Based Inverse Model for Identifying Transmission Line Network Parameters from Tabulated Frequency-Domain Data

In this paper, a neural network based inverse model is presented that can predict the values of the geometrical, physical, and material parameters of a given transmission line network from its tabulated frequency-domain responses. The proposed inverse model consists of two parts – a forward neural network that emulates the frequency-domain responses of the transmission line network under test as analytic functions of the geometrical, physical, and material parameters and a particle swarm optimizer (PSO) that minimizes an appropriate loss function to derive the requisite values of the network parameters.