Inverse artificial neural network modeling for metamaterial unit cell synthesis

Abstract

Inverse artificial neural network (IANN) modeling is used to synthesize a metamaterial unit cell. The ANN approach is best suited to capture the mapping relation between inputs and outputs given a training dataset. In this work, the IANN method is applied to synthesize a metamaterial unit cell for which the inverse model is ill conditioned. A perfectly trained ANN is capable of generating the input data for a desired set of outputs for a fixed set of synaptic weights. In this work, a prior knowledge based input with difference (PKI-D) method is used to perform the inverse mapping. Although the split-ring resonator (SRR) is widely used to realize metamaterials for different applications, no closed-form formula is available to determine its design parameters. The goal of this work is to determine the dimensions of a SRR for a desired frequency and permeability value. The work is carried out in two phases. In the first phase, the inverse model is developed to estimate a single design parameter, then in the second phase, a multiple-parameter estimation model is developed. The average error for the PKI-D model is found to be 0.059. Backpropagation is used to train (solve) the input vector for unknown parameters.