THz-TDS Parameter Extraction via Machine Learning

Terahertz time domain spectroscopy is a powerful analytical technique for investigating a wide range of material samples, however calculating refractive index values is not a trivial process and can be computationally intensive. Here we investigate the use of machine learning to simplify and replace certain aspects of THz TDS data analysis. We train a neural net to convert THz TDS data into complex refractive index values by creating a large training set based on simulated data. The resultant neural net is computationally efficient and extracts TDS parameters with a high degree of accuracy. Further, we propose that by investigating the trained network we may gain new insights into THz parameter extraction procedures.