Principal Component Regression for Small-Sample Microwave-Microfluidic Chemometrics Without De-Embedding

Broadband dielectric spectroscopy allows for label-free analysis of biological material. However, the data can be challenging to interpret. In this work, Principal Component Regression (PCR) is introduced for the concentration extraction of multiple constituents of a liquid using microwave-microfluidics. S-parameters are measured from 0.5 to 18 GHz on a microwave-microfluidic chip with intentionally introduced error, loaded with solutions having various concentrations of Sodium Chloride (NaCl) and glucose. The mean absolute percentage error for the predictions of the concentration of NaCl and glucose was 2%. It was found that PCR can be used for the prediction of multiple constituents of a liquid with a limited number of training samples, and without needing to de-embed the measurements.