Machine Learning Assisted Temperature Insensitive Microwave Photonic Sensor Based on Single Microring Resonance

Abstract

We propose a machine learning (ML) assisted athermal microwave photonic (MWP) sensor with high resolution based on a single optical microring resonance. Through the MWP sideband processing, the superposed optical resonance shifts caused by the variation of measurand of interest and the undesired temperature perturbations are transformed into the shift of an ultra-deep notch in the radio frequency (RF) spectrum with high resolution. Simultaneously, the concurrent nonlinear change of the optical resonance extinction ratio is manifested as the variation of DC bias of the modulator. To eliminate the temperature interference in real time, the ML algorithm of support vector regression (SVR) adept at nonlinear regressions is established, through the training with RF notch frequencies and exploited DC bias voltages. As a proof-of-concept, the relative humidity (RH) sensing based on a polymethyl methacrylate coated silicon microring resonator as the sensor probe is performed at different temperatures. Over 50 dB high RF rejection ratio, indicating the high sensing resolution, is achieved during the whole sensing process. Despite the small training datasets, the established SVR-based model effectively eliminates temperature induced variances and guarantees a mean absolute error of 1.30% RH.