Performance Upgradation of Microwave Photonic Filtering Interrogation Using Gaussian Process Regression

Abstract

The Gaussian process regression (GPR), a powerful machine learning tool, is introduced to upgrade the microwave photonic filtering interrogation, with improved demodulation speed and accuracy. In a fiber Bragg grating (FBG) based microwave photonic filtering interrogation system for strain sensing, the GPR is employed to learn the relationship between the frequency response of microwave photonic filter and the strain applied on the sensing FBG. Compared with the traditional direct-notch-detection method, the proposed method can achieve better measurement accuracy under the sparsely sampled frequency response, whilst the interrogation speed is greatly improved. More importantly, the well-trained GPR model remains valid for the filter frequency response with large notch depth fluctuation, greatly increase the tolerance to device parameter deviation and ambient changes. This work demonstrates that the machine learning algorithms will provide a new avenue for microwave photonics filtering interrogation with the improved performance.