High Performance Interrogation Technology Based On Microwave Photonics Filter Assisted By Machine Learning

Microwave photonics interrogation is a novel demodulation technology that combines optical domain assisted sensing perception and microwave domain demodulation.The performance bottleneck of traditional optical domain demodulation is expected to be broken through advanced microwave signal processing technology.However,the reported microwave photonics interrogation techniques still have tradeoff problems in terms of measurement accuracy,interrogation speed,and the number of demodulated sensors.In order to improve the performance of microwave photonics interrogation technology,the investigation on notch microwave photonic filter(MPF)and passband MPF demodulation technology has been performed.The major contributions are as follows:(1)An advanced machine learning algorithm called Gaussian process regression(GPR)is introduced into the notch-type MPF-based interrogation system,to overcome the trade-off between measurement accuracy and demodulation speed of traditional demodulation methods.In a proof-of-concept experiment,a two-tap microwave photonic filter system consisting of two fiber Bragg grating(FBG)was constructed for strain sensing.Unlike the traditional direct-notch-detection method where the demodulation is enabled by reading the notch frequency shift,the proposed method is to create a GPR model to learn the relationship between the whole frequency response of MPF and the strain.The experimental results show that the well-trained GPR model can accurately obtain the strain even in the case of a coarsely sampled frequency response,which is difficult to achieve in the traditional direct-notch-detection method.Meanwhile,for a given measurement accuracy,the relaxation of the sampling resolution requirements effectively improves the interrogation speed.The measurement time,training time,testing time,and root mean square error(RMSE)of our proposed method at the resolution of 10 MHz are 0.5 s,1.2 s,1.6 ms,and 1.296 (?),respectively.More importantly,the well-trained GPR model exhibits strong robustness to the variations of notch depth.(2)A fast demodulation algorithm based on Buneman frequency estimation is proposed for single passband-type MPF interrogation technology.In a proof-of-concept experiment of displacement sensing based on Mach-Zehnder interferometer(MZI),the proposed demodulation method has the advantages of low computational cost,high speed,and can accurately obtain the center frequency of the passband.And the RMSE less than 10.31 μm is achieved at 151 MHz sampling resolution.Moreover,in order to achieve simultaneous demodulation of multiple interferometric sensors through the passband microwave photonic filter demodulation technology,a multi-passband MPF demodulation system combined with extreme learning machine(ELM)has been established.The results show that the proposed scheme can accurately demodulate the center frequency of each passband of multiple sensing response passbands with different overlapping degrees,and the RMSE less than 11.82 μm is achieved at 10 MHz sampling resolution.This thesis provides a certain guiding significance for the research of demodulation algorithm based on microwave photonic filtering,and a new solution was given for realizing new demodulation algorithm with high precision,high speed and high capacity.