High Resolution and Speed Demodulation for Optical Fiber FabryPerot Magnetic Field Sensor

To get rid of the dilemma between the speed and the resolution of optical fiber Fabry-Perot (F-P) magnetic field sensor (MFS) demodulation algorithm and to address the problem of difficult detection in the low magnetic field area, the compact parallel dual neural network (PDNN) with wavelet decomposition (WD) was proposed for the first time to demodulate the magnetic field. PDNN consists of the prediction neural network (PNN) and the error compensation neural network (ECNN). PNN is used to predict the true magnet field and ECNN is designed to predict the error between the true value and predictive value. With the combination of the dual networks, PDNN achieves a higher accuracy and resolution compared with traditional series single networks. In addition, to reduce the computational cost of PDNN, a dimensionality reduction method based on WD is developed. It is demonstrated according to simulation and experiments that PDNN obtains stronger anti-interference capability, less computational cost, and higher demodulation resolution after the process of WD. Three-layer WD is used to extract the features of F-P MFS spectral signal, which can keep 13.8% signal energy and reduce the spectral data points by about eight times. PDNN with WD achieves 0.004 Gs demodulation resolution with 1 kHz in low magnetic field, which is applicable to all fields requiring high speed and resolution.