Optimized Feedforward Neural Network for Multiplexed Extrinsic Fabry-Perot Sensors Demodulation

A method based on feedforward Neural network (FNN) is proposed to demodulate multiplexed extrinsic Fabry-Perot interferometer (EFPI) sensors. The FNN accepts the superposed spectra of multiplexed EFPI sensors and returns the strain information on each sensor. A strain sensing experiment is carried out to verify the performance of the method. In the experiment, an optical spectrum analyzer (OSA) with a sampling interval of 0.1 nm was used to obtain the spectra of two EFPI sensors, whose cavity length are around 160 μm and 180 μm, respectively. A few spectra of two EFPI sensors under special strains were collected to train the FNN. The well-trained network was used to demodulate the arbitrary strains applied on each sensor. The experimental results indicate that the well-trained FNN shows a better performance than the fast Fourier transform (FFT) method. The crosstalk of the FNN is as low as 4.5241 μϵ in the strain range of 0 to 1020 μϵ, while the FFT method is failed to extract the strain information from the superposed spectra. The influences of the signal to noise ratio (SNR) and the sampling interval of the spectra are studied through simulations. The simulation results show that the demodulation performance mainly depends on the SNR of the spectra. Furthermore, we can reduce the cost and increase the demodulation speed by using a spectrometer with a larger sampling interval, since the sampling interval influences the performance slightly. The proposed method provides a new approach to enhance the sensing capacity of the EFPI sensing networks.