Demodulation method for vibration sensors of ultra-weak fiber Bragg grating using a Self-Referencing signal based on DBNs with ensemble learning

The vibration signal with high signal-to-noise ratio (SNR) is difficult to be obtained in the conventional methods owing to the influence of environment and equipment in ultra-weak fiber Bragg grating (UWFBG) distributed vibration sensor. In this paper, a novel demodulation method using a self-referencing signal based on deep belief networks (DBNs) with ensemble learning is proposed, to enhance the SNR of demodulated signal. The reference interferometer can obtain the information of environmental noise and system noise. The DBN is employed to construct a dynamic correlation model between the interference signal and vibration signals. By extracting multidimensional features, the model learns how six output interference signals are changed with vibration in measured environment. To further promote the performance of single DBN model, an ensemble learning method is used to learn the knowledge of five different DBN models. Experimental results demonstrate that in the range of 10 Hz to 1000 Hz, the ensemble DBN model can achieve vibration measurement with higher SNR. The proposed method can reduce background noise by 22.8 dB in 1 kHz vibration environment, and the SNR is 79.9 dB. It can be drawn conclusion that the proposed method can weaken the influence of environmental noise and inconsistent splitting ratio of 1 × 3 coupler, to make demodulation result with higher SNR in vibration sensor demodulation.