Parallel Computation Technology for Distributed Optical Fiber Sensing System

Distributed optical fiber sensing (DOFS) system exists the issue of large amount of data with the improvement of sensing distance and sample rate, which result in poor real-time performance of the DOFS system. A parallel computation technology based on graphics processing unit (GPU) is presented to speed up the data processing and improve the real-time performance. The key to fast parallel processing of data lies in the allocation of multiple threads to execute the same instruction to reduce the time of data operation. As the conventional data processing method for phase sensitive optical time domain reflectometer (Phi-OTDR) system, the differential accumulation algorithm can realize vibration location and improve the SNR of the system. The validity of this concurrent processing technology for the algorithm is verified and experimented. The computation time spent on GPU is only 1.6 mu s, which is 21250 times faster than Central Processing Unit (CPU) for the 2020m length of optical fiber. This technology significantly accelerates the differential accumulation algorithm. When the amount of data is huge, the traditional short time Fourier transform based on Brillouin optical time domain reflectometer (STFT-BOTDR) system has a longer data demodulation time with the serial operation. Moreover, the algorithm can be processed in segments without interfering with each other. Based on these characteristics, this parallel processing technology based on GPU can be selected to accelerate the data demodulation process and greatly improve the execution efficiency of the algorithm. For the same Brillouin scattering time domain signal and the number of cumulative averages is 8192, the time of signal demodulation which is performed on GPU is 116.3 s, which is 25 times faster than that of CPU. In conclusion, the parallel computation technology based on GPU can speed up the data processing and enhance the real-time performance for the DOFS system.