Fast shape reconstruction based on GPU parallel computation in optical frequency domain reflectometry

A fast shape reconstruction optical frequency domain reflectometry (OFDR) system that employs graphics processor unit (GPU) to parallelly calculate the cross-correlation wavelength shift of the two outer cores of a standard multicore fiber (MCF) was proposed and demonstrated. The effect of the number of threads in each block on the time consumption was studied. The consumption time was 50.86 ms, when the simulated data size and the number of threads per block were 10 MS and 256, respectively. Compared to the serial computing with CPU, the time consumption was reduced by nearly 21 times for 2D and 3D fast shape sensing. At the spatial resolution of 5 mm, the maximum reconstruction errors for 2D and 3D shape sensing were 3.23% and 2.47%, respectively, with MCF lengths of 47.1 cm and 43.6 cm.