Nonlinear error compensation for microfiber knot current sensors based on artificial neural network

To obtain a good linear response, the nonlinear error of the current sensor based on the micro knot ring (MKR) is first predicted by a deep belief network (DBN) and further used to compensate for the response curve. According to the structural parameters of the MKR, the current, the output light intensity, and the working process, the DBN was used to forecast the nonlinear intensity errors of the MKR. Thousands of current-intensity error data pairs acquired from 60 MKR current sensors are used for the network training and the error-prediction ability of the DBN is evaluated by the root mean square error (RMSE). The DBN with a single-restricted Boltzmann machine has the best prediction performance. From the statistics of the 20 best prediction results, the RMSE in the currentrising and current-falling processes are 0.0257 +/- 0.0072 dB and 0.0212 +/- 0.0062 dB, respectively. For both working processes, the average Pearson correlation coefficients of the response curves are over similar to 0.9950 after compensation, which are much higher than the ones of the raw data. Compared with the traditional linear fitting method, the nonlinear error of the MNF sensor could be accurately forecasted by the DBN according to the structure of the sensing unit and the response curve can be effectively linearly compensated on time. This may provide an effective and flexible method for promoting performance improvement of the MNF sensors with different sizes in the application.

Automated summary

To achieve a good linear response, a deep belief network (DBN) is used to predict and compensate for the nonlinear error of a current sensor based on micro knot ring (MKR). By utilizing the structural parameters, current, output light intensity, and working process of the MKR, the DBN accurately forecasts the nonlinear intensity errors. The DBN with a single-restricted Boltzmann machine shows the best prediction performance.