Modeling of ringdown cavity maladjustment based on the extreme learning machine.

Abstract

In the cavity ringdown technique, cavity maladjustment may affect the accuracy of measurement. The effect of maladjustment is complex and nonlinear. Therefore, a neural network method of an extreme learning machine is presented to model the nonlinear relation between the intracavity loss and the cavity maladjustments. This method was tested by two-dimensional angular scanning simulations and experiments. After dataset training, this method finely predicted the intracavity loss at certain maladjustments. The root mean square values of the prediction deviation were about 0.27ppm in simulation and about 0.44ppm in experiment. This method was further applied to a cavity ringdown system for high reflectivity measurement. The measurement uncertainty was improved from ±0.0025% to ±0.0019%.