Interplay between physics and statistics for modeling optical fiber bandwidth

In multimode optical fiber, the refractive index of the glass is varied radially in such a way that a light pulse propagates through multiple modes, or paths, as it travels. We investigate the prediction of fiber transmission capacity, specifically bandwidth, by using refractive index profiles of glass preform rods measured in a manufacturing environment before the rods are drawn into optical fiber. By closely linking empirical and theoretical approaches to modeling, we demonstrate the feasibility of predicting bandwidth despite the finding that profile measurements are grossly inaccurate. Empirical components of the modeling involve extensive preprocessing of raw measurements that approximate continuous functions, followed by spline fitting and specialized cross-validation for model assessment. Modeling from physical theory is based on solutions of Maxwell's equations. This article shows how a blending of physics with statistics provides conclusions that could not be obtained separately from either approach-namely, good predictions are possible, but profiling equipment needs to be upgraded to remove substantial measurement biases.