Problems in bandwith measurement and a suggested solution

A recent series of bandwidth measurements by several Bell Labs and Western Electric locations on the same set of fibers (a bandwidth measurement round robin) has addressed and clarified some of the problems that cause disagreement between test sets. Three measurements out of twelve had a large standard deviation (based on the results of all participants), which raised the overall ratio of the standard deviation to the mean (
\sigma/\mu
) to 12 percent. These three fibers had double-peaked outputs; the resulting transfer functions had either a plateau or an undulating behavior around the 3-dB level: changes in the launch conditions or source spectral width could shift the plateau or undulations from above the 3-dB level to below it, producing large changes in measured 3-dB bandwidth. Reducing measurement variations for such fibers will require closer control over the launch conditions and, for short wavelength measurements, the source spectral width. In addition, using the 3-dB bandwidth from a Gaussian fit to the transfer function instead of the transfer function itself greatly reduces measurement variations and in some cases provides a better prediction of the bandwidth of fiber concatenations.