Improving Optical Fiber Bandwidth

Jim Landwehr Scott Vander Wiel

Multi-mode Fiber

Lucent produces optical fiber as a central element of high speed, high volume data networks. Though multi-mode fiber is more difficult to produce than its single-mode cousin, it remains an economical choice for local area networks because its large core diameter is compatible with inexpensive connectors and light sources.

We have been analyzing data from multi-mode fibers and preforms to understand how subtle changes in a refractive index profile can degrade fiber bandwidth. Our modeling work will facilitate fine-tuning the lathes that produce multi-mode preforms.

Refractive Index Profiles and Bandwidth

The glass core of a multi-mode fiber has a graded refractive index that decreases radially from the center. The precise shape of this refractive index profile is critical to the fiber's information carrying capacity as measured by bandwidth. Refractive index profiles are routinely measured on glass preform rods before they are drawn out into fiber. A preform consists of many concentric layers of glass produced sequentially using an MCVD (modified chemical vapor deposition) process. Index gradations are constructed by varying the chemical doping in successive layers of glass. The overall shape of the refractive index profile and its layer structure are evident in the idealized profile measurement shown below.

$An idealized refractive index profile$

Both the overall shape and the layer structure potentially influence the bandwidth of fibers drawn from a preform. Optics theory predicts that a near-quadratic refractive index profile will produce high bandwidth fibers. The theory, however, is ill-suited to profiles with layer structure or larger scale departures from the ideal shape. For this reason profiles need fine-tuning and the work is somewhat empirical.

Methods for Tuning Lathes

Preform rods are fabricated on lathes. Two main challenges are present in using preform profiles to fine-tune a lathe.

Characterizing the typical profile shape for the lathe in question.
Developing a model to predict fiber bandwidth for various profile shapes.

If our predictions are reliable, then the model should indicate how to modify a lathe's target profile to improve bandwidth.

We have developed a scheme for processing several profile measurements routinely made on each preform. The major steps are

radial scaling,
subtracting an ideal quadratic profile,
smoothing the result, and
averaging the several smooths for each preform.

Our studies of processed profiles show that a large fraction of the variation from one to the next on a given lathe is of a simple multiplicative form that has little impact on bandwidth. After adjusting for this source of variation, our processed profiles reveal clear and consistent differences among the lathes.

We have also developed a model to predict bandwidth from (processed) profile measurements. The model decomposes profile variations into their principle components and fits a second order response surface to several of the largest components. For this work we employ statistical model building and diagnostic tools such as cross-validation, Cp plots, influence plots, and residual analyses.

We are still working on the model but we expect that it will usefully guide the tuning of lathes. The feedback time using profile measurements will be much shorter than the present tuning method.