Modeling draws a bead on manufacturing tolerances

Joe Shiefman

Significant drops in spending and increased competition have made the optical telecommunications components market a very difficult place to do business. Once spending levels rebound, the winners will not necessarily be those companies with the cleverest component concepts because, for most component types, there will be many companies offering products with specifications that meet the market's needs. From within these companies, those with the most inexpensive and most manufacturable designs will be the winners.

This tolerancing is accomplished by varying alignments, shapes, and material properties of the system elements. The resulting effects on system specifications, such as polarization-dependent loss, insertion loss, return loss, crosstalk, and polarization-mode dispersion, are then examined. Tolerancing procedures highlight weaknesses in the original design. The next step is to change the original design to remove these problems. This iterative process of alternating tolerancing and design revisions continues until the most "buildable" design is obtained.

Modeling software has several advantages. Building and modifying a model is usually much faster in software than in a lab. In the lab it is not only time-consuming to assemble the system elements, it also takes significant time to obtain them. Additionally, modeling the system in software is less expensive since it does not involve purchasing the system elements or the support equipment necessary to hold and align them, and to measure the system input and output. The parameters of both the system elements and the state of the light going into the system are also more controllable in software than in the lab. In particular, the issues of time and expense become magnified in such an iterative approach.

It then enters a half-wave plate, which is the switching element. If the optic axis of the wave plate is aligned in a direction halfway between the two polarization components that emerge from calcite #1 (a rotation of 45°), the ordinary polarization (X-pol) will switch to extraordinary (Y-pol) and vice versa (see path #1 in Fig. 1). In this case, both beams will exit calcite #2, reflect off the two right-angle prisms, and be directed to the top output collimator, which focuses the beam into its companion fiber.

On the other hand, if the optic axis of the wave plate is aligned in the same direction as one of the two polarization components that emerge from calcite #1 (a rotation of 0°), the polarization of the two beams will be unchanged and they will continue to move further apart as they propagate through calcite #2 (see path #2 in Fig. 1). In this case, both beams miss the right-angle prisms, get recombined by calcite #3, and are fed back into the path #2 output fiber by its collimating lens. The compensating glass that is positioned just before calcite #3 is introduced to add additional path length to the bottom beam in path #2, which has a shorter optical path through the rest of the system than the top beam in path #2.

The software tolerancing procedure is demonstrated by examining the effect of a rotation of the optic axis (OA) in calcite #1 on the system insertion loss. The OA in calcite #1 is nominally in the Y-Z plane. The type of OA rotation examined here is for a rotation of the OA out of its nominal Y-Z plane such that it acquires some X component.

This rotation of the OA, relative to its nominal location in the system Y-Z plane, can arise either from the OA not being oriented correctly, relative to the reference side of the calcite block, or from a mechanical misalignment where the block is mounted with a rotation relative to its nominal system orientation. The mechanical misalignment can be due to any of several possible causes—such as bad mechanical parts, poor assembly of those parts, or the assembly method used to fix the calcite block to the mechanical fixture.

These results can then be plotted (see Fig. 2). The insertion loss in this study is limited only to that due to the calcite OA rotation. It does not include Fresnel losses due to surface reflections. The beam propagation was started in the collimated space after the input collimating lens and the coupling overlap integral was calculated in the collimated space just before the output collimating lens. Therefore, the insertion loss also does not include losses due to aberrations in the beam from the collimating lenses. Certainly, both of these other factors contribute to the insertion loss and also must be toleranced and budgeted into the design check process.

The insertion losses for this design are 0.1 dB for +1.25° and -1.34° of OA rotation, 0.2 dB for +1.77° and -1.96° of OA rotation, 0.3 dB for +2.17° and -2.46° of OA rotation, and 0.4 dB for +2.52° and -2.90° of OA rotation.

Therefore, since it does not include other sources of insertion loss, for most components the acceptable tolerance on this type of OA rotation in calcite #1 will be in the 1° to 3° range. To determine how easy it is to achieve 1° to 3° of OA rotation, the optical engineer must obtain a specification on the accuracy of the OA orientation relative to the reference side of the calcite block from its manufacturer. The reference side of the calcite block is determined by how it is mounted in the component.

Additionally, the optical engineer must determine specifications for the flatness and rotation of the mechanical reference surface onto which the calcite block mounts. The method of mounting is also important since this can also induce rotation. Once these questions are answered one can determine whether this rotation of 1° to 3° of OA is easily achievable, or whether it will require expensive parts or difficult assembly, or lead to a low yield. If there is a problem, the design will have to be modified and retoleranced.

First it ensures that the numbers obtained in the tolerancing run are valid. For instance, in the example at hand, one would like to know why the plot (in Fig. 2) shows an asymmetry of insertion loss between plus and minus rotations.

The following analysis looks at the case of 10° of OA rotation and an input beam of 45° polarization using a picture of the energy emerging from calcite #1 (see Fig. 3, top). The original beam has split into two beams after exiting the calcite. The beams are separated not only in Y but also in X, and the bottom beam contains 73% of the total energy.

FIGURE 3. Starting with 45° polarized light, the energy in a plane just after calcite #1 (top) is greater in the bottom of two beams for the case of +10° OA rotation. Both beams emerge in linear states rotated by 13.8° from their nominal X and Y states (center). After passing through the half-wave plate, the polarization states of the two beams get flipped about the 45° axis (bottom).

Since the OA of calcite #2 is not rotated out of its nominal position, the normal polarization modes for a beam traveling in the Z direction into calcite #2 are X and Y. Therefore, the two beams entering calcite #2 each split into their X and Y components, which gives rise to four beams exiting calcite #2.

FIGURE 4. A logarithmic picture (left) of the energy in a plane located just after calcite #2, for the case of +10° of OA rotation in calcite #1, shows four beams. The four beams are generated by calcite #2, which breaks the two beams that exit the wave plate into their X and Y polarization components (right).

The polarization map of the four beams exiting calcite #2 also supports this argument (see Fig. 4, right). This map also shows that the two beams that touch in the center Y region are out of phase with each other. This is why the polarization map shows an elliptical polarization state in the overlap region rather than some intermediate linear state.

At this point one can also see that the two beams at the extreme +/- Y values will miss the right-angle prism and will not arrive at the path #1 output as intended for this switch mode. This 5.7% (1.5% + 4.2%) of the flux that is lost at this point is the first contributor to insertion loss due to the OA rotation.

The remaining two beams at the entrance plane to the path #1 output collimator are also pictured (see Fig. 5, top). As given previously, the X polarized beam has 69% of the original energy and the Y polarized beam has 25.3% of the original energy. Unfortunately, because the two beams are displaced relative to one another, it is impossible to couple all of this 94.3% of the original flux into the output fiber.

FIGURE 5. In the presence of +10° of calcite #1 OA rotation, the energy in a plane just prior to the collimating lens of path #1 shows two beams (top). The beam with the least energy is located in the nominal position and therefore couples well into the output fiber. The beam with the most energy is displaced from the nominal position and therefore couples poorly into the output fiber. The two beams are in orthogonal X and Y polarization states (bottom) and have different phases.

In this example, no attempt to improve the coupling into the output fiber was made. In most components, coupling improvement could be obtained during component assembly by translating the output collimator/fiber pair as a group. This coupling improvement by translating the output collimator/fiber pair could easily have been added to the model.

Software analysis is extremely informative and useful. It not only tells what the change in performance is, but why it is changing. The analysis in this article represents less than one day of work by an experienced user and no material or equipment expense beyond the cost of the software and a computer. Clearly, in the process of arriving at a best design, there are significant cost and time benefits to analyzing and tolerancing telecom components using software.

Joe Shiefman is an optical consultant with Shiefman Consulting, 2530 N. Falling Water Court, Tucson, AZ 85749. He can be reached at [email protected].