Subaperture Stitching Interferometry for Large Optics

Picture a 300 mm flat mirror sitting on the bench, with an interferometer that can only see a patch roughly 100 mm across. So how do you verify the flatness of the entire surface? That’s the exact challenge subaperture stitching interferometry was developed to handle. We’ve run this kind of job plenty of times in our own lab, and once you break the steps down, the method is a lot less intimidating than most people assume.

What subaperture stitching interferometry actually does

The underlying idea is straightforward. Instead of capturing a large or steeply curved surface in one shot, you measure it as a series of overlapping small patches. Software then mathematically joins those patches into a single full-surface height map.

A useful comparison is photographing a long wall section by section and then assembling the shots into a panorama. Every individual frame is sharp, and the finished image covers the whole wall. Optics works the same way, except the frames are phase measurements and the panorama is a precise height map.

The real benefit is that a compact interferometer can characterize optics far larger than its own aperture. The cost is more alignment effort and longer processing. In return, you skip buying a reference optic matching the part’s size, which is usually expensive and often impossible to source.

Why a single capture falls short

An interferometer compares light reflected off your part with a known reference wavefront. What you see is a pattern of fringes, those alternating bright and dark bands. On a gently curved surface the fringes are wide and easy to read.

On a steep slope they bunch together. Once the spacing drops below what the camera pixels can resolve, the detector loses them, and that data is gone. Field of view and pixel count together set a hard limit on what any single capture can record.

The traditional fix is a full-aperture reference, but for a 1 m mirror that reference is large, costly, and rarely available off the shelf. Stitching becomes the practical route whenever that limit sits well below the size of the part you need to measure.

The workflow from start to finish

The process is more routine than it sounds. First, divide the surface into a grid of sub-apertures, deliberately overlapping each neighbor. That overlap is what lets the software work out how one patch relates to the next.

Then measure each sub-aperture by moving either the part or the interferometer head so every patch lands in front of the optics. Each patch gets fitted to a common reference, and the solver computes the relative offsets and tilts between neighbors. Finally, everything merges into one height map.

Most of the real effort lives in the solve and the merge. A weak solve shows up as visible seams in the output, so we always inspect those seams before accepting the map.

Choosing a sensible overlap

Overlap is the setting that most often determines whether a stitch succeeds. Too little leaves the solver with weak constraints, and any error gets amplified at each join. Too much simply adds scan time without a meaningful gain in accuracy.

From our experience, somewhere around 10 to 30 percent overlap between adjacent sub-apertures handles most jobs. The right figure depends on how steep the surface is. Steeper regions call for denser sub-apertures, which reduces residual slope error but stretches the total scan time.

That is a deliberate trade-off to make for each part. Don’t treat any number as a universal default.

Where positioning errors come from

Stage errors are the usual culprits. Lateral and axial motion mistakes often show up in the stitched map as spurious power or astigmatism the part never actually had. If a clean-looking result suddenly displays a faint saddle shape, check the stages before blaming the optic.

Thermal drift during a long multi-patch scan can bias later sub-apertures relative to earlier ones. Vibration from nearby equipment adds noise that the solver can’t distinguish from real surface features.

None of these problems is exotic, but they build up over a long acquisition. That’s why we keep the room quiet and let the stages warm up before any serious run. For background on controlling vibration, our notes on vibration isolation basics for a small optics lab describe the setup we rely on.

A real case: a 300 mm flat with a steep off-axis segment

Here’s an example from our own bench. We needed to characterize a 300 mm flat along with a steep off-axis segment using a small interferometer. We sized the sub-aperture grid so each patch stayed comfortably within the fringe-resolving limit, set overlap near 20 percent, and sampled more densely around the steepest zones.

The first pass, run without tilt compensation in the solver, left obvious residuals. The largest errors clustered at the part edges and on the steepest slopes, exactly where trouble tends to appear. Adding a tilt-compensation step reduced those residuals considerably, with only a modest increase in processing time.

The lesson for us was that edges and steep regions deserve the most attention when you plan the scan.

Stitching versus a full-aperture reference

Stitching earns its place when a reference optic would cost more than the measurement project itself. For a one-off large part, buying a full-aperture reference rarely makes financial sense.

The math changes for a repeatable production part. Once a full-aperture null setup is in place, it can be both faster and more accurate. When we weigh the two options, we look at part size, slope, required uncertainty, throughput, and budget. Those answers usually settle the decision without much debate.

Verifying a stitched result before you rely on it

A stitched map can look convincing even when it’s wrong, so verification isn’t optional. We run a calibrated artifact or a known-shape test part through the entire pipeline. If the artifact comes back with the correct form, we have reasonable confidence in the process.

We also repeat the measurement across several part rotations. True surface features stay fixed as the part turns, while systematic artifacts tend to shift or change. Comparing those runs helps separate the two.

Last, we write an uncertainty budget so the stitched figure goes out with limits we can defend if questioned. Teams that also check measurements against coordinate systems may find our piece on laser trackers for large volume metrology in spacecraft assembly a useful comparison.

Common mistakes that waste time

The biggest error is accepting solver output without looking at it. Always inspect the seams visually before trusting a map. A seam caught at this stage takes minutes to fix, while one found after the report is written costs far more.

Environmental drift and vibration are the next traps, since long acquisitions give small disturbances plenty of time to accumulate. Inconsistent sub-aperture sizes also make convergence harder. Choose a grid, keep it regular where possible, and stick with it for the whole job.

Frequently asked questions about subaperture stitching

Is subaperture stitching accurate enough for high-precision optics?

It can be, provided you control alignment, drift, and overlap carefully. For the tightest tolerances, many teams still rely on a full-aperture null test. Stitching works very well for characterization when the budget is limited.

How many sub-apertures does a typical large mirror need?

There’s no single answer. It depends on part size, surface steepness, and the usable field of your interferometer. Start with a coarse grid, check the residuals, and add density only where errors concentrate.

Can I stitch data captured on different days?

You can, but it carries risk. Temperature shifts, stage repeatability, and setup changes between sessions all add error. If you have no choice, measure a known artifact at the start of each session and fold that drift into your uncertainty budget.

Does stitching work on aspheric surfaces?

Yes, and aspheres are among its strongest applications because their slope changes across the aperture. The usual approach is to stitch the patches and then compare the result against the design surface, paying close attention to the steep outer zones where residuals tend to be largest.