Colour Cast in the Corners: Separate the Design from the Batch

Ask for three records before the rest of the modules are built. First, the lens’s chief ray angle curve and the sensor’s CRA specification shown together on one chart, so the design match is visible rather than asserted. Second, the assembly record for this lot: the spacer or back focal length setting, the tilt and de-centering measured after assembly, and the IR-cut filter lot. Third, a white target shot on one module from this lot beside the same shot from the approved sample module, same light, same settings, with the corner values written down. Then run one test in your own lab that costs nothing: rotate the lens a half turn and shoot again. What the shading does next tells you whether to blame the lens or the sensor side.

What the chief ray angle has to do with corner colour

A chief ray is the ray from a point off the centre of the frame that passes through the middle of the aperture, which is the definition given in the reference on the chief ray. In a small camera module, the pixels at the corners of the sensor sit under microlenses that are deliberately offset, and each sensor is specified with the angle at which light should arrive at those corners. If the lens delivers light at a different angle than the sensor expects, the corner pixels gather less light and, more visibly, they gather it unevenly across the colour filters. The result is a colour cast that grows towards the corners, usually with a magenta or green tint, while the centre of the frame stays neutral.

That is why the symptom is colour rather than sharpness. A focus error blurs the whole frame; a CRA mismatch colours the corners. So the first question is a design question, and the answer is a pair of curves on one chart: the lens design’s CRA against image height, and the sensor’s specified CRA over the same heights. If those two curves are close, the design is not the problem and the investigation moves to the batch. If they are not, no amount of assembly care will fix it, and the lens choice has to change.

The sample was fine, so the design did not change

Your approved sample did not show the cast, which means the design curves were compatible on the day it was built. What changes between one production lot and the next is not the design; it is the assembly. The spacing between the lens and the sensor, the tilt of the lens axis relative to the sensor plane, the de-centering of the elements inside the barrel, and the filter that sits in front of the sensor all move a little from lot to lot, and all of them push the actual ray angles away from the design values. That is why asking only for the CRA curve would leave you with a document you already trusted.

So split the request in two. The curves confirm the design is right for this sensor. The assembly record explains the difference between the sample and this shipment. Ask for the measured spacer or back focal length, the tilt and de-centering figures from the alignment step, and the filter lot number for the modules in this container. If the maker cannot produce the assembly record per lot, that inability is itself the answer to why one lot behaves differently from another.

Two tests you can run yourself

The first is a white target reading. Shoot a uniform white or neutral grey target with the module under the same light and settings you use in production, then read the colour values at the centre and at each corner. You are looking at two things: the brightness falloff from centre to corner, and the colour shift. A falloff with no hue change is relative illumination, which every lens has and which many systems correct. A hue shift, with the corners reading further from neutral than the centre, is the CRA family of problems. Write the values down and photograph the screen or the chart: centre and four corners, on this lot and on the sample module.

The second is the rotation test, and it is the one that ends the argument. Rotate the lens a half turn in its mount, keeping the sensor and the light unchanged, and shoot the target again. If the shading rotates with the lens, so the corner that was cast is now clean, the asymmetry is in the lens or its mounting. If the cast stays on the same corners of the sensor, the lens is not the source; either the module assembly is tilting the lens, or the sensor’s own shading behaviour is the dominating effect. Either way, the two photographs are a much stronger statement than a claim, and they take ten minutes to produce.

Colour shading is not the same as a dark corner

Buyers often send one photograph and call the problem vignetting, and suppliers answer a vignetting question because that is what they were asked. Keep the two apart in your message. Relative illumination is about how much light reaches the corner; colour shading is about how the corner reads. A module can have almost no brightness falloff and still show a clear hue shift at the corners, and that is the version your customers will notice on a white surface. Say which one you are seeing, with the values from the white target, and the maker will answer the right question.

If the lens has an adjustable iris, shoot one frame stopped down and compare it with the frame wide open. Corner colour that improves markedly when the lens is stopped down points at the angles the extreme rays take, which is consistent with a CRA or de-centering issue. Corner colour that does not change points somewhere else, such as the filter or the sensor’s own calibration. Two frames, same target, ten seconds apart, and the diagnosis narrows before anyone ships anything back.

The record before the rest of the run

Before the remaining modules are assembled, ask for the three records in one pack, with the lot number on them: the design CRA curves together, the assembly record for the lot with spacing, tilt, de-centering and filter lot, and the paired white target shots from the sample and from this lot. Ask the maker to state what they will change for the next lot, and how it will be verified. A shim change, a re-centred element, or a different filter lot each leads to a different verification, and the verification is the part that tells you whether the next lot will be the same or not.

Keep the material for the next order as well. If the answer is that the assembly has to be held to a tolerance the current fixture cannot hold, that is a change in the product or the process, and it belongs in the specification for the next build the way a new drawing would. The custom side of this work, described under custom optics, is where a fixture or a design change is normally quoted, and a batch that cannot be held to the required tolerance is exactly the case for it. The general method for matching a lens to a sensor is set out in the lens selection guide, which is worth re-reading before the next platform is specified.

What to do with the modules you have

Sort them with the white target test rather than by eye. Modules whose corners read close to neutral go forward, and the rest go into one group with their serial numbers recorded. Then decide whether lens shading correction in software is an acceptable route for this product: a calibration that flattens the corners can make a marginal module usable, and it can also hide a variation that will grow as the production spread widens. Make that decision deliberately, with the values in front of you, rather than letting it be made by whoever tunes the first unit.

Keep one module from each group as the record, with its white target photographs and its serial, and keep the rotation test photographs with them. On a camera module, the difference between a design that is wrong and a lot that drifted is invisible in a single image and obvious in a pair of them. The pair you take now is what makes the next discussion short: it either shows the shading rotating with the lens or staying with the sensor, and the maker has to answer the specific question that picture asks.

Write the hold the same day

Send one message to the person who can hold the production line. Name the order, the lens and the sensor. Say the sample module was neutral and modules from this lot cast colour at the corners, and state the hold in one sentence: no further modules assembled or shipped until you have the records and the test results. Name them. First, the lens CRA curve and the sensor CRA specification drawn on one chart. Second, the assembly record for this lot: spacing or back focal length, tilt, de-centering and filter lot. Third, the white target shots from the sample module and from this lot, with centre and corner values. Fourth, the rotation test, showing whether the cast moves with the lens or stays with the sensor.

A release is your note that the curves match within the agreed tolerance, that the next lot will be assembled to a stated change, and that a module built that way passes the white target comparison against the sample. A single CRA curve, a promise to sort the modules, and a software calibration offered as the fix do not release the hold.

Keep the charts, the assembly records and the paired target shots in the order folder with the lot numbers. Corner colour is the kind of defect that a customer notices on a white wall and a test bench can miss, and the file from this lot is what makes the next one measurable rather than arguable.