The dimensions passed. The image did not. 

How a systematic registration bias inside a conforming sensor-assembly chain displaced the circular field of view 

The image was sharp, but it was not centered 

The image was clear and stable.

The circular field of view was not centered on the reference overlay.

On a 32-inch 4K display, the video software generated concentric circles and crosshairs. The pattern provided a fixed reference against which different camera handpieces could be compared.

It was not an optical-bench measurement.

It was an assembly-level screening method: simple enough to repeat, but sensitive enough to reveal a field displacement that was difficult to judge from the image alone.

The circular field consistently appeared above and to the right of the reference center.

The symptom was visible.

The responsible boundary was not.

Figure 1. Concentric-circle overlay used to evaluate field-of-view centering on a 32-inch 4K display. The circular field remained clear but was displaced upward and to the right of the reference center.

Adjustment kept the first design moving 

The first response was practical.

The sensor structure was made adjustable. With the reference overlay visible, the sensor position could be moved incrementally until the circular field returned toward the center.

The method worked.

It allowed individual handpieces to be aligned and kept development moving while other product issues were being addressed.

But adjustment was containment, not closure.

The result depended on adjustment resolution, operator judgment and whether the sensor remained in position during final fixation.

More importantly, it corrected the finished image without identifying why the field had moved.

The problem remained open over roughly six months of development. This was not six months of uninterrupted investigation. Other issues took priority, individual units could still be aligned and local improvements were introduced as the design evolved.

The product moved forward.

The centering problem remained inside it.

A second handpiece changed the question 

The adjustable approach remained workable until the same field-centering symptom appeared in a second camera-handpiece architecture.

That recurrence changed the meaning of the problem.

It could no longer be treated as one isolated machining error, one poorly assembled unit or one weakness unique to the first handpiece.

Two architectures were exposing the same type of failure.

The common issue was more likely to exist somewhere in the relationship between:

• the optical mounting axis;

• the machined handpiece datums;

• the PCB location;

• the sensor assembly;

• and the final circular field.

At that point, continuing to adjust completed units was no longer an adequate answer.

The centering chain itself had to be investigated.

Ten handpieces showed the same directional pattern 

A controlled screening test was performed on ten camera handpieces using the same 32-inch 4K display, the same optical configuration and the same concentric-circle reference.

For each unit, the horizontal and vertical displacement of the circular field was measured and converted through the internal image-to-mechanical calibration method.

The magnitudes varied.

The direction changed very little.

All ten circular fields were displaced upward and to the right.

The first mechanical checks passed 

I began at the most direct mechanical boundary: the structure locating the sensor assembly inside the handpiece.

If the locating holes, mounting surfaces or PCB-support features were displaced, the sensor assembly could move relative to the optical axis even when the rest of the imaging chain was correct.

The relevant locating dimensions were inspected.

They were within specification.

The next boundary was the front optical interface.

I checked the relationship between the C-mount internal thread (1″-32 UN-2B) and the mechanical datum structure used to locate the sensor assembly. The applicable drawing tolerances for radial location, runout, and centering were specified at 0.02 mm. 

Those relationships also met their requirements.

The measurements were not wrong.

The handpiece locating features had passed.

The optical mounting interface had passed.

Neither inspection identified enough error to explain the repeated field displacement.

The dimensions had answered the local questions written on their drawings.

They had not yet answered where the complete field center would land after assembly.

A shorter optical path removed several suspects 

The complete imaging chain contained several interfaces:

• the rigid endoscope;

• the optical adapter;

• the C-mount (1-32 UN-2A/2B) interface ;

• the machined handpiece components;

• the sensor-support structure;

• the PCB;

• and the image sensor assembly.

Any of them could influence the final field position.

To shorten the chain, I designed a dedicated short-path eyepiece adapter.

Its purpose was not to reproduce the complete clinical optical path. It was to remove everything that did not need to be present.

Critical alignment features on the adapter were controlled with 0.02 mm limits for position, runout and radial centering. This created the shortest practical path between the camera handpiece and a controlled optical reference.

The handpieces were tested again with the same display overlay.

The displacement remained.

The fixture did not identify the root cause by itself.

It removed several plausible causes at once.

The longer rigid-endoscope path and its intermediate interfaces could no longer explain the common directional pattern.

The remaining investigation moved back toward the sensor assembly.

The measurements converged on the sensor-assembly chain 

The next measurements examined the relationship between:

• the PCB locating structure;

• the external sensor package;

• the package-placement reference;

• and the mechanical datums used by the handpiece.

The inspected results remained inside their local requirements.

There was still no single rejected dimension.

But the investigation had now removed much of the external optical path, confirmed the key machined interfaces and reproduced the same directional pattern across the screened population.

The unresolved relationship lay between the PCB-based mechanical reference and the active image center.

That relationship contained more than package placement on the PCB.

It also included:

• PCB fiducial definition and recognition;

• package-to-PCB placement;

• sensor-package dimensional variation;

• internal die-to-package registration;

• and the position of the active pixel array relative to the external package reference.

The available measurements did not independently separate external package placement from internal die-to-package registration.

A conforming package could be placed correctly on the PCB while its active pixel array still occupied an allowed offset inside the package.

Conversely, the internal die could be well registered while the package-placement distribution sat toward one side of the PCB placement window.

Both conditions could produce similar screen-level evidence.

It would therefore have been too strong to assign the entire failure to the SMT placement process alone.

What the evidence supported was a broader conclusion:

A systematic registration bias existed somewhere inside the sensor-assembly chain, between the PCB-based reference and the active image center.

One bias set the direction; the legal tolerances set the spread 

This distinction explained the ten-unit pattern.

The common registration bias established the dominant displacement direction.

The remaining permitted variations determined how far each completed unit moved from that biased center.

Those remaining terms included:

• machined handpiece location;

• optical-mount alignment;

• PCB position;

• adapter runout;

• package placement;

• and internal sensor registration.

They did not all need to reach the same side of their limits.

They only needed to vary around a population whose functional center had already been displaced.

That is why every field moved upward and to the right, while the displacement magnitude still differed from unit to unit.

The drawings were controlling the spread of several local features.

They were not controlling the common offset of the complete active-image relationship.

Every part could pass its own inspection.

The assembled field could still miss its functional center.

A small sensor-side movement was large on the display 

The mechanical displacement appeared small when viewed beside a machined housing.

At the sensor plane, it was not optically small.

Image-sensor pixels are measured in micrometers. A movement of only a few hundredths of a millimeter can therefore move the displayed field by many pixels.

In this system, a sensor-side adjustment on the order of 0.10 mm produced a large and immediately visible movement on the 32-inch 4K display.

The mechanical correction direction was not inferred directly from the screen direction.

Optical inversion, software rotation and coordinate transformation can separate display coordinates from PCB coordinates.

A known sensor movement was applied.

The resulting field movement was observed.

That experiment established the mechanical direction required to compensate for the displayed bias.

The design nominal was shifted 

The ten-unit screening provided a common field-displacement vector.

That vector represented the center of a biased population, not ten identical units.

I shifted the design nominal for the sensor position incrementally in the empirically determined compensating direction.

New samples were built and evaluated using:

• the same 4K display;

• the same concentric-circle overlay;

• the same optical configuration;

• and the same centering method.

The correction was not completed in one step.

The nominal position was moved.

Samples were built.

The resulting field position was checked.

The remaining offset informed the next iteration.

This was fundamentally different from adjusting every finished handpiece.

Individual adjustment corrected one unit after assembly.

Changing the design nominal moved the expected center of subsequent builds.

Compensation did not prove that the process had been recentered 

The design change compensated for the registration bias observed during the investigation.

It did not, by itself, prove that the underlying manufacturing process had been recentered.

That distinction mattered.

A design compensation depends on the observed bias remaining sufficiently stable.

Changes in placement equipment, program settings, fiducial recognition, sensor lot, package construction or supplier process could alter that relationship.

The bias could become smaller.

It could move.

It could reverse.

The compensated design therefore could not rely only on the assumption that the historical offset would remain unchanged.

The assembly-level centering check remained necessary.

The concentric-circle overlay was retained not merely as an adjustment aid, but as a verification method for the assumption behind the compensation.

The design nominal had changed.

The final field still had to prove where it landed.

Control was rebuilt around the field center 

The answer was not to tighten every mechanical tolerance indiscriminately.

That would have increased machining cost, inspection burden and supplier risk without directly controlling the active image center.

The functional chain was instead treated as one relationship:

Optical mounting axis
→ machined handpiece datums
→ PCB locating features
→ sensor-package placement
→ die-to-package registration
→ active image center
→ circular field-of-view center

The short-path eyepiece adapter remained available for separating handpiece alignment from the longer optical chain.

The machined and optical features continued to be inspected against their drawing requirements.

The display overlay supplied the assembly-level evidence that those local inspections could not provide by themselves.

Verification returned to the same field reference 

The corrected builds were evaluated under the same conditions that had exposed the original problem:

• the same display;

• the same concentric-circle reference;

• the same optical configuration;

• and the same field-centering method.

Before the change, the screened handpieces shared a clear upward-and-right population bias.

After the design-nominal correction, that common directional bias did not recur within the corrected verification set.

The evaluated circular fields met the internal assembly-level centering criterion without relying on individual sensor adjustment as the primary production method.

This was not a full optical qualification program or a statistical process-capability study.

It was a focused verification of the failure mode that had driven the work.

The final field position became controllable only when the center was treated as a property of the complete optomechanical chain—not of any one drawing.