Why lens MTF falls off design magnification, and how to get it back

A lens that tested beautifully on the bench can disappoint on your line: focused perfectly, lit correctly, still soft where it matters. The usual cause is not a defective part. The lens was optimized at one magnification and working distance, and your station sits at another. This guide explains why MTF collapses away from that design point, works the conjugate math so you can predict it with your own numbers, and compares the three design answers — range-balanced, fixed-magnification and floating-element — with a decision for each case.

The short answer

A lens delivers its printed MTF only near the conjugate set it was balanced for: one magnification, one working distance, one sensor format, one spectral band. Refocus the same optics to different conjugates and every aberration term shifts by its own amount, so the balance that produced high contrast at the design point stops being the right balance. Contrast at high spatial frequencies falls first, and it falls faster the farther you move.

The decision follows. If your station has a fixed part and fixed working distance, order a lens whose design magnification matches yours instead of refocusing a general-purpose one. If the field of view must change by a factor of two or more, specify a floating-element design that re-spaces its internal groups, or swap lenses per part family. If you stay inside a rated range, a range-balanced FA lens is the right trade. The worked example below puts numbers on all three cases.

What a lens is optimized against

A designer fixes a target before optimizing: one magnification, one working distance, one field size, one wavelength band. The optimizer then spends its correction budget — surface curvatures, glass types, element spacings — suppressing spherical aberration, coma, astigmatism, field curvature and chromatic error at exactly that conjugate set. MTF is the scorecard: image contrast as a function of spatial frequency, measured in line pairs per millimeter at the sensor.

The conjugates follow from two distances and the thin-lens relations:

Here s is the object-side distance, s' the image-side distance, f the focal length and m the magnification. No aberration scales the same way with both distances. Spherical aberration and coma follow the aperture and the conjugate ratio. Astigmatism and field curvature grow with field height. Chromatic focus shift follows the spectrum. Change the magnification and each term moves by a different amount, so the balance is gone even though nothing about the glass has changed.

Because one custom lens per station is not affordable, most catalog lenses are balanced across a rated range of working distances and magnifications instead of at a single point. That 'good everywhere, best nowhere' compromise is exactly right for general factory work, and exactly wrong when an application needs near-maximum contrast at one frequency.

Worked example: refocusing a 0.33x line-scan lens to 1x

Take a line-scan station: a 12k sensor array with 5 µm pixels and a 50 mm lens designed for m = 1/3.

Start at the design point. The array measures 12,000 × 5 µm = 60 mm, so the Nyquist frequency is 1/(2 × 0.005 mm) = 100 lp/mm: one line pair spans two pixels, and one millimeter holds 100 of them. The lens only has to hold contrast up to 100 lp/mm, because the sensor cannot represent anything finer. The field of view is 60 mm ÷ (1/3) = 180 mm — the same working distance and field of view formula that sizes a blister station. Object-side sampling follows too: 5 µm ÷ 0.3333 = 15 µm on the part, which is how you calculate object space pixel size from magnification. The geometry: s = 50 × (1 + 1/3)/(1/3) = 200 mm working distance, s' = 50 × (4/3) = 66.7 mm image-side distance.

Now the part changes and you refocus to m = 1. The field of view drops to 60 mm ÷ 1 = 60 mm. The working distance halves to s = 50 × 2 = 100 mm. The image-side distance must grow from 66.7 mm to 100 mm: Δs' = f(m2 - m1) = 50 × (1 - 0.333) ≈ 33 mm. A C-mount focus barrel typically travels a few millimeters, so this move needs extension rings or a macro lens before the optics are even discussed.

Focused perfectly, the lens still objects. At f/2.8 the diffraction cutoff is 1/(0.00055 mm × 2.8) = 649 lp/mm at 550 nm; call it 650. A perfect lens would still pass about 0.80 MTF at 100 lp/mm: v/cutoff = 100/650 = 0.154, and MTF = (2/π)[acos(0.154) - 0.154·√(1 - 0.154²)] = 0.637 × (1.416 - 0.152) = 0.80. Diffraction is not the problem. The aberration balance tuned for 3:1 conjugates is now working at 1:1, and the correction that delivered 100 lp/mm contrast at 0.33x is spent on different terms at the new position. Refocus removes defocus; it does not re-strike the balance. That is why the same glass can test superb at its design point and mediocre 100 mm closer.

Run your own station through the same relations:

What moves the loss, and how fast

Pixel pitch sets the bar. A 5 µm sensor asks for 100 lp/mm, a 3.45 µm sensor for 145 lp/mm, a 2.5 µm sensor for 200 lp/mm. An off-design curve falls with frequency, so the same refocus that survived at 100 lp/mm can sink at 145. As sensors add pixels, the range over which a range-balanced lens holds acceptable contrast shrinks — the Nyquist limit and camera resolution guide works out which sensor a given lens can serve.

Field height decides where the loss shows first. Astigmatism, field curvature and lateral chromatic error grow toward the corners, so an off-design lens loses the corners long before the center. If your feature sits at the field edge, read the corner curve, not the on-axis one. MTF is contrast at a frequency, which is exactly what lens contrast and resolution: evaluating fine detail covers.

The iris trades one loss for another. Stopping down trims the aperture-dependent terms — spherical aberration and coma — and recovers some contrast, while the diffraction ceiling comes down with it: at f/8, 1/(0.00055 × 8) ≈ 227 lp/mm, still comfortably above 100. Expect modest gains; astigmatism and field curvature ignore the iris. The spectrum matters as well: a lens run under white light plus NIR carries an extra chromatic focus shift away from its design band.

Three design answers to one problem

A floating design keeps the glass and changes the air between element groups. Spacing is one of the designer's own variables, so letting the mechanics adjust it recovers much of what a fixed spacing must give up: same elements, re-optimized balance at each stated magnification. It is still not a custom lens. Performance falls off quickly outside its own settings, and each magnification is a separate assembly, which is why floating lenses cost more than a plain FA lens and exist in more variants.

Datasheets that state one magnification and one working distance are telling you the design point is the product:

The LS-TCL-0574 is the measurement-grade version of this logic: an object-side telecentric lens specified at 0.4x and 227 mm WD for a 1.2-inch sensor. A telecentric design has the least refocus freedom of all, because its magnification is part of the correction — you order the magnification the gauge needs rather than dialing one in.

For stations that live inside a rated range, the range-balanced FA lens stays the economic answer, and most factory inspection at moderate pixel sizes falls there. The classic shape of that choice is a 35 mm FA lens for 2/3-inch, 6 MP class sensors:

Browse machine vision lenses by sensor format and nominal magnification, or filter both at once in the lens finder.

A decision for each case

Fixed part, fixed working distance, measurement or fine-defect work. Decision: buy the design point. A telecentric or fixed-magnification lens at exactly your m and WD, not something refocused near it. The mismatch costs contrast precisely at the frequencies measurement algorithms sample.

One camera, several part sizes, field of view changing 2x or more. Decision: a floating-element design if changeovers are routine, a lens change per part family if they are rare. Do not refocus one range-balanced lens across a 3x magnification span; the ends of the span are exactly where its balance fails.

General inspection, working distance stable inside a rated range, pixels 3.45 µm or larger. Decision: a range-balanced FA lens is sufficient. Confirm the working distance sits inside the rated range rather than at its edge, and that the published curve reaches your Nyquist frequency.

Line scan is the first case in different clothes. A 60 mm array at 0.33x means a 180 mm field of view, and line-scan lenses are specified by that design magnification for a reason: 5 µm pixels leave no spare contrast to spend.

Checks before you order

• Compute Nyquist from pixel pitch, 1/(2p), and write it on the station spec. It is the only frequency the lens must serve.
• Read the MTF curve at Nyquist, at your field height, at your magnification. Published curves are plotted at nominal, so if yours differs, ask for the curve at your m — how to read lens performance curves for vision inspection explains what the axes and field heights mean.
• Check focus travel with Δs' = f(m2 - m1). If the barrel cannot cover the gap between design m and yours, the refocus does not physically exist; you need a different lens, not more force on the ring.
• Re-check distortion at the final focus setting if the task measures. Distortion is specified at the design conjugates and can shift slightly when you refocus; geometric vs TV distortion sorts out which datasheet number applies.

When the usual answers are wrong

Just refocus it. Focus removes defocus only. A perfectly focused off-design lens can still miss Nyquist contrast by a wide margin, as the worked example shows.

Add extension rings and it will reach 1x. Rings supply the 33 mm of image-side distance; they do not re-strike the aberration balance. For sustained high magnification, a macro lens for machine vision, chosen by magnification is the right part.

Buy the sharpest lens in the catalog. Sharpness is not portable. A lens superb at 0.33x is superb there; the question is never which lens is sharpest but which is sharpest at your magnification.

The published curve is conservative. It is published at nominal. Off nominal, the truth is worse, not better.

One-minute checklist

• Write down sensor format, pixel pitch, field of view and working distance first; m and Nyquist follow from them.
• Nyquist = 1/(2p) in lp/mm. The lens needs contrast there, not everywhere.
• Check MTF at Nyquist, at your field height, at your magnification — not the nominal curve alone.
• Compute Δs' = f(m2 - m1) and compare it with the barrel travel you actually have.
• Field of view must change by 2x or more: floating-element design or a lens per FOV, not one refocused lens.
• Measurement station: telecentric at your exact magnification.

If you are sizing glass now, the lens configurator walks focal length, sensor and working distance in order, and the lenses topic page collects the guides behind every check above. Bring a datasheet curve and your station numbers — magnification, working distance, pixel pitch — to the contact page, and we will read the design point with you.