Build thread - at home measurement of camera UV spectral response

[JMC]

Some data to share :)

 

I took the advice and downloaded Rawdigger to have a look at the RAW files for the images I was making. From that I got the R, G, B and G2 values from within the part of the image where the exit port of the integrating sphere was, and averaged them together to get an average response for the sensor. I then corrected this for the output of the integrating sphere based on my earlier work (this is to correct the for varying intensity of the light source as a function of wavelength). What I end up with is a graph of relative sensitivity of the sensor/lens/filter combination vs wavelength. This is a relative scale, not absolute units, but twice the score should be about twice as sensitive etc, based on a number of assumptions.

 

Enough of the chit chat. What does it show? I have 2 sets of data here. First a normal unmodified EOS 5DSR with Rayfact 105mm lens, and compared with the monochrome multispectral EOS 5DSR with the same lens. Secondly comparison between my ACS modified EOS 7D mk1 (with their UV filter installed, but the Bayer filter remains), vs the EOS 5DSR monochrome multispectral conversion, again both with the Rayfact lens.

 

1. Normal vs monochrome multispectral EOS 5DSR

Here's the graph of relative sensitivity vs wavelength.

 

What can we see here. As expected the unmodified camera is sensitive to light above 400nm, but at 400nm and below it's flatlining in the noise. The monochrome multipspectral conversion, shows slightly more sensitivity in the visible region, but instead of the sharp cutoff at 400nm, the sensitivity slowly drops as the wavelength decreases, down to 280nm, which is the limit of where I can confidently measure. Given the way the graph is dropping for the monochrome conversion it doesn't look like there'd be much sensitivity below 300nm anyway, so the WG280 glass over the front of the sensor looks to be a good option.

 

2. Bayer filter plus UV filter vs monochrome camera plus Baader U

Not quite a single variable experiment, but an EOS 7D mk1 with ACS proprietary UV filter and Bayer filter still installed, vs monochrome multispectral EOS 5DSR (with Bayer filter removed) plus Baader U filter.

 

Both filters look similar in their response, but the monochrome conversion is significantly more sensitive. MaxMax told me to expect about 2.5 stops more sensitivity to UV from removal of the Bayer filter, and that is pretty much what I'm seeing here, given the relative scores of the 2 systems.

 

Overall I'm pretty happy with this as a first attempt at analysis. It's taken into account the changes in the light source variability as a function of wavelength, and seems to give differences and measurements which are in keeping with what's expected.

[Cadmium]

This may be a little off topic, but so both UV pass filters show some transmission above 400nm?

Maybe I am not understanding something here.

[JMC]

It looks that way from the graph Steve, but keep in mind my monochromator, and how wide the peaks are for each of the 20nm chunks. If for instance there was something getting through at 395nm or maybe even 390nm it may give a signal at 400nm. At 420nm and above, like 300nm and below (for the second graph) I'd say that was down in the noise though. I must dig out the Baader U transmission curve I measured before and have a close up look around 400nm.

[JMC]

A more geeky experiment today. 3 plots, no filter, B+W UVIR cut filter, and Baader UVIR cut filter. All taken with the monochrome multispectral EOS 5DSR, Rayfact 105mm UV lens, f4.5, 30s, ISO1600. The aim was to look at what happens when adding in the UVIR cut filters to the overall measured sensitivity.

 

Firstly, the measured relative sensitivity curves for no filter, B+W and Baader UVIR cut filters.

 

As expected the UVIR cut filters drop the light getting through in the 380-420nm region, but they are different to each other. They also are different in how much light the let through in the blue visible part of the spectrum, with the Baader UVIR cut reducing the visible light the least (it is closest to no filter).

 

I've measured transmission of these two filters previously on an Analytical Perkin Elmer spectrometer. Here's the transmission spectra between 280nm and 480nm.

 

They have very different cutoffs in the transmission spectra - The Baader UVIR cut filter reaches 50% transmission around 418nm, and B+W filter at around 391nm. Also the Baader UVIR lets more light through in the visible part of the spectra.

 

Now, can the transmission characteristics of the filters, be used to model the data in the first graph - the measured sensitivity from the camera as derived from the RAW files. For this I took the data from the filter transmission curves, and split it into chunks. I calculated average transmission for each of these chunks. For 480nm I averaged between 475 and 485, for 460nm, averaged 455 to 465 etc down to 280nm. I did this to try and take into account the wavelength spread of the light let through by the monochromator in my setup, but was an arbitrary decision about how wide to go. I then took these 'average' transmission values and multiplied them by the sensitivity of the 'no filter' setup in the first graph, to get modelled curves for the effects of the two filters. I've plotted these modelled curves, against their measured ones in the graph below.

 

The modelled curves are pretty close to the measured ones, which is good to see. Feeling suitable geeked out now :)

[Andy Perrin]

Feeling suitable geeked out now :)

Heh.

 

I think you proved the concept, anyway. You could get a more exact result for the modeled curve (without arbitrary assumptions) by integrating up the product of the filter transmission and the sensitivity and the monochromator curves, then dividing by the integral of the monochromator curve (for each center wavelength).

 

I don't know if you care, though? Was this just meant as a quick check?

[JMC]

Yeah, the modelling piece was just a quick check really. I'd rather measure things than model them where possible.

[JMC]

Another experiment, and a surprising result this time. I compared 3 lenses, the Rayfact 105mm UV lens, a Canon Eos 40mm pancake lens and finally a Canon Eos 50mm f2.5 Macro lens. I set each to f4.5, and the tests were done at ISO1600 and 30s exposure, and looked at them on my moncochrome multispectral Eos 5DSR, and I got the following;

 

The y axis is relative sensitivity of the camera/lens combination. Now this didn't make sense to me at first. The available information for the Rayfact lens says it has about 70-75% transmission over this whole wavelength range. However for the same aperture on the lenses, the two Canon lenses in the visible region are letting in over twice the amount of light. Obviously they can't be 140% transparent. Then I realised about the setup of the experiment. With my rig there is a baffle which I can rest the front of the lens against - it helps me with repositioning, and keeping stray light at bay. To use the Rayfact at the close distance that the equipment needs, I have to set it at pretty much the close focus limit. This then changes the effective f-stop of the lens, as the aperture is further away from the camera. Also the camera is further away from the light source. As a result the light getting through drops and the overall response of the sensor decreases. This happens to the greatest extent for the Rayfact, and very little for the other 2 lenses being tested (the 40mm pancake and 50mm Macro) as they only change size by a small amount. The 'effective' f-stop for the Rayfact is now smaller than the other 2 lenses, and hence the observed difference in relative sensitivity.

 

I guess that the more experienced photographers out there will not be surprised by this at all, but it did surprise me, especially given the extent of it. It does make lens comparison difficult as well, although it is still possible to see where the response drops to zero, and get an idea of usefulness of a lens.

[OlDoinyo]

Have you tried to plot a curve using a pinhole aperture rather than a lens? I realize this would require longer exposure times, but the result might be an interesting bit of reference data.

[JMC]

Have you tried to plot a curve using a pinhole aperture rather than a lens? I realize this would require longer exposure times, but the result might be an interesting bit of reference data.

I'd need a much more powerful light to be able to use a pinhole. At the moment I am at 30s at f4.5 and ISO1600. I have a pinhole I've used before which is f180. Quick calculation and I'm up around 40,000s to take get approximately the same exposure. I'm presuming there is a way of doing this without the lens - just on the bare sensor, but that's a bit too technical for a simple guy like me :)

[ulf]

Could it be possible by not using the integrating sphere, but placing an optical fiber from the monochromator with the end in a controlled way close to the pinhole?

[JMC]

Could it be possible by not using the integrating sphere, but placing an optical fiber from the monochromator with the end in a controlled way close to the pinhole?

Maybe, but I'm guessing I'd need a new adapter to fit on the exit slit of the monochromator to attach the optical fiber to. I would also need to buy a new fiber as well, as my one is permanently attached to the spectrometer. Given how accurate that fiber would need be aligned that wouldn't be something I could make myself unfortunately. Alternatively I suppose I could just place the collimator next to the output slit and see whether that works. If it comes to taking the system to bits again I'll give it a go.

[ulf]

Oh, I thought the monochromator had a SMA-connector at the output. My misstake.

 

Be aware of that there is a wavelength dependent focus-shift in the collimators due to the dispersion of the fused silica.

It might not be big enough to cause problems within UV only.

I have seen problems I suspect are caused by this when trying to measure transmission of lenses and including visual wavelengths.

[JMC]

No, unfortunately it just has a slit Ulf, no connector. Thanks for the advice on the collimators, I didn't know that.

[Andrea B.]

Still reading with interest!

 

There is so much to learn about all this! Whew!

I still harbor the hope that someday I could actually try this myself.

[JMC]

Here's a bit more of an extreme manipulation of the data, but is in itself interesting. The idea here is to back calculate to get some idea of lens transmission as a function of wavelength. To do this I assumed the Rayfact 105mm transmitted the same amount of light at each wavelength from 280nm to 480nm. This is a reasonable assumption (at least as an approximation). I then worked out the multiplication factor to correct the scores at each wavelength back up to what it was at 480nm. I then normalised it to 1, so this gives me a information on the lens transmission as a function of wavelength, and removes the impact of the sensor losing sensitivity at the shorter wavelenths. So basically the Rayfact would have a score of 1 at each wavelength as it is assumed to transmit the same from 280nm to 480nm. I then repeated this for other lenses I've assessed, using the correction factors derived from the Rayfact;

 

Canon Eos 40mm pancake lens

Canon Eos 50mm macro lens

Lithagon Enna 35mm f3.5 lens

Soligor 35mm f3.5 lens

Asahi Ultra Achromatic (UAT) 85mm f4.5

 

This assumes then at each of the lenses transmits the same amount of light at 480nm (not ideal, but good enough to allow comparison). This is what the plot of them all looks like;

 

What's this telling us? Firstly, ignore the points at the far left 280nm and 300nm. This is pretty much down in the noise for the measures, and to see them rising up again is an artifact of the data analysis rather than an actual increase in transmission. This important thing is to start at 480nm and work down to shorter wavelengths. Secondly, the UAT is similar to the Rayfact as expected - it has good transmission at the really short wavelengths. Thirdly, not all the other lenses are the same. The Canon 50mm macro lens is the worst here for UV - it drops the fastest as you go into the UV, and is pretty much useless below 380nm. The Canon 40mm pancake lens is better, and I've used this quite a bit on my UV cameras (ideal as it retains autofocus). The Soligor is a little better than the Eos 40mm pancake lens, letting more light through at 360nm. The Lithagon seems to go the deepest into the UV, and gets to 320nm before it levels off. I believe the Lithagon Enna lenses have been discussed before as being good for UV photography, and this would seem to bear that out.

 

Now, I'm not saying this replaces actually lens transmission measurements on a dedicated spectrometer, but I thought it was an interesting way of looking at the data from my setup so wanted to share it.

[Shane]

Maybe, but I'm guessing I'd need a new adapter to fit on the exit slit of the monochromator to attach the optical fiber to. I would also need to buy a new fiber as well, as my one is permanently attached to the spectrometer.

The output of a monochromator resulting from a diffraction grating and bandwidth slit is rectangular. Using a round fiber optic directly on the slit results in significant output loss. An appropriate lens can be used to focus this a little better but significant improvements can be made using a fiber optic with a rectangular end (Oriel used to make these and may still possibly make them).

[JMC]

Ok thanks Shane.

[JMC]

Added a new lens to the mix today - the UV Sonnar. I had to get some extension tubes in, as the close focus for the UV Sonnar is 1.8m and my rig is only about 50cm long. I've treated it in the same way as the other lens and calculated an approximate transmission curve for it from the data (and added it to the previous graph). These are normalised to the same transmission at 480nm, so is not a "% transmission curve". Here's the updated graph with the UV Sonnar added;

 

Good thing is it looks to be behaving similarly to the Rayfact and UAT lens - transmitting well down to atleast 280nm. I must admit I didn't know quite what to expect with it. I spoke with Zeiss, to try and find out some more about the lens, and they said something quite curious to me. This is what they said;

 

"In total Zeiss produces 370 of the UV- Sonnar. And especially 4 items for the NASA directly with some customization. Older UV- Planar lenses were developed for extraterrestrial application at NASA, yes. But the Hasselblad UV Sonnar 4,3/ 105 was produced for all kind of scientific purposes. It has a correction for blue and near UV , where normal glass show zero transmission. The transmission of all Zeiss UV lenses was limited at 320nm, more or less. The lens design contains a mixture out of calzium fluorite and ordinary glass with good UV Transmission . One very rare lens additionally use quartz glass. With that and no normal glass you can reach 200nm."

 

I've added the underlining, but it was that that really surprised me, I thought they were all CaF2 and Quartz, rather than UV glass. My transmission curve adds to the confusion as it doesn't seem to drop below 320nm and carries on like the Rayfact and UAT lenses. Maybe this mysterious lens will keep some of its mysteries for a while yet.

[Andrea B.]

I have from Marco Cavina's website that the Zeiss UV-Planar, for example, uses LiF2, CaF2, Schott F1 flint glass and Schott FK-3 fluor crown glass.

 

My UV-Planar supposedly has about 25% transmission at 300 nm reaching 70% transmission at 325 nm. I don't have a chart, but it seems like the 320 nm mentioned above is a reasonable cut-in point to mention.

 

Thing is, it seems that Zeiss made some variations in these UV sets. I have one statement that a UV-Planar is 5 elements in 3 groups and another which says 6 elements in 4 groups.

Well, maybe Dr Klaus has additional info?

 

 

It has a correction for blue and near UV, where normal glass show zero transmission.

I think that Zeiss was trying to say that the UV-Sonnar is aberration corrected between near UV and blue?

But it is a bit of a mystery as to what is "normal" glass. :)

[JMC]

Today we have a Canon vs Nikon showdown. I have 2 cameras which have been modified by Advanced Camera Services Ltd in the UK for taking UV images - a Nikon d810 and a Canon Eos 7D mk1. Both of these have the same ACS filter in them. I've run both of them on my rig and assessed response as a function of wavelength, keeping everything else the same (lens Rayfact 105mm, f4.5, ISO6400, 30s exposure, Canon images saved as CR2 raw, Nikon as 14bit lossless compression). Data was analysed in Rawdigger. Here's what the 2 look like;

 

There looked to be differences between them, so I then took this, and modelled them as normal distributions - ok, perhaps not the best model, but I do not know the transmission curve of the ACS filter. This is what the modelled curves look like;

 

It definitely looks like the 2 cameras aren't behaving identically - the Nikon has max sensitivity at about 376nm, and the Canon lower, at about 371nm. The Canon has lower max sensitivity, but looks to go a bit deeper than the Nikon and be sensitive over a slightly wider range of wavelengths. it was interesting - when I took photos of a white PTFE tile in sunlight with the 2 cameras, they looked to have almost identical sensitivity. The areas under the curves here would bear that out, as they are very similar. But the distribution of sensitivity as a function of wavelength is different between the two. I'm guessing this is down to i) differences in sensor sensitivity and ii) differences in Bayer filter composition/thickness.

 

As an aside, I think the signal at 420nm in the modelled curves is an artifact of the forced normal distribution in the model, rather than a real effect as the raw curves look more asymmetric than true normal distributions.

 

EDIT - graph captions updated at the request of Andrea.

[Andy Perrin]

Could use a Weibull distribution- they can be skew.

[Shane]

It definitely looks like the 2 cameras aren't behaving identically

 

Why would you expect them to behave identically. It's like comparing an apple with an orange, sure they are both fruits but they don't taste the same.

Nikon and Canon may both use CMOS type sensors but they have architectural, dimensional and materials differences, all of which effect the transmission/absorption of light.

[JMC]

 

 

Why would you expect them to behave identically. It's like comparing an apple with an orange, sure they are both fruits but they don't taste the same.

Nikon and Canon may both use CMOS type sensors but they have architectural, dimensional and materials differences, all of which effect the transmission/absorption of light.

 

When I imaged a PTFE tile in sunlight, the two cameras were almost identical in their behaviour, with practically the same brightness of the tile for a given exposure. Hence my going in hypothesis was that their response as a function of wavelength in this test would also be almost identical. Given the absence of credible data on the subject, the best way I can learn is through testing.

[Shane]

Given the absence of credible data on the subject, the best way I can learn is through testing.

Certainly nothing wrong with that :)

[Andrea B.]

But I don't like it that those two cameras have an internal filter about which we know nothing? With that internal filter we really aren't learning anything about the sensors! We are learning instead about that internal filter. What the heck did ACS use for a filter there anyway? It does seem to pass quite a lot of blue.

 

I worry that viewers might be mislead because the charts are not labeled to indicate that the cameras are filtered. Perhaps you could add such a label in the legend? "Canon 7D with ACS Filter", or something similar? What do you think about this, Jonathan??