Jump to content
UltravioletPhotography

Foveon Sensor Transmission Charts


dabateman

Recommended Posts

EDITOR'S NOTE: This discussion of the Foveon sensor was split from the topic about Bayer sensors seen here: Bayer filter transmission in UV and IR. Dabateman's first paragraph below refers to that topic.

This topic contains Foveon sensor transmission charts gathered from several sources. Their accuracy cannot be guaranteed by UVP as I'm sure readers will understand, especially in examples where bench grade spectrometric equipment has not been used.


 

Wow, Thank you Johnathan. That curve is very interesting. Not only in the UV part, but the IR section as well.

We are basically only using 360nm to 400nm section of our filters, based on the sensor sensitivity. Something I will have to think more on as I also look at lens options.

 

MaxMax used to have the spectral curves for sensors on their website and the link is still there. I found the version I saved. I have included them below. NOTE, these are copied from LDP website. I did not take these or know how they were captured.

 

post-188-0-05977800-1526937139.jpg

 

post-188-0-33546000-1526937165.jpg

 

post-188-0-37305900-1526937175.jpg

 

post-188-0-71660300-1526937195.jpg

 

Andrea, if there is any reason for why these should not be here, then please Delete them. I don't want to cause any problems. I grabbed these in 2009, Just as I save most interesting images I find on the internet.

Link to comment

The curves I have for sigma cameras is from published work.

The Sigma Sd1m response was published in: Fent, L. And Meldrum, A. A Foveon Sensor/Green-Pass Filter Technique for Direct Exposure of Traditional False Color Images. Journal of Imaging. 2016 (2) 14.

 

 

The Sd14 curves I have were published in a thesis:

DETECTING NEAR-UV AND NEAR-IR WAVELENGTHS WITH THE FOVEON IMAGE SENSOR by Cheak Seck Fai, 2004.

post-188-0-67296300-1526940758.png

post-188-0-47377900-1526940809.png

post-188-0-02342700-1526940826.png

post-188-0-40604400-1526940844.png

Link to comment
I read C S Fai's paper and have some serious concerns about the equipment and methodology he used for the UV.
Link to comment

I think the method used to determine camera response was valid. Using a calibrated weak light source, a monochrometer to isolate wavelengths and the camera. This is the set up Jonathan, here is using. The difference is that in the under 450nm region the student boosted the blue channel output by using a specific white balance setting, as we do when shooting photos. This is specified and was calibrated. So I think these response curves are valid. However, the latter images in the thesis to add significant, with photographing soldiers in brush with filters is a different matter.

Not to take over this post, but to add some comments on foveon sensors. The first commercial sigma camera Sd9, was a bare sensor with fixed uv/ir cut filter on the sensor. This is not easily removable. The next Sd10 added micro lenses to the surface of the sensor to aid in better imaging and the block filter was now more easily removed with some screws from the mount. The Sd14 had increase in pixel density keeping the micro lenses, the block filter now pops out and is the last adobe software supported camera. The Sd15 used a metal layer on the surface of the sensor, to improve high Iso response. This not only allowed for better high isos, but blocks the uv usage of this camera. The Sd14 is only usable at base iso 100, above that is just gain with lots of noise. The Sd1 had a completely different redesigned larger sensor. The Sd9 to sd15 use 1.7 crop sensor, where as the Meril cameras used a 1.5x crop sensor. This sensor also has the metal layer on top to improve higher iso. I am not sure the uv sensitivity of the Merils. The latest Quattro sensors are also a complete redesign, with top layer being highest mega pixel, and lower layers, much lower resolution. I am not sure about the uv sensitivity of these cameras. One day if the cost comes down I would like to compare the Sd14 to the newer Quattro. The newer Sdq has live view and an Evf. At least one new forum member here has the Quattro, but I don't know if they have any other Sigma camera for comparison.

 

Also I think the sigma curve is similar to Jonathan's. There is a up swing in the peak to 400nm, but dropping into noise below 350nm.

Link to comment

Thanks for the info Da Bateman. I will hold my hand up to knowing very little about the Foveon sensors. They are however a very different construction to the standard type of sensor, and I wonder what the comparison between them would tell us. If I understand it right then with a Foveon sensor, to hit a red pixel, the light has to pass through both the blue and green parts of the detector as well. Which I'm guessing leads to the rather strange looking response curve in Figure 17 that you posted, where it looks as though red light also triggers a strong response in blue and green. I'm also confused why Figure 17 and Figure 2 look so different.

 

According to David Prutchi in his book, he says that with Foveon sensors all the UV would essentially be absorbed in the blue level so that camera would essentially be monochromatic. Is that borne out in practice?

 

So who in the UK has a full spectrum modified Sigma for me to test (and can they be fitted with the Nikon Rayfact Lens)?

 

Andrea, Da Bateman. Given the differences between the standard and Foveon sensors I feel this warrants a different thread. Given the different construction and the lack of a Bayer filter for the Foveon the whole concept of what it is measuring and how would be a big discussion in itself.

Link to comment

Yes, I will split the thread when I get a moment. :D

 

Later: The Foveon sensor now has its own topic here.

Link to comment

Jonathan, how a Foveon sensor "sees" color information has been an argument for more than 10 years. In practice all layers are getting wavelength data. If you're careful you can calibrate the sensor to specific spectrum sections, as in very expensive laboratory careful for use on microscopes and spinning disc confocal systems.

 

Uv on the other hand I think actually is on the first layer and why I would like to test the Quattro sensors. Sigma has changed the sensor architecture in the Quattro, to be mostly a single calibrated layer, with some additional information from lower layers.

I just found out that I can rent one from Lens Rentals for $70, for 7days. Since the conversion is simple and reversible, I may try this. I would like to compare these cameras to get an idea. The main problem with the Sd14, is that its an IR camera and you need to use it at ISO 100. Any IR leak becomes obvious. So uv has been hard. But now with discrete Led lights, I have been getting much better results.

 

Sadly Sigma cameras do not take the Nikon mount lenses. They do take easily M42 lenses. So your 85mm m42 pentax lens will work. If you get your hands on a SD14, can be bought used for $200 to $500, the crop factor is 1.7x. The Quattros, Sdq crop is 1.5x and the SDQH crop is 1.3x. So an SDQh and your 85mm will get close to same field of view as the Canon and 105mm Rayfact.

 

Link to comment
Thanks Da Bateman. Yes I had a horrible feeling the Rayfact wouldn't be an option for it, but as you say the UAT 85mm would be usable. I shall look to see if someone round me has a Sigma to try out. Like you I'm also interested by the Quattro sensor, given how they have changed the design for that. Please keep us posted with any more Foveon findings.....
Link to comment

David, can you explain pls -- why is SD14 an IR camera which needs to be used at ISO 100? I'm curious. Thx.

 

I wonder if the Lens Rentals folks would remove the filter for anyone wanting to rent a Sigma that way??

 

The Sigma FFD = 44 mm. (I was just curious about that too.)

Link to comment

I can verify that Foveon sensors capture almost all UV data in the top layer. What reaches the lower layers is mostly out of band bleed-through.

 

These cameras must be used at low gains because their high-gain performance is so wretched. Even in the visible, using settings above 200 is not recommended, though settings up to 1600 are available.

 

The information does not come off the sensor neatly separated into RGB channels. The task of reconstructing these channels is handled by mathematical formulae.

Link to comment

@Andrea,

I will do a controlled multiple spectrum series and provide the exposure values. But you get an idea from the SD1 and SD14 spectal curves above. The Sd1m is centered at 700nm, and the SD14 has very broad non-zero IR response above 700nm. All layers pick up the IR bands. Sigma has now redesigned the AR coatings on their lenses to drop hard at 830nm. The new global vision Sigma lenses are now known for very poor IR use.

As OlDoinyo said the noise level above ISO 200 is very bad. The SD15 and SD1m have improved higher ISO values than the SD14 with a metal substrate on the surface, but this also diminished the UV response.

 

As for removing the internal dust uv/ir block filter, it is super easy. You just pop it out. Its slightly harder to put it back in, as you have to gently apply lower pressure to snap it back in. I applied too much pressure and broke my first block filter on my SD14. But I have since removed and placed it back many times without any issues.

Link to comment
I'm thinking I would like to put a Sigma summary in the Sticky where cameras are discussed. We would need a list of the Sigma Foveon-sensored cameras. And a short discussion of the general characteristics applicable to UVIR photography. If David and/or Clark write up something, then I would credit them in the Sticky. I consider the Stickies collaborative knowledge and will always credit contributions. Or, if a member does not want their name in the Sticky, then I will respect that also. :)
Link to comment

Sure I can come up with something. I may also test the Quattro sometime this summer.

I also have no issues with my name being added, so readers can ask me questions. But I can't list any affiliations, other wise it will take about 10 years for approval. Sadly that is not a joke.

Link to comment

I was kindly sent an SD14 to test using my camera sensitivity setup. It hadn't been used for years, so I ordered a few parts needed (battery charger and Sigma to M42 adapter so I can fit my 85mm UAT on it). It suffered from what seems to be a common SD14 problem - it didn't work, so I had to open it up and replace a couple of the earthing points inside.

 

The raw files (X3F files) are very different to Bayer filter ones, and while RawDigger opens them I had to try and figure out how to get usable R, G and B data from the which I eventually did be subtracting the RGB values from black background from my test images from RGB values from the exit port of the integrating sphere in my rig. So I ended up with a set of R, G and B values for 300nm to 800nm in 20nm intervals. Once corrected for the variation in light source intensity as a function of wavelength, this is what I got.

 

post-148-0-05294700-1528028782.jpg

 

This was done with no internal filter (it wasn't supplied with the camera, so cannot do a with and without comparison). Each colour channel is made up of 2 scans. One from 300nm up to 480nm, and then another from 440nm to 800nm. This is done because I switch off my deuterium light source for the 440nm to 800nm part.

 

This is a first attempt at the measurement of the SD14 for me, so I need to redo it. Not least because of the strange feature in the red channel at 420nm. This was definitely in the file, but I have no idea what it would be. It would be nice to compare it with one of the newer Quattro sensors, so will have a word around and see whether anyone round here has one I can borrow.

 

The values should in no way be compared to my Bayer filter camera data, the X3F files were processed differently.

 

In general terms it looks like Figure 2 from Dabatemans post of the 21st March, but that does state it has been modified (although how it has been modified I don't know). It definitely looks like the majority of sensitivity between 350nm and 400nm is in the blue channel. Not sure why the red channel seems to retain some sensitivity even at short wavelengths. Perhaps it has the gain turned right up in the camera firmware as it is the lowest part of the sensor? Also as for whether the signal still being seen below 350nm is real or an artifact I don't know. If I ever get a 320nm laser I'll test it again :)

 

EDIT - now the updated chart from 5th June 2018 based on new findings

 

Given I discovered a problem with IR in my UV scans, below is the new chart from 280nm to 800nm, where the UV section has been corrected for the IR leak. This supersedes the graph above;

 

post-148-0-91071300-1528204818.jpg

 

The strange bump at 420nm has now gone, which was found to be related to IR contamination from the measurement setup.

Link to comment

This is great, thank you.

The Sigmas don't need modification. Thus why the figure I posted from someone else's thesis (not mine) says no camera filter. The sigmas have a dust blocking filter in the mount that is the ir/uv blocking filter for all sigma cameras after and including the SD10. So the only modification is poping out the filter. The SD9 was the only one without micro lenses on the sensor. But the SD9 has the block filter glues on to the sensor and needs modification. The dust blocker on SD9 is just passive glass.

So its good that it matches previously published data. This means your both doing something right.

Great work Jonathan. Now I really hope you do get a SDQ, remember all you need to do is carefully pop the dust block filter out. So you could rent one also.

 

Update, I forgot to mention that you can still buy the dust block filter direct from Sigma on their website for $30 until supply runs out. In 2011 sigma stopped making the SD10 block filter and at the beginning of this year they said they have stopped making the block filter for the SD14/SD15 cameras.

The SD1m has a larger sensor size 1.5x vs the previous models being 1.7x. I am not sure if the SDq and SD1m use the same block filter, but the sensor size is the same.

The SDQH has a 1.3x size sensor.

Link to comment
Those graphs imply that even the blue channel dies below 360 nm. That would hint at some attenuation by the cover glass or microlens assembly, so to say that these cameras could not benefit from modification might have to be qualified a bit for UV, at least (they clearly do not need it for IR.) Usually, CMOS-type sensors have some response down to 300nm or so. I would be curious to see a filter-array test run with one of these cameras and a pinhole or dedicated lens.
Link to comment

@OlDoinyo,

I plan on it as soon as I get some sun. I got my grating working and have a pinhole pro lens. So I plan on testing a full spectrum, Baader venus filter, 1.5mm U330 with 2mm S8612 filter and my Zwb1 with S8612 filter on the pinhole pro to see the effect on the solar spectrum.

 

However, let first make some incorrect assumptions and assertions.

Jonathan provided the output of off the shelf Canon, full spectrum converted and monochrome converted. Now lets eye ball that graph and this one above to get % range in our minds.

The full spectrum Canon looks just like the Sigma above, dropping off at 350nm.

Now lets incorrectly assume that the silicon is the same between the cameras. The only difference now is the color dye layer.

Both have micro lenses on sensors. So the excellent uv response of the monochrome Canon is due to removal of the micro lenses, and not due to the dye layer using these incorrect assumptions.

Link to comment

Da Bateman

With the Sigma Foveon sensors, we don't know the transmittance properties of the cover glass, the microlenses, or the top layer of silicon above the blue receptor.....?

The Sigma Foveon dust cover, come hot mirror, has a steep cut-on at 420nm.

Obviously with enough exposure some measure of UV A will be recorded.

Cheers

Col

Link to comment
I have some difficulty believing that a regular, converted camera is inherently incapable of recording below 350-360 nm. Some of the Sparticle tests on this forum clearly show response down to 320 nm with some cameras, and I am reasonably sure that my own camera can resolve chromatic detail in the 300-350 range, given a suitable optic. There is a very large difference in chromaticity between images taken through the Steinheil 50 and those shot through the Tamron 21 (which cuts out around 350-360.) One would expect no difference if the camera itself rolled off in the 350-360 range.
Link to comment

Col,

Yes exactly. Sorry I wasn't clear in my previous post. Looking at Jonathan Canon curve:

http://www.ultraviol...attach_id=12261

It reminded me of the published sigma curve I have posted in post 2.

I always thought the 6x advantage of a monochrome camera was due to removal of the dye layer. But I think based on similar curve trends its actually due to the removal of the micro lenses.

This may make more sense as its a chunk of glass that focuses the light to the sensor.

Why I like this revelation is: if MaxMax can remove the micro lenses from an SDq sigma camera, then you would have great monochrome uv detection. As Gary R has shown, the monochrome +B setting is best. But your camera is still capable of colour images due to the depth detection of the sensor. However, the amount of loss of detail due to removal of the micro lenses may be an issue. People argued on both sides between the SD9 and SD10, only difference being the micro lenses.

 

@OlDoinyo,

I am not going to argue with Jonathan's data, just accept it as is. Maxmax published the exact same thing and people here made the same arguments, nothing being resolved, only removal of the MaxMax curves.

So I do think Cadmium is seeing 320nm with his Sparticle and I do think Jonathan sees a sharp cut off at 350nm. However, member JCDOWDY could argue better than me why Jonathan gets the data he gets based on the limits of his test system.

I am more interested in the global trends. As to make any conclusions you have to accept a lot of assumptions.

Different test lenses having similar character, not unrealistic.

Diferent sensors having similar silicone response.

Link to comment

EDIT - this has now been updated with recent data as of 5th June 2018.

 

It seems as though as always sharing spectral sensitivity data has resulted in some passionate discussion. I shall add my 2 cents about some of the comments and hopefully clear some things up.

 

The plot that I shared above is for the camera with the dust/IR filter removed, and using the Asahi 85mm UAT lens. This has been corrected for IR contamination. To start here's the graph of the part of the data between 280nm and 480nm;

post-148-0-28811200-1528203606.jpg

 

The blue response drops drastically as expected, and the red and green responses are already very low, especially by 400nm. It looks as though the majority of the sensor response is from the blue and between 350nm and 400nm for the UV. There is a wiggle at 340nm where corrected plot here drops below zero. I am not sure why that is as the rest of the line doesn't go below zero. Perhaps this is the degree of error/uncertainty in the measures.

 

However we do not use these cameras without additional filtration. I can combine the response graph with the Baader U transmission profile split into 20nm chunks, and if I do so I get this;

post-148-0-88434700-1528203748.jpg

 

As the Baader U has the bulk of its transmission between 380nm and 320nm the huge response from the sensor in the blue at 400nm becomes inconsequential. As I mentioned above 'the majority' of the image would be from the 350-380nm region, the key word being 'majority', not 'everything'.

 

The 'wiggle' at 340nm is now unfortunately amplified, as that is where the transmission of the Baader U is greatest at this scale (20nm chunks). However other than that it looks like the image would be mainly in the blue channel and not in the red or green.

 

Now again this does assume the light source is constant across the 300nm to 420nm region, which of course for sunlight it is not. The sunlight is more biased towards the long wavelength end.

 

These types of sensitivity graphs need to be considered along with the nature of the light source being used and any additional filter (and the lens transmission, depending on what you have), otherwise they are just a graph of a function of the sensor response under an idealised scenario.

Link to comment

Thank you Jonathan,

I was going to add that the way you conduct your tests is a light captured on the sensor. Do dynamic range and signal to noise play a big part in your results? We users of the Sigma cameras know signal to noise is low on the Sigmas. Probably just at the Rose criterion. Raw Digger web site has an article on baseline exposure compensation and the SD14 has the largest value pushing it to 1.25 stops. So I am assuming noise is possible which could be seen as signal depending on how you determining the dynamic range cut off for your studies.

 

That all being said, do you still have the invisible U filter? I would be curious if you do to see if you get anything out of the sigma.

Without a proper lens, I got a cheap 313bp50 filter. I will test it with my pinhole pro lens, and maybe some others on my cameras. It cuts off at 350nm, so I should see something on something.

Link to comment

Dabateman, I'm not sure how I wold quantify the signal to noise really, but from looking at these and the images from the Canon and Nikon cameras, the Sigma images appear noisier to me. Once the light source has gone through the monochromator, and the integrating sphere and then is imaged, there isn't much light light left, so I have to run the camera at high ISO and long acquisition times. In the UV I ran the Canon and Nikon at ISO6400 and 30s acquisition time.The SD14 doesn't go that high so it was ISO1600 and 30s. I had some luck with the SD14 by reducing the resolution and that seemed to reduce the background noise in the black parts of the image. The Sigma looked noisier at 1600 than the Canon and Nikon did at 6400. That of course is a little unfair, as the SD14 is not quite old, so perhaps the new Quattro sensors would be better. Gut feel, you'd still get something out of the Sigma down there, but it would be extremely long acquisition times.

 

I do not have the 308nm filter anymore unfortunately - it was a loan one and I had to send it back.

Link to comment

First I must apologise for the excessive grammar and spelling mistakes in my last post. I have to read these things before posting.

Wow pushing the Sigma to 1600 is something I would never do. I would rather use long exposure at ISO 100. The cable release is the same as some Canon cameras, just a 2.5mm jack.

So definitely noise is an issue and may fully account for your red channel signal.

 

No it's totally fair to say noise is an issue with the Sigma cameras. I would say the noise at 400 ISO should be more than a Nikon camera at 6400. Anyone whom argues differently has really been drinking the coolaid.

Link to comment

I'm not sure about the noise accounting for the Red. As I thought noise could be a problem, as I was going through each image I measured red, green and blue in two areas. Firstly the exit from the integrating sphere, and secondly a black area towards the corner of the shot. I then subtracted the black area values for each channel from the ones from the exit port from the integrating sphere before doing anything else. So the noise should be corrected for.

 

Interestingly there didn't seem to be any significant different between the noise values for each of the 3 channels.

 

My remote release adapter arrived today. I'd like to repeat this at a lower ISO when I get the time. Problem is at lower ISO it's going to take me 2-3 hours to take all the shots because of the longer exposure time needed. Maybe I'll just try it for the UV region given the visible data had better signal to noise.

Link to comment

Please sign in to comment

You will be able to leave a comment after signing in



Sign In Now
×
×
  • Create New...