Chronos 2.1 noise study

[mklinger]

I spent a fair amount of time trying and understand the noise characteristics of my Chronos 2.1.  From previous work, I determined that there are really only 4 native horizontal resolutions that should be used from a speed perspective: 1920, 1472, 1152, and 832.  Anything else is the same as cropping in post.

I wanted to better understand the noise of each of those base resolutions as well as if vertical resolution had any effect.

Unfortunately, it does look like vertical resolution effects noise level and patterns, especially at the higher gain levels.

My process for all of these runs were as follows (all done via the web interface):

  1. All tests were done with the lens cap covered at all times and the camera had warmed up for 30+ min.
  2. Select the resolution, gain level, and max frame rate
  3. Do two black level calibrations in a row
  4. Save the 100th image as a DNG file
  5. Load the raw images into Photoshop
  6. Overlay the images, flatten the layers, and then boost the global image gain equally across all tests


I also ran a test to see if speed itself made a difference.  It looks like it does somewhat.

I'm not sure what my conclusions from all this testing are.  1920 and 1472 are definitely the best resolutions to use and keeping gain to 0 or 6 dB is ideal.  I was a bit disappointed at what a mess the 832 resolution was even at 0 dB.   The noise seems very inconsistent and all over the place.  I would use 1152, but only sparingly.

Let me know what you think!  This type of testing should be reproduceable fairly easily.  It might be interesting to see if different sensors behave differently. 

My camera is using version 0.6.0 of the software and reports FPGA Revision 3.24 and PMIC Firmware Version 11.

I can zip up all the DNG files I used if anyone is interested in looking at them directly.

[Nikon1]

[...]
 This type of testing should be reproducible fairly easily.  It might be interesting to see if different sensors behave differently.
 [...]

 From how i understand it, the Noise produced by any of the Chronos Sensors is like an Human Fingerprint of Sorts. Meaning that if you would know the Noise "Fingerprint" of every Sensor made, you should be able to figure out which Sensor shot what Footage (given you have Plenty of uncompressed Footage with enough Noise compared to "Image Signal" visible) with some Effort put into it. So, yes, each sensor should behave different. But only in terms of the Actual Noise Patterns itself, overall Noise Behavior should be roughly the same across Sensors of the Same Production Batch i assume.
 .
 Regards Everything else you said, i can confirm everything you wrote here about your testing. Done Plenty of tests Similar to this myself, but never quite in such a Scientific manner. Results where about the Same, but my focus was more on Full Resolution/ 1280x720 and lower Framerates (25, 30, 50, 60, 100, 120, 200, 240, 400, 500), as well as Full Framerate at 1080p and 720p Resolutions, and also the Higher end of Framerates. The Point of my testing was to see how to best shoot with as little Light as Possible (available Light Shooting in "Low Light" conditions, just in case if i ever needed to, cause i at some point probably will. also generally just good to know the true Limits of your Equipment in my opinion), and how much Framerate i would still be able to use in different Light Conditions. Naturally, i tested gain and digital Gain Combinations and also tried to see how much brighter i can still make the Footage in post.
 .
 From my tests the noise would behave about as one would expect, besides some aspects of it. I found that Shutter Speed and Framerate do also influence Noise quite a bit, which i didn't initially expect really. especially well below like 240fps (cant completely remember, just a guess) shutter speed changes noise noticeable, and low Framerates seem to have had a lot more Noise in it on my tests iirc.
 .
 I didn't really use a "black frame" to judge the Noise, but rather a "grey Frame", so a neutral Color with some exposure Level to it, or on later tests something that would cover the Full Dynamic Range of the Sensor and do a Pan across the Thing to see how different areas of the Sensor Respond to different Exposure Levels.
 While you still will get plenty of information out of a test like you did, comparing the Noise on a blocked Sensor, i personally didn't care all too much about that, as i can and usually will crush the Blacks anyways, especially if noise seems to show up there.
 So if you want to really get to know your sensor better, i recommend you to test the Full Dynamic range from Black to fully Clipping White.
 Found some characteristics/ problems of my Sensor, one wouldn't really notice otherwise.
 As an Example, My Sensor in particular has some weird vertical line of pixels about 2/5th of the way from the Right in the Lower half, that would have a weird red tint to it sometimes. But that would only show up in some Specific exposure Ranges/ on certain settings. Firmware 0.6.0 improved that by a lot, on 5.1 and before it was very visible and generally rather bad.  From my tests i found that if you expose carefully, and have Plenty of light overall (also using 0dB Gain helps, on latest Firmware higher gain is mostly save to use regards that, but is still visible sometimes), that usually never shows up.
 In terms of Noise, Light generally does help a lot with the Noise, but Harsh Lighting is still problematic. So somewhat evenly lit, bright Scenes will give the Best image Result.
 Now that i know it does that, i can stay away from Conditions that would probably make it show up. Also i know to look out for it, and will usually notice immediately if it shows up. Thats what i took out of my tests.

[mklinger]

Thanks for the reply!  Yeah, this was just an attempt at a scientific noise study, so the black frame is the best way to do that.  To scientifically measure dynamic range across the entire range of sensor options would be a huge task.  Also, I kinda already have an idea about the dynamic range... it goes from fair to really bad  

Personally, I don't think the Chronos is the right tool for the job at shooting anything below 300 fps.  Something like the Blackmagic URSA Mini can shoot 4K at 120 fps and 300 fps at 1080p.  Cheaper options like a GoPro Hero9 or a modern phone will give better results at 240 fps and below.  300 fps to 1000 fps is a bit of a no-man's land with current cameras.  The Chronos is really the only reasonably priced option that is a pretty awesome 1000-4000 fps camera but the base ISO of 500 is somewhat limiting.   To get good results with the Chronos in the 10k+ fps range is really challenging, but it's fun to play around with.

I have a feeling the way the camera is windowing down the resolution and adjusting framerate/shutter is very complex at the hardware level.  It probably involves changing a whole bunch of amplifier gains that effect the noise and noise patterns quite a bit.

I was wondering if it'd be possible to have an option to shoot the color Chronos in BW to try and reduce the noise or perhaps increase the base ISO.  I know the BW version of the Chronos uses a completely different sensor, but it'd be interesting if the color sensor could be tweaked in software in a way to get some of the benefit of the BW hardware.

P.S.  Look into shooting 1472x720 instead of 1280x720.  I think you'll find there is no difference in speed and you'll have an extra 192 pixels to play with framing your shot in post.

[Nikon1]

Personally, I don't think the Chronos is the right tool for the job at shooting anything below 300 fps.  Something like the Blackmagic URSA Mini can shoot 4K at 120 fps and 300 fps at 1080p.  Cheaper options like a GoPro Hero9 or a modern phone will give better results at 240 fps and below.  300 fps to 1000 fps is a bit of a no-man's land with current cameras.  The Chronos is really the only reasonably priced option that is a pretty awesome 1000-4000 fps camera but the base ISO of 500 is somewhat limiting.   To get good results with the Chronos in the 10k+ fps range is really challenging, but it's fun to play around with.

 While you couldnt be more right with that statement i have to politely inform you, that i own a camera that shoots up to 60fps in FullHD and 120 at 720p, and a Chronos.  Nothing else really. I also dont plan on upgrading or buying into an other Camera-System anytime soon. So, sure there are way better ways to do this, but while i have the 2.1, that can shoot these Framerates, i might as well use it for that, if you know what i mean. Then there is also the thing with low Light, i talked about in the Previous reply. There is just times, when i only got the Chronos on hand when i am somewhere, and there just isnt enough light for good Quality 1000fps, and 1000fps would be way too slow for what i am shooting anyways. So at least for me anywhere between 60 and 1000fps is a realistic usecase for the 2.1, even if it probably wont make any sense for anyone else.

I have a feeling the way the camera is windowing down the resolution and adjusting framerate/shutter is very complex at the hardware level. 

Yes. i very much assume it is.

I was wondering if it'd be possible to have an option to shoot the color Chronos in BW to try and reduce the noise or perhaps increase the base ISO.  I know the BW version of the Chronos uses a completely different sensor, but it'd be interesting if the color sensor could be tweaked in software in a way to get some of the benefit of the BW hardware.

not quite sure what you are trying to do there, but you could kinda run a Color Chronos in BW mode by setting every value on the custom Color Matrix to the Same Value (ideally use 1, tweak it a bit if you want to change contrast), see image attached.
 also the DNG data should still be per-Pixel Data, meaning undebayered, so using Software like PIPP (https://sites.google.com/site/astropipp/) you will still see what is basically a monochrome Image. I used PIPP to mess around with RAW files Quite a bit, helps you to understand what is going on really.
 anymore than that, and you would need to go and actually remove the Bayer filter Layer on the sensor itself, which i wouldnt recommend for multiple reasons on a camera like this, the main one beeing, that there isnt even a point in that, as Krontech literally sells Monochrome ones...
 For Reference and/ or anyone that has never heard of that:
 https://stargazerslounge.com/topic/166334-debayering-a-dslrs-bayer-matrix/page/1/
 .
 Noise Characteristics should still be unchanged from how i understand it however, but you gain a bit of Light from removing the Filter Layer (or just getting a Monochrome Version).
 Monochrome Cameras should however have less noise anyways, because of that bit of more light, and the camera generally doing a lot better in terms of noise if it has plenty light.
 If you dont mind a casual Magenta/ pink tint on some parts of the image because of some IR Light, you could also just remove the IR Filter, that will also give you quite a bit more light, and help with the noise. Will mess with your Flange Focal distances and your color however.
 but i guess the only way to know how the Monochrome actually does in terms of Noise, is to test one or ask someone who owns/ works one.
 Only Person i know that uses one is Lauri from Hydraulic Press Channel. Username on the Forum is hydraulicpresschannel. He should know, he uses a Monochrome 1.4 alongside Color Models. I dont know, if he still has/ uses his Color 1.4, but he at least had them in use right next to each other for quite a while in the Past.
 .
 

P.S.  Look into shooting 1472x720 instead of 1280x720.  I think you'll find there is no difference in speed and you'll have an extra 192 pixels to play with framing your shot in post.

 I might will at some point, but until now i just didnt bother remembering that quite specific Resolution. Standard 16:9 resolutions i can Quote from Memory, even if you where to wake me up in the middle of the Night and ask me, that one i would have to write down somewhere or something to remember. I guess i could try 1440x720, as that would be something, i will be able to remember no problem, but 1472px wide is quite the odd number to remember.
 .
 

Yeah, this was just an attempt at a scientific noise study, so the black frame is the best way to do that.  To scientifically measure dynamic range across the entire range of sensor options would be a huge task.

 Yes, thats why i never bothered to do a full on test worth publishing anywere, cause the amount of work that would take is just not fun anymore to do in my spare time.

[mklinger]

Yeah, I'm not talking about getting the best BW image from the camera as it is, that's not too difficult in post.  What I was wondering if there were some hardware level adjustments that can be done in the firmware to allow for better noise characteristics if you are only concerned with BW.  It's definitely not the same as using the BW sensor with no Bayer filter, but I was just speculating that maybe some of the amplifier gains could be tweaked (or turned off) if it could be run in single color mode or something like that.  This would require firmware changes, but it'd be neat option if it helped.   If you check the Chronos manual, they show the spectral response of both the color and monochrome sensors.  The monochrome is much better (almost double - thus the base ISO at 1000 instead of ISO 500), but if you look at the color graph, the green response is better than either the red or blue.  It might be possible to just read the green channel from the color sensor and tweak the gains to make it a pseudo BW sensor with slightly better noise.

I hear you on remembering the resolutions.  There are only 4 you need to remember with the Chronos 2.1:  1920, 1472, 1152, and 832

It would be nice if there was a way to program the default drop-down resolutions and base them on those 4 values, but I understand from a marketing standpoint why they want the standard HD and SD numbers.

[Nikon1]

It would be nice if there was a way to program the default drop-down resolutions and base them on those 4 values, but I understand from a marketing standpoint why they want the standard HD and SD numbers.

Yes, i agree, user Customizable Resolution/ Framerate Presets would be a very neat upgrade to the overall Camera for one of the Next Firmware versions!

[Nikon1]

Yeah, I'm not talking about getting the best BW image from the camera as it is, that's not too difficult in post.  What I was wondering if there were some hardware level adjustments that can be done in the firmware to allow for better noise characteristics if you are only concerned with BW.  It's definitely not the same as using the BW sensor with no Bayer filter, but I was just speculating that maybe some of the amplifier gains could be tweaked (or turned off) if it could be run in single color mode or something like that.  This would require firmware changes, but it'd be neat option if it helped.   If you check the Chronos manual, they show the spectral response of both the color and monochrome sensors.  The monochrome is much better (almost double - thus the base ISO at 1000 instead of ISO 500), but if you look at the color graph, the green response is better than either the red or blue.  It might be possible to just read the green channel from the color sensor and tweak the gains to make it a pseudo BW sensor with slightly better noise.

 .
 i did read this a bunch of times, and still dont understand where you are going with this...?
 about the Monochrome sensor beeing more sensitive, yeah, thats kind of the point, why they even sell it that way.
 Dont want to be rude here, but I am not sure if you realize, even the UV-/IR-Cut Filter in front of the Sensor makes you loose like 10 to 20% of Light reaching the Sensor in the Visible Spectrum, a lot more in Invisible Spectrum (just a rough number as an estimate from very quick research, dont know the real numbers on whatever Filter they actually use in the Camera itself, so please dont quote me on that). Now that is basically just a Transparent bit of glass with not much to it for the naked eye. The Bayer Layer is a full on color filter in front of the Sensor, so of course its gonna make your sensitivity worse. about one Stop of light loss is about normal of what a Bayer Layer will eat up of your Light, which is pretty much the Difference between ISO 500 and 1000. The Graph in the Datasheet is, if you look carefully, the Characteristics of the Sensor itself. I am rather sure they just took that straight from the Sensors Datasheet and put it in there. So up to that point, its the Physical Sensor itself, we are talking about, nothing to do with the Camera at all yet.
 So that one Stop of loss in sensitivity is what you pay, if you want/ need Color on your Camera. Otherwise even Krontech themselfs highly recommend you to get the Monochrome Version, if you dont need color.
 .
 About the Green beeing more sensitive, also, thats kind of the Point of a Bayer Pattern Filter. If you take a look at the Filter pattern, you will notice that there are literally TWICE as many pixels assigned to capture Green Color than for blue and Red. So having twice the Area to capture said color, its kind of obvious, why it would be more sensitive.
 While there are a bunch of other types of patterns around, The Bayer Pattern is as far as i am aware by far the most widely used one. Green actually needs to be the most sensitive, because of the way human eyes perceive Light.
 .
 The other thing i could imagine you might mean, is that Krontech might try to drive the Lines / Rows of pixels at different gain/ voltage/ whatever (i dont understand the Hardware down to THAT level to know how you would actually do that if you wanted to) Settings because only every other line will have red/Blue Pixels in it, but i dont know if that is even Possible with the Sensor Hardware.

[mklinger]

It's the amplification of the signal which causes the noise.  I'm just suggesting/speculating that if it were possible to read just the green channel off the sensor, you might be able to not amplify the blue/red channels and get a cleaner overall signal.  I have no idea if the hardware can do that. 

Of course it'd be better to also have a monochrome version.  It would also be nice to have a Phantom TMX 7510. 

[Nikon1]

It would also be nice to have a Phantom TMX 7510.

Damn, that thing is fast, but i would rather not want to know what that thing costs, i assume. Didnt check on Vision Research lately.
 For me it would probably be something more like a Phantom 4K and some PC that could handle editing and storing the Footage that thing produces (was actually some clips off a Phantom 4K that got me into Highspeed Cameras initially), if money wasnt a issue, but i get what you mean by that.
 .
 About the Green Chanel thing, maybe look into the PIPP Software, i assume it can do that. As far as Black Calibration goes, we are kinda Stuck with whatever the Guys at Krontech are able to get out of the Camera, but with PIPP or similar Software you can do quite a bit in terms of Debayering and what to do with which channels. I never extracted a Single Channel with it, but i think it should be possible. When using 0dB gain, you should have the Cleanest signal the Camera can provide in theory.
 .
 When i find time i might look into it a bit.
 

[Nikon1]

So here is some more Thoughts on the Whole Topic on Studying and Analyzing the Noise Profile of the Sensor.
 Did think about it for a bit more, and i feel like layering a bunch of Frames would give an good average of the Sensors Static Noise, also basically eliminating Random Noise from the Analysis mostly.
 More of a Proof of Concept than any meaningful test, i did Let the Camera Warm up, do Black-Calibration, waited a bit more and did another one. Then Recorded for a bit with Lenscap on, so "Black" frames. Saved like 300 frames to sure have enough as DNG.
 Then Importing the DNGs into Photoshop as Layers and setting Layer blending Mode to "Add" Layer after Layer, until the Histogram would indicate that i am Clipping, then go back one.
 Firmware 6.0, FullHD, 1kfps, 357° 0dB and 12dB respectively.
 Ended up blending 29 Frames that way for 0dB and 17 Frames for 12dB Gain.
 Also did some Similar Thing Stacking a bunch of frames from a recording at Temperature without prior Black-Calibration (this was my first test, was just playing around at that point, no idea how many layers that was).
 To make any useful conclusion from that or come up with any usable "Noise-Fingerprint" for Software Noise Reduction i assume you would still need at least a similar Stack of Frames from the Sensor hard Clipping to White and probably ideally something in the Middle of the Exposure Range.
 Doing any Test of a Scale to yield meaningful Data from that would take quite a bit of time, which i currently dont have, but maybe this Bit of Information will help anyone else to come up with something meanwhile, so i share it anyways here.