Reading mode: Light Dark

Foveon(SD-14) and higher noise-why?

Go to last post
C

Coyote_Cody

Senior Member
Messages
1,166
Solutions
1
Reaction score
127
Location
FuckOffDPR, US
If you have paid any attention to Canon and how they have put a A/D and amplifier on each pixel site, and argurably Canon is the most noise free high ISO DSLRs (yes with a strong AA filter), I was wondering if the Foveon has this setup and if not why not? (yes I know there would be 3/pixel but 14mp is 14mp, 1 layer or 3).

Most have to agree the Foveon has the best per pixel quality, or at least equal to or better than any bayer/hybrid I have seen, so if they could conquer the high ISO noise and really have a tiff/jpeg quality that is undeniable for a given (large) size/ISO, there would be no 'mine better - yours not' debate! (IMOH) Foveon would be the winner (and gain wider acceptance)!

So does the Foveon have the A/D-amp per pixel OR is the high ISO issue due to getting enough light in the desired light band equalized to the desired amount per layer?

Thanks!
--
The Light is Right - Take that Shot!!

Check out my photo galleries @
http://picasaweb.google.com/TauPhoto
 
Coyote_Cody said:
If you have paid any attention to Canon and how they have put a A/D
and amplifier on each pixel site, and argurably Canon is the most
noise free high ISO DSLRs (yes with a strong AA filter), I was
wondering if the Foveon has this setup and if not why not? (yes I
know there would be 3/pixel but 14mp is 14mp, 1 layer or 3).

Most have to agree the Foveon has the best per pixel quality, or at
least equal to or better than any bayer/hybrid I have seen, so if
they could conquer the high ISO noise and really have a tiff/jpeg
quality that is undeniable for a given (large) size/ISO, there would
be no 'mine better - yours not' debate! (IMOH) Foveon would be the
winner (and gain wider acceptance)!

So does the Foveon have the A/D-amp per pixel OR is the high ISO
issue due to getting enough light in the desired light band equalized
to the desired amount per layer?
Click to expand...
Aplifying the response of a single pixel does not lower noise.
Coyote_Cody said:
Thanks!
--
The Light is Right - Take that Shot!!

Check out my photo galleries @
http://picasaweb.google.com/TauPhoto
Click to expand...
 
Coyote_Cody said:
If you have paid any attention to Canon and how they have put a A/D
and amplifier on each pixel site, and argurably Canon is the most
noise free high ISO DSLRs (yes with a strong AA filter), I was
wondering if the Foveon has this setup and if not why not? (yes I
know there would be 3/pixel but 14mp is 14mp, 1 layer or 3).
Click to expand...
Canon's current DSLR sensors do not have A/D per pixel.

-Z-
 
Coyote_Cody wrote:
snip
Coyote_Cody said:
So does the Foveon have the A/D-amp per pixel OR is the high ISO
issue due to getting enough light in the desired light band equalized
to the desired amount per layer?

Thanks!
Click to expand...
snip

Although my previous life is partly in electrical engineering, I honestly don't claim to know the internal electronics of the Foveon as implemented in the SD14 for example. Based upon the prior comments, I'm guessing it may not have the A/D amp per pixel and if not that one reason may be the complexity of layering that technology in an essentially 3 layer sensor which must pass light through all 3 layers.

As for why noise, there has been speculation on several threads. It seems that since the Foveon depends upon the degree to which light of specific wavelengths penetrates the silicon of which it is made, and since apparently the sensitivity to each of the primary colors is preferential but not exclusive by depth, some essential math must be done to separate the primaries from the somewhat intermingled color at each level - something which the Foveon argubly does rather well. However as in any signal to noise separation exercise it is likely IMHO that when the signal (proportional to light intensity) in a given layer - primarily sensitive to a particular color - is low (dim light) and/or the ISO is raised, the likelihood of getting noise in both the total S/N and also in separating the colors to create the R, G, and B channels, increases. Now take that rationale as highly speculative.

Regards,
--
Ed_S
http://www.pbase.com/ecsquires
 
Thanks for the only useful response. What you said is what I tried to infer in my statement, and you are totally correct that it is a mathmatical subtraction used to get the desired color curve out of each layer.

I do not think though, that the 3 layers prohibits an A/D & or amplifier at each cell site, but it does complicate and reduces each cell's (pixel) light gathering area probably. Since the chips are very exact, alignment would NOT be an issue.

Thanks!
Ed_S said:
Ed_S said:
So does the Foveon have the A/D-amp per pixel OR is the high ISO
issue due to getting enough light in the desired light band equalized
to the desired amount per layer?

Thanks!
Click to expand...
snip

Although my previous life is partly in electrical engineering, I
honestly don't claim to know the internal electronics of the Foveon
as implemented in the SD14 for example. Based upon the prior
comments, I'm guessing it may not have the A/D amp per pixel and if
not that one reason may be the complexity of layering that technology
in an essentially 3 layer sensor which must pass light through all 3
layers.

As for why noise, there has been speculation on several threads. It
seems that since the Foveon depends upon the degree to which light of
specific wavelengths penetrates the silicon of which it is made, and
since apparently the sensitivity to each of the primary colors is
preferential but not exclusive by depth, some essential math must be
done to separate the primaries from the somewhat intermingled color
at each level - something which the Foveon argubly does rather well.
However as in any signal to noise separation exercise it is likely
IMHO that when the signal (proportional to light intensity) in a
given layer - primarily sensitive to a particular color - is low (dim
light) and/or the ISO is raised, the likelihood of getting noise in
both the total S/N and also in separating the colors to create the R,
G, and B channels, increases. Now take that rationale as highly
speculative.

Regards,
--
Ed_S
http://www.pbase.com/ecsquires
Click to expand...
--
The Light is Right - Take that Shot!!

Check out my photo galleries @
http://picasaweb.google.com/TauPhoto
 
markspan said:
Aplifying the response of a single pixel does not lower noise.
Click to expand...
Actually it can, at least on a single pixel basis, but it is a 'conditioned' amplification, which, in the case of a Foveon, could boast the total cell signal before the subtraction process to get the desired spectrum of light from that cell. As you know there is RGB input/light in all the cells/pixels, all 3 layers, and the desired signal must be subtracted to arrive at the desired curve/spectrum. If you want to know about this see Foveon's pdf files concerning their extraction process.

Thanks!

--
The Light is Right - Take that Shot!!

Check out my photo galleries @
http://picasaweb.google.com/TauPhoto
 
Coyote_Cody

If I said or implied that the Foveon structure "prohibits" A/D I stand corrected. I meant to say that (speculation only) the difficulty of doing so may have (in the designers' and developers' view) made it unattractive.

If you found my response of any use you're more than welcome.

Kind regards,
--
Ed_S
http://www.pbase.com/ecsquires
 
Coyote_Cody said:
If you have paid any attention to Canon and how they have put a A/D
and amplifier on each pixel site, and argurably Canon is the most
noise free high ISO DSLRs
Click to expand...
Foveon's high ISO noise is more complicated than simply A/D and amplifiers. This sensor is really "quantum mechanics" with photons penetrating silicon at different depths. All layers need to be saturated with enough amount of photons to form a correct picture. At high ISOs layers are poorly saturated due to low light and also some layer can be much less saturated than others. With Bayer sensor, this data would be enough to amplify and do further interpolation which affects only neighboured pixels. At least we know exactly that blue pixel records blue, red records red, and green records green, and nothing else. With Foveon sensor, colors are distributed across layers and some photons get stuck in "wrong" layers thus making a kind of color noise by itself. When there is many photons, this can be ignored. When there is little photons, this becomes a problem, as valuable sparse data can miss the needed layer thus creating wrong color (With Bayer it will create either proper color or nothing). Furthermore, color transformation matrices are applied to the entire set of data, thus making some extreme values even more extreme and sometimes they get totally out of control.
 
Thanks, if you read Foveon's pdf that explain their chip, there response curves to each layer, which loosely follow the desired (very non linear I may add) output for each layer, but as you say there is RGB in ALL 3 layers and mathmatical extraction (DSP) is used to extract out the desired 'photon' :)) values.

I can't help but wonder if there can be more specific doping in each layer for that color band and/or, per pixel A/D and amplifiers which could help the noise problem.

The Foveon chip is a wonderful chip but it does need more pixel output and lower noise to 'takeover' the world like many of us would like, but to do that it has to be 'kick butt' better and yet it is not (although wonderful but....).

On a related topic, it will be interesting if Nikon ever develops its RGB patent, I am confused why Nikon has not got at least a full sensor (35mm) yet in a pro camera, maybe they are developing the RGB chip, only they know! I am sure there are full sensor CCDs made, maybe too pricey.

Thanks again!
dr.noise said:
Coyote_Cody said:
If you have paid any attention to Canon and how they have put a A/D
and amplifier on each pixel site, and argurably Canon is the most
noise free high ISO DSLRs
Click to expand...
Foveon's high ISO noise is more complicated than simply A/D and
amplifiers. This sensor is really "quantum mechanics" with photons
penetrating silicon at different depths. All layers need to be
saturated with enough amount of photons to form a correct picture. At
high ISOs layers are poorly saturated due to low light and also some
layer can be much less saturated than others. With Bayer sensor, this
data would be enough to amplify and do further interpolation which
affects only neighboured pixels. At least we know exactly that blue
pixel records blue, red records red, and green records green, and
nothing else. With Foveon sensor, colors are distributed across
layers and some photons get stuck in "wrong" layers thus making a
kind of color noise by itself. When there is many photons, this can
be ignored. When there is little photons, this becomes a problem, as
valuable sparse data can miss the needed layer thus creating wrong
color (With Bayer it will create either proper color or nothing).
Furthermore, color transformation matrices are applied to the entire
set of data, thus making some extreme values even more extreme and
sometimes they get totally out of control.
Click to expand...
--
The Light is Right - Take that Shot!!

Check out my photo galleries @
http://picasaweb.google.com/TauPhoto
 
The question, and I don't think we know the answer yet, is this issue a fatal flaw in the Foveon design with respect to noise and ISO performance or can it be overcome? The Foveon is the first to be developed and many times the first design is not the best - but the simplest to implement at the time.

I would expect to see other approaches to a full color caputre sensor in the near future and these designs will have the advantage of seeing the Foveon design and its related problems.

Many times the first design is not the one that ends up as the best solution.

Hopefully we won't have to wait too long for Fujifilm, Sony, Canon, Nikon, etc. to release a color capture sensor.

Truman
dr.noise said:
Coyote_Cody said:
If you have paid any attention to Canon and how they have put a A/D
and amplifier on each pixel site, and argurably Canon is the most
noise free high ISO DSLRs
Click to expand...
Foveon's high ISO noise is more complicated than simply A/D and
amplifiers. This sensor is really "quantum mechanics" with photons
penetrating silicon at different depths. All layers need to be
saturated with enough amount of photons to form a correct picture. At
high ISOs layers are poorly saturated due to low light and also some
layer can be much less saturated than others. With Bayer sensor, this
data would be enough to amplify and do further interpolation which
affects only neighboured pixels. At least we know exactly that blue
pixel records blue, red records red, and green records green, and
nothing else. With Foveon sensor, colors are distributed across
layers and some photons get stuck in "wrong" layers thus making a
kind of color noise by itself. When there is many photons, this can
be ignored. When there is little photons, this becomes a problem, as
valuable sparse data can miss the needed layer thus creating wrong
color (With Bayer it will create either proper color or nothing).
Furthermore, color transformation matrices are applied to the entire
set of data, thus making some extreme values even more extreme and
sometimes they get totally out of control.
Click to expand...
--
Truman
http://www.pbase.com/tprevatt
 
Truman Prevatt said:
The question, and I don't think we know the answer yet, is this issue
a fatal flaw in the Foveon design with respect to noise and ISO
performance or can it be overcome?
Click to expand...
If photons hit wrong layers by pure random probability, I think this is fatal. The only solution - give more light!

If sensor design can be refined in order to improve layer hit accuracy, than we shall see our brightest days.
Truman Prevatt said:
The Foveon is the first to be
developed and many times the first design is not the best - but the
simplest to implement at the time.
Click to expand...
Foveon staff is unique. They are very clever people and this solution is as simple as genius. I still believe they can overcome all the problems.
 
dr.noise said:
If photons hit wrong layers by pure random probability, I think this
is fatal. The only solution - give more light!
Click to expand...
Yes it is weighted random, but this is in principle no problems. Compare to an overlapping RGB filter (as in Bayer CFA) that also uses weighted random in the overlap area.

But - the strong conversion from X3 to RGB costs in noise. So - yes more photons.
dr.noise said:
If sensor design can be refined in order to improve layer hit
accuracy, than we shall see our brightest days.
Click to expand...
Some doping of silicon maybe?

--
Roland
 
I was hoping they would - but it seems they didn't. This problem still persist in the SD14. Either the Foveon is going to be a great 100 ISO sensor or some progress needs to be made. There is nothing wrong with an 100 ISO sensor - it should just be sold that way.

Hopefully other options will come along or Foveon will craft a solution. Many times it is not the most clever design that is the best design. It also is normally not the first design that ends up being the best.

Truman

dr.noise wrote:
curacy, than we shall see our brightest days.
dr.noise said:
dr.noise said:
The Foveon is the first to be
developed and many times the first design is not the best - but the
simplest to implement at the time.
Click to expand...
Foveon staff is unique. They are very clever people and this solution
is as simple as genius. I still believe they can overcome all the
problems.
Click to expand...
--
Truman
http://www.pbase.com/tprevatt
 
All,

It's kind of interesting. This noise issue is, and is not, it seems to me a matter of ISO. At the risk of what may (if correct) be incredibly obvious, the SD14 often does pretty well at high ISO in various situations. Frits recently posted his cormorant photo handholding a moderately long lens at ISO 800 in good light. Sandy F posted some seemingly nice photos on her Native American "Pow Wow" series shot at 800 in dim light. And on tripod, lower ISO does indeed provide respectable low light performance as by Hardy. I haven't had occasion to do a lot of low light/high iso stuff myself as yet, but somehow I'm seeing considerable good work by the camera across ISOs. Even with high ISO slide and negative films low light shooting was not without its flaws. Am I being insufficiently critical of the SD14's ISO/low light performance?

Regards,
--
--
Ed_S
http://www.pbase.com/ecsquires
 
I have some film with grotesque noise in dark areas when I scan it with Nikon Coolscan 5000. I read somewhere that if the exposure isn't right that causes noise. That's a bit like saying don't expect much at ISO 6400 when you thought it was ISO 400, 800, or 1600.

Converting to black and white is sometimes an option, and that will convert the colored noise into something like film grain, as someone here just said.
 
Sandy got great ISO 800 results with P mode and center exposure metering and AWB.

Center metering might be important for minimizing noise in important areas.

WB might be important, particularly with night shots with huge tonal ranges. Center metering with those night shots might be most important for noise reduction.

What do you think?
 
I've gotten some great ISO 800 shots and some that were pretty poor. There seems to be a lot more variability than with other cameras. In good light, my ISO 800 results are better than what I get with the 5D under the same conditions. I've found that the SD14 ISO 800 shots cannot be "pushed" at all though. Take a shot that needs even +1 EV exposure correction, and it goes from good to terrible. Not sure why that is, but I've learned to get exposure right up front when using higher ISO or you may end up with a mess. It's not "linear" like other cameras. I can push ISO 800 5D shots quite a bit before I see the noise take over. Not the case with the SD14.

--
Mike
Author: Qimage, Profile Prism
http://www.ddisoftware.com
 
Mike,

As I said, I haven't had the chance to do much by way of SD14 low light/high ISO myself, but that sounds like a fair assessment. Along the way several have offered the perspective that in some ways the SD14/Foveon acts more like film. I have no problem giving the 5D props for its performance in those situations. At the same time if one has the '14 it sounds like it can perform adequately if the exposure is close to correct at high ISO/low light. While I might not buy one for that express purpose, having chosen one for general shooting it's perhaps not a fatal flaw.

Regards,
--
Ed_S
http://www.pbase.com/ecsquires
 
Mike Chaney said:
I've gotten some great ISO 800 shots and some that were pretty poor.
There seems to be a lot more variability than with other cameras. In
good light, my ISO 800 results are better than what I get with the 5D
under the same conditions. I've found that the SD14 ISO 800 shots
cannot be "pushed" at all though. Take a shot that needs even +1 EV
exposure correction, and it goes from good to terrible. Not sure why
that is, but I've learned to get exposure right up front when using
higher ISO or you may end up with a mess. It's not "linear" like
other cameras. I can push ISO 800 5D shots quite a bit before I see
the noise take over. Not the case with the SD14.
Click to expand...
Several people say this. It is quite strange actually.

Lets take an image that is correctly exposed - with the high lights just touching MAX value. Lets say that this looks very nice - from shadows to high lights.

Lets now take an image that is 1 stop less exposed. This image should look fine everywhere - except maybe in the darkest areas. And if you make a dark print, without compensating for under exposure - then it shall look just as fine as the first one - but darker.

If that is not true - something is very weird.

--
Roland
 
You must log in or register to reply here.

Keyboard shortcuts

f
Forum
m
My threads
Mobile site
About
Editorial content
Cameras & Lenses
Community
All content, design, and layout are Copyright © 1998–2025 Digital Photography Review All Rights Reserved.
Reproduction in whole or part in any form or medium without specific written permission is prohibited.
When you use DPReview links to buy products, the site may earn a commission.
©GPS Media - Guides, Products, Services.
Back
Top