From: Bob Newman on
On 21 Jul, 00:32, "Roger N. Clark (change username to rnclark)"
<usern...(a)qwest.net> wrote:
> Bob Newman wrote:
> > On 20 Jul, 22:41, "Roger N. Clark (change username to rnclark)"
> > <usern...(a)qwest.net> wrote:
> >> Bob Newman wrote:
> >>> Yes, this is confusing, but working in one unit only also causes
> >>> confusion.  Particularly concerning 'read noise' in electrons, when
> >>> it's a voltage referred noise.
> >> Scientifically, it is better in my opinion (as a scientist) to
> >> work in units that relate to the scene.  That could be watts/sq meter,
> >> photons, or in this case electrons which can be directly
> >> correlated to the photons captured.  In my scientific work,
> >> we always work in units related to the subject, not voltage in
> >> the instrument.
>
> >> Roger
>
> > That's fair enough. The problem is, when it gets in the way of people
> > conceptualising what's actually happening. In this case, the photo-
> > electron referred noise figure is distinctly unhelpful in working
> > one's way through the actual noise sources, and the variously
> > amplified versions of them which appear in the final signal.
>
> But we consumers can not measure these engineering parameters unless
> we disassemble the cameras.  We can only measure what is produced
> by the camera output for a given known input.  Having zero
> initial information it is pretty impressive that we can derive
> the photon count in the first place, as long as we can get access
> to raw data.
I really do agree it's impressive, and a very useful body of work.
However, I think you need to understand the limitations of such data,
and how far you can stretch it in terms of analysis of trends from it.
For instance, I have had your site quoted at me as irrefutable
evidence that the 5D has a higher FWC than the D3. Certainly, most of
your graphs show the 5D outperforming the D3. Does it? I don't know,
but some of your figures are taken from Peter Facey, yet in his direct
comparison, the D3 has a higher FWC. Because of the constraints, these
figures are not exact, and it's fruitless to talk about
interpretations which are in the noise, unless you can average a very
large numebr of such datapoints to reduce the noise ;-)
> > One of
> > your major criticisms of John's stuff is based on an assumption that
> > 'read noise', referred to photo-electrons, remains approximately
> > constant.
>
> I never said read noise is constant, at least between sensors.
> What I said was there was NO correlation of read noise with pixel size.
> Read noise is independent of ISO because ISO is a post sensor read gain.
> Read noise varies a lot between different sensors because of different
> methods for reading the signal and suppressing noise.
I believe there is a weak corelation in theory (in that input referred
read noise is clearly scaled by the cell capacitance, which is weakly
related to cell area). The best you can say is that there is no
evidence in your data to support that, not that there is no
correlation.
> > I cannot see why this should be so, since the translation
> > from what it is (a noise current in the amplifier stage) to apparent
> > photo-electrons must be due to the cell capacitance, which is at least
> > loosely related to cell area. The real truth is that a lot of the
> > assumptions are based on observations of sensors which are not things
> > that occur naturally in inevitable configurations. What you are
> > observing is the result of conscious design choices, to use them as
> > evidence of an inevitable trend is hardly 'scientific'.
>
> You might have a look at:http://learn.hamamatsu.com/articles/ccdsnr.html
OK, but it's not particularly illuminating when we get to the level of
discussion we're at now.
> The pixel and sensor size issue is well known in systems design
> for scientific instruments, on aircraft, telescopes, and spacecraft.
> See, for example,http://en.wikipedia.org/wiki/Etendue
>
> There are scientific reasons for observed trends.
I don't think anyone is arguing about the science (apart from my read
noise scaling thing, which might have been dealt with by Emils
response, but I think not). It's the engineering consequences of that
science. Everyone is at least agreed that the photon shot noise is, in
the end an area based, not a pixel based thing. Where the disagreement
is, is whethere engineering solutions may exist which allow you to
utilise that fact to make a camera which can bot produce high
resolution and, when you want it, low noise at smaller magnifications.
In most of the applications you deal with, the engineering constraints
are very different. For instance, spacecraft rarely use processor
designs less that 20 years old, and data processing is at a premium.
An imaging system design which produces humungous amounts of data, but
rrequires a lot of signal processing to release low noise images from
it is hardly ever going to be a goer. By contrast, consumer cameras
use commodity, Moores law empowered processors and memory. Often,
piling in processing can be a better solution to fancy high cost
optics.
> Roger- Hide quoted text -
>
> - Show quoted text -

From: Bob Newman on
On 20 Jul, 23:45, ejmartin <ejm_60...(a)yahoo.com> wrote:
> On Jul 20, 9:37 am, Bob Newman <bob.csx...(a)gmail.com> wrote:
Back to this:
> Carrying this analysis a step further, can we assume that Nf is
> thermal noise?  Then <V^2>=kT/C, and so at high gain (thus dropping
> the effects of Nb) one has
>
> (Cs/Qe)*(Sqrt[kT/Cs] +q Nm +q Nb/Gi)
>
> Cs should be proportional to the collection area, as this gets
> asymptotically small the input-referred noise should scale according
> to this formula as the sqrt of the collection area, ie with the linear
> size of the pixel.  Actually it would decrease somewhat faster than
> that, for a given level of technology the size of the support
> electronics is fixed and the collection area will decrease *faster*
> than linearly with the pixel spacing.  We can make the input referred
> read noise as small as we want if we let the photosensitive area go to
> zero!
OK, so at least we're agreed that input referred noise does scale down
with pixel size, although I don't think you should have that Cs in the
quotient of the first noise term. The limit you talk about at the end
sounds right, no signal = no noise, which seems better to me than no
signal = lots of noise.
> If the collection area is Ac and the support electronics occupies Ae,
> and the pixel spacing is d, one has d^2=Ac+Ae.  The FWC goes as Ac,
> the read noise as sqrt[Ac], and the DR per area is (see above post)
>
> DR/area ~ const * Ac/(sqrt[Ac] * d) ~ const * sqrt[1-(Ae/d^2)]
>
> So with these assumptions  -- fixed area requirements for support
> electronics -- DR per area goes down as the pixels are shrunk.  One
> can only decrease pixel spacing and maintain DR per area if the
> support electronics shrinks in proportion to the pixel size, which
> makes a lot of sense.
This assumes that the pixel is simpy not being scaled. Since the noise
of the source follower (and for that matter, all the other
transistors) is a shape (L/W) based thing, rather than size based
thing, there is no reason for the support electronics to not be simply
scaled with the pixel, so long as they haven't reached a limit at
which the standard noise models break down (for instance, when
tunneling through or across the gate becomes an issue) what we don't
know is how close current sensors are to those limits. If we restrict
our argument within sensible limits (like will the D3x be a Good Thing
or a Bad Thing) then John's test would seem to indicate that even
within a 2um active pixel there is space for the transistors to work
well and for it to collect some light. Therefore we might conclude
that in the range 8-4um there is still scope for scalability. Whether
designers choose to use it, or what design parameters they work to, is
a different issue.
> One might also worry that there are input referred noises that might
> not scale.  Can we be sure that there are no constant sources of input-
> referred noise, for instance noise in the 4T arrangement that reads
> out the photoelectron count?- Hide quoted text -
One might worry about that. The transistor noises (for instance, the
reset noise and the noise from the transfer transistors) are shape
related, so don't scale. Those will favour sensors with higher FWC,
which might be one reason why designers might tend to keep FWC as
large as possible, and therefore not get the scaling benefits with
regard to read noise. I think there's little doubt that large pixels
will have a higher DR at an image level than small ones, it's the
slope that's really under discussion. I think it's probably pretty
flat.
> - Show quoted text -

From: ejmartin on
On Jul 21, 8:04 am, Bob Newman <bob.csx...(a)gmail.com> wrote:
> On 20 Jul, 23:45, ejmartin <ejm_60...(a)yahoo.com> wrote:
>
> > On Jul 20, 9:37 am, Bob Newman <bob.csx...(a)gmail.com> wrote:
> Back to this:
> > Carrying this analysis a step further, can we assume that Nf is
> > thermal noise? Then <V^2>=kT/C, and so at high gain (thus dropping
> > the effects of Nb) one has
>
> > (Cs/Qe)*(Sqrt[kT/Cs] +q Nm +q Nb/Gi)
>
> > Cs should be proportional to the collection area, as this gets
> > asymptotically small the input-referred noise should scale according
> > to this formula as the sqrt of the collection area, ie with the linear
> > size of the pixel. Actually it would decrease somewhat faster than
> > that, for a given level of technology the size of the support
> > electronics is fixed and the collection area will decrease *faster*
> > than linearly with the pixel spacing. We can make the input referred
> > read noise as small as we want if we let the photosensitive area go to
> > zero!
>
> OK, so at least we're agreed that input referred noise does scale down
> with pixel size, although I don't think you should have that Cs in the
> quotient of the first noise term. The limit you talk about at the end
> sounds right, no signal = no noise, which seems better to me than no
> signal = lots of noise.> If the collection area is Ac and the support electronics occupies Ae,
> > and the pixel spacing is d, one has d^2=Ac+Ae. The FWC goes as Ac,
> > the read noise as sqrt[Ac], and the DR per area is (see above post)
>
> > DR/area ~ const * Ac/(sqrt[Ac] * d) ~ const * sqrt[1-(Ae/d^2)]
>
> > So with these assumptions -- fixed area requirements for support
> > electronics -- DR per area goes down as the pixels are shrunk. One
> > can only decrease pixel spacing and maintain DR per area if the
> > support electronics shrinks in proportion to the pixel size, which
> > makes a lot of sense.
>
> This assumes that the pixel is simpy not being scaled. Since the noise
> of the source follower (and for that matter, all the other
> transistors) is a shape (L/W) based thing, rather than size based
> thing, there is no reason for the support electronics to not be simply
> scaled with the pixel, so long as they haven't reached a limit at
> which the standard noise models break down (for instance, when
> tunneling through or across the gate becomes an issue) what we don't
> know is how close current sensors are to those limits. If we restrict
> our argument within sensible limits (like will the D3x be a Good Thing
> or a Bad Thing) then John's test would seem to indicate that even
> within a 2um active pixel there is space for the transistors to work
> well and for it to collect some light. Therefore we might conclude
> that in the range 8-4um there is still scope for scalability. Whether
> designers choose to use it, or what design parameters they work to, is
> a different issue.> One might also worry that there are input referred noises that might
> > not scale. Can we be sure that there are no constant sources of input-
> > referred noise, for instance noise in the 4T arrangement that reads
> > out the photoelectron count?- Hide quoted text -
>
> One might worry about that. The transistor noises (for instance, the
> reset noise and the noise from the transfer transistors) are shape
> related, so don't scale. Those will favour sensors with higher FWC,
> which might be one reason why designers might tend to keep FWC as
> large as possible, and therefore not get the scaling benefits with
> regard to read noise. I think there's little doubt that large pixels
> will have a higher DR at an image level than small ones, it's the
> slope that's really under discussion. I think it's probably pretty
> flat.
>
> > - Show quoted text -

What puzzles me about your analysis is that you get the input-referred
read noise proportional to the sensel capacitance, with no appreciable
dependence of the front end noise Nf on sensel size; I would think
also that the FWC is also proportional to the sensel capacitance.
Then the DR per pixel is totally independent of the capacitance of the
sensel, which doesn't seem right to me. Moreover, it predicts DR/area
actually goes up in inverse proportion to the pixel spacing, which
also seems a bit goofy.
From: ejmartin on
On Jul 21, 8:04 am, Bob Newman <bob.csx...(a)gmail.com> wrote:
> On 20 Jul, 23:45, ejmartin <ejm_60...(a)yahoo.com> wrote:
>
>
> OK, so at least we're agreed that input referred noise does scale down
> with pixel size, although I don't think you should have that Cs in the
> quotient of the first noise term. The limit you talk about at the end
> sounds right, no signal = no noise, which seems better to me than no
> signal = lots of noise.

This of course assumes that there are no noise sources that are
constant as a function of pixel size. As you say below, reset noise
might be one of those. Is CDS 100% efficient, or only up to a few
electrons? And transistor noises at the pixel don't scale, according
to the summary you linked. So we're agreed *if* it can be shown that
there are no noise sources that don't scale.

> > One might also worry that there are input referred noises that might
> > not scale. Can we be sure that there are no constant sources of input-
> > referred noise, for instance noise in the 4T arrangement that reads
> > out the photoelectron count?- Hide quoted text -
>
> One might worry about that. The transistor noises (for instance, the
> reset noise and the noise from the transfer transistors) are shape
> related, so don't scale. Those will favour sensors with higher FWC,
> which might be one reason why designers might tend to keep FWC as
> large as possible, and therefore not get the scaling benefits with
> regard to read noise. I think there's little doubt that large pixels
> will have a higher DR at an image level than small ones, it's the
> slope that's really under discussion. I think it's probably pretty
> flat.

The calculations I did earlier in the thread show that the FZ50 does
about a half stop better per area than the 1D3 at base ISO using the
off-photosite limited DR figures that the 1D3 delivers. If one uses
the full sensor DR of the 1D3, the FZ50 pixels will have to have read
noise in the range of 1 electron, input referred, in order to match
the full sensor DR of the 1D3 when it's not crippled by downstream
electronics. Such a pixel is very close to a photon-counting device,
which would be great if it could be achieved, but that means that all
the potential noises from the pixel must be eliminated down to the
single electron level. I would be pleasantly surprised if that were
possible.


Also, a followup point to the comment about capacitance dropping out
of pixel DR in your model; it means the DR is independent of fill
factor (only affecting QE I suppose), which also doesn't sound right
to me.
From: Bob Newman on
On 21 Jul, 16:39, ejmartin <ejm_60...(a)yahoo.com> wrote:
> On Jul 21, 8:04 am, Bob Newman <bob.csx...(a)gmail.com> wrote:
>
>
>
> > On 20 Jul, 23:45, ejmartin <ejm_60...(a)yahoo.com> wrote:
>
> > > On Jul 20, 9:37 am, Bob Newman <bob.csx...(a)gmail.com> wrote:
> > Back to this:
> > > Carrying this analysis a step further, can we assume that Nf is
> > > thermal noise? Then <V^2>=kT/C, and so at high gain (thus dropping
> > > the effects of Nb) one has
>
> > > (Cs/Qe)*(Sqrt[kT/Cs] +q Nm +q Nb/Gi)
>
> > > Cs should be proportional to the collection area, as this gets
> > > asymptotically small the input-referred noise should scale according
> > > to this formula as the sqrt of the collection area, ie with the linear
> > > size of the pixel. Actually it would decrease somewhat faster than
> > > that, for a given level of technology the size of the support
> > > electronics is fixed and the collection area will decrease *faster*
> > > than linearly with the pixel spacing. We can make the input referred
> > > read noise as small as we want if we let the photosensitive area go to
> > > zero!
>
> > OK, so at least we're agreed that input referred noise does scale down
> > with pixel size, although I don't think you should have that Cs in the
> > quotient of the first noise term. The limit you talk about at the end
> > sounds right, no signal = no noise, which seems better to me than no
> > signal = lots of noise.> If the collection area is Ac and the support electronics occupies Ae,
> > > and the pixel spacing is d, one has d^2=Ac+Ae. The FWC goes as Ac,
> > > the read noise as sqrt[Ac], and the DR per area is (see above post)
>
> > > DR/area ~ const * Ac/(sqrt[Ac] * d) ~ const * sqrt[1-(Ae/d^2)]
>
> > > So with these assumptions -- fixed area requirements for support
> > > electronics -- DR per area goes down as the pixels are shrunk. One
> > > can only decrease pixel spacing and maintain DR per area if the
> > > support electronics shrinks in proportion to the pixel size, which
> > > makes a lot of sense.
>
> > This assumes that the pixel is simpy not being scaled. Since the noise
> > of the source follower (and for that matter, all the other
> > transistors) is a shape (L/W) based thing, rather than size based
> > thing, there is no reason for the support electronics to not be simply
> > scaled with the pixel, so long as they haven't reached a limit at
> > which the standard noise models break down (for instance, when
> > tunneling through or across the gate becomes an issue) what we don't
> > know is how close current sensors are to those limits. If we restrict
> > our argument within sensible limits (like will the D3x be a Good Thing
> > or a Bad Thing) then John's test would seem to indicate that even
> > within a 2um active pixel there is space for the transistors to work
> > well and for it to collect some light. Therefore we might conclude
> > that in the range 8-4um there is still scope for scalability. Whether
> > designers choose to use it, or what design parameters they work to, is
> > a different issue.> One might also worry that there are input referred noises that might
> > > not scale. Can we be sure that there are no constant sources of input-
> > > referred noise, for instance noise in the 4T arrangement that reads
> > > out the photoelectron count?- Hide quoted text -
>
> > One might worry about that. The transistor noises (for instance, the
> > reset noise and the noise from the transfer transistors) are shape
> > related, so don't scale. Those will favour sensors with higher FWC,
> > which might be one reason why designers might tend to keep FWC as
> > large as possible, and therefore not get the scaling benefits with
> > regard to read noise. I think there's little doubt that large pixels
> > will have a higher DR at an image level than small ones, it's the
> > slope that's really under discussion. I think it's probably pretty
> > flat.
>
> > > - Show quoted text -
>
> What puzzles me about your analysis is that you get the input-referred
> read noise proportional to the sensel capacitance, with no appreciable
> dependence of the front end noise Nf on sensel size;
Yes
> I would think
> also that the FWC is also proportional to the sensel capacitance.
> Then the DR per pixel is totally independent of the capacitance of the
> sensel, which doesn't seem right to me.
Why should it not be? What we have done is a simple scale, and none of
the noise sources (or significant ones) appears to be dimension
related.
> Moreover, it predicts DR/area
> actually goes up in inverse proportion to the pixel spacing, which
> also seems a bit goofy.
Whether it's goofy or not just depends on your preconceptions. If you
look at it another way, we're reading through a greater number of
'channels', why should that not cause less noise? Also, it also seems
to work in the limit, the one electron pixel, which requires little
pixel DR (note to self: have a look at the DR in a dynamic RAM cell)
to produce an image with zero read noise. Remember also that we're
talking here only about the read noise, shot noise controlled DR is of
course smaller per pixel and the same per area.
Still, if you don't believe it, find the hole in the reasoning.