MP wise, where's sweet spot on FF sensors...
/forum/topic/1155125/3

1       2       3      
4
       5       end

jcolwell
Registered: Feb 10, 2005
Total Posts: 20519
Country: Canada

mttran wrote:
some day soon, sony getting closer and closer to this P65 sample







Michael,

You stop that. Right now!

Incredible photo. Too bad Mirex came out with the M645 TS adapter first...

Cheers, Jim


BrianO
Registered: Aug 21, 2008
Total Posts: 8547
Country: United States

RobDickinson wrote: buckets have no meaningful place in sensor discussions.

WHAT?!? You've never heard of analogies? Even Einstein used "thought experiments" to work with various concepts. I suppose you'll say that boys bouncing balls on moving trains have no meaningful place in discussions of relativity theory.



jcolwell
Registered: Feb 10, 2005
Total Posts: 20519
Country: Canada

BrianO wrote:
RobDickinson wrote: buckets have no meaningful place in sensor discussions.

WHAT?!? You've never heard of analogies? Even Einstein used "thought experiments" to work with various concepts. I suppose you'll say that boys bouncing balls on moving trains have no meaningful place in discussions of relativity theory.


How fast are the trains going?



RobDickinson
Registered: Sep 25, 2009
Total Posts: 3390
Country: New Zealand

People who talk about buckets tend to forget that you can hold just as much water in a lot of smaller buckets as fewer larger buckets.

I understand its an analogy, its just one that really doesnt work.



StillFingerz
Registered: Jul 29, 2010
Total Posts: 3283
Country: United States

The best "sweet spot" is the sensor in hand and a bit of superb glass to make it shine



R.H. Johnson
Registered: Oct 08, 2006
Total Posts: 1977
Country: United States

alundeb, R.H. Johnson wrote:

the bigger the bucket the more water it will hold. the bigger the photo-site the more light it can absorb.

How about counting for the whole sensor and not individual pixels.

Canon 5DIII FWC: 67531 e- x 22 M = 1486 G e-
Nikon D800 FWC: 44972 e- x 36 M = 1619 G e-

http://sensorgen.info/

The sensor with smaller pixels can in this case absorb more light than the sensor with larger pixels.

RobDickinson:

People who talk about buckets tend to forget that you can hold just as much water in a lot of smaller buckets as fewer larger buckets.

I understand its an analogy, its just one that really doesnt work.

the bigger the bucket ie. photosites. i once read that larger deep welled photo-sites have a lower S/N ratio than smaller buckets thus they produce less and a more appealing noise pattern in the image. yes u can pack a full frame sensor with many small buckets but at what cost to image quality, noise pattern, or S/N? if you knew what your taking about you would not compare apples to oranges.

imho there is a point of diminished returns, the sweet spot. imo that is why Canon stepped back to 18 mega pixels in their flagship 1DX. the 1DX may be the camera i wished for a 5D classic sensor in a 1 series body with 6 extra mega pixel to boot.

i read about the pixel argument in a post on this forum being discussed by some of the FM heavy scientific types years ago. perhaps you gentlemen should search the archives find the post and educate your selves on the topic.



uz2work
Registered: Mar 04, 2004
Total Posts: 11682
Country: United States

R.H. Johnson wrote:


imho there is a point of diminished returns, the sweet spot. imo that is why Canon stepped back to 18 mega pixels in their flagship 1DX. the 1DX may be the camera i wished for a 5D classic sensor in a 1 series body with 6 extra mega pixel to boot.




I suspect that having 18 MP on the 1DX has very little, if anything, to do with image quality but that, instead, 18 MP is the largest number of pixels that Canon could put on the sensor and still have it shoot at 12 fps. Once Canon decided that the 1DX was going to be a jack-of-all trades camera to replace both the 1D and 1Ds series bodies, compromises needed to be made, and limiting the megapixel count to accommodate a high frame rate was one of those compromises.

Les



mttran
Registered: Nov 03, 2005
Total Posts: 6638
Country: United States

jcolwell wrote:
mttran wrote:
some day soon, sony getting closer and closer to this P65 sample


Michael,

You stop that. Right now!
...

Cheers, Jim


Yes, sir - i've heard you loud and clear but it won't stop my praying silently. Whoever gets this clarity IQ same resolution first ...will be the first dslr winner for most shooter here



BrianO
Registered: Aug 21, 2008
Total Posts: 8547
Country: United States

RobDickinson wrote: People who talk about buckets tend to forget that you can hold just as much water in a lot of smaller buckets as fewer larger buckets. I understand its an analogy, its just one that really doesnt work.

I disagree; I think it's a very good analogy. Most noise is a function of signal-to-noise ratio, which happens at the pixel level, and the amount of light gathered by the entire sensor has little to do with SNR; the amount of light gathered by each sensel does.

Let's say for example that a given sensor generates one photon's-worth of noise at each pixel. Let's then imagine a sensel so small that at -- say -- 1/250 second only 2 photons can get in at a given light level. Small bucket = 2:1 SNR.

Now let's say we have a sensor with the same number of total pixels, but each sensel is larger due to closer spacing (sensel pitch), etc. This one can gather 10 photons in 1/250. It has the same number of buckets, but larger buckets = 10:1 SNR = less noise.

Furthermore, let's look at exposure. Regardless of the number of pixels, each pixel has to be exposed to a certain level to accurately reflect the scene. Smaller sensels, because they gather less light in a given amount of time, would require higher ISO settings at the same shutter speed than a sensor with larger sensels. Again, it's the size of the light-holding buckets that affects exposure, not the number of them. The number affects the fineness of detail that can be recorded, but not the exposure level.

At least that's how I understand it.



BrianO
Registered: Aug 21, 2008
Total Posts: 8547
Country: United States


BrianO wrote: ...I suppose you'll say that boys bouncing balls on moving trains have no meaningful place in discussions of relativity theory.

jcolwell wrote: How fast are the trains going?

299,792,458 metres per second, as seen from either train or from a stationary point between them.



alundeb
Registered: Nov 06, 2005
Total Posts: 4243
Country: Norway

R.H. Johnson wrote:
i read about the pixel argument in a post on this forum being discussed by some of the FM heavy scientific types years ago. perhaps you gentlemen should search the archives find the post and educate your selves on the topic.


I was one of them.



alundeb
Registered: Nov 06, 2005
Total Posts: 4243
Country: Norway

BrianO wrote:
RobDickinson wrote: People who talk about buckets tend to forget that you can hold just as much water in a lot of smaller buckets as fewer larger buckets. I understand its an analogy, its just one that really doesnt work.

I disagree; I think it's a very good analogy. Most noise is a function of signal-to-noise ratio, which happens at the pixel level, and the amount of light gathered by the entire sensor has little to do with SNR; the amount of light gathered by each sensel does.

Let's say for example that a given sensor generates one photon's-worth of noise at each pixel. Let's then imagine a sensel so small that at -- say -- 1/250 second only 2 photons can get in at a given light level. Small bucket = 2:1 SNR.

Now let's say we have a sensor with the same number of total pixels, but each sensel is larger due to closer spacing (sensel pitch), etc. This one can gather 10 photons in 1/250. It has the same number of buckets, but larger buckets = 10:1 SNR = less noise.

Furthermore, let's look at exposure. Regardless of the number of pixels, each pixel has to be exposed to a certain level to accurately reflect the scene. Smaller sensels, because they gather less light in a given amount of time, would require higher ISO settings at the same shutter speed than a sensor with larger sensels. Again, it's the size of the light-holding buckets that affects exposure, not the number of them. The number affects the fineness of detail that can be recorded, but not the exposure level.

At least that's how I understand it.


Then you need to refresh your understanding.

The buckets are equally tall, you just have narrower buckets standing next to each other.

Smaller sensels do not require higher ISO at the same shutter speed.

If you are talking about smaller sensels distrubuted on the same area, they require exactly the same ISO.

If you are talking about a smaller sensor, it requires LOWER ISO to produce an image from the same perspective / subject distance, with the same field of view, the same depth of field and the same shutter speed, because you will use a wider F-stop to get the same physical aperture diameter( and hence DOF) with a shorter lens.



alundeb
Registered: Nov 06, 2005
Total Posts: 4243
Country: Norway

BrianO wrote:
Let's say for example that a given sensor generates one photon's-worth of noise at each pixel. Let's then imagine a sensel so small that at -- say -- 1/250 second only 2 photons can get in at a given light level. Small bucket = 2:1 SNR.

Now let's say we have a sensor with the same number of total pixels, but each sensel is larger due to closer spacing (sensel pitch), etc. This one can gather 10 photons in 1/250. It has the same number of buckets, but larger buckets = 10:1 SNR = less noise.

.


It goes like this:

Sensor 1 example: 50 sensels that can gather 2000 photons each, with read noise 1 photon. The photon shot noise at saturation will be sqrt(2000) = 44.72 photons. If we combine shot noise and read noise in qaudrature, we get sqrt(2000 + 1*1) = 44.73 photons.

S/N = 44.71

Sensor 2 example: 10 sensels that can gather 10000 photons each, with read noise 1 photon. The photon shot noise at saturation will be sqrt(10000) = 100 photons. If we combine shot noise and read noise in qaudrature, we get sqrt(10000 + 1^2) = 100.005 photons.

S/N = 99.995


If you resample sensor 1 to 10 pixels just by pixel binning, the signal will combine linearly and the noise will combine in quadrature. sqrt(5*(44.73^2)) = 100.02 photons.

S/N = 99.980




15Bit
Registered: Jan 27, 2008
Total Posts: 3780
Country: Norway

We seem to be discussing the maximum number of photons a sensel can hold (i.e. how big the bucket is). I think this is a red herring. From what i can see the big advances are at the low light end - as shown in the relative shadows performance of the D800 over the 5D3.

I would expect that a sensel has an activation threshold to overcome - a minimum number of photons that it can measure. And also the general performance of the sensel at low photon counts is important. What is this threshold? How does the measurement accuracy scale with photon count? It doesn't matter matter how many sensels you can bin together if none of them has received enough photons to measure a signal above the noise threshold - all you do then is bin noise, not data.

At very low photon counts fewer and larger sensels should outperform many smaller sensels with binning. Logically the shadow and high ISO performance of a sensor is going to be linked to this. Of course the clever thing that has been happening in recent years is that the sensitivity of the sensels has improved alongside their downsizing, and i think it is more this which makes higher density sensors possible, not so much the ability to fabricate at a smaller size.



alundeb
Registered: Nov 06, 2005
Total Posts: 4243
Country: Norway

15Bit wrote:
We seem to be discussing the maximum number of photons a sensel can hold (i.e. how big the bucket is). I think this is a red herring. From what i can see the big advances are at the low light end - as shown in the relative shadows performance of the D800 over the 5D3.

I would expect that a sensel has an activation threshold to overcome - a minimum number of photons that it can measure. And also the general performance of the sensel at low photon counts is important. What is this threshold? How does the measurement accuracy scale with photon count? It doesn't matter matter how many sensels you can bin together if none of them has received enough photons to measure a signal above the noise threshold - all you do then is bin noise, not data.

At very low photon counts fewer and larger sensels should outperform many smaller sensels with binning. Logically the shadow and high ISO performance of a sensor is going to be linked to this. Of course the clever thing that has been happening in recent years is that the sensitivity of the sensels has improved alongside their downsizing, and i think it is more this which makes higher density sensors possible, not so much the ability to fabricate at a smaller size.



All of this makes sense, except there is no activation threshold. We cannot count individual photons, only measure the average effect on the current, and that effect does not disappear below any threshold.



jorkata
Registered: Sep 02, 2009
Total Posts: 669
Country: United States

alundeb wrote:
All of this makes sense, except there is no activation threshold. We cannot count individual photons, only measure the average effect on the current, and that effect does not disappear below any threshold.


And what is this thing that they call 'noise floor' then .



alundeb
Registered: Nov 06, 2005
Total Posts: 4243
Country: Norway

jorkata wrote:
alundeb wrote:
All of this makes sense, except there is no activation threshold. We cannot count individual photons, only measure the average effect on the current, and that effect does not disappear below any threshold.


And what is this thing that they call 'noise floor' then .


The noise level with zero signal.
.



jorkata
Registered: Sep 02, 2009
Total Posts: 669
Country: United States

And of course that's the threshold that 15Bit is talking about.



RobDickinson
Registered: Sep 25, 2009
Total Posts: 3390
Country: New Zealand

and something sony is much better at than canon.



jorkata
Registered: Sep 02, 2009
Total Posts: 669
Country: United States

Overall, the bucket analogys is correct ... for ideal sensors.
But in practice, sensors are not ideal.

I'm waiting for DxO to test the 1DX.
The 1DX and 5DIII are made with the exact same technology - with only pixel size being different.
So, in theory, the 1DX and 5DIII should have identical (normalized) performance.
We'll see if this will be the case indeed.



1       2       3      
4
       5       end