Dynamic Range - Try it for yourself, conclude for yourself: 5D III vs. A7r

[Lee Jay]

The reason ISO 100 on a Canon has roughly the same DR as ISO 400 is because the read noise drops by about a stop with each progressive increase to ISO. Hence the flattening of the DR curve on Canon sensors.

Well, it's not that the read noise drops, it's just that all other data & noise off the sensor is amplified 4x compared to ISO 100, which lowers the impact of the read noise. So even though you're throwing away 2 stops of highlight range at ISO 400, 4x (2 EV) darker signal is now being amplified to be brought up to your SNR = 1 (or what have you) threshold, leaving DR largely the same. Cameras with very little downstream read noise don't need the sensor signal amplified to overcome the 30+ electrons of read noise in a Canon camera.

I think there is more to it than that. There is something else going on that flattens the curve relative to read noise at lower ISO, because once your past ISO 400, then you get that linear fit that matches the ideal DR curve. ISO 100 on a 5D III has 33.1e- RN. That is noise added to the signal AFTER the pixels are read (well, after they are amplified and shipped off the sensor, during read). At ISO 200, the pixels are amplified, then read. Since the RN is added down stream, logically, you should still have 33.1e-...the signal was already amplified, so it's still the same strength at this point as an unamplified ISO 100 signal. Similarly, at ISO 400, the pixels are amplified even more, then read. Since RN is added down stream, logically, you should still have 33.1e-. The information coming off the sensor always effectively represents a signal with the same strength, since it's being amplified there before readout.

Your explanation doesn't really explain why read noise drops with higher ISO settings. I don't know exactly why it drops, ...

Yeah, it does.

Let's say, for sake of argument, that 1V = 10,000e- at ISO 100.
Let's say the analog noise is 10mv. 10mV * (10,000e-/1V) = 100e-

Now, let's say that 2V = 10,000e- at ISO 200.
We still have 10mv of analog noise. 10mV * (10,000e-/2V) = 50e-

Got it?

 

[Marsu42]

One tricky thing is that absolute extreme ETTR can make processing tricky as most standard tone curves end leaving you with poor highlight separation and a lot of stock color profiles are twisted so you can get weird color tints and problems.

This is also an issue with Adobe's ACR process version 2012 which is "intelligent", unlike PV2010 before it.

In my experience this means that a shot filling the histogram left to right w/o having a nice and cozy "normal" structure with a bump in the middle is a hassle to postprocess. The shadows/highlight controls react in a different way and/or are not strong enough, they need to be dialed at least to 11. So you regularly end up tweaking the tone curve which takes more time.

 

[privatebydesign]

One tricky thing is that absolute extreme ETTR can make processing tricky as most standard tone curves end leaving you with poor highlight separation and a lot of stock color profiles are twisted so you can get weird color tints and problems.

This is also an issue with Adobe's ACR process version 2012 which is "intelligent", unlike PV2010 before it.

In my experience this means that a shot filling the histogram left to right w/o having a nice and cozy "normal" structure with a bump in the middle is a hassle to postprocess. The shadows/highlight controls react in a different way and/or are not strong enough, they need to be dialed at least to 11. So you regularly end up tweaking the tone curve which takes more time.

You can dial it to 11, just dial it to 10 and export as a TIFF, then reimport and put on another 10.

 

[Sporgon]

First, it's definitely an Exmor. Sony calls it an Exmor in their marketing materials:

http://www.sony.net/Products/di/en-us/products/vq5f/index.html

There are a couple different types of Exmor. There are two versions of the full frame, as well as the BSI Exmor R and RS. I believe the A7s uses a dual-CDS design, one which amplifies at the pixel, performs analog-CDS, then does the CP-ADC and digital-CDS. I don't know why they use the dual CDS approach, according to Sony's original paper on Exmor a number of years ago, they went to digital CDS because analog CDS was a source of banding noise. Anyway...the sensor in the A7s is indeed an Exmor.

Sure, the A7s does have some higher read noise (relative to other Exmor sensors) at base ISO. Isn't it's base ISO at 80? I thought that at ISO 100, it's RN dropped to around 6e-. Also, even though it's base ISO RN is higher, it's still got considerably more DR than any Canon sensor there. I think Sony had to make some tradeoffs at low ISO to achieve the incredible high ISO DR, which seems to take a larger and larger lead over competitors the higher you push ISO.

So what you are saying is: All Exmors are equal but some are more equal than others ?

 

[Lee Jay]

The reason ISO 100 on a Canon has roughly the same DR as ISO 400 is because the read noise drops by about a stop with each progressive increase to ISO. Hence the flattening of the DR curve on Canon sensors.

Well, it's not that the read noise drops, it's just that all other data & noise off the sensor is amplified 4x compared to ISO 100, which lowers the impact of the read noise. So even though you're throwing away 2 stops of highlight range at ISO 400, 4x (2 EV) darker signal is now being amplified to be brought up to your SNR = 1 (or what have you) threshold, leaving DR largely the same. Cameras with very little downstream read noise don't need the sensor signal amplified to overcome the 30+ electrons of read noise in a Canon camera.

I think there is more to it than that. There is something else going on that flattens the curve relative to read noise at lower ISO, because once your past ISO 400, then you get that linear fit that matches the ideal DR curve. ISO 100 on a 5D III has 33.1e- RN. That is noise added to the signal AFTER the pixels are read (well, after they are amplified and shipped off the sensor, during read). At ISO 200, the pixels are amplified, then read. Since the RN is added down stream, logically, you should still have 33.1e-...the signal was already amplified, so it's still the same strength at this point as an unamplified ISO 100 signal. Similarly, at ISO 400, the pixels are amplified even more, then read. Since RN is added down stream, logically, you should still have 33.1e-. The information coming off the sensor always effectively represents a signal with the same strength, since it's being amplified there before readout.

Your explanation doesn't really explain why read noise drops with higher ISO settings. I don't know exactly why it drops, ...

Yeah, it does.

Let's say, for sake of argument, that 1V = 10,000e- at ISO 100.
Let's say the analog noise is 10mv. 10mV * (10,000e-/1V) = 100e-

Now, let's say that 2V = 10,000e- at ISO 200.
We still have 10mv of analog noise. 10mV * (10,000e-/2V) = 50e-

Got it?

I'm still not sure that fully explains it.

Of course it doesn't. But it explains why it's not a constant, as you said it should "logically" be above.

 

[Marsu42]

More equal? ??? I don't know what that means.

You really don't? It's from George Orwell's Animal Farm, a fable about the communist revolution. After the animals get rid of their jailers and start into a bright new era with all animals being equal, it turns out that the top cadres are "more equal", murder their rivals and enslave the tamer animals again. Always a good quote

https://en.wikipedia.org/wiki/Animal_Farm#Plot_summary