Testing my GH2's sensor

Started Feb 20, 2011 | Discussions thread
Kopend05
Contributing MemberPosts: 868
Like?
Re: Testing my GH2's sensor
In reply to bg2b, Feb 23, 2011

bg2b wrote:

For amusement, I've been doing some tests on my GH2's sensor; here are a few random observations.

"Real" ISOs are probably just 160, 320, and 640. 200, 400, and 800 show odd artifacts (gaps or near gaps) in the RAW histograms. 250, 500, and 1000 show them even more. Everything beyond ISO 1000 is completely artificial; they're obviously just digital scalings of lower ISOs. So if you're a RAW-only type, there's no point in going past ISO 640 (or maybe 800, which is probably the least artificial-looking of the x00's). All you're doing is just costing yourself highlight headroom without gaining any signal-to-noise ratio (SNR) in the shadows.

I derived the SNR curves using the procedures described in these references:

http://www.clarkvision.com/imagedetail/evaluation-1d2/ (seems to be offline at the moment)
http://theory.uchicago.edu/~ejm/pix/20d/tests/noise/index.html

Here are the characteristics of ISOs 160, 320, and 640:

ISO 160
Well capacity 12350 electrons
3.1 electrons/RAW unit (electrons = detected photons)
Model read noise 6.5 electrons

(The "model read noise" is from a numeric fit of the measured SNR curve to a simple "total noise = photon shot noise plus constant read noise" model)
Per-pixel "useful" dynamic range (DR for SNR 12dB) 8.5 stops
Per-pixel DXO-like dynamic range (DR for SNR 0dB) 10.8 stops

Full SNR curve (The colors indicate samples from different pixel colors; there are two greens, green1 and green2. The horizontal line is at 12dB, and the diagonal line is the SNR limit due solely to photon shot noise.):

ISO 320
Well capacity 6200 electrons
1.6 electrons/RAW unit
Model read noise 4.0 electrons
Per-pixel useful dynamic range 7.8 stops
Per-pixel DXO-like dynamic range 10.4 stops
Full SNR curve:

ISO 640
Well capacity 3250 electrons
0.8 electrons/unit
Model read noise 3.5 electrons
Per-pixel useful dynamic range 7.0 stops
Per-pixel DXO-like dynamic range 9.7 stops
Full SNR curve:

I didn't measure the true saturation ISOs, but according to DXO, ISO 160 corresponds to a saturation ISO of 167, and going by the well capacities for 320 and 640, I think they must also be labeled accurately.

If you want to compare to DXO's "print" graphs, you need to add 0.5 stop to the DR numbers.

The camera's read noise performance seems actually pretty good--not state-of-the-art at base ISO, but certainly no slouch. Where it's lacking somewhat is in full-well capacity. I'd guess the limit is actually the 12-bit converter. A 14-bit converter with the 12K electrons from ISO 160 and the same read noise as ISO 640 would give excellent DR at high ISO. Maybe for the GH3...

The unity gain ISO is approximately ISO 640, with just below one electron per RAW unit. It would be ISO 500, but that's not a real ISO. Probably 640 is the best one to use when you're light-limited and you're not worried about highlight clipping, except maybe you have to consider thermal noise differences among the different ISOs. I haven't tested thermal noise performance.

You don't work for NASA do you ?

Reply   Reply with quote   Complain
Post (hide subjects)Posted by
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark post MMy threads
Color scheme? Blue / Yellow