ISO 160 better than ISO 125? Canon

Can someone confirm that this is real. It appears that Canon achieves ISO in different ways so that lower ISO's may exhibit more noise than higher? See DP thread (http://forums.dpreview.com/forums/readf ... d=19721647)

Thanks,
Paul

E.J. and others have commented on this type of situation before, although more from the standpoint of getting better DR capture around ISO 160. See http://www.naturescapes.net/phpBB3/view ... hp?t=69226 for example. So it appears to be a real phenomenon.

From my casual reading, sensors have a "native" sensitivity. All other sensitivities achieved by dialing in different ISO are really just performing digital signal processing, and hence impact both noise and DR.

Royce, I don't think different ISOs are doing DSP on the signal. E.J. can confirm, but I think it's analog--when ISO is increased, the gain is boosted prior to A-D conversion. (If it was digital, you'd be able to get the same results in post processing.) Boosting the gain decreases the signal-noise ratio.

-Paul

You're right. Something I read this morning indicated it was being done after, but a moment of thinking indicates that can't possibly be right for ISO. I'm not a hardware guy.

The difference between ISO 125 and 160 is so minimal as for it not to matter. The "native" ISO of Canon CMOS sensors is about ISO 160 so by going lower in theory you give up a bit of dynamic range but also get a bit of lower noise. Likewise by going to ISO 200 you lose a bit of dynamic range and it gets a bit noisier. But the reality of the situation is that these 1/3 stop ISO changes make no practical difference in noise or dynamic range. You will however start to notice differences with full stop ISO changes so for example ISO 320 or 400 is noticeably noisier than ISO 160 or 200 and ISO 50 has significantly less dynamic range than ISO 160.

I've read some claims that the multiples of the native resolution are "better" than the adjusted ones. Based on what EJ said, would we gain an (admittedly small) advantage then by trying to use multipes of 16 when we can conveniently do so? 160, 320, 640, hopefully never higher than that?

Paul Skoczylas wrote:Royce, I don't think different ISOs are doing DSP on the signal. E.J. can confirm, but I think it's analog--when ISO is increased, the gain is boosted prior to A-D conversion. (If it was digital, you'd be able to get the same results in post processing.) Boosting the gain decreases the signal-noise ratio.

-Paul

The thread being referred to addresses an issue that is probably specific to the 30D. John Sheehy (I suspect an astrophotographer, based on the software he is using) analyzed the direct RAW data from the 30D at different 1/3 stop ISO increments and discovered that the only native ISO's (those which are being amplified as analogue signals to the appropriate gain) are ISO100, 200, 400, 800, 1600. The intermediate ones are achieved by doing a simple divide/multiply on the digitized data *after* A/D conversion. Consequently there is more shadow noise at ISO125 (secretly ISO100 pushed by 1/3 stop) than there is at ISO160 (ISO200 lowered by 1/3 stop EC). It seems that the intermediate ISO's on the 30D have less information in the RAW file than the 'standard' ones. For instance the pushed ones have less headroom in highlights, and more noise in shadows.

After he first announced this a couple of months ago, people sent him RAW files from other Canon cams such as the 5D, and the same analysis seemed to indicate that on these cameras the signal amplification is done properly, prior to A/D conversion. Somewhere in the thread referenced by the OP here there is a link to the previous DPReview thread where all this started.

Remember that the 30D is an upgraded 20D which didn't have 1/3 stop ISOs. So I can totally believe that Canon took a short cut on the 30D to provide 1/3 stops.

Emil, that explains the saw-tooth character of Sheehy's initial graph, as noted in the DPR thread. Reading back, this is also where I got the misfired thought that some digital processing was being done in the processing of ISO sensitivity. But not as a general case, just for the 30D, and just for the +/- 1/3 stop settings.

Does it matter then whether I work at say ISO 200 and underexpose by 1/3 (and adjust in conversion) to achieve 260 or work directly at ISO 260?

Paul

Paul--apparently it doesn't matter on the 30D, but it does matter on any other camera with 1/3 ISO stops.

-Paul

Paul & Paul, it looks like it does matter on the 30D as well, but for different reasons. (Paul G, I assume you meant ISO 200 and 160 respectively, not 200 and 260?) According to this guy John Sheehy, although ISO 160 and ISO 200 + 1/3EV meter the same, the ISO 160 RAW file is "mangled"...