Pixel density - can the playing field be leveled???

Started Jun 6, 2009 | Discussions thread
Steen Bay Veteran Member • Posts: 7,418
Re: Per area read noise?

ejmartin wrote:

Steen Bay wrote:

ejmartin wrote:

bushi wrote:

Daniel Browning wrote:

b) not that often that you are shooting in perfect light,

c) not that often that you are shooting in strong light, but without high dynamic range in the scene

and in both b) and c), the less dense sensors tends to perform much better than the high resolution ones.

I think you can make a good case for b), but c) is less obvious. For example, at any given spatial frequency, the LX3 has a full stop more dynamic range than the 5D2. D3X has half a stop more, though, IIRC.

..how could be? according to dxomark comparison, LX3 have a clear edge in both b) and c) over G10 (let's don't overcomplicate things and lets stick to the same size sensors/similar class cameras)

The LX3 beats the G10 by a good margin in read noise and pattern noise in the RAW data, which accounts for its substantially better SNR at a fixed scale:


I don't think one can unequivocally attribute that better performance to lower pixel density, however, since the read noise per pixel is much lower with the LX3 (5.6 electrons for the LX3, 8.3 for the G10). Seems to me that Panasonic simply makes a better chip, independent of density.

That's interesting! The G10/LX3 pixel density is 34 vs 24 MP/cm2 and the read noise per pixel is 8,3 vs 5,6 electrons, meaning that the G10s read noise per area is more than twice the LX3s, namely 282 vs 134 Me/cm2. And even if the G10 had a better sensor with the same per pixel read noise (5,6 electrons) as the LX3, then the G10s per area read noise would still be 42% higher (190 vs 134 Me/cm2) because the G10 has 42% more MPs per area.

Per area read noise scales with the square root of pixel density, since noises add as RMS (the square root of the sum of squares of individual photosite noises). Thus the per area read noise scales with the pixel spacing, not the pixel density. The G10 read noise per pixel would have to be about 17% better than the LX3 (4.8 electrons) for it to have the same read noise per area.

For the G10 to match the LX3s per area read noise, its read noise per pixel should be 3,95 electrons. Then the G10 and LX3 would have the same per area relationship (assuming same efficiency) between read noise and shot/photon noise, and have the same DR and high-ISO performance, isn't that correct? But is that even possible, for the (same generation) smaller pixel to have a lower per pixel read noise than the larger pixel?

Mal Practice seems to think that one can scale the electronics in a way that the read noise per area stays the same. I don't see any evidence of that in manufactured products. For instance, the Canon line of DSLR's has, within a generation, about the same noise per pixel. For a long time it was stuck at about 4 electrons (+/-0.3 for sample variation) -- for the 20D/30D (6.4µ), 40D (5.7µ), 1Ds2 (7.2µ), 1D2/1D2N (8.4µ), 1D3 (7.2µ) and 1Ds3 (6.4µ); the latest generation 5D2 (6.4µ), 50D (4.7µ) seems to achieve a read noise of about 2.5-2.7 electrons (BTW I'm taking these figures from Roger Clark's collation of various sources). So read noise seems to correlate in the Canons better with sensor era (and so one might infer, with the manufacturing process) than it does inversely with pixel spacing.

OK, my per area read noise calculation wasn't quite correct, but the conclusion is still the same: that the smaller pixel must have a lower per pixel read noise than the larger pixel, in order to attain the same per area DR and noise (?) And you don't think that that's possible?

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