In which ways, and why, are smaller sensors more efficient than larger?

Started 5 months ago | Discussions thread
Anders W
Forum ProPosts: 17,062Gear list
Re: apparently everything I thought I knew is wrong
In reply to Great Bustard, 5 months ago

Great Bustard wrote:

Anders W wrote:

Great Bustard wrote:

William Porter wrote:

Great Bustard wrote:

Yes -- the cliché is wrong. FF attains less noise than smaller formats either by using a longer shutter speed at base ISO (e.g. 24mm f/8 1/100 ISO 100 on FF vs 12mm f/3 1/100 ISO 100 on mFT) or using a more shallow DOF in lower light (e.g. 50mm f/1.4 1/100 ISO 1600 on FF vs 25mm f/1.4 1/100 ISO 1600 on mFT).

For equivalent photos (same DOF and shutter speed), the same total amount of light falls on the sensor for all systems, thus the noise advantage goes to the system with the more efficient sensor.


Well, well. It's always bracing to discover that everything I thought I knew was wrong! Bracing, and in this particular case, rather exciting. Thanks to you and Anders for helping me get the gist here.

Could be worse. I mean, imagine you found out that Miley Cyrus used a stunt double for twerking. That would be rather more upsetting, I would think.

Now I hesitate to ask, but I have to know... I was under the impression that you and Anders disagree about something technical. But unless I've really understood him (VERY possible), I think he'd agree with what you say above, indeed, it's pretty close to what he said to me himself. Am I wrong?

The disagreement between Anders and myself is on sensor efficiency as a function of format. My claim is that sensors of the same generation have remarkably consistent QE (Quantum Efficiency -- the proportion of photons falling on the sensor that release electrons) regardless of brand, sensor size, or pixel size. I admit that there are most certainly exceptions, however, such as when a new tech is being introduced, such as BSI.

My claim about the QE of sensors for a given generation is based on sensorgen's calculations from DxOMark data. If the sensorgen figures are wrong, then my assumption of same QE for sensors of a given generation (as a general rule) will also be wrong, unless the errors are small or sensorgen consistently over or underestimates the QE for all the sensors.

With that in mind, then my entry into this thread:

explains my position, but Anders feels that smaller sensors, as a general rule, are more efficient, regardless of pixel count.

Anders doesn't just "feel" that smaller sensors tend to be more efficient than larger. He claims to have shown that such is the case (for a reasonably representative sample of current sensors).

Let me throw out some illustrative facts in order to see if this can take us further towards a resolution of the problem. As you know, the E-M5 and the D4 have exactly the same QE according to sensorgen (53 percent) and just about the same pixel count (16 MP). As you also know, we should expect a decline of about 3 dB in max SNR for every doubling of ISO (as long as QE is actually constant and things like read noise and fixed pattern noise can be ignored).

Now consider the following data series for the two bodies and ask yourself how they square (or not) with the above-mentioned premises. The figures are measured ISOs, max SNR, and the difference in max SNR compared to the prior observation in the series. What would be your thoughts?

This is *exactly* what I wanted to discuss!



75 47.4

151 45.6 1.8

301 43.4 2.2

601 41.0 2.4

1192 38.4 2.6

2356 35.5 2.9

4838 32.4 3.1

9508 29.4 3.0

17854 27.2 2.2

35734 24.1 3.1

69939 20.7 3.4

139250 17.1 3.6


107 42.2

214 40.0 2.2

394 37.4 2.6

782 34.7 2.7

1489 32.0 2.7

3024 28.8 3.2

5953 25.8 3.0

11848 22.4 3.4

Please excuse my lack of cognitive capacity at the moment. Could you first walk me through what you are saying it shows? I know you might be thinking, "WTF?! I laid it out plain as day -- is he being evasive on purpose?!" Trust me -- I'm just stupid right now.

No problem.

Let me walk you through some of the questions I have. Does each corresponding line from each camera represent the same exposure? For example, does "301 43.4 2.2" from the D4 correspond to the same exposure as "394 37.4 2.6" from the EM5?

No it doesn't. It means that the E-M5 receives about 0.4 EV less exposure.

'Cause if it did, then the D4 would receive 4x as much light as the EM5, and thus have half the photon noise, and photon noise should absolutely dominate read noise at that exposure.

Thus, we would expect the D4 to have twice the SNR, which would be about right if your SNR values are dB, I think. (I get confused on the dB measure, 'cause sometimes 1 stop = 3 dB and other times 1 stop = 6 dB, depending on what you're measuring, which has always been confusing for me.

The difference in SNR should be about 6 dB between the D4 and the E-M5 at the same exposure if the sensors were equally efficient. As you can see, it is significantly less than that at lower ISOs. Higher up the range, however, the difference starts to approach the one we would expect at equal efficiency.

If so, then we would also expect that if the D4 sensor and EM5 sensor were equally efficient, then the SNR for the D4 would match the SNR for the EM5 two lines further up (e.g. the SNR for the 10th line on the D4 would match the SNR for the 8th line for the EM5, if the sensors were equally efficient).

Anyway, please be kind if I'm completely off.

You aren't off at all. What I wanted to say in general (apart from the comparison between the D4 and the E-M5 already discussed above) is the following:

The difference figures in the last column indicate that for both cameras, and especially the D4, the difference in SNR as we half the exposure (going to the next higher ISO) is significantly less than the 3 dB figure we would expect under the premise I mentioned (constant QE, no fixed pattern noise, trifling impact of read noise), especially in the lower part of the ISO range. Since we know read noise to have a trifling impact in that range, it follows that either QE isn't constant across ISOs and/or there is significant fixed pattern noise. This may be at least part of the explanation why my results for normed max SNR differ from Bob's QE figures.

For the purpose of measuring sensor efficiency, I think the implication is that we should move away from the simplified view provided by a focus on QE (as a constant for each sensor) and read noise. Instead I would suggest, that we use four measures to capture that efficiency. The three measures I have already constructed plus a fourth focusing on "normed max SNR" at higher ISOs. As I hope/think you agree, the two DR measures are good indicators of efficiency with regard to shadow rendering at low and high ISO and the two SNR measures are good indicators of highlight rendering at low and high ISO. So we cover variations in efficiency across the ISO range as well as between shadow and highlight rendering. What do you think?

 Anders W's gear list:Anders W's gear list
Panasonic Lumix DMC-G1 Olympus OM-D E-M5 Olympus E-M1 Panasonic Lumix G Vario 14-45mm F3.5-5.6 ASPH OIS Panasonic Lumix G Vario 7-14mm F4 ASPH +21 more
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