Total Light Equivalence and Moores Law

Started Jun 7, 2012 | Discussions thread
Forum ProPosts: 27,817
Re: Total Light Equivalence
In reply to Detail Man, Jun 12, 2012

Detail Man wrote:

bobn2 wrote:

coroander wrote:

Detail Man wrote:

It is surely true that the phrase "all other things being equal" is not always (or even frequently) the case. But it is unclear to me whether anybody else has missed that obvious point.

My point is not that different sensors are different, but that smaller sensors are better.

You have not established that.

I have explained clearly how i think equivalence should be altered to account for these differences. I am not advocating throwing the whole thing out, however its utility when altered appropriately starts to approach zero (in terms of the inclusion of ISO and noise into the equation.)

It always had ISO and noise in the equation, just considered sensibly.

When you include sensor size, ISO and noise in your "theory", then you must account for any significant relationships between sensor size, ISO and noise.

You have not established those relationships such that they show what you say they do,


It seems like we wandered into those discussions not all that long ago on a previous thread.

Dynamic range

Dynamic range is the ratio of the largest and smallest recordable signal, the smallest being typically defined by the 'noise floor'. In the image sensor literature, the noise floor is taken as the readout noise, so DR = Q [max] / Noise [readout].

The measurement here is made at the level of a pixel (which strictly means that the DR of sensors with different pixel counts is measured over a different spatial bandwidth, and cannot be compared without normalisation) .

If we assume sensors with the same pixel count but different sizes, then the pixel area will be in proportion to the sensor area .

If the maximum exposure (amount of light per unit area) is the same then both the maximum signal and the read noise reduce in proportion to the pixel (and therefore the sensor) area, so the DR does not change .

If the comparison is made according to DOF limited conditions, so that the exposure of the larger sensor is reduced in proportion to the area of the sensor (and pixel, for sensors with equal pixel count) then Q [max] is constant, and the read noise ( Noise [readout] ) falls with the sensor area, leading to a higher dynamic range for the smaller sensor .

Now I thought that the statement (directly above) related (only) to a pixel-level analysis of Dynamic Range, only the light per unit area (per pixel in this case) is being taken into account - and not the total light ?

Remember the specification there that the sensors have equal pixel counts. Therefore the light per pixel and the total light are in proportion. This is another strict scaling thing, strictly scaling a sensor results in a sensor with the same number of pixels, only smaller.

This business continues to fascinate me. Perhaps bobn2 could (again, sorry) confirm or deny that my particular understanding expressed in this paragraph is sound ?

I am still a bit unclear how a pixel-level analysis of DR, or (alternatively) a sensor-level analysis of DR can happily coexist ? Might they not return a different result (as the Wikipedia text quoted above might indicate) ?

DR is bandwidth dependent. In that case the bandwidths were equal because the sensors had equal pixel counts.

Assume that the periodic noise components are minimal, and the distribution of the read/dark noise sources (at either level of analysis) are random (with a Poisson distribution).

Perhaps bobn2 (or other players in these discussions, if able to specifically state the elements involved in their thinking) could speak directly to my thoughts above ?

Poisson noise power increases with frequency, so is bandwidth dependent as above. Emil Martinec did a nice demonstration a while back, your googling proclivity might be able to locate it.

-- hide signature --


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