At 3:32 AM +0200 8/3/04, Listar wrote:
>Date: Sun, 01 Aug 2004 22:57:41 -0400
>From: W Shumaker <om4t@xxxxxxxx>
>Subject: [OM] Re: Noise versus pixel area in image sensors
>
>All cameras I have seen with smaller pixels perform worse.
Agree.
>Any improvements to semiconductor technology that reduces noise will
>still have lower noise with larger pixels.
Yes, but improvements in semiconductor technology will reduce that
noise, and eventually it won't matter in practice, even though it
will always be true that larger pixels have less noise.
>Heat is a real noise generator,
>so CMOS might have a technological edge by being lower power.
Yes, if they can solve the self-noise problems soon enough to matter.
>What is "self-noise"?
Everything other than shot noise (which is also called photon noise).
You give a good list in the following items.
>Fixed pattern noise can be removed by algorithms
>and auto-calibration techniques.
At the expense of increasing the other kinds of noise. That said, it
seems to be a tradeoff that camera makers willingly make. I don't
know how raw RAW images are. Do they use these algorithms, or just
gives us the data straight from the imager? Probably varies from
make to make, model to model.
>Reduction of feature size, the main
>improvement to semiconductors, does not in itself reduce noise.
The maximum gain-bandwidth achieved for a given amount of DC power
per transistor also increases by leaps and bounds, and this improves
the amplifiers and other analog components. As the processes get
better (mostly, the material becomes cleaner and the geometry better
controlled), the gain per stage increases and the absolute and
relative noise contribution of the amplifiers decreases. Flicker
(1/f) noise in particular is reduced by cleaner materials and
surfaces.
The cost of the components also drops. Someone else (Tim Hughes?)
pointed out that one cause of the large self-noise in digital cameras
was that the cost model didn't yet support use of the best of
amplifiers; this sounds right, but will surely change.
>Heat generates phonons (crystal vibrational waves). Amplifiers have
>noise, A/D's have noise, etc. There will always be noise, larger will
>always be better because you have more signal to work with, no matter
>what the technological improvements. We've had this discussion before.
Yes. The fundamental issue is when the absolute noise levels no
longer matter in practice for a specific application, photography.
>I have never heard of photon noise?
It's also called shot noise, and is caused by the fundamental
granularity of light. The photons arrive randomly, so the average
rate varies randomly. This is the fundamental limit to how low the
noise in an image can be.
>I've heard of shot noise, thermal
>noise, 1/f noise, noise figure of an amplifier, fixed-pattern noise, kT/C
>switching noise, dark current, substrate noise, ... indeed there are
>lots of noise sources. Improvements will be made but the noise
>tradeoff with pixel size is with us to stay I'm afraid. Bigger buckets
>simply carry more water, no matter how many holes you plug.
So, we must always drink our beer from large buckets? Agree.
Joe Gwinn
>At 09:25 PM 8/1/2004, Joe Gwinn wrote:
> >There have been a number of discussions about the tradeoff between
>>pixel area and noise, the basic rule being the larger the pixel the
>>less the noise.
>>
>>In <http://luminous-landscape.com/reviews/cameras/olympus-e1.shtml>,
>>under "Inside the 4/3 system", the claim is made that Olympus' choice
>>of less than a full-frame (35mm) sensor will corner the four thirds
>>system. The problem is that the argument assumes that photographic
>>CCD sensors and amplifiers are already fully developed, which isn't
>>true just yet. Current sensors are limited by self-noise, and so the
>>area-noise tradeoff will improve as semiconductor technology
>>improves. When sensors are limited by photon noise, then and only
>>then is further improvement impossible. This will be a while
> >unfolding.
>>
> >Joe Gwinn
==============================================
List usage info: http://www.zuikoholic.com
List nannies: olympusadmin@xxxxxxxxxx
==============================================
|