Rochester-based image sensor company ADVIS presented its ideas on-line. I'm not a big fan of their approach, but it's nice to see it presented in a collection of papers. It's also nice to see ADVIS web site up again.
Someday I would like to better understand the low light performance of this approach. Seems to me that kTC noise is continuously (discretely) injected via the feedback process but I don't know enough about the details to understand if this is a little problem or big problem. -EF
If you mean sigma-delta sensor from the first paper, yes, there is kTC noise, but its power should be proportional to the signal (square root SNR, that is). So, for zero signal there should be zero noise. For large signals there should be some added value to the shot noise, somewhat similar to noise factor in avalanche diodes. I believe with right design this noise factor can be made negligible.
For me the show stopper is pmos and nmos combination in the pixel. This makes it unusable for small pixels and low fill factor for large pixels. Hopefully they can modify their idea and make nmos-only pixel.
This is not a great way to have technical discussion, but it seems to me that the fedback signal is like a continuous (albeit discrete time) reset and if one divides the imaging into frames then the reset noise is present at the beginning of the frame, even if there is no signal in the current frame.
In addition, I don't know how the feedback charge is created but probably there is additional noise in its creation (e.g. switched cap or fill/spill).
My understanding is they do not reset pixel every frame. A properly done sigma-delta loop should run freely, without forced resets.
But even if we assume they do forced resets every frame (I do not believe this, but let's assume just for the sake of discussion), the comparator output would be zero at zero signal. So, this way they basically cut out all small signals together with noise - there will be zero signal and zero noise until signal is large enough to cross the comparator threshold. Only then, at that large signal we'll see both signal and kTC noise and all other noises they might have. The bigger signal, the more comparator switches, the more noise accumulating happens.
So, even in the unrealistic frame reset case their noise would manifest itself as some excessive noise factor on top of the shot noise, rather than signal-independent noise floor.
I did not say there was a reset. I said it was like a reset. Anyway, bad place for a technical discussion and all I will say is that I don't agree with you. I think there is a problem with low light level imaging using this technique. Someone will have to show me the data to convince me otherwise. -EF
Well, I should have said "difficult" rather than "bad". But, a blog-comment area is difficult because it is slow and generally not conducive to sketches, figures, etc., and especially because it is unforgiving if someone, like me, says something dumb. It lives forever long after the context and whirl of the discussion is gone. There is no eraser for this white board. The other thing that drives me insane is anonymous people in the discussion. Why should I put my reputation on the line when other people hide behind a pseudonym? Sorry, that is just the way I see it.
So, we should have an image sensor workshop for face to face real time discussion. Maybe in June sometime? :)
I understand what you are saying. You are right, sketches possibility would help. I can think about solutions of this, but they all quite inconvenient and time consuming.
On the other hand, being anonymous has its advantages. People can openly say things they really think, what they can not say under their real names. Sometimes these things are right, sometimes wrong, but we would never heard them, if this site would require a registration under the real name.
So, it's quite hard to find the right balance. But what prevents you from being anonymous or use nick name in these discussions?
Someday I would like to better understand the low light performance of this approach. Seems to me that kTC noise is continuously (discretely) injected via the feedback process but I don't know enough about the details to understand if this is a little problem or big problem. -EF
ReplyDeleteIf you mean sigma-delta sensor from the first paper, yes, there is kTC noise, but its power should be proportional to the signal (square root SNR, that is). So, for zero signal there should be zero noise. For large signals there should be some added value to the shot noise, somewhat similar to noise factor in avalanche diodes. I believe with right design this noise factor can be made negligible.
ReplyDeleteFor me the show stopper is pmos and nmos combination in the pixel. This makes it unusable for small pixels and low fill factor for large pixels. Hopefully they can modify their idea and make nmos-only pixel.
This is not a great way to have technical discussion, but it seems to me that the fedback signal is like a continuous (albeit discrete time) reset and if one divides the imaging into frames then the reset noise is present at the beginning of the frame, even if there is no signal in the current frame.
ReplyDeleteIn addition, I don't know how the feedback charge is created but probably there is additional noise in its creation (e.g. switched cap or fill/spill).
-EF
My understanding is they do not reset pixel every frame. A properly done sigma-delta loop should run freely, without forced resets.
ReplyDeleteBut even if we assume they do forced resets every frame (I do not believe this, but let's assume just for the sake of discussion), the comparator output would be zero at zero signal. So, this way they basically cut out all small signals together with noise - there will be zero signal and zero noise until signal is large enough to cross the comparator threshold. Only then, at that large signal we'll see both signal and kTC noise and all other noises they might have. The bigger signal, the more comparator switches, the more noise accumulating happens.
So, even in the unrealistic frame reset case their noise would manifest itself as some excessive noise factor on top of the shot noise, rather than signal-independent noise floor.
I did not say there was a reset. I said it was like a reset. Anyway, bad place for a technical discussion and all I will say is that I don't agree with you. I think there is a problem with low light level imaging using this technique. Someone will have to show me the data to convince me otherwise.
ReplyDelete-EF
OK, let's stop the discussion here. By the way, why do you think it's a bad place for the technical discussions?
ReplyDeleteWell, I should have said "difficult" rather than "bad". But, a blog-comment area is difficult because it is slow and generally not conducive to sketches, figures, etc., and especially because it is unforgiving if someone, like me, says something dumb. It lives forever long after the context and whirl of the discussion is gone. There is no eraser for this white board.
ReplyDeleteThe other thing that drives me insane is anonymous people in the discussion. Why should I put my reputation on the line when other people hide behind a pseudonym? Sorry, that is just the way I see it.
So, we should have an image sensor workshop for face to face real time discussion. Maybe in June sometime? :)
-EF
I understand what you are saying. You are right, sketches possibility would help. I can think about solutions of this, but they all quite inconvenient and time consuming.
ReplyDeleteOn the other hand, being anonymous has its advantages. People can openly say things they really think, what they can not say under their real names. Sometimes these things are right, sometimes wrong, but we would never heard them, if this site would require a registration under the real name.
So, it's quite hard to find the right balance. But what prevents you from being anonymous or use nick name in these discussions?
This comment has been removed by a blog administrator.
ReplyDeleteWow, someone posted above with the same initials as me 12/5/09 09:21 but it was not me. Nice impersonation.
ReplyDelete-EF