Baffling Pentax KP Read Noise

The Pentax K1 II has draw a great deal of attention lately because of strong Noise Reduction (NR) applied starting at ISO 640.

This would appear to be a strategy they are using in lieu of Dual Conversion Gain (DCG) with several of their recent models.

However, my Pentax KP results are baffling:



Note the drop at ISO 400, which looks like DCG, followed by the drop at ISO 640, which looks like NR.

I'm unaware that Pentax uses DGC in any of their cameras; I suppose this might be two different strengths of NR.

--
Bill ( Your trusted source for independent sensor data at PhotonsToPhotos )

So Jim Kasson (thanks Jim) confirmed my observation regarding Noise Reduction.


No NR


Still no NR despite drop in Read Noise


Ditto at ISO 500


Prominent NR at ISO 640 as expected

I considered digital scaling less than uniity but the histograms don't support that hypothesis.

--
Bill ( Your trusted source for independent sensor data at PhotonsToPhotos )

bclaff said:

So Jim Kasson (thanks Jim) confirmed my observation regarding Noise Reduction.


No NR


Still no NR despite drop in Read Noise


Ditto at ISO 500


Prominent NR at ISO 640 as expected

I considered digital scaling less than uniity but the histograms don't support that hypothesis.

--
Bill ( Your trusted source for independent sensor data at PhotonsToPhotos )

Click to expand...

Interesting. How many other makers do this?

--
Reporter: "Mr Gandhi, what do you think of Western Civilisation?"
Mahatma Gandhi: "I think it would be a very good idea!"

bclaff said:

The Pentax K1 II has draw a great deal of attention lately because of strong Noise Reduction (NR) applied starting at ISO 640.

This would appear to be a strategy they are using in lieu of Dual Conversion Gain (DCG) with several of their recent models.

Click to expand...

I think you should be careful not to label the noise you are measuring as "read noise", esp. once image-based algorithmic noise reduction is applied. It is a matter of semantics, perhaps, but to me read noise is the intrinsic circuit read noise of the image capture process.

The more we can keep the technical terms aligned to those of image sensor technologists, the better, in my opinion.

Eric Fossum said:

bclaff said:

The Pentax K1 II has draw a great deal of attention lately because of strong Noise Reduction (NR) applied starting at ISO 640.

This would appear to be a strategy they are using in lieu of Dual Conversion Gain (DCG) with several of their recent models.

Click to expand...

I think you should be careful not to label the noise you are measuring as "read noise", esp. once image-based algorithmic noise reduction is applied. It is a matter of semantics, perhaps, but to me read noise is the intrinsic circuit read noise of the image capture process.

The more we can keep the technical terms aligned to those of image sensor technologists, the better, in my opinion.

Click to expand...

You're right of course, strictly speaking it's something like "observed read noise" or "system read noise"; but generally in most dpreview fora such a distinction would probably be lost and perhaps even confusing (except perhaps on this forum).

Regards

Eric Fossum said:

bclaff said:

The Pentax K1 II has draw a great deal of attention lately because of strong Noise Reduction (NR) applied starting at ISO 640.

This would appear to be a strategy they are using in lieu of Dual Conversion Gain (DCG) with several of their recent models.

Click to expand...

I think you should be careful not to label the noise you are measuring as "read noise", esp. once image-based algorithmic noise reduction is applied. It is a matter of semantics, perhaps, but to me read noise is the intrinsic circuit read noise of the image capture process.

The more we can keep the technical terms aligned to those of image sensor technologists, the better, in my opinion.

Click to expand...

Eric, I haven't been very consistent, but I've more and more been calling this "dark-field noise". Is that any better?

In the case of in-camera processing, the actual read noise is unknowable (but maybe guessable with some assumptions from results at ISO settings where no such filtering is applied) by the experimenter who only makes test exposures and looks at raw files, is it not?

Jim

Would it be churlish to ask for units on the y-axis? Are they normalised to total power?

Extrapolating the NR response by eyeball, it looks like it would have fallen off by about 3dB at Nyquist. That seems like quite a gentle filter to me (but then the last filter I made was machined out of a big lump of aluminium and had to achieve 140 dB rejection!) Is that substantial in terms of NR in a typical photographic image?

Sorry - I can't think of any clever explanation for the double step in noise, except dual conversion gain.

J.

Joe Pineapples II said:

Would it be churlish to ask for units on the y-axis? Are they normalised to total power?

Click to expand...

Decibels. Consider the denominator arbitrary, although it is algorithmically determined so that the graph is easy to plot and see. It varies from graph to graph.

Member said:

Extrapolating the NR response by eyeball, it looks like it would have fallen off by about 3dB at Nyquist.

Click to expand...

About three dB from the value fairly close to zero:

Member said:

That seems like quite a gentle filter to me (but then the last filter I made was machined out of a big lump of aluminium and had to achieve 140 dB rejection!) Is that substantial in terms of NR in a typical photographic image?

Click to expand...

It's typical, in my experience, but Bill has more experience with more different cameras than I.

Jim

--
http://blog.kasson.com

JimKasson said:

Joe Pineapples II said:

Would it be churlish to ask for units on the y-axis? Are they normalised to total power?

Click to expand...

Decibels. Consider the denominator arbitrary, although it is algorithmically determined so that the graph is easy to plot and see. It varies from graph to graph.

JimKasson said:

Extrapolating the NR response by eyeball, it looks like it would have fallen off by about 3dB at Nyquist.

Click to expand...

About three dB from the value fairly close to zero:

JimKasson said:

That seems like quite a gentle filter to me (but then the last filter I made was machined out of a big lump of aluminium and had to achieve 140 dB rejection!) Is that substantial in terms of NR in a typical photographic image?

Click to expand...

It's typical, in my experience, but Bill has more experience with more different cameras than I.

Click to expand...

It's on a long list of things I'd like to do to produce the type of curves that Jim does with my workflow.

Jim, just to double-check; this shape of curve is "typical" of other noise reduction you have seen?
I ask because this type of 2D FT is not typical:



Regards

--
Bill ( Your trusted source for independent sensor data at PhotonsToPhotos )

JimKasson said:

Joe Pineapples II said:

Would it be churlish to ask for units on the y-axis? Are they normalised to total power?

Click to expand...

Decibels. Consider the denominator arbitrary, although it is algorithmically determined so that the graph is easy to plot and see. It varies from graph to graph.

Click to expand...

Okay, thanks. So 0 dB doesn’t correspond to anything in particular, and is different between graphs.

For “horizontal” and “vertical” plots, do you sum along one dimension to collapse the plot to 1D?

J.

bclaff said:

Eric Fossum said:

bclaff said:

The Pentax K1 II has draw a great deal of attention lately because of strong Noise Reduction (NR) applied starting at ISO 640.

This would appear to be a strategy they are using in lieu of Dual Conversion Gain (DCG) with several of their recent models.

Click to expand...

I think you should be careful not to label the noise you are measuring as "read noise", esp. once image-based algorithmic noise reduction is applied. It is a matter of semantics, perhaps, but to me read noise is the intrinsic circuit read noise of the image capture process.

The more we can keep the technical terms aligned to those of image sensor technologists, the better, in my opinion.

Click to expand...

You're right of course, strictly speaking it's something like "observed read noise" or "system read noise"; but generally in most dpreview fora such a distinction would probably be lost and perhaps even confusing (except perhaps on this forum).

Click to expand...

Well, it is always a detective story - backing out sensor performance from data that has gone through the whole system - and honestly I am not sure of how you do that (even though I suspect you have documented it someplace..but my morning coffee only takes 15 min to drink). But, I think it is up to you to give this an appropriate name. "Input-referred camera readout noise" comes to mind, or "camera readout noise" for short.

Of course even shorter is "readout noise", but now we are in "read noise" territory.

What do you think about "camera readout noise," which in an ideal camera system, would be limited by the sensor readout noise?

JimKasson said:

Eric Fossum said:

bclaff said:

The Pentax K1 II has draw a great deal of attention lately because of strong Noise Reduction (NR) applied starting at ISO 640.

This would appear to be a strategy they are using in lieu of Dual Conversion Gain (DCG) with several of their recent models.

Click to expand...

I think you should be careful not to label the noise you are measuring as "read noise", esp. once image-based algorithmic noise reduction is applied. It is a matter of semantics, perhaps, but to me read noise is the intrinsic circuit read noise of the image capture process.

The more we can keep the technical terms aligned to those of image sensor technologists, the better, in my opinion.

Click to expand...

Eric, I haven't been very consistent, but I've more and more been calling this "dark-field noise". Is that any better?

In the case of in-camera processing, the actual read noise is unknowable (but maybe guessable with some assumptions from results at ISO settings where no such filtering is applied) by the experimenter who only makes test exposures and looks at raw files, is it not?

Click to expand...

hi Jim, please see what I just typed to Bill moments ago. Just to be picky on "dark-field noise" you don't mean FPN, and also "read noise" in sensors is always measured in the dark so I am not sure dark-field as a modifier helps distinguish sensor from camera system.

Anyway, you guys (gals too) do an amazing job of characterization, esp. given that you cannot easily separate the camera and the sensor. I am just being fussy about the wording. I am not even sure any other sensor technologists read this forum so maybe I am the only one who cares. Also, in-pixel CG changes do affect sensor read noise as most of the read noise comes from that first SF transistor. Additional gain just gains that up.

Correlated multiple sampling (CMS) can also reduce read noise and I wonder when sensor manufacturers will start to use that as an option, and how that will be reflected in the measurements you make. Good thing to simulate, perhaps!

see, for example, Shoji Kawahito's paper on CMS:

Noise reduction effects of column-parallel correlated multiple sampling and source-follower driving current switching for CMOS image sensors

bclaff said:

JimKasson said:

Joe Pineapples II said:

Would it be churlish to ask for units on the y-axis? Are they normalised to total power?

Click to expand...

Decibels. Consider the denominator arbitrary, although it is algorithmically determined so that the graph is easy to plot and see. It varies from graph to graph.

bclaff said:

Extrapolating the NR response by eyeball, it looks like it would have fallen off by about 3dB at Nyquist.

Click to expand...

About three dB from the value fairly close to zero:

bclaff said:

That seems like quite a gentle filter to me (but then the last filter I made was machined out of a big lump of aluminium and had to achieve 140 dB rejection!) Is that substantial in terms of NR in a typical photographic image?

Click to expand...

It's typical, in my experience, but Bill has more experience with more different cameras than I.

Click to expand...

It's on a long list of things I'd like to do to produce the type of curves that Jim does with my workflow.

Jim, just to double-check; this shape of curve is "typical" of other noise reduction you have seen?
I ask because this type of 2D FT is not typical:

Click to expand...

The amount of relative attenuation at f/fs = 0.5 is typical. The shape of the curve is distinctly atypical. Usually the lp curves are monotonic.

Jim

--

JimKasson said:

bclaff said:

JimKasson said:

Joe Pineapples II said:

Would it be churlish to ask for units on the y-axis? Are they normalised to total power?

Click to expand...

Decibels. Consider the denominator arbitrary, although it is algorithmically determined so that the graph is easy to plot and see. It varies from graph to graph.

JimKasson said:

Extrapolating the NR response by eyeball, it looks like it would have fallen off by about 3dB at Nyquist.

Click to expand...

About three dB from the value fairly close to zero:

JimKasson said:

That seems like quite a gentle filter to me (but then the last filter I made was machined out of a big lump of aluminium and had to achieve 140 dB rejection!) Is that substantial in terms of NR in a typical photographic image?

Click to expand...

It's typical, in my experience, but Bill has more experience with more different cameras than I.

Click to expand...

It's on a long list of things I'd like to do to produce the type of curves that Jim does with my workflow.

Jim, just to double-check; this shape of curve is "typical" of other noise reduction you have seen?
I ask because this type of 2D FT is not typical:

Click to expand...

The amount of relative attenuation at f/fs = 0.5 is typical. The shape of the curve is distinctly atypical. Usually the lp curves are monotonic.

Click to expand...

Thanks for the clarification. That makes sense given what I've seen.

Regards

--
Bill ( Your trusted source for independent sensor data at PhotonsToPhotos )

Joe Pineapples II said:

JimKasson said:

Joe Pineapples II said:

Would it be churlish to ask for units on the y-axis? Are they normalised to total power?

Click to expand...

Decibels. Consider the denominator arbitrary, although it is algorithmically determined so that the graph is easy to plot and see. It varies from graph to graph.

Click to expand...

Okay, thanks. So 0 dB doesn’t correspond to anything in particular, and is different between graphs.

For “horizontal” and “vertical” plots, do you sum along one dimension to collapse the plot to 1D?

Click to expand...

Essentially. Here's a section of the Matlab code, with the operational lines indicated:





The variable TR is already in the frequency domain. The code was originally written by DPR member DSPographer many years ago, and I have modified it since.

Jim

--

Hello,

As a disclaimer, I'm not a sensor chip designer (although I do have a related patent on electron leakage reduction for night vision sensors), I'm an analog/RF chip/system designer. So please take my comments lightly and don't rip me to shreds :-|

With the pixel Source Follower (SF) as the first amplifier in the sensor voltage channel chain, this seems as a major source of electronic induced noise (like the LNA in RF receiver systems) that can't be reduced with post gain (sets the lower overall system "noise figure"). Since the SF has less than unity voltage gain, the system input referred noise is rather high, higher than if it were a Common Source (CS) which has potentially higher than unity voltage gain.

I've captured a screen shot of the Aptina mentioned White Paper, shown below. The left is Figure 5 from the white paper and the right side is a modification shown in orange. The SF amplifier is replaced with a CS amplifier by using a PMOS instead of a NMOS device and the source is returned to the RS line which I assume is a higher voltage when active than supply Vaa_PIX.

The voltage output Vout is inverted since the SF is a follower and the CS is an inverting amplifier. Vout can swing a higher peak to peak voltage than the SF which may be beneficial with less post amplification required and since the gain magnitude may be greater than unity the overall system input referred noise could be lower as well. No additional control, power nor signal lines are required either.

Anyway, this was just a thought. Please let me know what you think.

Best,


Pixel Level Common Source Amplifier Concept shown in Orange

--
~Mike~

mawyatt2002 said:

Hello,

As a disclaimer, I'm not a sensor chip designer (although I do have a related patent on electron leakage reduction for night vision sensors), I'm an analog/RF chip/system designer. So please take my comments lightly and don't rip me to shreds :-|

With the pixel Source Follower (SF) as the first amplifier in the sensor voltage channel chain, this seems as a major source of electronic induced noise (like the LNA in RF receiver systems) that can't be reduced with post gain (sets the lower overall system "noise figure"). Since the SF has less than unity voltage gain, the system input referred noise is rather high, higher than if it were a Common Source (CS) which has potentially higher than unity voltage gain.

I've captured a screen shot of the Aptina mentioned White Paper, shown below. The left is Figure 5 from the white paper and the right side is a modification shown in orange. The SF amplifier is replaced with a CS amplifier by using a PMOS instead of a NMOS device and the source is returned to the RS line which I assume is a higher voltage when active than supply Vaa_PIX.

The voltage output Vout is inverted since the SF is a follower and the CS is an inverting amplifier. Vout can swing a higher peak to peak voltage than the SF which may be beneficial with less post amplification required and since the gain magnitude may be greater than unity the overall system input referred noise could be lower as well. No additional control, power nor signal lines are required either.

Anyway, this was just a thought. Please let me know what you think.

Best,


Pixel Level Common Source Amplifier Concept shown in Orange

Click to expand...

hi Mike,

Thank you for sharing your CS idea. Generally a CS amp will work but any improvement is offset by a few factors. (1) CS-drain is about equal to Vout so you get gate-drain Miller capacitance (2) current fluctuations in CS (origin of 1/f noise) are also amplified in this configuration and (3) implementing PMOS in the pixel is not convenient due to layout rules. It is a common suggestion (usually NMOS CS) and some people have made it work, although not really with any compelling advantage. Thus it has not made it to widespread use. Actually, I am not sure anyone has tried PMOS CS - perhaps the reduced intrinsic 1/f noise will help.

Anyway, try simulating this and see what you get!

Eric

Eric Fossum said:

mawyatt2002 said:

Hello,

As a disclaimer, I'm not a sensor chip designer (although I do have a related patent on electron leakage reduction for night vision sensors), I'm an analog/RF chip/system designer. So please take my comments lightly and don't rip me to shreds :-|

With the pixel Source Follower (SF) as the first amplifier in the sensor voltage channel chain, this seems as a major source of electronic induced noise (like the LNA in RF receiver systems) that can't be reduced with post gain (sets the lower overall system "noise figure"). Since the SF has less than unity voltage gain, the system input referred noise is rather high, higher than if it were a Common Source (CS) which has potentially higher than unity voltage gain.

I've captured a screen shot of the Aptina mentioned White Paper, shown below. The left is Figure 5 from the white paper and the right side is a modification shown in orange. The SF amplifier is replaced with a CS amplifier by using a PMOS instead of a NMOS device and the source is returned to the RS line which I assume is a higher voltage when active than supply Vaa_PIX.

The voltage output Vout is inverted since the SF is a follower and the CS is an inverting amplifier. Vout can swing a higher peak to peak voltage than the SF which may be beneficial with less post amplification required and since the gain magnitude may be greater than unity the overall system input referred noise could be lower as well. No additional control, power nor signal lines are required either.

Anyway, this was just a thought. Please let me know what you think.

Best,


Pixel Level Common Source Amplifier Concept shown in Orange

Click to expand...

hi Mike,

Thank you for sharing your CS idea. Generally a CS amp will work but any improvement is offset by a few factors. (1) CS-drain is about equal to Vout so you get gate-drain Miller capacitance (2) current fluctuations in CS (origin of 1/f noise) are also amplified in this configuration and (3) implementing PMOS in the pixel is not convenient due to layout rules. It is a common suggestion (usually NMOS CS) and some people have made it work, although not really with any compelling advantage. Thus it has not made it to widespread use. Actually, I am not sure anyone has tried PMOS CS - perhaps the reduced intrinsic 1/f noise will help.

Anyway, try simulating this and see what you get!

Eric

Click to expand...

Eric,



Thanks, very familiar with 1/f noise and the Miller effects, always try and use Cascode stages to help when possible.

If the output of the PMOS CS amplifier was a current rather than voltage, which would be routed thru the NMOS RS device, with a variable effective load a variable voltage gain referenced to the pixel could be achieved. A common use low noise Cascode Stage could make the impedance as seen from the PMOS drain very low thus minimizing the Miller effect and isolating a variable transconductance stage for variable voltage gain control.

Yes, I probably should try and get this simulated sometime.

Thanks for the reply and notes.

Best,

--
~Mike~

Thanks. I speak Matlab.

J.

bclaff said:

The Pentax K1 II has draw a great deal of attention lately because of strong Noise Reduction (NR) applied starting at ISO 640.

Click to expand...

Does the processing applied by the KP appear to be different (in nature and/or strength) to what the K-1 II is applying?

DPReview praised the KP for its "excellent high ISO performance in both Raw and JPEG" and only mentioned suspected RAW baking in passing. The K-1 II, on the other hand, got slapped for its processing. I'm wondering whether that's due to different processing or a different assessment of the same situation.

FWIW, I'm against any in-camera destructive RAW baking in any shape or form.

Of course there are uncontentious system-noise reduction strategies performed within the sensor, but post-A/D "denoising" should be left to out-of-camera processing, AFAIC.

Am I right in assuming that the "accelerator" chip Pentax has been employing could most likely be replaced by out-of-camera processing? Given the "closed black boxes" modern Sony sensors appear to be, I have difficulty believing that the "accelerator" chip leverages any sensor-internal data or processes.
Potentially, the "accelerator" chip could make use of the equivalent of "dark frames", etc., i.e., data that is normally not available outside a camera (but could be provided as secondary data), but from what I've seen in the analyses conducted so far, it seems that the "accelerator" attempts "beautification" with a nearest-neighbour smoothing component as part of its data massaging.

I hope this is not considered to be an off-topic post.

BTW, @bclaff, have you noticed my proposal to use "image stacking" for analysis purposes?
While FT plots and power spectra are useful in detecting image manipulation, when applied to images of pure noise they are not informative regarding the retention of signal. In other words, if some image manipulation did a miraculous job of almost not harming signal but only combating noise it may deserve less scolding than an alternative that simply attenuates high spatial frequencies.
I'll be the first to argue that in general it is impossible to distinguish signal (here meant to be "information" present in a scene) from noise (given the stochastic nature of light itself) by evaluating a single image, but could it still be useful to compare denoising strategies by averaging over many images (in an image stack) and then evaluating which denoising strategy supports the recovery of signal through averaging better than others?

BTW, attempting to work with scenes that have content (as opposed to "lens cap" shots which I'm assuming you are using) could circumvent potential optimisations by some manufacturers.

Sony, for instance, had a line of CD players that switched off analogue circuitry upon detecting a "zero" stream of signal. This led to phenomenal dynamic range measurements which were, of course, of no practical relevance. As soon as any information was fed to the A/D converters, the noise floor was significantly raised by the re-activated circuitry.

Are we sure something similar is not happening with "lens cap" images?

--
http://www.flickr.com/photos/class_a/