Jump to content

  •  

CNers have asked about a donation box for Cloudy Nights over the years, so here you go. Donation is not required by any means, so please enjoy your stay.

Photo

CMOS sensors for EAA

  • Please log in to reply
132 replies to this topic

#51 jimthompson

jimthompson

    Apollo

  • *****
  • Posts: 1088
  • Joined: 06 Oct 2009
  • Loc: Ottawa, Canada

Posted 31 August 2017 - 12:18 PM

Finding good publicly available technical data on new sensors is getting harder and harder.  Not 100% relevant but perhaps a good indicator may be to look at the data that can be found on the older RGB CCD sensors.  Attached are some plots I put together of sensitivity data extracted from some old Sony datasheets.  By calculating the area under the different curves we can get an estimate of the relative sensitivity of each RGB channel to the unfiltered monochrome sensor.  For the 4 sensors in the attached image I calculate the following:

                        R            G             B             RGBG average

ICX205            45.5%     32.9%     20.4%     32.9%

ICX274            44.0%     32.6%     19.3%     32.1%

ICX285            55.0%     38.1%     19.7%     37.7%

ICX618            61.5%     34.0%     15.5%     36.3%

 

There is some variation between sensors, but they are all still pretty similar.  How different the new CMOS sensors are in terms of spectral response, I don't know.  I suspect they have responses more like the ICX618 which has better response to IR.  On average though it looks like an RGB sensor has about 1/3 the sensitivity of its monochrome counterpart.  As a side note, some CCD sensors still used for EAA today (eg. Lodestar X2C) use a CMYG Bayer matrix.  When you do the same calculation as above you get an average sensitivity closer to 1/2.

 

Best Regards,

 

Jim T.

Attached Thumbnails

  • RGB CCD sensitivity plots.png

  • will w, roelb and elpajare like this

#52 Astrojedi

Astrojedi

    Soyuz

  • *****
  • Posts: 3801
  • Joined: 27 May 2015
  • Loc: SoCal

Posted 31 August 2017 - 12:30 PM

I think JPK is really just doing back of the envelope math which is in the ballpark but as per Olecuss and Curtis could be refined. The difference boils down to sensor QE vs. effective QE (which takes Bayer filter losses into account). The effective QE for most color sensors is typically <50% because of this reason. The exception being the 224 sensor which is why I would love to have a mono version of that sensor.

 

I agree the debayered mono version may be "acceptable", but you lose one of the key benefits of a mono sensor i.e. the much higher sensitivity vs. the color sensor (unless QHY has found a way to debayer without removing the microlenses). You still get the benefit of higher spatial resolution though.



#53 A. Viegas

A. Viegas

    Mercury-Atlas

  • *****
  • Posts: 2989
  • Joined: 05 Mar 2012
  • Loc: New York City/ CT

Posted 31 August 2017 - 12:50 PM

I don't want to sound like a curmudgeon but why would you bother to go through all that work to scrape off the bayer matrix when there are so many new CMOS sensors coming out every year.  The demand for low light security applications continues to be very strong so we can expect to see many future years of new sensor development and that will invariably include some very awesome mono sensors too.   So you put all this effort in to debayer a color sensor to only see a new even better mono sensor coming out in 6 months.   Waste of Time!   just my penny.gif penny.gif

 

Al



#54 jimthompson

jimthompson

    Apollo

  • *****
  • Posts: 1088
  • Joined: 06 Oct 2009
  • Loc: Ottawa, Canada

Posted 31 August 2017 - 12:52 PM

I agree with Hiten, it would be very interesting to know if QHY has found a way to remove the colour filters without affecting the microlenses.  This seems like a big challenge considering how the sensors are constructed.  Are they actually physically removing the Bayer filters or are they doing some signal processing to back out an effective monochrome image?

 

Attached image is copyright Sony.

 

Regards,

 

Jim T.

Attached Thumbnails

  • CMOS cross section.png


#55 ccs_hello

ccs_hello

    Voyager 1

  • *****
  • Posts: 10245
  • Joined: 03 Jul 2004

Posted 31 August 2017 - 05:34 PM

QHY does not.

Once upon a time, I told him an Internet urban legend about someone use a high-power UV light to bleach the die in CFA.

He tried that and fully cooked the sensor but the CFA dye is not even affected.

 

No one can do that without scraping off the microlens layer as well as buried AR (anti-reflection) layer before CFA (color-filter array) is scrapped off.

Buying the factory fresh mono image sensor is THE way.

 

View this as a layered cake, say a chocolate cake buried three layer below surface.

A kid may not like chocolate at all.  However, there is no way to get rid of it without digging in.


Edited by ccs_hello, 31 August 2017 - 07:31 PM.

  • roelb likes this

#56 OleCuss

OleCuss

    Vanguard

  • *****
  • Posts: 2280
  • Joined: 22 Nov 2010

Posted 31 August 2017 - 05:52 PM

It's not that simple, the QE number applies to the photons that hit the photocell, the photons have to make it past the Bayer filter first.

I'm still trying to sort out all the QE stuff and I'm not sure it is consistently used when it comes to camera sensors.

 

But what little digging I managed showed that the ASI1600 manual indicates the QE of the monochrome sensor is right around 60%.

 

I believe the Olympus E-M1 uses the color version of the sensor (but a little different packaging) and it is listed here:  http://www.sensorgen...sOM-D-E-M1.html as having a QE of 48%.

 

This suggests that there is, indeed, a significant difference in the QE between the color sensor and the monochrome sensor.  This would then suggest that the color sensor's QE was measured with the Bayer matrix in place.

 

Assuming this is correct, then the QE drop for this particular sensor by moving from monochrome to OSC is about 20% of the QE of the monochrome sensor.  Definitely a penalty but fortunately not as much as a 67% drop would be.

 

I would not be surprised if I got a fact or logic wrong in all that (it's been a long day), but at the moment that seems to me to be how it sorts out for that particular sensor.


Edited by OleCuss, 31 August 2017 - 05:53 PM.


#57 Censustaker

Censustaker

    Apollo

  • *****
  • Posts: 1251
  • Joined: 26 Mar 2014
  • Loc: Utah

Posted 31 August 2017 - 06:18 PM

Btw, on the SharpCap facebook page Robin is saying the next release is going to have a feature to help measure sensor sensitivity for yourself:

 

21034312_1949262582012386_2212108337938177694_n.png

 

 

 

This is the result of the new automatic Sensor Analysis function in SharpCap 3.1 measuring the characteristics of a ZWO Astronomy Cameras ASI174MC sensor.

All you have to do is point the camera at something with even illumination, select the Sensor Analysis tool and cover/uncover the camera when prompted.

 



#58 ccs_hello

ccs_hello

    Voyager 1

  • *****
  • Posts: 10245
  • Joined: 03 Jul 2004

Posted 31 August 2017 - 07:02 PM

This chart has nothing to do with absolute QE measurement.

Actually, it is completely decoupled with any type of relationship in between

the incoming photon flux and the final ADU values.

 

QE measurement is all about the O/E conversion gain (optical to electrical conversion.)

 

Since no Q/E is given,

it only reflects that how the photo-electrons once collected in quantum-well will be mapped to the final outcome, the ADU value.

 

Because the measurement protocol has both flat and dark, the RN is also measured.  With full well and RN, on a specific gain setting, the DR (how many steps) can be calculated.

 

Clear Skies!

 

ccs_hello



#59 Relativist

Relativist

    Cosmos

  • *****
  • topic starter
  • Posts: 8151
  • Joined: 11 Oct 2003
  • Loc: OC, CA, USA

Posted 31 August 2017 - 07:02 PM


It's not that simple, the QE number applies to the photons that hit the photocell, the photons have to make it past the Bayer filter first.

I'm still trying to sort out all the QE stuff and I'm not sure it is consistently used when it comes to camera sensors.

But what little digging I managed showed that the ASI1600 manual indicates the QE of the monochrome sensor is right around 60%.

I believe the Olympus E-M1 uses the color version of the sensor (but a little different packaging) and it is listed here: http://www.sensorgen...sOM-D-E-M1.html as having a QE of 48%.

This suggests that there is, indeed, a significant difference in the QE between the color sensor and the monochrome sensor. This would then suggest that the color sensor's QE was measured with the Bayer matrix in place.

Assuming this is correct, then the QE drop for this particular sensor by moving from monochrome to OSC is about 20% of the QE of the monochrome sensor. Definitely a penalty but fortunately not as much as a 67% drop would be.

I would not be surprised if I got a fact or logic wrong in all that (it's been a long day), but at the moment that seems to me to be how it sorts out for that particular sensor.

That's only part of the story. The QE number is taken of the light getting through the filter, this does not take into account the filters effect, both need to be considered when trying to gestimate the number of electrons you'll get from a given light flux.

This is one of the reasons we EAAers are interested in a mono+OSC system, the luminance channel coming from a mono camera and applied to the entire OSC field could likely significantly speed acquisition of views.

#60 ccs_hello

ccs_hello

    Voyager 1

  • *****
  • Posts: 10245
  • Joined: 03 Jul 2004

Posted 31 August 2017 - 07:08 PM

TL; DR

 

About QE

https://www.cloudyni...l/#entry6039895



#61 Relativist

Relativist

    Cosmos

  • *****
  • topic starter
  • Posts: 8151
  • Joined: 11 Oct 2003
  • Loc: OC, CA, USA

Posted 31 August 2017 - 07:17 PM

TL; DR

About QE
https://www.cloudyni...l/#entry6039895


You should update that thread for CMOS :-)

#62 ccs_hello

ccs_hello

    Voyager 1

  • *****
  • Posts: 10245
  • Joined: 03 Jul 2004

Posted 31 August 2017 - 08:08 PM

CMOS image sensor thread by Jon Rista

https://www.cloudyni...-level-example/



#63 Astrojedi

Astrojedi

    Soyuz

  • *****
  • Posts: 3801
  • Joined: 27 May 2015
  • Loc: SoCal

Posted 01 September 2017 - 03:01 PM

To add to the discussion here is a summary of current Sony CMOS sensors ranked by mV/micron sensitivity (i.e. normalized for pixel size). There is some information missing here as I could not find official sources for it. But if you have any insights please feel free to let me know or comment directly on the thread.

 

Notes:

 

SNR1 - Lower is better - standardized measure by Sony. Our internal measurements have shown that this is a reliable indicator of low light performance. Only downside is that it is not normalized for pixel size.

 

mV Sensitivity - Higher is better. But this has to be evaluated in conjunction with read noise. If two cameras have the same mV sensitivity the one with lower read noise will have superior low light capabilities.

 

Normalized mV sensitivity - This tries to get at the real / true sensitivity of the sensor per micron independent of pixel size. The most important measure in my opinion. A higher sensitivity sensor can have smaller pixels and deliver higher spatial resolution with smaller scopes without excessive increase to exposure time. Like mV sensitivity it has to be evaluated in conjunction with read noise. 

 

Clipboard01.jpg


Edited by Astrojedi, 01 September 2017 - 03:31 PM.

  • OleCuss, roelb and elpajare like this

#64 Relativist

Relativist

    Cosmos

  • *****
  • topic starter
  • Posts: 8151
  • Joined: 11 Oct 2003
  • Loc: OC, CA, USA

Posted 01 September 2017 - 03:54 PM

Shouldn't it be by pixel area?

#65 Astrojedi

Astrojedi

    Soyuz

  • *****
  • Posts: 3801
  • Joined: 27 May 2015
  • Loc: SoCal

Posted 01 September 2017 - 04:45 PM

Shouldn't it be by pixel area?

 

Curtis, your question gives me the opportunity to address another very common misconception on these forums.

 

The sensitivity of a sensor is determined by QE and read noise. I am using the mV / micron measure I derive above as a proxy for QE. Hence combined with read noise it is a much more accurate measure of sensitivity. The role of pixel size is a little bit different as I explain below.

 

Pixel size itself is not the real measure of a sensor's sensitivity rather it is the lever that sensor manufacturers use to achieve a certain dynamic range and SNR. Basically, a low QE high read noise sensor will require larger pixels to generate the same SNR / dynamic range in the same exposure time as a sensor with high QE and low read noise.

 

The concept that pixel size = sensitivity was a rule of thumb which was useful for CCDs as the QE was low and read noise was typically very high. So larger photosites (i.e. sensor pixels) were required to collect enough photons to overcome read noise. The only way to increase "sensitivity" was a larger pixel. Unfortunately, the significant drawback was that you needed larger and larger scopes to image smaller objects due to the poor spatial resolution of sensors with very large size pixels. This was an arms race and the winners were astro vendors. You still need larger scopes to collect more light but now with my C8 I can image detail in ARP galaxies which would just not have been possible a few years ago.

 

With the advent of high QE and very low read noise CMOS you can have significantly smaller pixels and still achieve the same SNR / dynamic range. This provides much better spatial resolution (sampling) and much shorter exposures. It also allows imaging at longer f ratios without significant loss of SNR.

 

At this point a smart guy will ask me... well larger pixels should still improve SNR, right? Yes, that is correct but with the very low read noise of CMOS you can bin pixels in software and still achieve the same SNR. The photosite does not need to be designed to overcome read noise. You don't need physically large pixels.

 

A case in point. The 290mono can easily match or exceed my Lodestar mono in terms of sensitivity with better spatial resolution. One has 2.9 micron pixels and the other has 8.2x8.4 micron pixels - more than 8 times larger by area!

 

For this reason my ideal astro sensor would be a 1 micron pixel size, <0.5e read noise, 90%+ QE sensor. Then depending on the target and focal ratio I can decide what sampling I want using software binning. In fact this is an underappreciated paradigm shift. If I am an imager I don't need to decide what sampling I should use at acquisition time. I can make that decision in post processing. I can also use the extra spatial resolution data as an input to deconvolution algorithms to exceed detail possible with today's sensors. 


Edited by Astrojedi, 01 September 2017 - 05:12 PM.

  • mikefulb and roelb like this

#66 Relativist

Relativist

    Cosmos

  • *****
  • topic starter
  • Posts: 8151
  • Joined: 11 Oct 2003
  • Loc: OC, CA, USA

Posted 01 September 2017 - 05:08 PM

Hence my suggestion that we try out the Rasberry Pi NOIR camera for EAA, they are 1.12 micron pixels. We could really test the capabilities of modern CMOS technology.

I'll be ordering it myself next week.

#67 Astrojedi

Astrojedi

    Soyuz

  • *****
  • Posts: 3801
  • Joined: 27 May 2015
  • Loc: SoCal

Posted 01 September 2017 - 05:19 PM

Hence my suggestion that we try out the Rasberry Pi NOIR camera for EAA, they are 1.4 micron pixels. We could really test the capabilities of modern CMOS technology.

I'll be ordering it myself next week.

 

I cannot find any sensitivity benchmarks on the Sony website for the IMX219 sensor. Also read noise is unknown. But it seems to have a 2x2 analog binning mode.

 

http://www.sony-semi...4/imx219_e.html

 

Not all CMOS sensors are made equal. But at $30 probably worth a shot... some interesting possibilities with the Pi especially if paired with the Pi touchscreen. 


Edited by Astrojedi, 01 September 2017 - 07:02 PM.


#68 Relativist

Relativist

    Cosmos

  • *****
  • topic starter
  • Posts: 8151
  • Joined: 11 Oct 2003
  • Loc: OC, CA, USA

Posted 01 September 2017 - 07:47 PM

The Rasberry Pi camera has been used for AP and guiding. The lens can be removed so we can use it in prime focus mode. I'll be testing at f/2 with a 0.5x reducer. Here is a starting thread.

https://www.raspberr...016aa8d82589e8e

#69 Relativist

Relativist

    Cosmos

  • *****
  • topic starter
  • Posts: 8151
  • Joined: 11 Oct 2003
  • Loc: OC, CA, USA

Posted 01 September 2017 - 08:18 PM

On the pixel area issue, there are two reasons, one is that not all pixels are square. The other is that, understanding your stated goal of having a sensitivity gauge for a given pixel instance analogous to QE is the area if the pixel that will accumulate a charge, which builds up a voltage that is read. Here, things like HCG mode will make a difference.

I'm not so sure about the 1 micron size as ideal, but it's worth testing, that's for sure! For example, with the 1.12 micron pixels of the IMX219 and my 100mm Skyscanner, if I can reduce it to f/2 that will have a about a 1.16 arcsec per pixel resolution. This is worth testing as I said, maybe v3 will be even more sensitive. In addition there are other optical effects to consider when using smaller pixels.

By the way, here is a link to the datasheet for the imx219.

https://www.reddit.c...t_for_the_sony/

#70 ccs_hello

ccs_hello

    Voyager 1

  • *****
  • Posts: 10245
  • Joined: 03 Jul 2004

Posted 01 September 2017 - 09:00 PM

IMX219 is a rather special Exmor-R sensor designed for smartphones.

Its tiny pixel pitch (1.12um) dictates its charge to voltage conversion gain has to be high (i.e., always HCG.)

 

In addition to its usual digital 2x2 or 4x4 like-color summing (some call it binning...), it has something unusual:

2x2 "Analog (Special) Binning" mode, which AFAIK is vertically charge-domain binning and horizontally voltage-domain binning.

 

BTW, its "sensitivity" is 205 ADU (under the 10-bit A/D mode.)



#71 ccs_hello

ccs_hello

    Voyager 1

  • *****
  • Posts: 10245
  • Joined: 03 Jul 2004

Posted 01 September 2017 - 10:14 PM

Additional info:

 

Common use case on that special 2x2 analog binning mode is 1280x720 output format.

 

================

 

Also, max analog gain is 10.66x (or 20.56 dB approx) while

max. digital gain (numerical multiplication, which is just a pure math) is 15.85x (or 24 dB approx.)

 

Net gain (AFE + post A/D multiplication) is 168.96x gain (or 44.56 dB approx.)



#72 Relativist

Relativist

    Cosmos

  • *****
  • topic starter
  • Posts: 8151
  • Joined: 11 Oct 2003
  • Loc: OC, CA, USA

Posted 01 September 2017 - 10:48 PM

Is HCG mode confirmed in the datasheet? It didn't pop out at me as obvious when I saw the circuit page, but my mind is preoccupied at the moment.

#73 ccs_hello

ccs_hello

    Voyager 1

  • *****
  • Posts: 10245
  • Joined: 03 Jul 2004

Posted 01 September 2017 - 11:09 PM

What I meant is that its conversion gain is set to be very high, and is a fixed value.

It is as high as HCG in an IMX224.  It is typical for those small-pitched image sensor (per pixel photon flux too low under the photon-starvation shooting condition.)

 

Another clue is it's just a 10-bit A/D imager.  A larger-pitch IMX224 image senor will have 12-bit A/D capability and 72 dB max gain (roughly 60 dB gain if it's a 10 bit sensor.)



#74 Relativist

Relativist

    Cosmos

  • *****
  • topic starter
  • Posts: 8151
  • Joined: 11 Oct 2003
  • Loc: OC, CA, USA

Posted 02 September 2017 - 02:34 AM

The IMX 219 can also do 2x2 at HD resolution as well, which is fine for most users (4K displays not withstanding - get an A7s at that point).

So, with binning the focal ratio could be kept at f/4 for 1.16 arcsec per pixel for a 100mm scope.

#75 ChrisFC

ChrisFC

    Ranger 4

  • -----
  • Posts: 368
  • Joined: 20 Nov 2015
  • Loc: Sydney, Australia

Posted 02 September 2017 - 03:04 AM

Amp glow reduction circuitry in action. I got  the 224MC ages ago.

 

I know these are under different conditions etc, but here's the Trifid

 

First is with the 224MC version1 board. Taken May 2016.

Second is after I got it upgraded with the v3 board with amp glow reduction mod. Taken July 2017. Amp-glow across bottom and right a bit.

 

Neither had darks subtracted. Both taken in sharpcap, stacked and stretched

.

Attached Thumbnails

  • TrifidNebula 2016-05-04 224MCv1 board.jpg
  • TrifidNebula 2017-07-28 224MCv3 upgrade.jpg

Edited by ChrisFC, 02 September 2017 - 03:18 AM.

  • ccs_hello, Brian_S, roelb and 2 others like this


CNers have asked about a donation box for Cloudy Nights over the years, so here you go. Donation is not required by any means, so please enjoy your stay.


Recent Topics






Cloudy Nights LLC
Cloudy Nights Sponsor: Astronomics