57 replies to this topic

[jfrech14]

So I have continued to do some more tests on my 1600 cool and decided to investigate linearity since they claimed it is perfectly linear and I haven't seen any linearity graphs. Well... it is pretty dang linear. Attached is a graph with single image statistics. So in order to investigate the linearity to a greater extent, better statistics will be needed. But, it is pretty solid. I also decided on the listed ZWO driver settings because I found that it minimized my offset (minimized clipping in bias frames as well), and I purposefully shot for a DR slightly above the 4095 so that any fluctuations will not take my DR below 4095 and it also keeps other statistics minimized that would cause artifacts.

 

Since I am a student at an imaging science department we have integrating spheres and other tools to measure QE, so I am going to see if I can measure my system's QE with each filter so I can try to plan ahead more carefully since clear skies are proving to be a real rarity. Also... why not ? haha

 

Let me know if this helps any of you possibly interested in photometry since a big issue is linearity and it doesn't seem to be a big issue here.

 

Clear skies,

Josh

Attached Thumbnails

• 

[jfrech14]

Also would like to add that image statistics are particularly bad due to using single, uncalibrated images.

[Ken Sturrock]

Moved to CCD Imaging....

[gregj888]

Josh,

 

The concern I (many) have with CMOS and Photometry are changes due to the internal calibration.  I don't know what if anything forces a re-cal on the 1600.  On the QHY5Lii a change in shutter will usually do it as will a few other capture changes.  If the internal re-cal isn't tripped, CMOS should be as or more linear than CCDs I would think

 

Is it possible to take say 10 (or 100) image sets and change the exposure to something a lot different, then back to the original setting a few times and see how the sets cluster?  Be good to do the same thing starting and stopping the software.

 

Using one set as a flat for another is a good way to see if patterns emerge.

 

Thx,

 

Greg

[jfrech14]

Moved to CCD Imaging....

Dang.. last time you moved it to equipment general I thought so I put it here haha. I will get the hang of it one day...

[Ken Sturrock]

Moved to CCD Imaging....

Dang.. last time you moved it to equipment general I thought so I put it here haha. I will get the hang of it one day...

Nope. Last time, I did the exact same thing. No problem though, Josh - it makes me feel needed.

[Jon Rista]

Josh,

 

The concern I (many) have with CMOS and Photometry are changes due to the internal calibration.  I don't know what if anything forces a re-cal on the 1600.  On the QHY5Lii a change in shutter will usually do it as will a few other capture changes.  If the internal re-cal isn't tripped, CMOS should be as or more linear than CCDs I would think

 

Is it possible to take say 10 (or 100) image sets and change the exposure to something a lot different, then back to the original setting a few times and see how the sets cluster?  Be good to do the same thing starting and stopping the software.

 

Using one set as a flat for another is a good way to see if patterns emerge.

 

Thx,

 

Greg

I'm not sure why the recal would affect linearity to any detrimental degree. The sensor pattern from DSNU is extremely small on the ASI1600, much smaller in amplitude than even the read noise, which is already very low. Even if that pattern changes (and it does so slowly, I have found that I can reuse masters for a month and a half to two months before any change in pattern is large enough to affect calibration), it is of such a miniscule magnitude relative to the full range of the sensor, I can't imagine it affecting much more than the data right around zero. 

[jhayes_tucson]

Nice work Josh!   That looks really good.

 

John

[Alfredo Beltran]

Really interesting Josh.

 

it would be nice to know the measured QE.

 

Regards

 

Alfredo

[drmikevt]

Can someone please comment on why linearity is or isn't a good thing and why we want to know if our cameras follow a linear exposure graph?  Is it just so that we can predict needed exposure?

 

Just tryin to learn me some science...

 

Mike

[jfrech14]

Josh,

 

The concern I (many) have with CMOS and Photometry are changes due to the internal calibration.  I don't know what if anything forces a re-cal on the 1600.  On the QHY5Lii a change in shutter will usually do it as will a few other capture changes.  If the internal re-cal isn't tripped, CMOS should be as or more linear than CCDs I would think

 

Is it possible to take say 10 (or 100) image sets and change the exposure to something a lot different, then back to the original setting a few times and see how the sets cluster?  Be good to do the same thing starting and stopping the software.

 

Using one set as a flat for another is a good way to see if patterns emerge.

 

Thx,

 

Greg

 

I must admit I was not aware of any active internal calibrations so I can't really comment on that. If you just mean consistency between switching around settings and expecting things to be as they were last time you used that setting, I haven't seen any issues other than download issues in Maxim DL. I have read so many people claiming that any CMOS camera does a self calibration but my time on google in tech documents only point that Sony does this which begs me to think that it isn't a CMOS thing but just an engineering thing since CMOS sensors are used in DSLRs and you want to avoid having to do calibrations. But I do not believe the 1600 is a sony sensor. But, I am emailing Sam from ZWO to get an official verdict. Looking at the electronics of a CMOS itself, nothing seems to strike me as requiring this internal calibration and I can see the benefit of it for a camera that is usually used as a guide camera like the QHY5Lii, but not one meant for longer exposures. So, if you have more info on this I would love to give it a read and I will let you know what Sam tells me

 

 

 

 

Josh,

 

The concern I (many) have with CMOS and Photometry are changes due to the internal calibration.  I don't know what if anything forces a re-cal on the 1600.  On the QHY5Lii a change in shutter will usually do it as will a few other capture changes.  If the internal re-cal isn't tripped, CMOS should be as or more linear than CCDs I would think

 

Is it possible to take say 10 (or 100) image sets and change the exposure to something a lot different, then back to the original setting a few times and see how the sets cluster?  Be good to do the same thing starting and stopping the software.

 

Using one set as a flat for another is a good way to see if patterns emerge.

 

Thx,

 

Greg

I'm not sure why the recal would affect linearity to any detrimental degree. The sensor pattern from DSNU is extremely small on the ASI1600, much smaller in amplitude than even the read noise, which is already very low. Even if that pattern changes (and it does so slowly, I have found that I can reuse masters for a month and a half to two months before any change in pattern is large enough to affect calibration), it is of such a miniscule magnitude relative to the full range of the sensor, I can't imagine it affecting much more than the data right around zero. 

 

I must admit that I don't know what is being discussed here because DSNU is something I had thought is present in every sensor and I was unaware of an internal calibration in any CMOS sensors because having them just doesn't make sense for a long exposure astro camera in my opinion since packing on more transistors for a SOC kind of design will just add heat to the sensor which is one disadvantage of CMOS already. It makes sense if it is a Sony thing, but if it is universal I must say I have a lot of research to do haha. Anyway, I am also confused about his reference between "recalibration" vs "internal calibration". I would figure any filtering/math done is on an analog level and as long as a reference voltage is maintained then nothing in the backend electronics should shift except some CDS changes before each readout. Time to hit the books for me haha.

[jfrech14]

Nice work Josh!   That looks really good.

 

John

Thanks, John!

 

Do you have any knowledge of the calibrations mentioned above? 

[jfrech14]

Can someone please comment on why linearity is or isn't a good thing and why we want to know if our cameras follow a linear exposure graph?  Is it just so that we can predict needed exposure?

 

Just tryin to learn me some science...

 

Mike

It is definitely helpful in determining ideal exposure times but that heavily depends on the skies too so it is a small step in the process in that area. Linearity of a detector is important in photometry because it makes the correlation of your data back to the actual flux onto your detector that much easier. Also, for relative photometry, as a star pulsates or does what it does, the response could be different at those two points and the signals are not directly comparable. You can apply a curve to fix this in post processing but the problem is that if it is nonlinear, each pixel might be nonlinear to different amounts and the correction would have to be very elaborate to perfectly linearize a nonlinear sensor in processing. So, best to have a linear detector or to make sure you are only exposing in the linear region of a nonlinear detector... Still learning all the nitty gritty stuff myself but it sure is fun

[Jon Rista]

The ASI1600/QHY163 use a Panasonic sensor, not a Sony sensor. The sensor is much like the Sony CMOS sensors, though, with per-column ADC and probably some kind of advanced CDS units that self-tune each column to minimize banding FPN. Who knows what else this particular sensor may tune.

 

In my testing, there does seem to be slow drift in the bias pattern over the long term. I have not nailed down an exact rate of drift, or whether there may be something specific that triggers a change. I have been more interested in getting image data than testing as of late. But it does seem as though between one and two months, the patterns can shift just enough that calibration is no longer perfect. It isn't bad, per se...however there might be some slight residual pattern after calibration is performed.

[jfrech14]

The ASI1600/QHY163 use a Panasonic sensor, not a Sony sensor. The sensor is much like the Sony CMOS sensors, though, with per-column ADC and probably some kind of advanced CDS units that self-tune each column to minimize banding FPN. Who knows what else this particular sensor may tune.

 

In my testing, there does seem to be slow drift in the bias pattern over the long term. I have not nailed down an exact rate of drift, or whether there may be something specific that triggers a change. I have been more interested in getting image data than testing as of late. But it does seem as though between one and two months, the patterns can shift just enough that calibration is no longer perfect. It isn't bad, per se...however there might be some slight residual pattern after calibration is performed.

Ah, thanks for the clarification I am not sure how you figured out it is Panasonic. Maybe I didn't look hard enough or just gave up when ZWO said they agreed not to give much detail about the sensor.. Well that sure is a bummer and I get why Greg was saying that is a concern for photometry. But seems like that can be fairly negligible if the photometric target is  sampled well and calibration is maintained regularly. Looks like I have a new cloudy nights project so I can give Greg some feedback too haha

[Jon Rista]

Well,  I am pretty sure I found a datasheet for the color version of the sensor. Someone, forget who, speculated that it was the same sensor used in a couple Olympus cameras, so I started there, and found a sensor that had exactly the right specs, and it was a Panasonic 4/3 16mp sensor (which makes sense...those two manufacturers capitalized on the 4/3 market). I am actually assuming that the mono version is some alternative version of the color sensor. That could be totally incorrect...but it's the best we had to go on at the time. 

[jfrech14]

Well,  I am pretty sure I found a datasheet for the color version of the sensor. Someone, forget who, speculated that it was the same sensor used in a couple Olympus cameras, so I started there, and found a sensor that had exactly the right specs, and it was a Panasonic 4/3 16mp sensor (which makes sense...those two manufacturers capitalized on the 4/3 market). I am actually assuming that the mono version is some alternative version of the color sensor. That could be totally incorrect...but it's the best we had to go on at the time. 

 

 

Josh,

 

The concern I (many) have with CMOS and Photometry are changes due to the internal calibration.  I don't know what if anything forces a re-cal on the 1600.  On the QHY5Lii a change in shutter will usually do it as will a few other capture changes.  If the internal re-cal isn't tripped, CMOS should be as or more linear than CCDs I would think

 

Is it possible to take say 10 (or 100) image sets and change the exposure to something a lot different, then back to the original setting a few times and see how the sets cluster?  Be good to do the same thing starting and stopping the software.

 

Using one set as a flat for another is a good way to see if patterns emerge.

 

Thx,

 

Greg

 

According to Sam, the issue that he Sony and Aptina sensors have with this internal calibration is not present in the 1600mm, so any drifting in Bias may be a deeper and more worrisome problem?

[tolgagumus]

What method did you use to figure out the linearity? And I am not sure there is a claim that 1600 is linear. When I did my beta test, I ran it through the linearity inspector in CCDAP. I found that it was linear up to 17000 ADUs.

Sent from my SAMSUNG-SM-G870A using Tapatalk

[jfrech14]

What method did you use to figure out the linearity? And I am not sure there is a claim that 1600 is linear. When I did my beta test, I ran it through the linearity inspector in CCDAP. I found that it was linear up to 17000 ADUs.

Sent from my SAMSUNG-SM-G870A using Tapatalk

It's quite possible that those results are the product of using different settings that I have used. I used settings to optimize full well vs dynamic range in the reality it has a 12bit adc. Sam told me the sensor is incredibly linear and I tend to agree. Is it possibly that ccdap was using electrons rather than ADU because ADU is an arbitrary unit based on many settings in the camera. But my data support the claim it is linear and I will be adding data to increase the confidence of the data points. I used a flat source to expose the sensor to different amount of light using varying exposure times from nearly  bias length to around 4 seconds focusing on the regions where nonlinearity are most likely to occur. 

[tolgagumus]

What method did you use to figure out the linearity? And I am not sure there is a claim that 1600 is linear. When I did my beta test, I ran it through the linearity inspector in CCDAP. I found that it was linear up to 17000 ADUs.

Sent from my SAMSUNG-SM-G870A using Tapatalk

It's quite possible that those results are the product of using different settings that I have used. I used settings to optimize full well vs dynamic range in the reality it has a 12bit adc. Sam told me the sensor is incredibly linear and I tend to agree. Is it possibly that ccdap was using electrons rather than ADU because ADU is an arbitrary unit based on many settings in the camera. But my data support the claim it is linear and I will be adding data to increase the confidence of the data points. I used a flat source to expose the sensor to different amount of light using varying exposure times from nearly bias length to around 4 seconds focusing on the regions where nonlinearity are most likely to occur.

I used the unity gain for testing. CCDAP has a procedure to test sensor linearity. And it gives a report at the end. It uses adu counts. For photometry people use 3-4 short exposures and stack instead of using a single long exposure.

Sent from my SAMSUNG-SM-G870A using Tapatalk

[GTom]

Keep up the good work Josh! I would be very much interested in the QE of the monochrome version - at any wavelength, because relative QE is published on zwo's website. This is a decisive factor for me, if I should buy the narrower fov 414ex or grab the asi.

[yawg]

 

What method did you use to figure out the linearity? And I am not sure there is a claim that 1600 is linear. When I did my beta test, I ran it through the linearity inspector in CCDAP. I found that it was linear up to 17000 ADUs.

Sent from my SAMSUNG-SM-G870A using Tapatalk

It's quite possible that those results are the product of using different settings that I have used. I used settings to optimize full well vs dynamic range in the reality it has a 12bit adc. Sam told me the sensor is incredibly linear and I tend to agree. Is it possibly that ccdap was using electrons rather than ADU because ADU is an arbitrary unit based on many settings in the camera. But my data support the claim it is linear and I will be adding data to increase the confidence of the data points. I used a flat source to expose the sensor to different amount of light using varying exposure times from nearly  bias length to around 4 seconds focusing on the regions where nonlinearity are most likely to occur. 

 

Josh, to determine the linearity of a device you need to plot the residuals, not just find an R2 of a linear fit of the data.  could you plot the residual and see what you get?  without the residual plot, we dont' know if we're actually seeing linearity or not.

[jfrech14]

 

 

What method did you use to figure out the linearity? And I am not sure there is a claim that 1600 is linear. When I did my beta test, I ran it through the linearity inspector in CCDAP. I found that it was linear up to 17000 ADUs.

Sent from my SAMSUNG-SM-G870A using Tapatalk

It's quite possible that those results are the product of using different settings that I have used. I used settings to optimize full well vs dynamic range in the reality it has a 12bit adc. Sam told me the sensor is incredibly linear and I tend to agree. Is it possibly that ccdap was using electrons rather than ADU because ADU is an arbitrary unit based on many settings in the camera. But my data support the claim it is linear and I will be adding data to increase the confidence of the data points. I used a flat source to expose the sensor to different amount of light using varying exposure times from nearly  bias length to around 4 seconds focusing on the regions where nonlinearity are most likely to occur. 

 

Josh, to determine the linearity of a device you need to plot the residuals, not just find an R2 of a linear fit of the data.  could you plot the residual and see what you get?  without the residual plot, we dont' know if we're actually seeing linearity or not.

 

This was done with single exposures for each exposure length. I found nothing online showing any linearity data so I just did it as a quick show of whether the sensor is remotely linear. If you look at the data points, the deviations of each point from the regression are pretty low, but the uncertainty in the measurements are larger then the deviations from the regression. Therefore I didn't bother doing more in-depth statistics. I'm hoping I'll be able to doing better characterization after the semester ends and the integrating sphere is up and running. It's cloudy in Rochester in the winter, so I expect to have plenty of time to do it haha. But you have a valid point. R^2 doesn't say enough for someone characterizing their sensor for photometry. 

[GTom]

Well, its a very rude thing to wish clouds to an Astronomer, but this case...

[jfrech14]

Well, its a very rude thing to wish clouds to an Astronomer, but this case...

Hahah well... since the 8 position FW will be in my mailbox this weekend when I get home, my setup will be complete and I can get on it. Perhaps just set it up in my closet and let it run I am bringing my flat panel back with me so I can at least do some better measurements. As for QE... I HATE Sony's habit of producing relative QE. That is the number one thing I miss about Kodak/Truesense I think. They gave great qualitative information about their sensors.