New Larger Format CMOS Cameras - the other costs

[gregbradley]

The slight banding at the lowest offset disappears in the higher offset. That's good to know.

Greg.

[Martin Pugh]

Yes, I have the QHY600M and 600C.

I will try and get some darks at the same duration and temp and post up.

Originally Posted by AnakChan

Are you both talking about the ASI6200MM? I thought Martin had the QHY600MM, and in CN I've seen references of 50 for the ASI6200. Would the two have the same offset being IMX455's or would the implementation around it influence a difference in their offsets?

Anyhow, here's how my 300sec darks at -10C with 0 Gain with my ASI6200MM.

[Martin Pugh]

Greg...now I see what you are talking about.

yes, the QHY600 does have this, well, at 0degs it does. So I need to do more tests at a lower temp and various offsets, which I am doing.

Martin

Originally Posted by gregbradley

The slight banding at the lowest offset disappears in the higher offset. That's good to know.

Greg.

[Camelopardalis (Dunk)]

ZWO and QHY have typically used different scales for their range of possible gain values, so it’s entirely possible that their offset values are not portable between the two (although that does not necessarily mean not translatable)

What’s important re: banding is that the sensor is consistent. It would be interesting to see how the 6200 performs in this respect.

Some earlier Sony sensors suffer quite badly from inconsistencies caused by built-in (re)calibration from one power cycle to the next. The implementations of the 178 are a good of example of this and consequently can be more tricky to calibrate well.

[Martin Pugh]

I did get AIP4WIN to work once with these very large files, but now it crashes every time. That's if you can get the program to run on Win10 at all in the first place.

So maybe there is a script problem. I have seen screen shots of the script working with DSLRs.

Are there other programs that can measure these basic camera characteristics?

Originally Posted by Camelopardalis

Is it possible that the script doesn’t report with sufficient decimal places?

The thermal noise is very low on this sensor, an order of magnitude lower than the ASI1600. Quite astonishing.

[Camelopardalis (Dunk)]

I believe PixInsight has a CCDParmeters script.

[Martin Pugh]

That is the one I am using which does not report dark current.

[Martin Pugh]

So I did some extensive tests on the QHY600C yesterday and got AIP4WIN to work also.

I chose 2 gain and 3 offset values for each readout mode and these bias/dark frames were combined with flats taken at 3 ADU values (one near saturation, one midpoint and one low). The flats however, were taken at one gain/offset value and one readout mode only.

The gain values I chose were ones that I selected as being optimum and balanced against what I would like to get out of the camera in terms of read noise v full well v dynamic range.

Findings:
Offset - made no impact whatsoever across all images. I suspect this is because a different offset value would not manifest itself here, but rather in the dynamic range of the actual image.

QHY specify that the camera is linear all the way up to 70+K in readout mode 0 and 2 with a gain of 0.

Well of course, a gain of 0 is not much cop right? And if you look at the graphs on the QHY website, a gain of 0 has a very big impact on the other camera charateristics. So bit of a pointless marketing there methinks.

Indeed, in all of the 15 sets of images I took and ran through AIP4WIN, the flats at an ADU value of 15K, resulted in lower system gain, lower readout noise and dark current. In contrast, the flats with an ADU of 50K, had the exact opposite affect.

Temp was -10degs. Darks were 5mins duration (no amp glow).

The lowest dark current recorded was 0.0006 e/pix/sec.
The highest dark current recorded was 0.011 e/pix/sec

The lowest readout noise recorded was 0.16e
The highest readout noise recorded was 1 electron

So, what I cannot explain is why the flat ADU values had such a marked affect on the results - or does it simply mean that you should take your flats using the same gain/offset values and readout modes too!?

Of particular interest to me, is that NONE of the values I recorded match or come anywhere near the values seen on the QHY website for this camera.

So obviously, what I need to do is to take that data set and run it through a different program now and compare the results.

Comments, thoughts?

Martin

[Atmos (Colin)]

I would have thought that the flats would have to match the gain/offset of the rest of your calibration frames to get accurate results otherwise the underlying noise profile in the flats won’t match the darks or bias.

[ericwbenson (Eric)]

Originally Posted by Martin Pugh

Temp was -10degs. Darks were 5mins duration (no amp glow).

The lowest dark current recorded was 0.0006 e/pix/sec.
The highest dark current recorded was 0.011 e/pix/sec

Hmmm how is this possible? The dark current expressed in e-/pix/s is only dependent on the temperature of the silicon (and the design of the gate), or at least should be...

Are those ADU/pix/sec somehow or something else is amiss?

Best,
EB

[gregbradley]

That's what I would have thought. I haven't used one of these CMOS cameras but my FLIs have 2 digitisation speeds with different read noises.

I have had it happen where I forget to reset the digitisation to the lower read noise level (and slower downloads) after using it at the higher download speed to speed up focusing. Flats then do not work well in that scenario where the flats were taken with the slower low read noise digitisation but the lights were taken with the fast download with the higher read noise.

Also is it possible to have less than 1 electron read noise? What does .16 electron mean? Several pixels of zero and one with one electron in it?

Greg.

[Martin Pugh]

I totally agree. Getting less than 1 electron should not be possible and that was my first indication that AIP was outputting skewed results. But it appears to be a case of garbage in, garbage out as the test was flawed in any case as one set of flats was used, and it seems that flats should be taken at the same gain/offset (but not temp).

But those flats I do have would be applicable to some of the data so I will go back and inspect that.

[Martin Pugh]

Well, I went back and checked the data where the flats would apply, and still the read noise is less than 1e.

Then I forgot to check 'CFA' in the PI basic ccd parameters script and that too reported 0.003 e for readout noise. So perhaps AIP4WIN cannot interpret OSC data.

[Martin Pugh]

Just examined the FITS header of this data set.

For COLOURCCCD - it should have a non-zero figure if the image is from a bayer colour CCD. But it is set at 0.

I dont know how this would be interpreted in software though.

[Martin Pugh]

Just ran a set of images (with the flats that matched) through PI basic ccd parameters script.

While PI does not output dark current value, the readout noise now correlates with the QHY spec i.e in Photo Mode with a gain of 26, the readout noise by spec is 2.7e and I measured 2.48e and a gain of 0.38.

I then ran another set with darks/bias that did not match the flats and
again the readout noise matches that of the spec at 1.7e and a gain of 0.34.

The downside is that a third set using the Ultra Full Well mode and a gain of 20 (fullwell 69000), PI failed miserably saying the readout noise was 0.3e and the fullwell was 2300K.

So who knows. I dont think I like these CMOS cameras...too much room for error and you seemingly have to go to extraordinary lengths just to identify a usable set of parameters and to calibrate an image.

[gregbradley]

Originally Posted by Martin Pugh

Just ran a set of images (with the flats that matched) through PI basic ccd parameters script.

While PI does not output dark current value, the readout noise now correlates with the QHY spec i.e in Photo Mode with a gain of 26, the readout noise by spec is 2.7e and I measured 2.48e and a gain of 0.38.

I then ran another set with darks/bias that did not match the flats and
again the readout noise matches that of the spec at 1.7e and a gain of 0.34.

The downside is that a third set using the Ultra Full Well mode and a gain of 20 (fullwell 69000), PI failed miserably saying the readout noise was 0.3e and the fullwell was 2300K.

So who knows. I dont think I like these CMOS cameras...too much room for error and you seemingly have to go to extraordinary lengths just to identify a usable set of parameters and to calibrate an image.

You seem to be doing pioneering work here Martin.
I guess we have been spoilt as SBIG and other companies just put out their cameras already optimised for gain etc. Its a lot simpler.

I would suggest simplifying it down to 2 settings each with its own set of calibration files. Or perhaps even just one set optimised for the RASA.

I am not sure why the CMOS camera makers aren't doing the same with their product. To enhance flexibility?

Greg.

[Martin Pugh]

The pixel sizes obviously have their advantages but FWIW, I intend to bag a couple of CCD cameras before they are no longer available.

cheers
Martin

[Camelopardalis (Dunk)]

Martin, your experience does seem overly complicated, certainly more so than my (smaller) CMOS cameras (each ZWO btw).

I don’t fully understand the logic of QHY’s gain modes. ZWO just plots it along a linear scale, and at gain 100 the camera switches to HCG mode, with the subsequent drop in read noise. FWIW, the ZWO camera doesn’t drop below 1e, although such a property is not unheard of - the IMX224 exhibits this at higher gain.

Since most CMOS chips (at least, those in the astro market) have on-sensor amplifiers, the noise introduced can be somewhat disconnected from the signal reading itself. This is the same as the ISO-invariance we see with Sony/Nikon DSLRs/mirrorless cameras.

Looking at the plots for the ZWO ASI6200, I’d probably be picking gain 100 pretty much every time...there seems to be a tiny amount of dynamic range sacrificed for enabling of HCG. In “traditional” use - from a dark site, especially with a fairly fast scope - then gain 0 would give maximum well depth and a read noise of ~3.5e...still pretty respectable. The best choice will depend on your preference with respect to dynamic range.

The offset value itself has a minimal impact on dynamic range...if you increase the offset such that the median pixel value for a bias frame shifts from 200 to 400, this is almost negligible difference with a well depth >50,000.

[Martin Pugh]

Just looked at the ZWO graphs, and yes a gain of 100 is very good in all respects and I wish QHY did that; and it appears a single readout mode too.

Notwithstanding, how can they recommend a gain of 0? Just what sort of SNR would you achieve with that?

Also, very happy to see dark current of far less than 1 e/pix/sec.

When I have been collimating my RASA using sharpcap and 1 second exposures, I have turned the gain down to something like 35 and the star has become extremely faint.

Clearly, ZWO’s implementation of this sensor differs significantly from QHY’s

[Camelopardalis (Dunk)]

I've just had a quick glance at the QHY graphs, and it looks like Mode #1 and gain 55 is the closest match to how ZWO has implemented their gain profile.

The QE of the IMX455 is very high, according to QHY. At gain 0, before HGC kicks in, you're getting the full well depth, and full read noise to boot - this may be an option for going long and deep? To be fair, the read noise on existing full frame CCDs is substantially higher, so your SNR will still be better with the QHY600, and this is all with the resolution of 3.75um pixels.

With HGC, it looks like you're sacrificing about half the well depth for roughly half the read noise...hence the dynamic range is comparable.

The dark noise on Sony's latest sensors is quite astonishing when cooled. Even the IMX183 is down to ~0.003.

There is some flexibility with the variable gain on CMOS cameras - personally, I usually stick to 2 or 3 different values, depending on where I am and what I'm imaging and what filter is in front of the sensor.