Let's Discuss the Problem with the IMX492 Sensor

[The Lazy Astronomer]

So, I have the ZWO294MM camera, and whilst searching around the internet trying to resolve some troubles with flats on my most recent image, l stumbled upon this post on the Sharpcap forum from Robin Glover.

I tried this out for myself, scope pointed at a wall in the middle of the day, and sure enough, at 120 gain and no filter, it was not possible to saturate all of the pixels, no matter the exposure time. The minimum value as reported by N.I.N.A. was around 58000 - this was the same for a 2 second exposure and a 2 minute exposure.

When l tried 119 gain (LCG mode), it was possible to saturate all pixels, and at gains of around 140+ it was also possible to fully saturate the sensor.

My questions, then, for the far more learned than I, are: should l not be using this camera in the gain range where it doesn't seem to work properly?

Would it be better to stick to the LCG mode and treat it more like a traditional CCD (high read noise, so longer subs needed, but huge full well depth)?

Or would it be better to use higher gains (no real change in read noise from 120 gain, so no change in sub length, but lower well depth and reduction in dynamic range).

Or, is it a non-issue and should I just continue imaging at 120 gain?

Thanks in advance!

Edited July 11, 2021 by The Lazy Astronomer

[wimvb]

I'll be following this thread with interest since I also have an ASI294MM. I'm also not happy with the uneven flats that won't correct the light subs properly. We have summer recess up here untill mid/end of August, but I will start testing later this month or early August.

[david_taurus83]

So, there's an area of gain that's effectively unusable? Can't use unity gain?

[wimvb]

39 minutes ago, david_taurus83 said:

So, there's an area of gain that's effectively unusable? Can't use unity gain?

The problem is that the section of gain settings is just at the sweet spot of where you want to image: low read noise, high dynamic range. I compared the specs to those of my old camera (ASI174MM-Cool), and without the hdr setting, the 294 has higher read noise than the 174, and barely higher dynamic range at gain 0, despite its 14 bit output.

I've looked at flats I took in April at gain 0. 0.5 seconds flats with an Optolong L filter in place, looked fine, while flats taken at gain 120 with an Ha filter had a weird pattern. I will take new flats for Ha at gain 0, or thereabout.

[vlaiv]

Does it matter that pixels won't saturate?

You can impose artificial saturation and clip pixels yourself at say 50000ADU?

Most of the time that is what is happening with hard saturation limit - real values are clipped either because of e/ADU and bit depth or they are clipped in firmware.

Pixel full well capacities are probably not very even due to manufacturing defects and maybe in this gain range - there is simply no clipping but you are rather reading out actual full well of each pixel.

1 hour ago, wimvb said:

I'm also not happy with the uneven flats that won't correct the light subs properly

Uneven flats are not issue if flat fielding works. I also have very strange pattern in Ha with ASI1600 and Panasonic sensor. It's sort of checkerboard pattern - however, flat fielding works very well regardless.

Why is it that yours is not working? Have you tried flat/flat correction? That would be - take a set of flats of one exposure length, take set of flats of another exposure length and correct first master with second master (do add flat darks for proper flat calibration).

[The Lazy Astronomer]

2 hours ago, david_taurus83 said:

So, there's an area of gain that's effectively unusable? Can't use unity gain?

It's not that it's unusable, far from it - I've been using it at unity gain for the last 6 months and have taken some pleasing images, and I've seen plenty of others using 120 gain with good results as well. This is why I'm not sure if it's a non-issue in the 'real world'.

1 hour ago, wimvb said:

I've looked at flats I took in April at gain 0. 0.5 seconds flats with an Optolong L filter in place, looked fine, while flats taken at gain 120 with an Ha filter had a weird pattern. I will take new flats for Ha at gain 0, or thereabout.

The weird narrowband flats I think are a feature of the sensor itself, rather than anything to do with gain settings. 

From what I've been reading, this camera seems to be very picky about calibration frames (lots of people having issues with flats).

In my last image, I had to reshoot temperature matched flats and flat darks (-10C) in order to get the flats to properly calibrate the lights. For 2 previous images though, I had used room temperature flats and flat darks and these had worked. Still no idea why suddenly the room temp flats didn't work, but I'm wondering whether it was somehow related to sensor non-linearity and maybe somehow that's affected by sensor temperature? (I don't know what I'm talking about 🙃)

1 hour ago, vlaiv said:

Does it matter that pixels won't saturate?

This is exactly what I'm thinking at the moment. I've read conflicting things - someone pointed out that if a non linearity was repeatable then it shouldn't be an issue, but Mr Glover seems to be of the opinion that the 120 - 190 gain range should be avoided entirely.

I suppose the only way to know would be to do some real world side-by-side tests at gain 0 and gain 120 to see if there's any effect on the image. If only the clouds would clear... 🤔

[vlaiv]

4 minutes ago, The Lazy Astronomer said:

but Mr Glover seems to be of the opinion that the 120 - 190 gain range should be avoided entirely

Did he give exact reason why? What won't work if there is no saturation / clipping in camera?

[wimvb]

10 minutes ago, The Lazy Astronomer said:

It's not that it's unusable, far from it - I've been using it at unity gain for the last 6 months and have taken some pleasing images, and I've seen plenty of others using 120 gain with good results as well. This is why I'm not sure if it's a non-issue in the 'real world'.

I’ve also taken pleasing images with my ASI294, but I can’t get the same even flat background that I got with my ASI174 (although, that camera has a smaller sensor). And that won’t allow me to go as deep as I’d want.

[wimvb]

15 minutes ago, The Lazy Astronomer said:

The weird narrowband flats I think are a feature of the sensor itself, rather than anything to do with gain settings. 

I also have a similar pattern in the RGB flats. But I got new unmounted RGB filters together with the camera, and the filters have non treated edges, so I can’t rule out reflections or scattering from those. I will get a set of ZWO filter rings before I set up my rig again, and do a proper flats study.

[The Lazy Astronomer]

12 hours ago, vlaiv said:

Did he give exact reason why? What won't work if there is no saturation / clipping in camera?

No, I must admit I've not actually seen any reasoning behind it, other than the point that he says the technical documents for the sensor (which I've not seen either) apparently advise against using the affected gain range (but again, there doesn't seem to be an explanation of any potential impact to images taken within that range).

Unfortunately, my understanding of the technicalities of the subject is rather limited, so I'm just left stuck thinking "does it really matter?" The persistent cloud and rain here over the last 3 or 4 weeks has meant I've not even been able to test it!

[The Lazy Astronomer]

Well, I tried some gain 0 imaging last week thinking I could enjoy those lovely deep wells, but completely overlooked the fact that at gain 0, it's 4e-/ADU, which translates to terrible results.

As a side note: I assume in order to get a comparable image to one taken at unity gain, I'd have to expose each sub for 4x longer to record the same ADU value for a given pixel (as it takes 4x as many electrons to bump up to the next ADU value)? Or is it not quite as straightforward as that? Even if it is as straightforward as that, I don't particularly fancy having to increase my imaging time 4-fold, especially with typical UK skies!!

So, after that disappointment, I thought about it a bit more, and I'm now erring towards it being a non-issue in real world use, assuming that the issue only presents itself in the upper part of the camera's range. My thinking here is that the vast majority of my data would sit in the lower 1/3 of the ADU range, and it's only really the stars which would be affected at the upper end.

So, is there some kind of analysis/experiment I could do to test the sensor response/linearity across it's full range, and what sort of results should l be looking for?

[wimvb]

To test for linearity, you need a fixed light source (flat panel?). You then take a series of exposures with increasing exposure time and measure and plot the ADU count. At short exposure time there is a floor, which is dependent among other things, on the offset you set in the capture software. At long exposure time you reach a ceiling, which is saturation or full well.

[The Admiral]

On 09/07/2021 at 00:09, The Lazy Astronomer said:

I have the ZWO294MM camera, and whilst searching around the internet trying to resolve some troubles with flats

I presume that you looked at CN, I know that there was much discussion on flats uniformity there.

Ian

[The Lazy Astronomer]

4 hours ago, The Admiral said:

I presume that you looked at CN, I know that there was much discussion on flats uniformity there.

Ian

Yes, I think I solved my problem with the flats by temperature matching everything. 

[The Admiral]

3 hours ago, The Lazy Astronomer said:

Yes, I think I solved my problem with the flats by temperature matching everything. 

I may be wrong but I also vaguely remember something about not using very short subs for flats.

Ian

[The Lazy Astronomer]

On 27/07/2021 at 06:56, wimvb said:

To test for linearity, you need a fixed light source (flat panel?). You then take a series of exposures with increasing exposure time and measure and plot the ADU count. At short exposure time there is a floor, which is dependent among other things, on the offset you set in the capture software. At long exposure time you reach a ceiling, which is saturation or full well.

Ok, so probably went about this the wrong way, but here's what l got:

I have 2 seperate graphs because I my light source was too dim and l realised after a little while I'd be sat there forever doing the higher adu values (I did this all manually, I'm sure there's probably a far more sensible way to this through software, but oh well it's done now...)

The first graph shows exposures ranging from 0.1s to 20s, in increments of 0.1s up to 1s, and then increments of 1s thereafter.

The second graph is with a brighter light source, exposure times ranging from 2s up to 7.5s in increments of 0.5s.

I used the mean adu value as reported in NINA's statistics pane.

The second graph stops slightly before the saturation point as when it nears it the curve shape flattens off, but the sensor never fully saturates (there is virtually no difference between a 9s exposure and a 20s exposure - both with mean adus of 65227.xx)

My reading of this is that with r2 values of >0.99 on both graphs, l can conclude that the sensor does respond in a linear fashion until just before saturation, at which point something weird begins to happen. Is that fair conclusion to make?

If so, then I would also conclude that in an imaging scenario, it should only really be brighter stars which are affected, yes?

If I've gone about this completely the wrong way, then more than happy for someone to point me towards a correct way of doing this. 

Full data table below for anyone interested. Note that from point 30 onwards, the brightness of the light source was increased:

ref	time (s)	mean adu
1	0.1	1798.56
2	0.2	1877.22
3	0.3	1960.96
4	0.4	2056.71
5	0.5	2133.31
6	0.6	2221.36
7	0.7	2296.38
8	0.8	2391.94
9	0.9	2440.63
10	1	2965.45
11	2	3908.2
12	3	4656.95
13	4	5508.67
14	5	6459.62
15	6	7173.07
16	7	8051.98
17	8	8919.21
18	9	9978.88
19	10	10712.89
20	11	11812.54
21	12	12750.33
22	13	13278.09
23	14	14279.82
24	15	15335.5
25	16	16066.85
26	17	17344.15
27	18	17835.57
28	19	19414.6
29	20	21087.09
30	2	22494.96
31	2.5	26243.66
32	3	30755.9
33	3.5	34568.44
34	4	38761.24
35	4.5	43473.7
36	5	46864.11
37	5.5	50186.24
38	6	54370.03
39	6.5	58330.17
40	7	61840
41	7.5	63741.89
42	8	65089.26
43	8.5	65502.59
44	9	65526.98
45	9.5	65527.04
46	10	65527.06
47	15	65527.12
48	20	65527.21