CMOS cameras and optimum exposure times etc.

[UKJay1971]

I am about to start using a QHY163m cooled camera (I currently use an Altair IMX178c uncooled camera so not unfamiliar to CMOS) and am busy reading all about gain / offset, noise etc.

I understand that guiding is seldom to never perfect in the real world however I always shoot shorter exposures and stack whereas my friend insists that longer exposures will always give better SNR, all things being equal. I should add that he lives in Nevada and so dark skies are an hour's drive away, whereas I live in light polluted West Yorkshire!

My real question is this: given perfect guiding, would long exposures at gain 0, offset 0 always be the preferred option, limited only by skyfog or are there instances where using gain and shorter exposures would produce better results?

[vlaiv]

Given real well behaved sensor (real in terms that it has some read noise, some dark current, and in general behaves / is real electronic device, and well behaved in sense that that all noise sources are well approximated by appropriate distributions - Gaussian and Poison and response is linear both in time and across the value range), for given total integration time, fewer longer subs will always yield better SNR than more shorter ones, for same conditions (sensor temperature, gain, offset, equal number of calibration frames, target position, atmospheric conditions,....).

But this relation is not a simple one since it depends on many factors. Using longer and longer subs leads to point of diminishing returns. This point usually depends mostly on amount of read noise of sensor. Since CMOS sensors generally have less read noise than CCD sensors - it is more feasible to use shorter exposures with CMOS sensors - you reach point of diminishing returns quicker than with CCD.

Also CMOS sensors encourage one to take shorter exposures, since they usually have lower bit ADC (12bit / 14bit vs 16bit ADC of CCDs).

There are some benefits of using shorter exposures:

1. Less data is discarded in case of poor guiding / wind / whatever ruins a sub

2. Advanced stacking algorithms work better when there is larger number of subs

3. For a given amount of time to shoot darks - more dark subs can be collected - less noise injected back during calibration

There are additional issues with using shorter exposures besides somewhat lower SNR:

1. Much more data to store and process

2. Large number of subs requires increased computational precision to avoid quantization / rounding induced noise / error.

 

As for gain and offset settings, it is never good idea to have offset set to 0. Stacking works predictably and does its job when noise distribution is of particular form/shape. If you set offset low you are running the risk of "clipping histogram to the left" and thus altering distribution of the noise samples. Ideally you would like to set your offset so that bias frames have really nice bell shape with a bit of room to spare on the left side of histogram.

Gain really depends on ADC bit count and full well capacity.

For example I have ASI178 that has 14bit ADC and lowest gain setting has ADU value below 1 - on this camera, I'm happy to use any gain setting. Other CMOS camera that I own is ASI1600MMC, it has 12bit ADC and unity gain is at 139. Once you start setting the gain lower than unity (ADU larger then 1) you start introducing quantization noise into the mix. Now problem with this kind of noise is that it is not well behaved noise - it does not follow distribution that is suitable for stacking. General rule here is that level of quantization noise should be equal or less than read noise level. So the lowest gain setting that I would use on ASI1600 would be 80 - that is ADU of 2 - loosing 1 bit of precision and in the range of 2e read noise on this setting.

Low gain and high ADU on cameras with 12Bit ADC is ok if there is plenty of signal per exposure so that noise term is not dominating SNR in single exposure - this is useful for daytime shooting / moon / sun. For deep sky faint things I would recommend either unit gain (to avoid quantization noise and use max of 12bit range), or perhaps go lower in gain if longer subs are required for some reason but only to a point where quantization noise is still comparable to read noise.

[UKJay1971]

Thanks for a detailed and easy to follow explanation Viaiv.

 

 

[Gina]

Having done many tests with the ASI1600MM-Cool I have found an exposure time of around a minute to be optimum with gains up to the maximum of 600 (60dB) for faint DSOs.  This is very different from a CCD of which I have had several years experience and which definitely would best with long exposures.  The short exposures definitely produce an enormous amount of data and this requires a very fast computer and 32bit processing.  I bought a fast gaming desktop PC running PixInsight under Linux Mint with SSD storage for this purpose. 

But the advantages, as mentioned above, of being able to capture data when there are passing clouds and not needing guiding I find makes quite a difference.  This camera is also twice the resolution of the CCD I was using (Atik 460EX mono) and overall I find it better than the ubiquitous Atik cameras.  You need to completely rethink your exposure regime if you've been used to CCD sensors.

[UKJay1971]

I already use an IMX178 OSC so am familiar with the CMOS ethos but I am moving into mono / filters / NB for the first time and so looking for guidance.

[vlaiv]

In this case, I think a good place to start would be:

Unity gain for that sensor, offset that does not clip histogram (on ZWO ASI 1600 I use 139 (13.9db) gain and 25 offset, not sure how QHY drivers mark these values), 1 minute exposures. Don't use bias frames, use a lot of darks (leave cam somewhere where temp is decent during the day and get a bunch of darks, like 64-128 range, even more if you have disk space / time). I tend to get x256 / x256 - flats / flat darks (no bias here either, I just use same exposure with flatbox and scope covered - I think it is somewhere around 3ms on my TS80).

If you already used 178 chip, and are used to it - just divide exposure you used with 4 (12bit ADC vs 14bit ADC  -2 bit difference = x4). Mind you, with 178 you probably used gain that is below unity, but this is offset by greater sensitivity of mono cam. This is for Lum, for RGB I would probably double exposure (but it's really not needed, you can keep it the same). For narrow band, you will likely want to go with longer exposures, for this I would use gain that gives ADU of 2, and something like 4-5 minutes of exposure for single sub. With these settings number of hot pixels (saturated ones) will likely increase, so you should probably dither when guiding.

For processing, just skip DSS, I can't give more advice on a good alternative since I use ImageJ for calibration / stacking (no out of the box workflow, and couple of custom written plugins, it's really not astro processing software).

HTH