# Can read noise make getting a 16-bit CCD camera pointless

8 posts / 0 new
CrossoverManiac
Can read noise make getting a 16-bit CCD camera pointless

Correct me if I'm wrong but the bit depth (A/D conversion) how much the CCD camera electronics can differienate between how full the well is based on the equation 2n-1 (12-bit=4,095, 14-bit=16,383, 16-bit=65,535).

If the full well capacity is 100,000 e-, then a 16-bit camera should have a Gain of 1.53 e/ADU.  But practically, this level of precision isn't possible if the dynamic range doesn't match it.  If the read noise is 10 -e, then the dynamic range is full well ÷ read noise.  For this hypothetical CCD camera, the dynamic range is 10,000 or 13.3 stops (equivalent to 13.3 bits).  the level of precision in the measurement is less that 1/6th what it's bit-depth is capable of.  IOW: a 16-bit camera with a dynamic range of 10,000 is no more precise than a 14-bit camera with the same dynamic range.  However, a 12-bit camera would have less precision than the bit-depth would not be able to parcel out the output of the CCD camera as precise as the dynamic range would allow.  I asked this because I was comparing the dynamic range the ASI183MM (12-bit) with 16-bit CCD cameras.

If there is some other factor that makes getting the 16-bit CCD camera advantageous?  Is this example wrong?  Let me know.

HQA
dynamic range

Hi Tim,

You have the argument mostly correct - the dynamic range is the key parameter for determining the necessary bit depth.  There are some other constraints, however.  One is the "granulation"; basically whether you resolve the noise properly.  The usual guideline is at least 2 bits of resolution in the noise, so if your dynamic range says 10,000, you need to resolve 20-40K values.  So a dynamic range of 10K needs ~15bits of resolution.

That said, your example is pretty close to the KAF1603 CCD sensor in dynamic range.  For the CMOS 183 camera, it has 15Ke- full well and 2e- readnoise, and so still needs about 15bits to do true justice, yet the ADC is 12bits.  AAVSOnet gets around that by (a) binning 2x2, yielding 2 more bits of ADC range, and taking several frames and stacking them to get more bits of range.  It is not a perfect solution, but makes a very inexpensive sensor nearly noise-limited.

The upcoming cameras with the IMX455 CMOS sensor will be nearly perfect, as they have larger well depth, low readnoise, and 16-bit ADC, at a very attractive price.  The other non-CCD choice is a camera using sCMOS sensors, that have larger full well, low readnoise, pseudo-16bit digitizing, but much higher cost.

There are many considerations when comparing sensors.  For example, the CMOS sensors can be read out extremely fast, decreasing the dead-time between exposures and making bright-star photometry possible.  At the same time, the CMOS sensors typically have higher dark current and most have amplifier glow that limits the long exposures.  Lack of a physical shutter means dark frames are more difficult to acquire.  CCD cameras tend to have higher read noise and higher cost for the same physical size, and some have residual bulk image problems that are a pain to correct.  One camera vendor may specialize in CMOS; another may have a great reputation for service.  Dynamic range is only one (but very important) consideration in your camera selection.

Arne

CrossoverManiac

Thanks for answering my question.  The reason why I'm asking is that I have an ASI183MM-Pro, a Starshoot G4 CCD camera (16-bit A/D), and I'm thinking about getting a Canon camera APS-C size sensor since it would be cheaper, have a bigger chip, controlled wirelessly, and run on an internal battery.  I don't have the UBVRI filters (yet) but and I was about to buy some before having second thoughts about it.

All of this is for finding exoplanets.

I orginally was going to use my Canon 6D and prime focus lens but, after stopping the lens down remove abberations like coma, I don't have much aperture to work with and, when I tried to take images of a standard field for transformation, the images were so faint they were only a few standard stars that were useful even after stacking twenty minutes of frames (that and the Canon 6D has some rather difficult vignetting issues).

I'm seriously considering using my astrophotography rig specifically for photometry.  I own a 6-inch f/5 Newtonian-Maksutov so I believe I have enough aperture now.  But now the question is which camera to use.

I'm getting mixed signals at AAVSO or, more specifically, strong diametrically opposed opinions.  On one hand, I hear that nothing short of a monochrome CCD camera with a UBVRI filter wheel will do while AAVSO has courses for DSRL photometry and now there's a 40 post thread on developing astronomy CMOS cameras for photometry.

Maybe I can be more specific - between CCD, astronomy CMOS, and DSLR (also CMOS not specifically made for astronomy), which would be the best one to use to observe and find new exoplanets?

HQA
best camera

Hi Tim,

Please excuse me, but I'm going to be flippant:  what you are asking is like asking whether a Mazda or a Lexus is the best car to drive to the store.  There are lots of things to consider when making camera choices.

You've started the right way:  choosing a research topic (finding exoplanets).  My first question is whether you are actually talking about surveying the sky to find new candidates, or whether you are going to use a candidate list from, say, TESS, and do time-series photometry to discern whether this is a true transit or some other beast (like a contaminating eclipsing binary)?  Those are two completely different topics, and require two different approaches.  I'll be honest and say that the possibility of discovering new exoplanets that are not seen by one of the existing surveys (KELT, TESS, Kepler, whatever) is extremely rare, and is something I'd not consider doing personally (even if I was younger!).  If you want to try, then I can give you some references on how other groups have approached the project.  Doing follow-up work is less exciting, perhaps, but doable with your setup.  There are lots of resources there too, including the AAVSO Exoplanet Section.  Let me know the answer to that, and then a proper reply can be made.

Once the project is defined, then you can decide what equipment you need to get results.  A 6-inch telescope is doable (several of the original exoplanet searches, such as XO, TrES, etc. used similar apertures), and any of those cameras are capable of results (I've seen transit light curves from DSLRs).  However, I'm not an exoplanet expert, and you really need to talk to one, as technique is even more important than equipment.

Arne

CrossoverManiac
Hunting for exoplanets is

Hunting for exoplanets is probably too much of an undertaking for right now.  So starting off with confirming exoplanet finds by TESS would probably be a good starting point.

HQA

Hi Tim,

Good!  Follow-up is straightforward and a good place to start, rather than trying your own detection survey.  Surveys are complex computer tasks, requiring searching through millions of light curves to find a handful of transits.

http://brucegary.net/book_EOA/EOA.pdf

That will give you great guidance as to the techniques required for this high-precision photometry.  Then I'd get involved with the AAVSO Exoplanet Section:

https://www.aavso.org/exoplanet-section

where some good follow-up projects can be found.  Right now, the main follow-up-networks (FUN) are for TESS and for KELT.  They have lots of targets and need many more volunteers to obtain light curves to eliminate imposters.  Your equipment should be adequate to participate in either network.

Best of luck on this endeavor!

Arne

CrossoverManiac
If you don't mind me picking

If you don't mind me picking your brain a bit.  I have a few additional questions to clarify things.

Dynamic range being a function of the read noise:  From what I understand about read noise, it is a fixed amount for a given sub.  So, every pixel of every sub will have x-number of electrons added to it from the read noise.  But the way its relationship with dynamic range, it acts the smallest unit on a ruler.  The accuracy of a 12-inch ruler is 1/16th inch while a pair of calipers can measure in much smaller units even down to 1/1000th of an inch.  But then some of the other literature like the SBIG manuals for their vintage systems said that the longer the sub is, the less read noise is a factor.  It seems like like the smaller notch on a ruler but rather the end of the ruler after it's been worn down and anything smaller than that worn down part can't measured, which brings me to my question.

Offset and that additional 2-bits added to the calculated dynamic range: My first CCD camera had an offset value.  IIRC, when I set the offset value to its maximum, I didn't receive any signal at all.  So, would it be accurate to say the offset represent some sort of high-pass filter and that the CCD pixel must receive a minimum amount of light in order to have a voltage response or is this done electronically with the PC just subtracting the voltage that is there but doesn't reach that minimum threshold.  Also, would that be related to the additional 2-bits added to a system in addition to the calculated dynamic range?

HQA
CCD offset

Hi Tim,

There are two ways to add an offset:  a voltage inserted before the analog/digital conversion takes place, and a numerical value added afterwards.  The inserted voltage method is done so that the analog/digital converter only sees voltage values from, say, 0 to 5 volts, and not some negative voltage from the signal chain for which it can't convert.  That is why bias frames for most new cameras have values like 800 or 1000 counts - just being safe and making sure the conversion voltage is in the right range.  For cameras like the older SBIG units, the driver took a raw converted pixel number and added a 100 constant value to it.  This value has no noise, but impacts the dynamic range slightly.  As you saw, setting a offset to, say, 10000, means pixels would range from 10000 to 65535, a more obvious truncation.

The 2 bits I was refering to might be best described with an example.  If your read noise is 16 electrons, you might want to set the gain to 4 electrons per count unit so that you properly sample the noise.

If your readnoise is, say, 16 electrons, then that noise is a major part of your signal for a faint source, like a 30-electron sky value in a short exposure.  For a star with 10000 electrons, the 16 electron readnoise is small compared to the 100 electron Poisson noise from the star.  That is why using longer exposures on a star so that it is closer to the full well is best, so that the measured noise is coming from the star and not impacted by noise in the sensor.

Arne