| article: choosing and matching a CCD or CMOS camera to a telescope - a buyer's guide |
Introduction
The effective performance of any CCD camera is very much influenced by the telescope it is coupled to.
On the other hand there are features like the type of read out electronics or the feature of Anti-Blooming Gates (ABG) which are independent of the telescope.
The following article tries to explain the influence of the practical use of all these features.
For just comparing the main features of common CCD cameras please click here.
We will have a look at the following topics:
- Pixel Array (Chip Size, Pixel Size, Pixel Count)
- Signal to Noise (Quantum Efficiency, Full Well Capacity, Read Out Noise, Dark Current)
- Anti-Blooming Gates (ABG) vs. Non Anti-Blooming Gates (NABG)
- Chip Type influencing Quantum Efficiency and Shutter
Pixel Array (Chip Size, Pixel Size, Pixel Count)
The bigger the size of the chip, the bigger is the Field of View (FOV) with a given focal length.
In general a big size of view is desirable.
Sky objects can be found more easily and the bigger show pieces of the sky will fit on the chip.
It is also good to have some stars around a galaxy or a nebula giving a more expressive photo.
But for purely scientific use like astrometry this is not important at all.
And should your major goal might be imaging of the planets you need less than a minute of arc as FOV.
The smaller a pixel on the chip, the higher the resolution of the final image will be - again with a given focal length.
Small pixel CCD cameras allow high resolution images with very moderate focal length.
Hence the optical tube assemblies will weigh less and this means that the mount must not be that massive.
On the other hand a small pixel can collect lesser electrons produced by the starlight.
And this means the Full Well Capacity is decreased.
The size of the single pixel is the collecting area for photons.
Smaller pixels can hence collect lesser photons in a given time of exposure.
See Signal to Noise later for more detail.
The bigger the chip size and the smaller the pixels the bigger will be the total count of pixels.
More pixels give more resolution to the image which is fine.
But the bigger the pixel count the bigger the size of the image files you will have to store on your computer.
In example, my HX916 has 1.34 millions of them.
A FITS file is then 2.6 MB big.
And for efficient image reduction you need a lots of them for only one final image.
If the pixel count in x and y (1300 x 1030 for the HX916) is bigger than what your computer screen is able to display then it will be hard to do the most important job of image reduction.
Please do not underestimate that!
Most of the brillant pictures on the Web nowadays are not bigger than 800 x 600 anyway.
And they can show amazing detail if done properly.
For print reproduction there is a need for more pixels of course.
To compare the most common CCD and CMOS cameras with various telescopes from 50 mm to 4000 mm of focal length please click here
Signal to Noise Ratio SNR (Quantum Efficiency, Full Well Capacity, Readout Noise, Dark Current)
The Signal to Noise Ratio (SNR) can be defined as the light from the object of interest compared to the noise (light and noise from other sources). What are the main sources of noise?
- Readout Noise
- Thermal Noise (Dark Current)
- Quantization Noise
- Sensivity Variation Noise (Spectral Noise)
- Photon Noise
The light of i.e. a star is reaching the sensitive area of a CCD camera.
To have a quantum efficiency (QE) of 100% each of the incoming photons should be transferred to an electron.
Unfortunatley this is not possible.
But while chemical film can only register about 5% of these photons good amateur CCDs can register 50% and the best ones now up to 90%.
That clearly means we can catch more signal level per time and hence the QE is related to SNR.
But amateurs do love to spend time with their hobby and so the QE must not be overestimated.
There are two other attributes which are limiting the CCD camera much more.
The readout noise is one of them.
Together with the Full Well Capacity.
The registered photons in a CCDs pixel are stored as electrons.
The number of electrons which one of these pixels can store is called the Full Well Capacity.
If we want to transfer the electrons to the computer (or storage device in case of a digital daylight camera) we must measure the number of electrons in each pixel.
After measuring this analog value it must be digitized to enable a computer to read it.
This digitizing is also called quantization and is introducing the quantization noise.
With a 16 bit AD converter this noise is so small that we do not have to consider it here.
With 8 bit it is horrible and with 12 bit it is quite limiting so just avoid these cameras for high quality imaging purposes.
Now, the higher the readout noise and the lower the Full Well Capacity the lower the maximum achievable SNR will be!
The Full Well Capacity is limiting the maximum exposure time just because the well is full.
So you have to stop collecting more signal.
Now the readout noise will be added to the signal.
And this ratio is limiting the SNR of a single exposure with a given CCD.
The HX916 has a very low readout noise of only 12 electrons.
On the other hand it has a low Full Well Capacity of 30,000 electrons.
In the absence of any other noise the maximum SNR would hence be
SNR(max, 1 frame) = 30,000 / 12 = 2500
Unfortunately there is the ever present thermal noise.
Thermal noise is introduced by the fact that a single well on the chip will spontaneously prodcue electrons caused by temperature movement of the silicon.
Because they are stored in the well too, the well will be filled even faster!
Or in other words we again have to stop exposure earlier and are not able to collect more signal.
The typical thermal noise of the HX916 is very low and about 0.1 electrons each second when the camera is cooled to 30 K below ambient temperature.
Cooling is helping to reduce the thermal noise (hence the name!) very much.
An exposure time of 10 minutes equals 6,000 seconds or a total of 600 electrons of thermal noise.
(If we are having additive noise, this noise must be added squared and overall squareroot must be taken.)
This is reducing the SNR to:
SNR(1 frame) = 30,000 / squareroot( (600*600 + 12*12) ) = 50
Obviously that is much lower!
The thermal noise can be reduced (but not be eliminated!) by subtracting the so called thermal frame.
The thermal frame is an exposure as long as the signal collecting exposure just with the shutter closed (to be precise: the bias frame must be subtracted to call it a thermal frame. Otherwise it is a dark frame).
It is therfore collecting only electrons from the dark current.
This process is a statistical one and variing from exposure to exposure.
Therefore we will be left with a uncertainty of the squareroot of thermal noise (according to the laws of statistics).
SNR(1 frame, thermal frame corrected) = 30,000 / ( squareroot(600) + 12 ) = 822
OK, much better!
Now the emission of photons in our star (signal) is a statistical process too.
Again we have an uncertainty - this time it is the square root of the signal and that is the photon noise.
The sensivity variation noise is the noise which is added by the nonlinear reaction to photons from pixel to pixel.
One can also say the QE of each pixel is slightly different.
In a modern CCD frame this is below 5% for an average pixel.
But there is a chance that there are several pixels which will be full of electrons very fast (hot pixels) or very slow (cold pixels).
If you are not working in science like photometry this usually can be ignored.
If it must be corrected it can be done by applying a flat field.
A final thought about the SNR:
Normally a single frame is not enough to give a good enough SNR.
That is because the faint outer parts of nebulae or galaxies are so dim and blend to the sky background.
The very best CCD photos are created by combining dozends - sometimes hundreds - of single frames.
In the absence of readout noise the SNR is increased by the square root of the number of exposures.
With average to very good CCD cameras we can roughly estimate
- 4 exposures is up to 2 times better
- 9 exposures is up 3 times better
- 16 exposures is up to 4 times better
- 25 exposures is up 5 times better
- and so on, depending on the readout noise of a certain camera...
It only depends on you and your patience!
Click here for more details about image combining and readout noise.
Anti-Blooming Gates (ABG) vs. Non Anti-Blooming Gates (NABG)
Blooming is visible in the final image as a (vertical) streak away from the stars.
This is not looking very nice and additionally is ruining the scientific value of parts of that picture.
Without ABGs the maximum exposure time is therefore restricted when a brighter star is in the FOV.
These stars can emitt much more photons than the nearby nebula or galaxies of interest and hence saturate the pixel more easy.
As we have seen restricting the time of exposure is harmful to the SNR.
In older literature you can read: The use of ABG is reducing the quantum efficiency of the chip as well as the linearity.
The good news is that with some mordern chips from SONY and KODAK this is (almost) not true anymore!
The ABG is like a drain pipe drowning away the electrons which are not fitting anymore into the pixel - or in other words exceeding the Full Well Capacity.
This drain pipe takes space from the sensitive area of the pixel.
Therefore the QE of the pixel is reduced.
The solution of the mentioned chips of SONY and KODAK is to place a microlens in front of the chip to transfer the light to the sensitive area.
Even if the sensivity might still be a bit decreased the fact that you can expose much longer (in the presence of bright stars) compensates for that.
The SONY chips for example can be exposed to 800x of the *normal* Full Well Capacity.
The lack of linearity is another topic.
For photometry purposes it is important that the pixel is linear from 0 to 100% of Full Well.
0 in the raw image cannot be achieved anyway (dark current, bias offset), but with proprer image reduction this can be corrected.
For a reason I did not find out yet the new SONY chips shall be linear to 90% of Full Well.
Hence they are even suitable for photometry and slightly only worse than their NABG competitors.
For *pretty pictures* this is not important at all.
For the ease of use and longer possible exposure times I prefer the microlens ABG chips.
Chip Type influencing Quantum Efficiency and Shutter
From ordinary film cameras we know that these instruments need a shutter to control exposure time.
Same is true for CCD cameras.
But while most of the chip types have to rely on mechanical shutters there is one exception to that rule: the interline frame transfer chip.
The interline frames transfer is controlling the exposure time just by electronics.
The trick is that for every sensitive line of pixels there is an unsensitive beneath to where the electrons are shifted when the exposure is ending.
Like for anti blooming gate this line is subtracting sensitive area.
But again with the use of micro lenses the loss of light is very low.
While the mechanical shutter CCD cameras can rarely go below 0.05 seconds, an electronic shutter can go to as low as appr. 0.005 seconds.
This is quite useful when imaging bright sources like the moon.
But more important is that mechanical parts are quite expensive compared to electronic parts.
So here it's possible to save money.
|
|
|