Why all this? Marketing and advertisement sayings “our CCD cameras are low-noise” is nice, but what that exactly means? Some parameters of a new CCD camera are more important then others for a potential camera buyer (based on everyone’s personal preference), but in the end, all of us would like to have a CCD camera that works perfectly well. By this I mean that the camera gives consistent and repeatable results and that bias frames contains pure Gaussian (completely random) source of noise. This means that stacking more single frames actually lowers the noise and increases the SNR (Signal to Noise Ratio) of the resulting image. That’s what all of us wants (apart from the World peace).
The purpose of this camera testing and comparison is to:
NOTE: even a high readout noise camera can produce nice results based that there is not any pattern noise (or other serious defect) in the image. Especially large format cameras (36x24mm or bigger) allows for use of a large aperture telescope (to capture more light) and let’s you down-size the final image resolution to some acceptable value (to hide the remaining noise in the low SNR background part of your image). I do myself own a CCD which has one of the worst CCD chips inside, but I use it for LRGB imaging only where the readout noise is not important.
This test focuses on camera readout noise which is the most important parameter for narrow band imaging (which is my favorite discipline), where the narrow band filter limits the sky background flux to some marginal (zero) value (it passes wanted signal, e.g. H-alpha and cuts off all the rest light spectrum). In this case the readout noise plays significant role in how clean the final image is. The narrower the filter (while the best is to have also highest possible filter transmittance along to the narrow bandpass) the higher demand on low RN (readout noise) camera – the more important this parameter becomes. On a dark sky this could mean that a difference in image cleanliness of a camera (A) having RN of 4e- and camera (B) having RN of 10e- by a factor of 6x (you would need to image 6 times longer with camera (B) in order to have the same clean image as from camera (A) – comparing pixel to pixel signal). The reality is much more complicated than this simplified case as there’s something like e.g. optimal subexposition duration based on the sky flux, filter and camera used. To simplify things, again, with camera (B) would be optimal to shoot narrow band images of a subexposition duration like 90 minutes (in order to get the most out of the camera) while with camera (A) you may end up with only 15-20 minutes per subexposition.
The last, but not least, purpose is to put more pressure on camera manufacturers in order to increase QC (quality control) and produce perfect-most cameras within the limits of every particular CCD device.
Method of Testing
To compare cameras against each other I have extended the “simplified” (but very precise) method closely described by Craig Stark in his document Signal to Noise Part 3: Measuring Your Camera. In brief, I use a set of 10 bias frames to create a master frame that is used for subtraction from single biases from which I compute StdDev value. Therefore I compute 10 numbers which are averaged so as the test result is precise. Then I use 5 flat field calibration frames (which I calibrate with the master bias) for gain computation. Because all ABG CCD cameras on the market today are linear at around 25000 to 30000 ADU levels an ordinary (correct) flat field frame is the only thing I need for this. Again, to increase the precision (and to check the consistency of results) I compute 4 values of gain from 5 flat field frames. Then the RN value is simply computed as gain multiplied by previously computed StdDev value. Last step is to compute TSN (Total System Noise) that is affected not only by readout noise (must be always higher) but also by thermal noise (and other sources). The TSN characterizes all noise sources in a raw bias frame together. If it is much larger than RN then there’s something wrong. Notice that with lower CCD temperature the TSN on the same camera is lowered. To make the calculations easy and straightforward I wrote a simple script in Octave (free version of Matlab) that runs on my Linux box.
Camera Test Results
Sorted by CCD detector and Readout Noise value in ascending order.
Last update: 26th August 2013
DOWNLOAD the results table!
NOTE on FWC – Full Well Capacity: the table might be a bit misleading on this as the number computed (65535 levels of an 16-bit A/D converter multiplied by computed gain of particular camera) may be effectively higher (or lower) than what’s real (true) full well/pixel capacity (saturation signal) that is a „system“ property of every particular CCD device. This real number is usually specified by SONY/Kodak for every chip. e.g. Kodak KAF-8300 has Full-Well 25500 e- based on Kodak’s datasheet. Therefore „optimal“ value of gain in order to get most out of the chip with 16bit ADC would be 25500 / 65535 = 0.389 e- / ADU. If the camera’s gain is higher than this „optimal“ number then the stars become sooner saturated (burned out white). On the other hand, if the camera’s gain is lower than this value then the dynamics is sacrificed. Neither case is wanted
Interpretation of Results
The Observation column contain my personal opinion on the obtained results. If I get consistent results in the computation log and the TSN is only a hair above RN (provided that the CCD temperature is below zero degrees of Celsius) then the camera works well = doesn’t have any problem. Whether RN meets the manufacturer’s specification or not is the other side of story. If the TSN is lower than RN (happened with some cameras) then any calibration with bias frame effectively introduces more noise into the image then it removes – this is bad. Also, if the histogram shape is anyhow different from a standard Gaussian curve then there’s a suspect on pattern noise (that may come from e.g. power source or any other electronical device in the surroundings of the camera may interfere). In such a case a look at the master bias frame could reveal more secrets on how bad the issue really is.
May you find that there are many cameras listed as having issues – the reason is simple – that’s exactly why I started to dig into noise measurement – because my first and second CCD camera I ever purchased had some kind of problem. Also, people who are willing to provide their calibration data for my measurement do so only in case they expect (and need to confirm) that the problem exists. Others probably rather do not want to hear the numbers as they are affraid that there could be anything wrong while they are happy with the camera they have. Though I would very much prefer the other approach: to test brand new cameras that you have just received. In case of a problem you could claim it back to manufacturer or dealer to resolve the problem. This may save you money and frustration.
How to capture calibration data needed for analysis?
If you find this article interesting and you own any kind of cooled astro CCD camera, please help with this research by providing your calibration frames. I’d be happy to do a test run for you. The only inconvenience that persist, is that you would have to upload the large archive file to some website and provide a download link to me – best contact is through the comments section below as comments needs to be approved prior being published.
To measure the readout noise of your CCD camera in an uniform manner (so as the results are comparable against each other) please provide a set of 10 single bias frames and 5 flat field frames in native FITs format.
- capture 10 consecutive single bias frames with cooling on, binned 1×1. If you have set-point cooling camera cool to at least -10 degrees Celsius for Sony chips and -20 for Kodak chips (the colder the better). If you can’t control the cooling, turn it on and wait for up to 30 minutes to stabilise. Make sure you shoot the frames in a dark room no matter if your camera has mechanical shutter or not in order to prevent any possible light to “sneak in”. Nosepiece is not light-proof unless you cover it with aluminum foil and fix with e.g. a rubber band.
Flat field frames:
- capture 5 consecutive flat field calibration frames (again, with cooling on) using the same binning (i.e. 1×1). A very good tool for this is an EL-panel or a light-box. Use sheets of white paper (or RED/H-alpha filter, if filter wheel is in place) to dim the light if needed. It doesn’t matter if you shoot real flats with optics attached to the camera or just plain flats without any optics involved, when the CCD is just lying on the surface of EL-panel. If you have a mechanical shutter be sure to keep the exposition of a single flat frame a bit longer (e.g. couple of seconds). Aim for around 25000 ADU levels (as almost every ABG camera stays in linear range at this level).
NOTE: you may use real calibration data that you already have on your hard drive. Some vignetting or dust motes on flat fields doesn’t matter as these frames will be center-cropped.