Astronomy 480 Characterizing a CCD

Objectives In this exercise, you will characterize the following properties of either the A) Kodak KAF-1603E/ME that is part of the ST-8XEA camera or the B) KAF-3200E that is part of the ST-10XME camera: the detector gain and readnoise the detector linearity and compare your results with the manufacturer specifications Critical Set-Up Instructions for both the ST-8XEA and ST-10XME: Report For this exercise, you will be writing up a brief report of your results. You do not need to include what you did during the observations other than to note that you followed these instructions, but you should include the procedure you followed in your analysis, all calculations, the results, and some interpretation of those results. Also include any relevant tables and plots, and answers to all questions embedded in these instructions, indicated by a . Exercise is based on Howell (2006), Section 4.3 Introduction The first charge coupled devices (CCDs) were produced just 30 years ago at Bell Labs in New Jersey. They have become the detector of choice for most astronomical observations in the UV and optical spectral regions, and are becoming increasingly popular for X-ray observations. There are a number of reasons for this popularity. The best CCDs have high quantum effciency ( ~ 90%), low readnoise (r ~ 2-3 e ), relatively uniform response, large dynamic range (D 100 dB) and excellent linearity.

The chip needs to cool down for at least 10 minutes before taking data. Your instructor should have taken care of this for you. When taking bias frames, check the counts using the display tab on the CCD control panel to get the mean counts. The shutter may not always close completely. When taking flats, tissue should be placed over the tube so that a diffuse light source is hitting the CCD. The pixels should not be binned, so make sure the binning number is 1 x 1.

Virtually all large observatories maintain websites describing the characteristics of their CCDs, and it is important to know which detector is right for your observing program. It may be wise to measure and test these characteristics during the course of your observing run, although we usually trust the telescope technicians at each observatory to know the instruments and detectors. I. Setting-Up the Camera and Computer Software Have the manual for operating Equinox and the SBIG camera handy, if needed. Get your computer and camera ready to go. Play around a bit with the instrument and software to get familiar with them.

[Exercise adapted partially from from http://www.physics.rutgers.edu/grad/629/lab2.pdf] characterize_CCD_2009.doc 1

4/10/09

You can set the number of automatic images that are taken. All of your saved exposures should be registered in a log book. This is a good habit to get into; something that is absolutely required when at a research telescope. II. Detector Gain and Readnoise Theory Every CCD relies on an amplifier to measure the electrons contained within the various detector pixels; readnoise is defined as the mean error contributed to a pixel by the amplifier. The amplifier signal is in turn digitized by an Analog-to-Digital (A/D) converter, and the gain is number of electrons per data unit produced in the conversion (i.e., e per A/D conversion units, or ADUs). The lower the gain, the smaller the signal entering the A/D converter for a given number of electrons in the pixel (i.e., a low gain refers to a higher a value of e /ADU).

We will be following Howell to determine the gain and the readnoise for our CCD. It is possible to calculate the gain and readnoise by measuring the mean and standard deviations of pixel values in two pairs of images that have much different signal levels: i.e., a pair of bias images versus a pair of flatfield images. In the bias images, the measured variation in the pixel values must arise solely from the readnoise of the amplifier for the simple reason that the image exposure time is zero seconds the pixels on the CCD are just read out. The standard deviation of this variation is the readnoise. Given a bias frame, it is a simple matter to calculate the readnoise in ADUs (Eqn. 1), but we are interested in its value in electrons. We will be taking 2 bias frames to form a difference frame; youll find out why shortly.

" read =

Read Noise (in ADUs) Gain

(1)

To convert from a readnoise measured in ADUs to one in electrons, we must also measure the amplifier gain. To measure the gain, we use exposures with much higher signal levels than the bias images, frequently dome or twilight or dark sky flatfield exposures. Once again, a pair of exposures is used to form a difference image, and the standard deviation is measured. In this case, though, the measured standard deviation contains a component from the pixel readnoise, and a random (Poisson) component due to the arrival time of the photons from the dome or sky. If the flatfield signal is strong enough, then the Poisson noise will be much greater than the read noise, and the standard deviation will be approximately:

" flat =

F # Gain (in ADUs) Gain

(2)

The signal coming from the amplifier is digitized by an analog-to-digital (A-to-D) converter and the gain is the number of electrons per data unit produced by the converter. This is inversely related to the gain of the amplifier (the terminology is a bit unfortunate). A lower gain for the amplifier means a smaller signal entering the A-to-D converter for a given number of electrons in the pixel and, thus, a large number of electrons per data unit. Conversely, a higher gain means fewer electrons per data unit.

Howell, Steve B., Handbook of CCD Astronomy, 2000 (Cambridge University Press), 2nd Ed., pp. 54, 72-73

The larger the gain, the larger the measured pixel-to-pixel deviation. The square root comes in as the Poisson noise is simply the square root of the signal.

Procedure for obtaining the frames and images for read noise and gain calculations
Take a series of bias frames and at least 2 flatfield images having the same exposure time. BIAS: With the tube to the CCD totally covered, and the save all option on, take the bias frames. Even though the shutter to the CCD never opens, you should have the cap on the camera. FLATS: When taking your flats, be sure that you have an evenly lit CCD; use a diffusing medium, such as a tissue, to spread the light around. Getting decent flatfield images is going to be a challenge (it is always a challenge, actually), since it is neither twilight nor are we in a dome. What we want is a diffuse source of light that will evenly expose the CCD and reveal any pixel-to-pixel variances. Now, you may want to uncheck the save file box while we experiment with different exposure times. What we want from our flatfield is enough counts so that we are in the shot noise or Poisson noise regime, but not so many that we may introduce non-linearity concerns. Work with the exposure times until you get a maximum somewhere around 30,000 for the ST-8XEA and 20,000 for the ST-10XME. Moving the cursor over the image under the display tab will give you a reading. Once youve determined the exposure time, take a couple of flatfield images, making sure you save them to disk! Calculations for gain and readnoise To find the gain: (F1 + F2 ) " (B1 + B2 ) , Gain = 2 2 # (F1 "F2 ) " # (B1 "B 2 )

(3)

where the mean pixel values within each image are designated F1, F2 , and B1, B2 , and the difference images by F1 " F2 , and B1 " B2 . Use the IRAF task imarith to subtract the frames (see instructions that follow after Eqn. 4). You can select any 2 bias and flatfield frames. Once we have the gain determined, we can calculate ! read noise: the ! Gain " # ( B1 $B 2 ) (4) Read Noise = 2 Give the readnoise in both ADUs and electrons. In calculating the means and standard deviations, be sure to use the same region of the image (say, a subraster of 100 x 100 pixels) for both the biases and flatfields. Try to avoid cosmetic defects such as bad columns, and regions which contain hot pixels or cosmic rays. You should use imstat in IRAF (where you could also specify a subraster). The following format within IRAF shows you how to specify a subraster or subregion on the chip (leave off the .fits and
4/10/09 characterize_CCD_2009.doc 3

substitute the pixel ranges for the subraster in square brackets, no spaces within braces): ecl> imstat filename[100:200, 100:200] Try different regions of the chip and the entire chip. Try different flatfields, if necessary, to get a feel for what is going on with the calculations. If your value is much different from the manufacturers, then check with your instructor to find out if youve done something wrong. Be sure to comment on your different tests and your results in your report. Introduction to imarith One of the easiest IRAF tasks you will ever use, yet one of the most important: imarith. Do phelp imarith and review the options for the parameters. Page down and review the rest of the help. You will note that imarith can become quite powerful especially note where you can operate on a subsection of the image. While you are at it, take a look at all of the tasks available within the imutil package. See any familiar ones?
IRAF PACKAGE = imutil TASK = imarith operand1= op = operand2= result = (title = (divzero= (hparams= (pixtype= (calctyp= (verbose= (noact = (mode = Image Reduction and Analysis Facility

Operand image or numerical constant + Operator Operand image or numerical constant Resultant image ) Title for resultant image 0.) Replacement value for division by zero ) List of header parameters ) Pixel type for resultant image ) Calculation data type no) Print operations? no) Print operations without performing them? ql)

IRAF can find the gain for you. Load the packages noao and obsutil by typing within IRAF or pyraf: noao.obsutil Use phelp findgain to figure out what this task does, and then use findgain to find the gain of this chip. Compare these results to your calculations. Note on statistics and IRAF: IRAF has statistical tests and fitting programs built in, based mostly on algorithms from Bevington & Robinson Data Reduction and error Analysis for the Physical Sciences. For the fit and the goodness of fit coefficients, see polyfit (phelp polyfit).

III. Linearity of the CCD Theory One the most appealing features of CCDs as astronomical detectors is their linearity, the degree to which the output signal is proportional to the incoming photons received by the detector. But, given enough photons, even the best chip in the world will eventually become non-linear and approach saturation. Our goal here is to find the level (the ADU limit) where the chip starts to become nonlinear, and then follow up that information with keeping within the linear regime with our observations. Each of our CCDs has a16-bit A/D converter. Theoretically, A/D saturation would at an output value of 216 = 65,536 ADUs (65,535 in reality, as count starts at 0). ST-8XEA: With a gain of 2.5 e /ADU, this would be 163,840 input photons (assuming a QE of 100%). However, according to manufacturers specifications, the pixel full-well capacity in 1 x 1 binning is around 100,000 electrons, making the full-well saturation occur at approximately 100,000 / 2.5 e /ADU or around 40,000 ADUs. ST-10XME: Gain is 1.3 e /ADU; ADU saturation in electrons = 65,535 ADUs x 1.3 e /ADU = 85,195 input photons (QE =100%). Full-well capacity is approximately 77,000 e , corresponding to ~59,000 ADUs. Take a look at the manufacturers specifications for these two chips and the information given on our other CCD camera, the ST-9XE, in question 4, page 6. Is there a relationship between the size of each pixel and the full-well capacity? What would you guess would change as far as saturation goes if you used 2 x 2 binning on any of these 3 CCDs? If the image of a star, for example, were to be saturated on the CCD by either of these two reasons, it would be noticeable. We would see a flat-topped star.

Profile of the image of a saturated star

http://arxiv.org/pdf/astro-ph/0502411

What a saturated star may look like on the CCD. The center part is dark because the counts have gone to negative numbers.
http://casu.ast.cam.ac.uk/

4/10/09

characterize_CCD_2009.doc

If we do not know the response of our chip, we may be working in the non-linear part of the CCD and not even know it. The stellar profiles and images may look perfectly normal and still contain totally inaccurate counts. Some CCDs (usually the most expensive ones) are manufactured such that one or the other of the two types of saturation is reached before the chip reaches saturation. If one is unsure about the linear range of a CCD that he or she will be using, then it is always wise to do independent measurements.#

Procedure for obtaining the images for the linearity determination

Increase the exposure time of the flatfield until you get saturation; it is best to have the saturation exposure at least 1 minute in length (90 seconds may be ideal) to get enough data points to fit. Then, expressing this exposure time as t, take a series of flatfield exposures of duration 1.00t,0.90t, 0.80t,0.70t, 0.60t, 0.50t, 0.40t, 0.30t, 0.20t, 0.10t, 0.05t, 0.025t, 0.0125t, 0.00625t, 0.003125t, and 0.0015625t. (The minimum exposure time for the camera 0.12 sec.) Be sure to keep an accurate record of your exposures on the observing log included in this exercise. Data Analysis Using the series of exposures taken above, plot your average counts in ADUs as a function of exposure time in seconds. Use any plotting program you feel comfortable with, and attach that plot to this exercise when turning the assignment in. Be sure to indicate your choice for the linear range of the CCD in ADUs. Questions (answer in narrative form within your report) 01. In 1 x 1 binning, does our detector reach full-well capacity or ADU saturation first? Explain. 02. What do you conclude about the linearity range of the detector? 03. From an observational or scientific standpoint, why is it important that one stay within the linear range of a light detector? 04. We have a different camera that uses a Kodak KAF-0261E CCD having 512 x 512 pixels (20 x 20 microns square) in the array. It has a 16-bit A/D converter, a gain of 1.6 e /ADU, and a full-well capacity of around 150,000 electrons. Using 1 x 1 binning, should we worry more about full-well capacity or about ADU saturation with this chip?

Howell, S., Handbook of CCD Astronomy, 2nd ed, pp. 55 - 59 (Cambridge University Press, 2006)

IV. Comparison with Manufacturers Values Compare your results with the listed quantities from the manufacture. Comment on discrepancies and put together a nice, one-paragraph summary of what you did in this exercise and what you learned.
Model ST-8XME -- Typical Specifications -- CCD Specifications CCD Kodak KAF-1603ME + TI TC-237 Pixel Array 1530 x 1020 pixels, 13.8 x 9.2 mm Pixel Size 9 x 9 microns Full Well Capacity ABG ~50,000 eFull Well Capacity NABG ~100,000 eDark Current 1e-/pixel/sec at 0 C Readout Specifications Shutter Electromechanical Exposure 0.11 to 3600 seconds, 10ms resolution A/D Converter 16 bits A/D Gain 2.3e-/ADU Read Noise 15e- RMS Full Frame Acquisition 3.7 seconds

4/10/09

characterize_CCD_2009.doc

Model ST-10XME -- Typical Specifications -- CCD Specifications

CCD Kodak KAF-3200ME + TI TC-237 Pixel Array 2184 x 1472 pixels; CCD Size 14.9 x 10 mm Total Pixels 3.2 million; Pixel Size 6.8 x 6.8 microns square Full Well Capacity ~77,000 eDark Current 0.5e-/pixel/sec at 0 degrees C. Readout Specifications Shutter Electromechanical Exposure 0.12 to 3600 seconds, 10 ms resolution A/D Converter 16 bits A/D Gain 1.3e-/ADU Read Noise 8.8e RMS Full Frame Download ~8.7 seconds

Grading Rubrick (hand in with paper)

25 pts possible Included basic introduction to the gathering of data (3) Read noise and gain calculations Gives values for all terms (2) Shows value of read noise (2) Explains logic (2) Compares calculations with IRAF findgain task (2) Linearity of the CCD Includes data table (2) Includes plot (2) Answered 4 questions (4) Comparison with manufacturers values (2) Embedded questions Commented on different tests and results (2) Discussed pixel size vs full-well and binning (2) *Be sure to include as many of your calculations as possible at each step. Score Comments

4/10/09

characterize_CCD_2009.doc