DSLR Star Pixel Saturation

Hi Tom,Pixel saturation is only one part of this. Certainly you want your exposure such that you stay below the saturation, but even more important is to stay in the linear region of the detector. THis usually means a maximum pixel count of around 35,000 ADU. To determine the linearity of your detect you must experiment. AIP4WIN book explain it fairly well, but basically get a fixed (constant level) small light source, Make a couple dozen exposures (do not stack) from say 0.1 seconds up to 10 seconds. Now plot the exposure times on the X axis (horizontal) and resulting maximum pixel count for each exposure on the y axis (vertical). You should end up with a plot that shows a striaight line up to a point and then starts to curve at a "knee" in the plot. The point where the line starts to curve is the limit of your linearity and you MUST limit your exposure times so you stay under that maximum ADU count.JeffCounting PhotonsHopkins Phoenix Observatoryphxjeff@hposoft.com

Thanks very much, Jeff.I'll run the test.Tom

Hi all! What surprises me tho is that saturated pixels will actually give a 65535 ADU count in AIP4win, as most DSLRs use 12 or at most 14 bit AD converters, so for example, images with my (admittedly low end) Olympus E-420 will give ADU counts between 0 and 4095 (12 bit). So I wonder whether perhaps the driver fpor your camera will "automagically" stretch the ADUs and I'm not sure this is good for photometry unless we can be sure the stretching is linear. I think it would be interesting to see what the raw ADU counts are when you select the "extract raw Bayer matrix" option in the DSLR settings of AIP4win . CS Heinz

Heinz,Hi,I have discovered a bug in AIP4WIN v2.3. When I open a RAW image under the DSLR Conversion Settings / 'No Conversion, Display Raw Bayer Array' the target star is oversaturated showing the 65535 ADU count. This does not make sense because the camera settings are the same as the satisfactory settings used with the previous version of AIP4WIN. By experimenting I found away around this which I cannot explain but it works. If I reselect the ' No Conversion, Display Raw Bayer Array' option and re save it and then open the image again, the image opens as it should with the ADU counts being what you would expect. Having made this step there is no need to make any further changes while AIP4WIN is open. Last night I proceeded to analyse my five sets of ten images of epsilon and got 3.311V, 3.307V, 3.313V, 3.349V and 3.326V. This averages at 3.321V and a standard error of 0.008. This seems very satisfactory.So version 2.3 has this opening glitch but can be made to work satisfactorily.I think that we will have to persuade the authors of AIP4WIN to produce a new version 2.3.1 which caters specifically for the needs of photometrists!Des Loughney

Hello Heinz, Thanks very much for your comments and thanks, too, to Des for his advice on the same subject. You got me wondering what AIP4WIN was doing behind my back. I'm still not sure what's happening, but Des' ideaformanipulatingthe DSLR Conversion Setting"No Conversion, Display Raw Bayer Array" seems to have made the needed corrections. My max pixel count is now 4095.It's alsoclear that my exposure settings, which gave midrange pixel counts previously, now show that my imageswere underexposed. On our next clear night I'll try camera settings similar to what you are using. Thanks again Heins, Des and Jeff for the great tips. Tom

Hi Des, et al--Opening raw DSLR images is optimized for "pretty picture" imaging. The default setting performs automatic white-balance and stretch into a 0-65535 range. Because white balance and auto-stretch are histogram based, star images will usually be clipped at 65535.For photometry, there are several considerations:1) ALWAYS set the de-Bayer algorithm to "Bilinear Interpolation."1a) DO NOT USE the other de-Bayer algorithm choices. They are designed to make nice-looking pictures, but the algorithms are inherently non-linear because they simultaneously convert the Bayer array, perform color smoothing, and sharpen the luminance component. Bilinear interpolation is strictly linear.2) DO NOT USE deBayer, Auto-white balance, Auto-stretch. This does the best job with pictorial image for people who "just want to get a picture." 0-65535 output range. Stars will usually be saturated.3) DO NOT USE white balance using camera settings. You don't know what the camera settings are because the camera sets them automatically. Output range depends on camera settings. They may be different for every image.4) You can use de-Bayer Using Scale Factors. Set the scale factors to 1.0 and 1.0. This images will appear somewhat too green. This is the native color balance of the camera. Split the image into R, G, and B color channels for multicolor photometry. If you use 1.0 and 1.0, the output will range from 0 to 4095 for a 12-bit camera.5) You can use de-Bayer Convert to Grayscale. This interpolates the native color balance of the camera to produce a grayscale image. This is what I would use if I were doing photometry with a DSLR. The output will run from 0 to 4095 for a 12-bit camera.6) You can use No Conversion, Display Raw Bayer array -- BUT the image is broken up by the Bayer pattern. The output will run from 0 to 4095 for a 12-bit camera. When you see the Bayer pattern, you can easily understand why undersampled star images lead to poor photometry in a DSLR images. If you use this option, it is best to set the camera out of focus so that stars are four to six pixels diameter.After you make a change to the DSLR conversion options, you must click "SAVE" to make them active.--Richard

Tom,Your exposures might still be saturated. Given what was said about most cameras using either a 12-bit or 14-bit analog to digital converter, you would theoretically expect to see maximum values of 2^12 - 1 = 4095 or 2^14 - 1 = 16383 for their maximum pixel values respectively.I worked with an undergraduate earlier this week on trying to find the saturation point of his Nikon camera. Saturation was, of course, dependent on star's brightness, but we found his camera saturated in about 4-seconds for eta Aur (V mag = 3.158). Capella saturated in less than 1-second.Unfortunately, he stacked the images before he obtained the count readings, so I can't quote ADU count values for his camera, but what was clear is that there was a turn-off from linear behaviour. I'll see if I can get his data and post a plot of it sometime in the next few days.I really would suggest following Jeff's instructions above. It's the quickest way of finding out how fast your camera saturates.Brian

Richard,Thanks for the information. Since reading the above I have experimented with the three methods than can be used for conversion. You recommend the de Bayer convert to Grayscale. I average stack 10 converted images and then split the stacked image and examine the green channel image with the single image photometry tool. I compare epsilon aurigae with eta aurigae. I examine the green channel image because, with a transformation coefficient determined empirically I can work out the V magnitude.What I have found that is that each of the three methods produces a different green channel image. By different I mean that the differential magnitude is different. It seems to imply that each conversion requires its own transformation coefficient. I would be grateful if you would recommend a conversion based on the green channel image it produces. Is there a conversion that produces an image that is more reliable in terms of recording the light that arrives at the camera sensors?I would be interested in the theory of what splitting the colours means for each of the three conversions.Thanks,Des Loughney

Hello Des--You did not explain in your previous message what you actually wanted. If I understand correctly, you wish to discard the red and blue components. You want a grayscale image made entirely from and only from the green pixels. Straight conversion to grayscale does not give this because it averages the contributions from all three channels.The most direct way to create a "green-pixel-only" image is:1) In Preferences, DSLR Conversion dialog, set De-Bayer, white balance using scale factors: set the scale factors to 0.0 and 0.0.2) In Preferences, DSLR Conversion dialog, set conversion to BI-LINEAR.3) Open the raw DSLR image. It should look green.4) Verify that the conversion is correct. Run the cursor over the image; the R and B values shown by the cursor should always be 0.0. If they are anything but 0.0 and 0.0, the scale factors are set to something other than 0.0 and 0.0.5) Convert to monochrome Color->Grayscale.I am curious why you stack ten converted images? Instead, measure the stars each image separately and tabulate the magnitudes of V, C1, and C2 in a spreadsheet. Compute the mean magnitude (which will be the same as averaging the images by stacking them) and the standard deviation (which you would otherwise not have). The empirical determination of the uncertainty in the magnitude determination gives more realistic result than you can get from an estimate of the CCD gain factor in the DSLR camera.--RichardPS Have you checked the linearity of you DSLR? You should be able to plot your transformed measured V magnitudes against the catalog V magnitudes of stars in the field.

Hi Richard, >1) In Preferences, DSLR Conversion dialog, set De-Bayer, white balance using scale factors: >set the scale factors to 0.0 and 0.0. That sounds surprisingly easy, I must admit I had never thought about this even tho it's kind of obvious when you read it, thank you! As to: Why stack in the first place, and not average individual measurements: I think the key reason is that the observation star for this campaign is very bright, even at its minimum (ca 3.8 mag) so to avoid saturation even with a slightly defocused image and low ISO levels, you have to choose a rather short exposure time, shorter than you probably want to avoid errors from atmospheric scintillation (IIRC it was recommended somewhere here to use at least 30 ... 60 seconds of light for a single measurement). Usually several (e.g. 5) stacks are examined per observation and that's where the averaging and SD comes in. CS Heinz

Hi Brian,Thanks very much for your recommendation. As it turns out, my problem all along was the "bug" that Des Loughney discovered (see above) in AIP4WIN. Apparently, every time you open the program you must "save" the settings under "Preferences", "DSLR Conversion Settings". Even though your previous selections are indicated in the radio buttons, the program has reverted to the default settings. Once I reset my preferences and hit the "save" button I started getting resonable "ADU" counts again.You mentioned your student was using 4 second exposures. Do you recall the ISO and F-stop?Thanks again, Tom

Hi Richard,Thanks very much for all your helpful advice.I tried your recommendation and it worked great for individual images. Each image was green and all pixel values were zero for red and blue. When, however, I tried to calibrate and stack images, the program reverted back to displaying color images.Is this what you would expect?Tom

Richard,I tried to create the green channel image as you suggested. I selected ' De Bayer, white balance using scale factors' and set the scale factors to 0.0 and 0.0.I also set conversion to BI-LINEAR. The scale factors would not stay set to 0.0 but always reverted to Red 2.11 and Blue 1.66. This was reflected in the opened images.I stack images to save time. To get the most precise results ( Standard Error 0.005 ) I stack five sets of ten images and then average the results. It would be very tedious analysing 50 different images. At the moment the time taken to take 50 images and analyse them is about 45 minutes.Des

Hello Tom--I don't have DSLR images on hand suitable for testing this. However, you must convert those green images to grayscale before you stack them. The green images are color images -- and they will stack as color images.I would appreciate one of you taking the time to explain your procedures from start to finish, with all the details. Tell me how you shoot the pictures, how you extract magnitudes from the images, and what you do with the magnitudes you extract. The forum approach tends to "atomize" what you're actually doing, so I find out from one person that you stack images, another that you shoot fifty images and stack them into ten groups of five, and another that you use the single image photometry tool.Doing photometry one star at a time is silly! The AIP4Win Multi-Image Photometry tool was designed to extract differential photometry measurements from series of hundreds of images automatically. I recently ran photometry on a set of 5000 CCD images (see http://www.wvi.com/~rberry/astronomy/xxcygni/xxcygni.htm). While it takes more work to do photometry with DSLRs than with CCDs, it can be done. Help me by explaining exactly what you do and exactly what you need the software to do.--Richard

Hello,OK, I can try to summarize my workflow which seems to be similar to what others do:Equipment:* Use unmodified Olympus E 420 (4/3 sensor) DSLR with settings like: White balance to full sunlight, noise reduction off, save images as raw, all fancy image enhancing settings turned to "off" or "neutral"* use lens with focal length 55mm, f/4* ISO 400 , exposure time set to 8 seconds, * image deliberately defocused by turning focus ring to max position (at infinity end) so star is imaged as a disk or rather a donut about 20..24 pixels in diameter * use a simple tripod, no tracking/guiding. shutter activated with remote control, imaging is delayed 6 seconds after flipping of mirror (pedantic)Exposure sequence:* Take 3 dark frames at the same settings as the light frames[* some of us take flats and bias frames as well, I don't, let's skip this for this example ]* Take 10 images of Epsilon Aurigae. Star is near center of field, but the very bright Capella should not be in it.repeat the exposure sequence (say) 5 times.For each exposure sequence independently, do the following:Analysis sequence* with a different third party program (here IRIS) I batch-convert all images (light and dark frames) to grey scale FITS images just from data of the green channel. If using AIP4WIN, I guess you would set DSLR conversion settings to manual white balance with red=blue=0 and then convert to greyscale (for all light AND dark frames). Then continue working with the grey scale images * in AIP4WIN, open Claibration setup dialog* select the three darkframes (already converted to greyscale)* select "average"* press "Process dark frames" ==> "green light" near button lights up * close dialog* select the deep sky autoprocessing dialog for stacking* select the 10 grey scale light frames as files to be stacked* pick the first as master frame* select "average" stacking* tick "calibrate" on the preprocessing tab* select two star alignment* tracking radius set to about 35 pixels* select two stars (I happen to use the variable and the first comparison star , they are a few hundred pixels apart)* select "manual" mode (kind of pedantic, automatic also works fine)* press Ok and watch AIPWIN do the stacking, in manual mode: confirm the alignment after each slave frame* visually inspect the stacked image for obvious faults * open the photometric "one image photometry" dialog* select appropriate radii for the aperture pattern. For me it's like 12, 14, 18 IIRC, so that the inner circle will completely cover the entire (defocused) star image (check with a rather low white point in the display settings, like white: >=10 ADU)* click on the image of the variable star to set the aperture * click on the image of the comparison stars (I usually use 3, two of them in close vicinity to eps Aur) * click on "get Magnitude"* cut and paste the output of AIP4win from the log window to a spreadsheet* Save the stacked image for future reference* Clear Dark frames in Calibration menu(repeat the analysis process for the remaining 4 series of frames)Statistical postprocessing* In Excel, use the instrumental magnitudes and a table of published magnitudes and color index for the comparison stars to get an estimation of the color Transformation Coefficient and the zero point. TC should be similiar to what was used last time, otherwise skip this measurement.* use these to compute the estimated magnitude for the variable and for the comparison stars (as sanity check) from instrumental magnitudes provided by AIP4WIN. If one of the comparison stars has an estimated magnitude very different from the published magnitude => skip this measurement(* I do not try to compensate for extinction, yet)* At the end, average the magnitude estimates for the variable from the 5 series and calculate the SD* If SD is resonably low, report average value to AAVSO. Otherwise exclude outliers or skip the whole observation (maybe there were clouds I didn't notice?)CSHeinz

Hi Richard,Thanks again for your insight. Here are the procedures I follow during an observation session:ImagingFor imaging epsilon Aurigae, I am using a Canon 20D with a 70mm lens and a fixed tripod. Currently, camera settings are 4 sec, F4.0, ISO 800. The lens is slightly defocussed by turning it against the stop. I begin my session by shooting 5 darks, then 30 star images then 5 more darks. Back inside I shoot 10 flats and 10 flat darks.ProcessingI have found that I need to take a minimalist approach to processing. If I attempt to do the more complex procedures, even stacking color images, AIP4WIN freezes on my computer. My processor may be inadequate for complex tasks. I’m using a laptop with an Intel pentium M 1.6 GHz processor and 512 MB of RAM. My OS is Windows XP Home Edition.These are the steps I follow:1. Open “preferences”, “DSLR & Bayer Conversion Settings” and select “No Conversion” and save the selection. I’ve found if I don’t do this each time I open the program, the settings revert to default values even though the radio button indicates, “No conversion”.2. Open each of the 60 images (30 calibration and 30 star) individually and select “color”, “split colors”. This is a real drudgery. It takes over an hour to do all 60. I save only the green channel of each image.3. Next, I select “Setup Calibration” and input the darks, flats and flat-darks for a standard calibration.4. I then select “Multi-Image”, “auto-process”, “deep-sky” to calibrate and stack the images. I do this four times. Once with 30 images to get an overall average. I also do three stacks of ten to obtain a standard deviation.5. I obtain a magnitude difference from the stacked images by using the single image tool. Epsilon Aurigae is my variable star and eta Aurigae is the comparison.I’m also using the single-image tool to calculate a Transformation Coefficient. I obtain an instrument magnitude for several other stars in the field. I subtract the tabulated magnitude for each star from the instrument magnitude. I then plot a graph with the stars color index on the x-axis and the magnitude difference on the y-axis. This plot is approximately linear and the slope of the line is the TC value.I hope this helps.Tom

Richard, Here is my procedure to obtain precision V photometry of epsilon aurigae when I used AIP4WIN v 2.1.10. It seemed to work very well and gave results, assuming a transformation co-efficient of 0.15, which were comparable to CCD estimates. A typical standard error was 0.007. My camera is Canon 450D.I experimented with settings which gave a signal to noise ratio of epsilon and eta Aur as stated by the single image photometry tool of over a 100. The settings used with an 85 mm Canon lens were exposure 5 seconds, ISO 200 and f5. Experimentation showed that the most consistent results required analysis of three sets of ten RAW images and ideally five.A dark master frame at these settings was obtained and used for calibration.Research has shown that the sensors in these Canon camera’s are linear, with specific settings, over a range of about 1.5 magnitudes.Ten images were selected with the multimage tool and average stacked. The single image photometry tool could not be selected for analysis of the average stacked image. The image had to be transformed into a Bayer Array image. This image could then be split into colour channels including the green channel. The green channel image could be analysed with the single image photometry tool. Using the transformation coefficent the differential magnitude could be converted to give a V magnitude. I have used the above method, over the last eighteen months, with a wide range of variables. It works well. I have used it on images obtained through a Canon 200 mm and 400 mm lenses and, with the right settings, produces good results down to magnitude 11.I have written a full refereed paper on the methodology which will be published in the Journal of the BAA. I can send you a copy if you give me your email address.Des Loughney

Richard, I have written a full refereed paper on the methodology I use which will be published in the Journal of the BAA.Attached is a copy minus the diagrams. Des Loughney

Tom,He is using ISO 400, F/5.6 (with the stock lens on a Nikon D300 DSLR) with a slight amount of defocusing (i.e. before the stars become doughnuts). Exposure times are always below 4-seconds (except in the saturation test). I've attached an image of the non-linearity test I had him do. I know it isn't quite right because his camera uses 12-bit encoding for the .raw files, but the plot shows values over the 4095-count limit, but it shows the general idea.

Hello Des, Tom, and Heinz--Many thanks for providing descriptions of your work flow. This gives me a much better idea of what you need from AIP4Win.Among my files I found a five-image sequence taken with my Canon 350D. Although the images are 120-second exposures with an 8-inch Newtonian, these allow me to test some new routines.The first change is that the DSLR settings are now saved properly.The second is that I have provided an option so that you can open a DSLR image as a green-pixel-only monochrome image.The third change is that you can use the Multi-Image Photometry Tool (MIPT) to run photometry on a sequence of images. This allows you to select a sequence of images and run photometry on them automatically. A sample output file is attached. I chose the stars at random.The fourth change is that the current incarnation of the Magnitude Measurement Tool (MMT). The MMT is more sophisticated than the MIPT. A sample output file is attached. You can import these output files into Excel and get means and standard deviations.I will do further testing, but then I would like a few volunteers to beta test the new routines. It would help if anyone who wishes to do beta-testing would join the AIP4WIN-Photometry Yahoo Group. Beta testing is sometimes frustrating and difficult, and I would prefer to restrict the discussion to just a few people until everything appears to be working correctly.--Richard

Hello Richard,Thanks for the suggestions and changes to AIP4Win, that should be really useful. I tried the MIPT with version 2.3.0, it's as easy as stacking and measurement, but in one step instead of two. The airmass calculation of MMT will simplify things further. Given the SD of the single exposure measurements, I guess it will be more than enough to shoot 30 light frames per night with my settings.I'll be happy to join the beta test group soon.Clear SkiesHeinz

Hello Heinz--It will take a bit more work to get these changes running smoothly, but I'll let you guys know when a reasonably stable beta is ready for testing. It could be a week or more since I have other things to work on also. If you're doing CCD photometry as well as DSLR photometry, you ought to be in the AIP4WIN-Photometry Yahoo Group.For context, MIPT is a "legacy" tool. I have left it in AIP4Win because quite a few people are familiar with it and don't want to change. The MMT is more versatile and produces more useful data because it can do any number of comp stars, the output includes airmass, you can get HJD, and output to the AAVSO Report format.--Richard

Richard,Thanks for the message. I am glad that a new version will be produced to make things easier for DSLR photometrists. I have joined the yahoo group and I hope I can help in the process.Perhaps you can advise me on what seems to be a problem I have AIP4WIN. I use the multiimage/ deep sky process to arrive at a stacked green channel image which I can analyse with the single image photometry tool. I have tried to use the Multi Image Photometry Tool but the process never gets started because a box always states ' Cannot subtract Darkframe from an image of different size'. This box never appears when I use the multiimage/ deep sky process. Similarly I cannot use the MMT. The same box comes up.Des

Hi Brian,Well, we've had several rainy days here so I finally got around to doing the saturation test on my Canon 20D (results attached). The camera was operating at ISO 800, F/5.6 with the lens at 55mm.It looks like the sensor remains linear at least to 3500 ADU and probably somewhat beyond. It is difficult to get a finer scale result because the minimum exposure increment is 1/3-stop.Thanks again for the advice.Tom

Hi Richard,Thanks very much for your invitation to participate in the beta test. Unfortunately, both my computer skills and my laptop's horsepower are probably not up to the task. I will join the Yahoo group, though, so I can follow your progress.Thanks again for all your assistance.Tom

Tom, I agree with you estimate of ~3500 ADU for this particular set of data. I modified your spreadsheet to include Possion noise (as a first-order approximation, I used sqrt(ADU), not accounting for the gain (see the discussion below)). I then fit a line to all of the data points except the 0.5 and 0.6-second exposures (#717 and 718) and then added the last two points onto the graph (but not part of the linear fit). Even with the Possion Error bars, the earlier points don't fall in-range of the error bars... implying there is something funny going on although it could just be background noise with such short exposures. What star did you use to generate the graph? You might try a dimmer star in the field-of-view so that you might see the knee in the graph. If you generate a few of these graphs for stars of different magnitudes, you could figure out the approximate knee location vs. magnitude and, through the original saturation graphs, predict where your camera will saturate for a given star. Good work! Brian

Hi Brian,I actually did the test inside with an artificial star. I made a pinhole in a sheet of foil and placed it over a night light. I see your point about trying longer exposures. For my next attempt I'll reduce the light intensity.Thanks again for the tips.Tom

Hello Brian,I am somewhat surprised to read you consider the Poisson noise (quantum fluctuation) being directly proportional to sqrt(ADUs). For me that applies to the number of particles, number of electrons to be counted at level of pixels. ADUs are only a rescalling of it through the gain of the read amplifier and the range of the ADC. To determine that noise level we should consider the number of electrons not the number of ADUs. If G is the gain in ADUs/electron the SNR should be: SNR= ADUs / (sqrt(G) x sqrt(ADUs))Do you agree ?Yours truly,Roger

Roger,It's really a question of what dominates the noise in the camera. As a first order approximation, we may assume that the noise of the camera is negligible in comparison to the signal in which the sqrt(ADU) is the correct answer for the SNR. Basically, in the quick analysis I did above I considered the noise akin to shot noise in electronics in which the noise is proportional to sqrt(N). I agree with you that the equation I gave above isn't appropriate in all situations and a better equation is given in Howell's paper on CCD observations:snr = ADU / sqrt(ADU + n_pix(N_s + N_d + N_r^2))where ADU is the (sky-subtracted) ADU count from the camera, n_pix is the number of pixels in the software aperature used to extract the data, N_s is the number of sky photons per pixel, N_d is the number of dark current photons per pixel, and N_r is the read noise of the camera (in electrons per pixel). There's a nice article on STSCI's website that discusses SNR in much more detail. Even better is the book entitled "The handbook of astronomical image processing" by Richard Berry, James Burnell (this comes with the AIP4WIN software package) or "Astronomical photometry" Arne A. Henden (director of the AAVSO) and Ronald H. Kaitchuck. Arne's book in particular goes deep into the statistics behind why we choose the distributions to represent the noise as we do.The problem with analysing CCD observations, especially when they are given as ADU counts, is that the noise of the system isn't always disclosed so you've got to start somewhere. Without a comprehensive way of characterizing his camera, I think the "best" estimate of the SNR simply comes from shot noise.I'm curious though, where did you get your equation? I've seen a few equations that incorporate the gain of the detector, but not that one. It appears to account for the gain "twice," therefore under-representing the SNR by a factor of sqrt(gain). Recall that if the other sources of noise become negligible, you equation should reduce to sqrt(ADU), however the equation you cite above doesn't. Is there a term missing or a typo? I'll see if I can dig up a copy of Berry's book to get the equations for when the gain is known and compare it with what you have above in the next few days.Brian

Hi Brian, Roger's equation can be found in section 8.2.2.4 of AIP. You have to consider the gain, otherwise it would be possible to get a better SNR just by counting each photo-elecron (say) 1 thousand times (higher gain). Or in other words: (e- : nr of photo electrons) SNR = e- / sqrt(e-) and replace e- = ADU / g where g is the gain factor, and you get Roger's equation. CU Heinz

Heinz,Thanks for the citation. You and John are, of course, correct. Perhaps after Tom finishes characterizing his camera, we can figure out the gain and revisit this topic. I think figuring out the correct SNR for DSLR cameras as a function of the various settings could be difficult, especially if the manufacturer doesn't disclose much information about the CCD. Isn't a typical gain somewhere on the order of 2?In the mean time, I've edited my original post to indicate that the equation is a first-order approximation that doesn't account for the gain of the camera as to not mislead our readers. Thank you for the correction.Cheers,Brian

Christian Buil has done many measurements on the characteristics various models of DSLR (mainly Canon)for astronomical applications including gain measurement (Which of course depends on the ISO setting you use) for example http://www.astrosurf.com/~buil/50d/test.htmhttp://www.astrosurf.com/~bui...

The equationsnr = ADU / sqrt(ADU + n_pix(N_s + N_d + N_r^2))It is not even a good first order approximation. The shot (photon) noise is not sqrt(ADU)or even close if the gain is significantly different to 1. Without the gain term it is just plain wrong. It is not even dimensionally correct. Robin

All right, I finally grabbed a copy of Berry's AID:In units of ADU, the noise is (equation 2.13):s_raw = 1/g * sqrt(s^2 + {s_d}^2 + {s_ron}^2)where s_raw is the total noise, s is the shot noise (in r.m.s. electrons), s_d is the dark current noise (in r.m.s. electrons), and s_ron is the readout noise (in r.m.s. electrons).

OK as far as it goes, but now the sky background term has disappeared. Have you seen Michael Richmond's on line calculator? It uses Steve Howell's equations but circumvents the needs to know the camera gain by calculating the actual number of photons you should be getting for a given magnitude, telescope and sky conditions.http://www.tass-survey.org/richmond/signal.shtmlIt does assume the camera thermal noise is insignificant though which might need to be checked for the case of an uncooled DSLR on faint objects (ie not eps Aur) from a dark siteRobin

Hi Brian,I'm afraid that I've procrastinated and not completed more testing. Once I realized that the "pixel saturation" was a false indication from AIP4WIN, I moved on to other image processing problems. I think I'm finally closing in on an acceptable methodology. I'll post my current approach as a new topic (DSLR Photometry Procedures) here shortly. I'd really appreciate any feedback you might have.Thanks, Tom

Hi--It is difficult to measure the gain of a DSLR because when you shoot flats, the Bayer array filter mosaic gives you a mosaicked flat, i.e., a non-flat. You must extract the red, green, and blue samples separately.However, you don't need precise values to get a reasonable handle on the photon noise level. Most DSLRs have full well capacity ~25,000 and 12-bit ADC (4096 levels), so the gain will be somewhere around 6 to 8 electrons per ADU. Read out noise is usually around 10 electrons rms, although recent models are better. These figures are probably close enough for most purposes.I believe one fo you has been using an automatic conversion setting in AIP4Win, so the tiny percentage of pixels in the stars images is scaled to pure white at 65,535. This is an appropriate setting for converting pretty pictures, but not for photometry. For photometry you must use a non-auto-scaling setting and bi-linear Bayer interpolation to get the actual pixel values.The most meaningful way to estimate the noise would be to determine the standard deviation in raw instrumental magnitude for one of the comparison stars. This account for Poisson noise, noise in the DSLR camera's electronics, sampling errors produced in the Bayer array, and software errors introduced by a non-linear de-Bayering algorithm.--Richard

Hi Richard, The gain of a DSLR camera varies significantly with the speed setting (from 100 to 1600 for example the gain changes by a factor of 16. This would give up to a 4 fold error in the S/N figure if a constant figure for gain is used.) Since most (all?) DSLR users take multiple fames to make measurement, the best estimate of the error would come from the standard deviation of a check star as you suggest. (edit: more specifically the SD of the values in the individual frames/sqrt(no of frames used) Robin

Hello Brian, My talk was only about the quantum fluctuation contribution to overall noise, excluding any other like read noise or fixed pattern noise (often said thermal noise, as an electronics guy I don't like that wording !) Next if you consider the overall process of comparative magnitude measurment it's even much more complex... I did try to reconciliate such analysis with the actual SD calculation on series of shots but the result was not very convincing... In my note I didn't refer to any particular book but simply made the following reasoning: - The physics says we can't count electrons with an accuracy better than sqrt(N), N being the number of electrons. - ADUs being the count at output of the ADC - G being the overall gain of read/sampling circuits, pre-amplifiers, amplifier, ISO setting, range of the ADC. It's clear: ADUs = G x N for each pixel, doing some math you get: sqrt(ADUs) = sqrt(G) x sqrt (N) ADUs / sqrt(ADUs) = sqrt(G) x (N / sqrt(N)) SNR = N / sqrt(N) = sqrt(N) = ADUs / (sqrt(ADUs) x sqrt(G)) or simply: SNR = sqrt ( ADUs / G ) That's only for the shot noise component. But in fact in most cases with present DSLR ( or DSC, I use a Canon G9 !) this component is far dominant even when you fill the pixel-well at 60~70% at low ISO. In my case shot noise is about 10 times the read noise at 100 ISO, 4 sec exposure and a 20 mm diameter aperture (for mag 3~4) One issue is to determine G, it's right, as said Robin you can find several cases being published in Christian Buil's notes, including the procedure he uses. It's also possible to make an estimate based on size of the imager, of the optics and Qe. I think such considerations are interesting to optinize our measurment conditions but not very effective for qualifying the actual results accuracy. I prefer to calculate SD on a series of 10 to 25 shots. Compared to such SD results the SNR methode suggested in the report form seems too much optimistic. Yours truly, Roger

Hello Robin, all,I think that in your "edit" note you touch one question I have for sometime. I don't stack images but make an automated evaluation of them (under DylogAPL with my own functions) that including the SD calculation for the series (10 to 25 images depending my courage...) Usually the convention in lab reports is to show the average of series of measurments and the sigma of the distribution of the series. It's not so usual I think to report the sigma of the average itself ( series SD / sqrt(n) as you mention) In our case, I agree with you, what is important is the distribution of the end result, not such of the series, by the way the SD of the average should be reported (The averaging of the series is here just a tool to reduce the influence of the noise, it doesn't mean the magnitude varies as the series and the SD would characterise this variability) What is the rule/convention for AAVSO ? Do we know ?Yours truly,Roger

(Anyone else find this forum software anoying in that it does not transfer the subject title onto subsequent replies or consistently thread them?)Hi Roger,I agree that what I use to call the "Standard Error of the Mean" but I believe is now better called "Standard Uncertainty" seems to be the logical choice but there does seem to be some confusion so I asked Arne to confirm on the AAVSO discussion forum. (see below) Hopefully this should clear up any confusion.BTW if anyone is interested in an easy to read general explanation of measurement errors and the types of sources, "The beginners guide to uncertainty of measurement" produced by the National Physical Laboratory is a pretty good gentle introduction.http://www.npl.co.uk/publications/uncertainty-guide/uncertainty-of-measu...Robin----- Original Message -----From: "arne"To: "Robin Leadbeater"Cc: Sent: Monday, December 28, 2009 3:02 PMSubject: Re: [AAVSO-DIS] Magnitude Error>> Robin Leadbeater wrote:>> Hi Arne,>>>> Could you clarify one point for me please. If one submits a value for>> the target star as the mean of several individual measurements,>> shouldn't the estimate of the error of this mean be the standard error,>> ie standard deviation of the standard star /sqrt(number of measurements)>> ? If not, can you explain why not?>>> Yes, it should be the standard deviation of the mean, calculated> as you describe above.> Arne>

Hi Robin,Thanks for your answer about SD reporting, we have the same view.About gain of DSLR amplifier you are right and I would note another gain variation factor, this time depending from camera type. That gain should also vary more or less as the inverse of the surface of the pixel (and its well depth by the way) Between my Canon G9 (~2x2 microns) and ie. a 5D (8.2x8.2 microns) or my D70s this is another factor 16 ! We have now a total of 256...Yours truly,Roger

Hi guys--Responding to the DSLR pixel saturation thread, the gain is usually the full well capacity in electrons divided by the number of gray levels (4095 for 12 bit images). The full well is something like 30,000 electrons and the gray levels is 4095, so the gain is ~8 electrons/ADU at the camera's lowest ISO speed.DSLRs use the whole full well capacity at their slowest rated ISO speed, so if the gain is 8 e/adu when the ISO is 200 (the minimum speed of your camera), to get ISO 400 the manufacturer uses only the lower half of the full well. This makes the gain ~4 e/adu at ISO 400. At ISO 800, only the bottom quarter is used, so it's ~2 e/adu, and at ISO 1600, it's ~1 e/ADU.In any case, several posts ago, someone noted that the most realistic way to find the uncertainty in the magnitude of eps Aur is to compute the standard deviation of a series of magnitude measurements. This is a much better way than trying to second-guess what Canon or Nikon is doing with gain and readout inside their Digic image processing chips. It --Richard

Hi Richard,I was wondering if the updates that you mentioned in your Oct 15 post have completed beta testing and are ready for distribution?Thanks, Tom Pearson