DSLR Transformation Coefficient

Hi Tom,The two methods are essentially identical. If we use the standard equationV_true = V_inst + Zeropoint + Coef * (B-V)_trueand use two stars, call them "1" and "2", then you getV_t1 = V_i1 + Zeropoint + Coef * (B-V)_t1V_t2 = V_i2 + Zeropoint + Coef * (B-V)_t2and then subtract the two equations(V_t1 - V_t2) = (V_i1 - V_i2) + Coef * [(B-V)_t1 - (B-V)_t2)]where the Zeropoint subtracts out since it is the same for both stars. Then useDes' notation:ED = OD + Coef * (EDC)orED - OD = Coef * (EDC)where ED = estimated difference (that is, the standard magnitudes)OD = observed difference (that is, using the instrumental magnitudes)and EDC is the estimated color difference (the difference of the standard (B-V) colors).Des' technique works ok to get you a rough idea of your transformation coefficient,which is good enough for these early observations with DSLRs. The more proper wayis to image some true standard stars with a wider range of color and in a smallerregion of the sky to avoid nearly all extinction variations; say looking at astandard cluster like M67, NGC7790, M11 or IC4665. We'll get around to thatin a few months.Arne

I think, too, that both methods are the same. However, I will have to amend the article I wrote on how to work out a TC. It is not just dependent on the Camera make/ the type of filter used in the Bayer Array.It is also dependent on the way a software computes the green channel image. The various options available for DSLR Conversion Settings within AIP4WIN 2.3 all give different green channel images of the same Raw image. These images will give different estimates of differential magnitude.The good news is that the transformation co-efficient, once determined for a specific setting ( such as the ' No Conversion, Display RAW Bayer Array ') in AIP4WIN 2.3 should be reliable.I look forward to the TCs of DSLRs being properly worked out as Arne describes.Des Loughney

Hi Arne, Thanks very much for your detailed explanation of the transformation coefficient equations. I see now how the “standard equation” and Des’s equation are the same. I tried your graphical method by plotting data from a stack of ten images taken of the Epsilon Aurigae region. I placed star color on the x-axis and the difference between instrument magnitude and tabulated magnitude on the y-axis. Unfortunately the plot yeilded a random scatter of points. This is the data I used:

I would appreciate any thoughts you might have on what I should try next? Thanks again for all your help. Tom

Tom,My comments on your method are as follows. When I am determining my TC I do five sets of ten images to get the best precision. I also found it vital to use the right settings with my Canon 450D. The settings must achieve a signal to noise ratio of over 100. In order to ignore atmospheric extinction affects the target stars must be close together. The stars to be analysed must be above 40 degrees above the horizon. For calibration purposes the nearer the zenith the better.The set of stars that I use is lambda Aur, rho Aur, HIP 24902, HIP 25143. With these I get consistent results. I compare lambda Aur with the other three and study how the actual difference varies from the predicted difference. I compared lambda with rho within five sets of ten images. The results I got were: 0.473, 0.428, 0.378, 0.392 and 0342. The average difference was 0.403. The predicted difference was 0.55. The difference between these two was 0.147. The difference in B-Vis 0.759. This gives a TC of 0.19 ( 0.147/0.759). This is a bit high but my TC of 0.15 is an average of the three pairs.In my calculations I do not use the instrumental magnitude of a star. I do not think that is important. What is important is the instrumental difference in magnitude compared with predicted difference in magnitude. The ' difference ' magnitude is the real issue.Des Loughney

Hi Des,Thanks very much for this info. I have 50 images taken recently with Auriga near zenith. I'll try to duplicate your procedure and see what I get. Have you ever tried Arne's graphical method? I may have misunderstood his directions.Tom

Early on Thursday morning ( Edinburgh, Scotland ) it was still and clear and epsilon was high up. I was able to determine the TC using AIP4WIN version 2.3. The TC was worked out using the setting, under the menu heading ' Preferences' - 'DSLR Conversion Settings' - ' No Conversion, Display RAW Bayer Array'. My camera settings with the 85 mm lens were 5 seconds exposure, f 3.5 and ISO 800. With these settings I was able to get a signal to noise ratio of over 100 with my calibration four stars - lambda Aur, HIP24902, HIP 25143 and HIP 25048. I did five sets of ten images. Each set was average stacked. The green channel of the stacked image was separated and analysed with the single image photometry tool. The TC with the later version of the software turned out to be 0.23. The TC with the previous version was 0.15. To obtain the correction for analysing epsilon I multiplied the TC by the difference in B-V between epsilon and eta which gives a correction of 0.157 which has to be added ( or as it is magnitudes we are dealing with subtracted ) from the green channel estimates of epsilon to give a V magnitude.After working out the new TC I took five sets of ten images of epsilon with the camera settings, using the 85 mm lens, of 5 seconds exposure, f5 and ISO 200. These camera settings gave a signal to noise ratio of around 120. First set: the difference between epsilon and eta was - 0.274. Assuming a V magnitude of 3.18 for eta this gave a magnitude of 3.454 for epsilon. Subtracting the correction of 0.157 the determined V magnitude was 3.297.Second set: the determined V magnitude was 3.296.Third set: the determined V magnitude was 3.289.Fourth set: the determined V magnitude was 3.269.Fifth set: the determined V magnitude was 3.234.The average was 3.277V. The Standard Error was 0.012. This result is consistent with CCD estimates during the same night.On less turbulent nights the standard error can be down 0.003. I have now been getting very good results for a year.It is very important that the stars are fairly high up and that the signal to noise ratio is at least a 100.Des

How do yo determine the signal to niose value? Your value for the TC seems quite big in compairsion with mine, but as your results seems fine I guess it is the right value for your setup. But it is a bit frustrating that the TC is so different between different software even if you got the same camera.

I determine the signal to noise value in this way. I open a sample image within AIP4WIN and examine the target star and comparison(s) with the Single Image Photometry tool. Under the ' Details' menu of the tool it quotes a signal to noise ratio. By changing the settings on your camera you can see how the S/N changes. It is frustrating that the TC varies between software and versions of software! I am thinking of contacting the authors of AIP4WIN to find out why this is the case.Des Loughney

Hi Des,It would be interesting to hear from Berry & Burnell on this. With the no-frills "DIY" de-bayerization with IRIS that was described here I now get a TC of ca. 0.13 for my Olympus after analyzing Eps Aur images, unfortunately the weather here doesn't allow to take a near-zenith calibration series...yet.So I guess all this is another reason NOT to archive your photometric raw material in the RAW camera file format... Think about it, some of us may still be around in 27 years and when we may want to re-evaluate our 2009 data with the most recent algorithms, we might not be able anymore to easily find hardware and software to consistently analyse 2009 camera raw files :-).CSHeinz

Des, you say that you got 0.15 for your TC with the older version of AIP and 0.23 for the newer version. Is this for the same exact set of data, or or two different sets? If two different sets, then this is normal. Each time you measure your coefficient, you will get a different value, since there is uncertainty in the measurement. Without an extinction correction, for example, you may be at a different airmass on the second determination, which will modify the results. It is best to calculate your coefficient(s) on several nights, then take the mean and standard deviation to understand the imprecision in the measurement.Heinz, my normal recommendation is to archive your data in FITS. FITS has actually been around longer than 27 years, and I expect it to survive a while longer. My oldest FITS image (of NGC 2903, still my favorite galaxy) dates from 1983. However, what won't remain for the next 27 years is the archive media, so be prepared to copy data every few years onto the current generation of backup media. Remember 7-track magnetic tape and punch cards? They were still in use in 1983. That is usually the time you think about renaming files or even changing format anyway.Tom (I think), the graphical method should work. I'm surprised at the scatter, though you must remember that the value is pretty close to zero and so it will look pretty random, just that it should be as close to the linear least squares line as your individual measurement uncertaintites (say, 0.03mag). You are seeing much larger scatter than that, which might indicate an acquisition or processing problem. What kind of flats are you using? Is there any systematic trend - there seem to be two basic zeropoint values, so for instance one set might be for stars near the center of the CCD/optical axis, and the other set might be for stars closer to the edges.I have another 6 CCD cameras to test, and then will get back to our Canon Xsi/450D and see what kind of coefficients and settings we need to use to get repeatable results. I'd say we'll have time to test in another week or so. One of our new CCD systems was just sent out to New Mexico and will be used to monitor epsilon Aurigae along with other bright variables, and should be on-line in a couple of weeks. Arne

Dear Arne,Thanks for your feedback. I determined the TC of 0.23 on a different set of data. It has been some time since I determined the TC of 0.15 with the older version of AIP4WIN. However, I think that the change is down to a change in the way AIP4WIN processes the data from the green filter pixels. I did many different images of different stars on Wednesday/ Thursday night. The new TC produced the most consistent results. AIP4WIN version 2.3 lays out a number of options under the heading of 'DSLR Conversion Settings'. I have analysed the same images with each option. Each option allows the separation of a green channel image. Each option resulted in a specific to the option magnitude differential. It seems to me that AIP4WIN manipulates the data coming through the green filters except in the option I chose to determine the TC which is ' No Conversion, Display Raw Bayer Array',I think that the reason the older TC is different from the newer one is that the older version of AIP4WIN manipulated the data but did not tell you. It did not offer the non conversion option. I look forward to learning of the results of your tests.The forecast is clear here for tonight and I may do another estimation - if it really does stay clear.Des

Hi Arne, Thanks very much for your advice. Unfortunately, because the calibration portion of my latest AIP4WIN update doesn’t work properly, I can’t subtract flat-frames now. I do have a dark-frame subtraction feature on my camera so, at least I can do that much. As an alternative, I thought I would try a larger selection of stars spread across the field to test your idea about systematic camera errors. I’ve attached two Excel files to illustrate the results. As you can see, there is a slight indication of a linear trend. There are, however, three stars, rho, mu and omega Aurigae that really spoil the result. The second excel file shows the regression line without those points. Unfortunately, the three stars are spread pretty evenly across the field so I can’t attribute their values to a systematic error. As a next step, I thought Iwould try imaging the Pleiades and testing a number of stars in that compact cluster. Do you think that might produce more reliable results? Best wishes, Tom

Do not subtract flat frames. Flat Frames are divided into the frame, dark frames are subtracted.Jeff

Hi Tom,I echo Jeff's comments: flats are divided, not subtracted. They are a different beast from dark frames. Looking at your data, it appears all of the fainter stars (V fainter than 5) are the ones that deviate the most. If you eliminate them, you can fit the remaining data pretty well. This would imply several possibilities to me: the DSLR has a non-linear response; something in your processing is effecting faint stars differently than bright stars; faint stars have poor signal/noise and are showing more scatter. Personally, I would see what I got with just the brighter stars, and then wait for anything further until I had learned how to flatfield images. That will improve your photometry by quite a bit.Arne

Hi Arne,I sure appreciate your advice on how to proceed. Sorry I mis-spoke about the flatfield images. Maybe I've been over confident in thinking AIP4WIN would handle the processing properly. As you know, Des Loughney discovered a bug in the "DSLR & Bayer Conversion Settings" selection of AIP4WIN. Apparently this window requires special handling to give accurate results. I now think it is the source of my problems, too. I've discovered the selections in this window were giving me false indications that my star images were saturated. I was trying to "fix" the problem by decreasing my exposure times to keep the ADU counts in mid-range. Using Des' procedure should allow me to get my exposures back on track. That's probably why the fainter stars in the graph were not working well. Hopefully, the new images I get will produce better results.Thanks again for all your insight.Tom

Hi again Arne,Well I finally managed to get the calibration portion of AIP4WIN to work for me. The program would freeze when I processed color images so I tried splitting the colors first and processing the green images. This worked well. I used 10 darks, 10 flats and 10 dark-flats. The calibration was applied to 30 images stacked. I was pleased (and surprised) with the results. I've attached an excel file and graph. The slope (TC?)of the regression line is 0.31. When I multiply this by the color difference between eta and epsilon (-0.148 - 0.537) I get -0.212. Am I right in assuming that this is the number I add to my measured magnitude for epsilon to get the "transformed" value?Thanks again for all your help.Tom

Hi Tom,This fit looks much more reasonable, though to be honest, a coefficient of 0.31 is quite large. Be sure to measure this on several nights to see what the standard deviation of the determination looks like. As for whether you add or subtract the 0.212, the simple way is to see which one gets you closer to the reported values. The calculation looks right; the sign just depends on the order that things got subtracted.I'm busy testing CCD cameras this week, but the weather is supposed to be clear in Boston. I'll try to get the AAVSO DSLR calibrated soon for comparison.Arne

Hi Arne,Yes, I was surprised by the coefficient, too. I plan to repeat the imaging and calculations on our next clear night. The adjustment of -0.212 did yeild a magnitude of 3.351 for Epsilon Aurigae which is similar to other "V" magnitudes now being reported.It will be interesting to see how your results compare.Tom

I got my TC around -0.1, but tht is using RGB and not only G, when using G I got arond -0.06 with my Canon 450D. I don't know ift it is a coinsidence, but the TC for converting V to Vt, used by the Tycho star catalog is -9/85=-0.106. I think Vt is more adapted to the human eye than V and the cameras should also generate pictures with the sam senibility as the human eye...