[mflaster]

Let's say my grays are all perfect from a color point of view across the entire spectrum, meaning they're all D65. That doesn't mean that my projector is necessarily calibrated correctly, right? I mean, there still has to be an appropriate amount of brightness for all the various IRE inputs, right?


Here's my question. Let's say I have a light meter and Avia. I put up 0% IRE, 10% IRE, 20% IRE, ... 100% IRE, and for each screen I measure the light output on my light meter.


What would be perfect results? Is it supposed to be linear, i.e. the light at 10% IRE minus the light at 0% IRE should equal 20 IRE - 10 IRE which should equal 30 IRE - 20 IRE etc.? Or should it be some kind of exponential function, i.e. a*(color ^ gamma) + b?


Thanks in advance!


Mike



------------------

Join the NYC-area HT club! Send me e-mail!

 

[Gary P. Gibbs]

You are correct. IRE to light output should be linear. Voltage to light output reqires the use of the gamma formula.




------------------

Gary


STOP DVI/HDCP!

DVI/HDCP! ~= HD-DIVX!!!

DO NOT SUPPORT JVC or anyone else who supports this!

 

[mflaster]

Thanks for the reply. So as long as my light meter measures lux or lumens, it's linear with the IRE level. I'm actually thinking about experimenting with the poor man's approach to calibration using one of these radio shack light sensors . (A pack of 5 is like $3 bucks...) I believe there's lots of talk about using them, hooked up to a multimeter, as a light sensor. So I need to figure out if their resistance is linearly proportional to the light they receive, or if there's a non-linear relationship. I have to go try to find some of those threads.


Also, as far as colors go, I could try to do two things. I could use the Avia colored filters in front of the sensor, except that my understanding is that these sensors have very different sensitivities to different wavelengths. So a perfect gray would *not* show the same readings for red, green, and blue.


Another possibility is to get some Kodak reference gray color cards. By eye, I would match the colors at various pure gray IRE patterns to a card.


A third possibility is a combination of those two. I match one gray by eye to a Kodak card, and then I see what the sensor reads for red, green, and blue. I then know to keep that same ratios for all the other grays.


Anyway, any advice/comments would be appreciated.


Mike



------------------

Join the NYC-area HT club! Send me e-mail!

 

[Mark Hunter]

Hi Mike,


This sounds like an interesting project, and you can't beat the price as far as experiments go!


The fun will *really* start once you start gathering some of that data. It's non-trivial to get from the raw data to something that can be applied to the projector for improvement, but it can be done. The mathematical technique that I had to use was Regression Analysis by the Method of Least Squares, which is something that I needed a *serious* refresher on myself (if in fact, I ever knew it in the first place).


It probably won't be too hard to get the luminance correct, as the adjustments are mostly linear, but keep in mind that different colors contribute different amounts towards the light output. Typically, Green will contribute more to light output than Red, and Red more than Blue. The percentages on my D-ILA are 66% Green, 33% Red and 1% Blue as a contribution towards the luminance.


I like your gray reference card idea. It sure beats the work that goes into the 'scientific' method!

 

[kromkamp]

That doesnt seem quite right, mark. The Color Space conversion algorithm for Luma is typically:


Y = 0.299*Red+0.587*Green+0.114*Blue


Now, allowing tolerance for individual units, 1% still seems awfully low for the blue component.


Andy K.

 

[adamgroth]

Jackpot!!!!!


Here is a great resource for using a photocell (from Radio Shack), a multimeter, and a spreadsheet (to enter your results) to calibrate the VW10HT.
http://home.pacbell.net/steve367/


This method had been "fine-tuned" over the past few months and seems to be raising alot of eyebrows in the 10HT community. I'm sure the methods could be adapted to work with many other non CRT displays.


The site also contains detailed instructions for constructing the light meter from the Radio Shack photocells.


Adam


[This message has been edited by adamgroth (edited 08-01-2001).]


[This message has been edited by adamgroth (edited 08-01-2001).]

 

[kem]

I decided it was time for me to stop lurking and get involved, since I'm trying out lots of different projectors and the experts are here.


I'm currently feeling the need to collect some quantitative data and do some reasonable gamma calibration during these evaluations. Also, it seems like an interesting project.


Mike, any luck on trying out the Radio Shack light sensors? I built a detector similar to the one Steve Smallcombe described, and have collected some data on the now famous LT150 that I'm currently evaluating. I will be collecting more data with Normal and Natural 2 tonight.


Mark, could you expand on how one might go about using regression analysis to determine the transfer function from Ohms to lux (or some other light relative value)? This transfer function appears to be quite non-linear. I'm reading various articles on gamma, light/luminance, etc. to get up to speed on this subject but would appreciate it if you could give me a kick in the right direction.


Thanks,

Kevin

 

[kem]

I guess I had an attack of the stupids last night. http://www.avsforum.com/ubb/redface.gif I was trying to figure out how to derive the Ohm to lux function from the CdS data sheets and wondering what that had to do with regression analysis. Obviously I was confused. Lack of sleep probably catching up with me, I guess. Also, even had I determined the function from the data sheets (which is possible), it would have not been very useful due to the response variance between different CdS photocells.


For those that are interested, the CdS photocells from RadioShack are made by Hamamatsu. You can find more info from their website . They also have a page where they describe the
basic characteristics of their CdS photoconductive cells.


Now that my head has been reattached http://www.avsforum.com/ubb/wink.gif I did a basic calibration of the homemade detector today by first collecting data using an old analog Sekonic light meter and the detector, and then fitting the data. The number of data points collected is not sufficient for a good fit, and the data from the Sekonic is not very accurate; however, this little experiment yeilded a reasonable transfer function. The resulting lux values were in the right ballpark. A better reference light meter is needed to make the estimated curve more accurate.


So, this leads to today's questions:


What reference light meter should I use to calibrate the CdS photocell? I expect they will be rather expensive. Is it possible to rent them? If so, where would be a good place?


I'll check around the Raleigh-Durham-Chapel Hill area this weekend to see what I can find.


BTW, if anyone is interested in the results of my experiments, please let me know. I'm mainly doing it for my own edification, and I'd be happy to share what I find.

 

[mflaster]

Yes, I'm very interested! Can you tell me the (approximate) relationship between resistance and lux? Or just give me the raw data...


Mike



------------------

Join the NYC-area HT club! Send me e-mail!


[This message has been edited by mflaster (edited 08-03-2001).]

 

[Mark Hunter]

I don't know how to apply this to a Rat Shack light sensor, but maybe it can be done with colored filters or something.


One reference that I found valuable was a paper on the conversions and some of the required math that can be found:

HERE (in PDF form)


I suppose that first you have to get the data from the light sensors to be in some sort of standard form first.


The paper referenced above touches on some of the bigger points in the mathematics. Although it focuses on CRT technology, the concepts are applicable to digital projection as well.

 

[kem]

Mike, the "Ohms to lux" function depends on the type of CdS photocell that you use and its individual characteristics. I chose the P722-7R photocell. I'm not sure if every package has the same contents, but if it does, I believe the P722-7R is the larger diameter one without the the glass cover. I'm going to try the others tonight to see if they give me better results at low IRE levels.


For the P722-7R CdS photocell, I fit the measurements that I took this afternoon to a simple power function, which has the basic form of E=k*(1/W)^p where E is illumination in lux and W is resistance in kOhms. I fit for k and p, and then scaled k based on the published photocell characteristics. I ended up with k somewhere in the range 42 to 168 and p being about 1.4. BTW, the reason that I used a power function was that it fit the data reasonably well (but not perfectly) and it was trivial to find the parameters.


After plugging all of the numbers in, I calculated lumens to be somewhere between 58 and 231 and a contrast ratio of 438:1. I think the lumens are quite low, which is probably due to either k being incorrect or using the wrong transfer function (probably the latter). I have a little more confidence in the contrast ratio since k cancels out of the equation and the room does not have perfect light control in the middle of the day.


A better solution is to first collect good reference data (any suggestions for a good device to use?) and then fit the data using regression analysis, which Mark suggested earlier. It (probably) won't give us a simple function like the power function I used above, but it will yield much more accurate results.


Mark, thanks for the pdf file! It looks very interesting. Using the AVIA color filters and the techniques describes in this paper, I think it should be possible to figure out color balance, but I need to do more research and give this topic more thought. This is the next task on my TODO list.


If you (or anyone else) know of other papers, technical journal articles, books, etc. that would help me understand the optical properties of photocells, projectors, etc., please send them along.


Kevin

 

[Guy Kuo]

Before you get too far.... IRE is linear to voltage, not light output. An ideal CRT gamma of 2.5 (giving an overall system gamma gain of 0.3 to compensate for dark field effect. That makes


Light output = c ((IRE -7.5)/ 92.5)^ 2.5


Where c is a proportianality constant, IRE is the input IRE signal level assuming a 7.5 IRE setup for black, and 2.5 is the display gamma.


The non-linear nature of the IRE to light output level is why AVIA has step patterns in monotonic IRE steps and also step patterns in brightness steps. The brightness steps are monotonic for light output assuming a CRT gamma response.




------------------

Guy Kuo
www.ovationsw.com
Ovation Software, the Home of AVIA DVD

 

[steve5097]

Before I saw Guy's reply, I thought I had it wrong (maybe both Guy and I are wrong). I was sure that IRE was a gamma function, so I basicly agree with Guy. One caution is that some devices are set up with 0 IRE as black, in which case his equation needs to be modified.



------------------

Steve

 

[kem]

Guy, Thanks for your insights! Your comments spurred me on to do more research -- mostly reading Charles Poynton's articles and some of his book. I think things are becoming somewhat clearer now (even through the foggy haze in which my brain currently resides [flu] http://www.avsforum.com/ubb/smile.gif ).


Here's what I'm planning for the first step...


As I mentioned earlier, I'm using a homemade light meter made from a Radio Shack CdS photocell plus a DMM. The problem is that the device is not calibrated, and to calibrate my meter, I believe I need a calibrated reference light meter. To calibrate for lux (or foot-candles), I plan to first measure a light source at various intensities using both my meter and the reference meter, and then fit my data to the reference data. I will end up with a function that transforms resistance (W) in Ohms to illumination (E) in lux, which (based on the data from the manufacturer) will probably look something like:


E = k1 * (1/W) ^ p


Let's call this the meter transform function, m(W). I'll fit for k1 and p. I might need to add an offset to make the fit tighter. With m(W), I will have a calibrated light meter (and hopefully saved myself a few $$$s in the process).


I will then measure the light output of the various projectors that I'm evaluating and calculate both lumens and contrast ratios for each one. The measurements will be made from full field gray test patterns generated directly from my HTPC. I chose this option since I can automate the data collection, which I can't do directly with the IRE steps on the AVIA DVD. From what that I've read, PCs do not apply a LUT gamma correction, so the RGB values should be linear in voltage just like the IRE patterns on AVIA.


As I understand it, projectors don't have a simple power transfer function like CRTs; however, they mimic a CRT-like transfer function behavior by using their own LUT. So, I should be able to fit the projector's light output data I collect to a CRT gamma function (as Guy pointed out above):


E = k2 * (g + e) ^ 2.5


where E is illumination in lux (caluclated from m(W) above) and g is the RGB value used to generate the gray pattern normalized to [0,1]. I should then be able to fit for k2 and e. If there are deviations from the CRT-like transfer function, I'd like to correct them (assuming there is a user adjustable gamma setting or a way to modify the LUT in the projector's service menu).


Well, that turned out to be much longer than I originally intended, but at least I got my current thoughts down and hopefully others will scrutinize my methods and correct me where I'm going wrong (or if there's an easier way!).


More later...


Kevin

 

[kem]

BTW, one complication to the plan is that it appears some projectors boost various light output ranges to produce a "better" picture (for some definition of better http://www.avsforum.com/ubb/wink.gif ). For example, NEC when in Natural 1 and Natural 2 gamma correction modes -- they boost the middle values from the preliminary measurements that I've taken.


These deviations from the standard 2.5 gamma transfer function for CRTs will make fitting the curves more difficult. I'm not quite sure what to do about them. However, since ideally I'd just like to be able to make adjustments to the projector's gamma LUT to match the CRT transfer function, my guess is that I don't need to worry too much about these settings; rather, I should just use the normal setting for my calculations (which for NEC seems to be the one that is closest to the CRT transfer function).


Kevin