Warning: All of this is purely theoretical BS and comes from someone with little experience with either a colorimeter or a spectroradiometer (a.k.a., a calibration idiot just trying to learn:))

There has been a brief discussion between UMR and myself in the A3000 settings a tweaks thread about the impact of xvYCC on color calibration, specifically the fact that the wider gamut (primaries are wider) of xvYCC makes calibration more difficult because of the "spikey" spectrum that can be produced.

For instance, the A3000 takes its "natural" primaries which are from the xvYCC (wider than Rec 709) space, and mixes them to make the Rec 709 "primaries" that HD uses. The difference between sets such as the A3000 is that the spectral distribution is different for say, Rec 709 "primary" red than would be for an HD set that is not xvYCC capable and whose "natural" primaries are (theoretically) Rec 709 to begin with. The spectral distribution for an xvYCC display shows a more "spiked" distribution, with more localized peaks and deep valleys in comparison to a non xvYCC display (UMR please correct me if this is wrong).

The result of the more localized peaks/valleys is that a spectroradiometer, taking readings at a discrete interval, can miss the peak/valley and lead to some amount of error. I believe that UMR has found a good amount of error between his PR spectroradiometer and his i1Pro.

After all that, my question is this: Could a tristimulus colorimeter (such as the Display LT) actually be more accurate for these wide gamut displays than a "low" end spectroradiometer if the colorimeter's filters are correct for the "spiked" portions of the spectral distribution?

I would appreciate anyone's comments, especially UMR's since he originally brought up the issue.

Thanks much in advance,
Casey

Theoretically you are correct, if the spectrograph does not have sufficient sampling to properly capture narrow peaks and you have a very carefully calibrated filter instrument specifically for that spectral distribution, it could be more accurate. There are ways to correct for undersampling but this requires doing your own XYZ integrals with the measured spectrum.

There are ways to correct for undersampling but this requires doing your own XYZ integrals with the measured spectrum.

Thanks for the input Zoyd, much appreciated. Do you have an example/info on the integrals you refer to? I don't see how you could mathematically correct for undersampling. Just curious.

The probe outputs XYZ are formed by integrating the measured spectra over wavelength weighted by the functions x-bar, y-bar, and z-bar established by the CIE. If your measured spectrum is undersampled you can try and reconstruct what you should have measured on a finer grid if you know something about what the spectra should look like. The simplest case I can think of is if you have data points that stradle a spectral peak and you know the peak should be gaussian shaped, you can calculate a best-fit gaussian that fits that data and reconstruct the entire feature. The more you know about what the spectral shape should look like the better you can model it and derive errors.

I believe that UMR has found a good amount of error between his PR spectroradiometer and his i1Pro.

The differences between the i1Pro and the PR-670 are due to the bandwidth difference of 10 nm for the i1 Pro vs the PR-670 having 5 nm. The PR-670 also has considerably more data points logged for greater optical resolution then the i1Pro which only returns data from 380-730 with 35 points being logged which is extremely coarse.

The i!Pro will not fair well when measuring displays with spikey spectra. If two peaks lie within 10 nm of each other then the i1Pro will not be able to resolve the peaks and an error in the data will take place. This will lead to a chromaticity error when converting from spectral to XYZ to xyY data. There are a growing number of displays with the various technologies currently available which pose problems for filter based instruments to provide accurate measurement data.

I have compared the i!Pro (10 nm), the PR-650 (3.8 nm), CS-1000 (1 nm) Minolta and a fourth instrument which we will be offering very soon which provides optical resolution of .195 nm per pixel which is five times better then the Minolta CS-1000 while providing 5 nm bandwidth. The differences are quite apparent when all four measure the same display and you view the spd curve (spectral distribution). Current filter based analyzers are not able to adequately assess wide gamut and spikey spectral displays as they are designed to work within a fixed colorspace. Whenever the displays colorspace is significantly larger then the instruments errors will take place.

The probe outputs XYZ are formed by integrating the measured spectra over wavelength weighted by the functions x-bar, y-bar, and z-bar established by the CIE. If your measured spectrum is undersampled you can try and reconstruct what you should have measured on a finer grid if you know something about what the spectra should look like. The simplest case I can think of is if you have data points that stradle a spectral peak and you know the peak should be gaussian shaped, you can calculate a best-fit gaussian that fits that data and reconstruct the entire feature. The more you know about what the spectral shape should look like the better you can model it and derive errors.

Ok, that makes sense. We can't really correct for the error, but we can make an educated guess.

The differences between the i1Pro and the PR-670 are due to the bandwidth difference of 10 nm for the i1 Pro vs the PR-670 having 5 nm. The PR-670 also has considerably more data points logged for greater optical resolution then the i1Pro which only returns data from 380-730 with 35 points being logged which is extremely coarse.

The i!Pro will not fair well when measuring displays with spikey spectra. If two peaks lie within 10 nm of each other then the i1Pro will not be able to resolve the peaks and an error in the data will take place. This will lead to a chromaticity error when converting from spectral to XYZ to xyY data. There are a growing number of displays with the various technologies currently available which pose problems for filter based instruments to provide accurate measurement data.

I have compared the i!Pro (10 nm), the PR-650 (3.8 nm), CS-1000 (1 nm) Minolta and a fourth instrument which we will be offering very soon which provides optical resolution of .195 nm per pixel which is five times better then the Minolta CS-1000 while providing 5 nm bandwidth. The differences are quite apparent when all four measure the same display and you view the spd curve (spectral distribution).

Got it. All of that makes perfect sense. Do you see a good bit of difference once you get past something like a PR-650 to the Minolta, or is it negligible for practical purposes?

Current filter based analyzers are not able to adequately assess wide gamut and spikey spectral displays as they are designed to work within a fixed colorspace. Whenever the displays colorspace is significantly larger then the instruments errors will take place.

I don't understand quite as much about how colorimeters work as I do about spectroradiometers. I didn't know that colorimeters were designed for a specific colorspace. What would something like the Display LT that seems to be popular use for a colorspace, sRGB?

Part of the reason I ask these questions is because I only have access to a colorimeter (I don't have the money or access to a good spectroradiometer). I would like to get my set ISF calibrated by someone with better equipment, but it doesn't make sense for me right now as I am starting to build a new house and will be moving the set to a different environment. In the meantime, I would like to get as close as possible with a colorimeter as I'm sure it has to be more accurate than something like Avia, even with the associated error.

It would be great if someone with a high end spectroradiometer could check the error on the Display LT with one of the new xvYCC sets, just so the rest of us could have an idea of the expected error. I think a lot of people are in the same boat as me for one reason or another, and while we don't expect perfect results, I think we can get closer with a colorimeter than we can eyeballing it with Avia/DVE.

Thanks very much for your input.

The differences between the i1Pro and the PR-670 are due to the bandwidth difference of 10 nm for the i1 Pro vs the PR-670 having 5 nm. The PR-670 also has considerably more data points logged for greater optical resolution then the i1Pro which only returns data from 380-730 with 35 points being logged which is extremely coarse.

The i!Pro will not fair well when measuring displays with spikey spectra. If two peaks lie within 10 nm of each other then the i1Pro will not be able to resolve the peaks and an error in the data will take place. This will lead to a chromaticity error when converting from spectral to XYZ to xyY data. There are a growing number of displays with the various technologies currently available which pose problems for filter based instruments to provide accurate measurement data.

I have compared the i!Pro (10 nm), the PR-650 (3.8 nm), CS-1000 (1 nm) Minolta and a fourth instrument which we will be offering very soon which provides optical resolution of .195 nm per pixel which is five times better then the Minolta CS-1000 while providing 5 nm bandwidth. The differences are quite apparent when all four measure the same display and you view the spd curve (spectral distribution). Current filter based analyzers are not able to adequately assess wide gamut and spikey spectral displays as they are designed to work within a fixed colorspace. Whenever the displays colorspace is significantly larger then the instruments errors will take place.


What entity is "we" that will be offering this new instrument?

Not to nitpick, but I will.

In spectroscopy 0.195 nm/pixel is the sampling rate (or pitch), not resolution. What you refer to as 5 nm "bandwidth" is actually the spectral resolution. Fundamentally, spectral resolution is the instrument response to a delta function in wavelength and is usually expressed as the full-width at half-max (FWHM) of this response. Bandwidth can also mean passband (380-730 nm) so it is not generally used as a term for resolution although you will hear it occasionally in reference to filter instruments. Also, historically bandwidth has been used in the frequency domain, not spectral. Resolution can also be expressed as resolving power at a specific wavelength, e.g. an instrument with 1 nm spectral resolution at 550 nm has a resolving power of 550.


I have compared the i!Pro (10 nm), the PR-650 (3.8 nm), CS-1000 (1 nm) Minolta and a fourth instrument which we will be offering very soon which provides optical resolution of .195 nm per pixel which is five times better then the Minolta CS-1000 while providing 5 nm bandwidth. The differences are quite apparent when all four measure the same display and you view the spd curve (spectral distribution). Current filter based analyzers are not able to adequately assess wide gamut and spikey spectral displays as they are designed to work within a fixed colorspace. Whenever the displays colorspace is significantly larger then the instruments errors will take place.

Not to nitpick, but I will.

In spectroscopy 0.195 nm/pixel is the sampling rate (or pitch), not resolution. What you refer to as 5 nm "bandwidth" is actually the spectral resolution. Fundamentally, spectral resolution is the instrument response to a delta function in wavelength and is usually expressed as the full-width at half-max (FWHM) of this response. Bandwidth can also mean passband (380-730 nm) so it is not generally used as a term for resolution although you will hear it occasionally in reference to filter instruments. Also, historically bandwidth has been used in the frequency domain, not spectral. Resolution can also be expressed as resolving power at a specific wavelength, e.g. an instrument with 1 nm spectral resolution at 550 nm has a resolving power of 550.

Zoyd,
Would you be kind enough to explain exactly what the sampling rate in nm/pixel refers to? I think I understand the spectral resolution, but the sampling rate is new to me.

Zoyd,
Would you be kind enough to explain exactly what the sampling rate in nm/pixel refers to? I think I understand the spectral resolution, but the sampling rate is new to me.

It's determined by the measurement grid and dispersion the instrument uses, so if a spectrograph has a 1024 pixel detector covering a dispersion of 400 nm it's sampling rate (or pitch, or slope) is 400 nm/1024 pixels=0.390625 nm/pixel.

... All of that makes perfect sense. Do you see a good bit of difference once you get past something like a PR-650 to the Minolta, or is it negligible for practical purposes?
...


Based on what I am seeing the final color performance when using the PR-670 is very consistent while most other products like the ones you mentioned are not. These differences are not negligible. With lessor instruments you must hand correct the colors when things are off. I had to do this with the EyeOne Pro and the frequency increased as we have moved to more saturated primaries. I have not found this to be necessary with my PR-670. ISF actually recommends hand correction in the course, but I think most people ignore it. As we enter the world of CMS hand color correction becomes very problematic and very high quality instruments are required if you want consistently good results. It is difficult enough to obtain a good D65 reference much less RGBCMY as well.

I have purchased many filter based units and have seen the results of numerous others. I do not hold out much hope for the low cost instruments or even high cost ones with poor resolution.

I have compared the i!Pro (10 nm), the PR-650 (3.8 nm), CS-1000 (1 nm) Minolta and a fourth instrument which we will be offering very soon which provides optical resolution of .195 nm per pixel which is five times better then the Minolta CS-1000 while providing 5 nm bandwidth. The differences are quite apparent when all four measure the same display and you view the spd curve (spectral distribution). Current filter based analyzers are not able to adequately assess wide gamut and spikey spectral displays as they are designed to work within a fixed colorspace. Whenever the displays colorspace is significantly larger then the instruments errors will take place.Can you give some idea of the size of the xy deviation with these instruments under specified conditions? Also, what's this mysterious 4th instrument you refer to?

What frustrates me with discussions like this is a noted lack of hard data.

What frustrates me with discussions like this is a noted lack of hard data.

I'll second that. Folks with higher end equipment seem unwilling to share data that would quantify the benefit (or lack thereof) of various probe technologies used to measure various display technologies. I can not see how such interaction would damage your investment since most of us are not interested in the commercial aspects of display calibration.

...
What frustrates me with discussions like this is a noted lack of hard data.


Here is some comparing the EyeOne Pro to the PR-670 for the last 10 displays I worked on. These included LCD, Plamsa, DLP and SXRD units. I have other data for various instruments, but the number of displays is much more limited.

I offered Sencore the opportunity to prove their new filter based unit was the equal of my PR-670 which they have been claiming. They said they would schedule a shootout at CEDIA with me, but never did. I suspect they knew what the outcome would be.

Here is some comparing the EyeOne Pro to the PR-670 for the last 10 displays I worked on. These included LCD, Plamsa, DLP and SXRD units. I have other data for various instruments, but the number of displays is much more limited.


Thanks, which displays generated your min/max values and do they correlate with unusually peaked spectral distribution functions? It would also be useful to include Y errors so that we can calculate deltaE(uv) for comparison.

It's determined by the measurement grid and dispersion the instrument uses, so if a spectrograph has a 1024 pixel detector covering a dispersion of 400 nm it's sampling rate (or pitch, or slope) is 400 nm/1024 pixels=0.390625 nm/pixel.

I see. So when we are measuring a certain color bar that should be a constant chromacity with regards to space, the spectral resolution would be much more important that the sampling rate, correct?

Here is some comparing the EyeOne Pro to the PR-670 for the last 10 displays I worked on. These included LCD, Plamsa, DLP and SXRD units. I have other data for various instruments, but the number of displays is much more limited.

I offered Sencore the opportunity to prove their new filter based unit was the equal of my PR-670 which they have been claiming. They said they would schedule a shootout at CEDIA with me, but never did. I suspect they knew what the outcome would be.

Thanks for the report UMR, I think that is what we have been looking for. This will at least give us an idea of how bad the error might be.

Do you have any "spikey" spectrums saved as taken with the PR-670?

filter based colorimeters can see outside standard gamuts just fine - they have to being that their optical filters are designed to match the eyes response function. In fact if you are not careful with filters they can see outside the visual spectrum into IR/UV. They don't do any spectral sampling such that they could miss a narrow peak. If you can see it - an optical filter that matches standard eye response can see it - it is all about how you stack optical filters to form the standard curve. If you had the perfect filters - you would only need three - one for each of X, Y, Z - but usually three filters will be inaccurate - so better to composite the curves using more filters.

you can read a bit of blurb here on "spectral fitting method" http://se.konicaminolta.us/products/product_brochures/cs_200.pdf

They are however calibrated because the filters may not be perfect - they are more or less at different points on the curve - so it is possible a spectral spike could hit a peak or dip in the error of the optical curve - which on a broaderband white light source these errors averaged out. . Their accuracy is solely determined based on how well their calibration matches the standard eye response function - not a standard gamut. The only way a standard gamut is involved is they may have used it to come up with the calibration. They could just as well use an expanded gamut display to calibrate against lab grade instruments or by using a standard light source with no display at all - they could use tunable lasers and spectral sampling to calibrate their filters if they wanted to - unlikely when they are $99 ebay specials....

I am still waiting for UMR to go back and recalibrate all of his customers since he apparently used innaccurate instruments for years before getting a PR670 - but I guess he is too busy convincing TV buyers to hire him because he is the only calibrator with accurate instruments...and that includes all the calibrators he sold inaccurate gear to...

They are however calibrated because the filters may not be perfect - they are more or less at different points on the curve - so it is possible a spectral spike could hit a peak or dip in the error of the optical curve - which on a broaderband white light source these errors averaged out. . Their accuracy is solely determined based on how well their calibration matches the standard eye response function - not a standard gamut.

This is really the only drawback to using filter instruments, because you can't exactly match the eye response functions you must calibrate them for each unique spectral distribution you want to deploy them on. To supply calibration coefficients for all the different technologies out there would price them out of the low cost market. (not to mention that ultrastable ion-assisted deposition filters currently run about 5K a pop)

Right it is not the filters are incapable of seeing spiky distributions as some think (because how can 3 see more than 256 or 1000) - it is that proper calibration is expensive. But there are those that will use ones that can be easily calibrated because they are more portable than NIST certified lab grade spectros with high resolution that they calibrate against - they just bring it in every month and tune it for their light sources to make sure field use is not drifting it - so it stays NIST traceable.

I certainly would take the linked KM colorimeter over a fleabay i1 spectro anyday.

I see. So when we are measuring a certain color bar that should be a constant chromacity with regards to space, the spectral resolution would be much more important that the sampling rate, correct?

They are both just as important and tied together, the sampling because you don't want to undersample a peak and miss important structure and the resolution because you don't want to wash out the structure and dump it into a different color bin. A general design rule is that the resolution should be 1/10 your narrowest feature and the sampling should be a minimum of 3 samples per resolution element (Nyquist says 2 but that is insufficient in practice). So if you have spectra with 10 nm wide peaks/valleys you should shoot for 1 nm resolution and at least 0.33 nm/pixel sampling.

Right it is not the filters are incapable of seeing spiky distributions as some think (because how can 3 see more than 256 or 1000) - it is that proper calibration is expensive. But there are those that will use ones that can be easily calibrated because they are more portable than NIST certified lab grade spectros with high resolution that they calibrate against - they just bring it in every month and tune it for their light sources to make sure field use is not drifting it - so it stays NIST traceable.

I certainly would take the linked KM colorimeter over a fleabay i1 spectro anyday.

I have seen the results of a KM CS-200 and and a CA-210 and I was unimpressed. NIST traceable is only to Luminant A which is meaningless for any of the light that comes off these displays. The problem with filter based units I have seen is that the designs do not match the XYZ curves close enough to work for multiple display types. They must be calibrated to a specific spectral power distribution to be accurate. Once you start trying to measure RGBYMC that falls apart. Note in the specification for the CA-210 it is for an LCD specifically with a limited CCT range.

...I am still waiting for UMR to go back and recalibrate all of his customers since he apparently used innaccurate instruments for years before getting a PR670 - but I guess he is too busy convincing TV buyers to hire him because he is the only calibrator with accurate instruments...and that includes all the calibrators he sold inaccurate gear to...

People knew what I had when I worked on there displays and the PR-670 was not available until I bought it. I waited in line behind many others to get one. Sticking with old technology is stupid when a better unit is available.

Thanks for the report UMR, I think that is what we have been looking for. This will at least give us an idea of how bad the error might be.

Do you have any "spikey" spectrums saved as taken with the PR-670?

Yes. I posted one on another thread. Search my posts you should find it.

Originally Posted by zoyd View Post
Not to nitpick, but I will.

In spectroscopy 0.195 nm/pixel is the sampling rate (or pitch), not resolution. What you refer to as 5 nm "bandwidth" is actually the spectral resolution. Fundamentally, spectral resolution is the instrument response to a delta function in wavelength and is usually expressed as the full-width at half-max (FWHM) of this response. Bandwidth can also mean passband (380-730 nm) so it is not generally used as a term for resolution although you will hear it occasionally in reference to filter instruments. Also, historically bandwidth has been used in the frequency domain, not spectral. Resolution can also be expressed as resolving power at a specific wavelength, e.g. an instrument with 1 nm spectral resolution at 550 nm has a resolving power of 550.

Zoyd,

Please take a look at the specifications of other manufacturers of spectroradiometers to see the terminology which they use and the values associated with them! I do not generally disagree with your statements however both PhotoResearch and KonicaMinolta specify their instruments using the terminology which I used in my post. I personally refer to the "Spectral Resolution" as "Optical Resolution" as it is simply dividing the number of pixels in the detector by the wavelength range of the instrument.

http://www.photoresearch.com/current/pr670.asp?type=2



TomHuffman

[QUOTE]What frustrates me with discussions like this is a noted lack of hard data.

Tom,

Attached are files made from my Viewsonic VP201b computer monitor which demonstrate the bandwidth and resolution of the various instruments which I mentioned in my earlier posting. I will try to find a better display example which which will show how bandwidth is important to measurement of spikey spectral displays. You will find that their are displays which will have spikes which may lie closer together then the bandwidth of the test instrument which will in turn make it impossible to display both peaks of the spectra in the display as they will overlap causing errors in the data to take place.



hwjohn

It would be great if someone with a high end spectroradiometer could check the error on the Display LT with one of the new xvYCC sets, just so the rest of us could have an idea of the expected error. I think a lot of people are in the same boat as me for one reason or another, and while we don't expect perfect results, I think we can get closer with a colorimeter than we can eyeballing it with Avia/DVE.

The results of this type of test would vary for every display which you compared the spectroradiometer to the EyeOne Display 2/ LT as the filter based products accuracy/error varies with each display.



What entity is "we" that will be offering this new instrument?

"We" refers to a new product offering to be released in the next few weeks by Progressive Labs. The spectroradiometer will be called the Microspec and we will provide complete specifications for this instrument as soon as we make it available with details on our website.

Here are some example spectra from my 670. Note the similarity of my LCD monitor to the CS-1000 graph.

Please take a look at the specifications of other manufacturers of spectroradiometers to see the terminology which they use and the values associated with them! I do not generally disagree with your statements however both PhotoResearch and KonicaMinolta specify their instruments using the terminology which I used in my post. I personally refer to the "Spectral Resolution" as "Optical Resolution" as it is simply dividing the number of pixels in the detector by the wavelength range of the instrument.


It's not unusual for manufacturers to be sloppy in their terminology and even though I understand what the intention is it's still incorrect. You can crack open any text on spectroscopy and find that the correct definitions are the ones I posted, see for example "Modern Classical Optics" by Geoffrey Booker, section 4.7 eqns. 4.4 and 4.5

note that the xrite folks get it right, their i1 brochure states an optical resolution of 10nm and physical sampling interval of 3.5 nm

Here are some example spectra from my 670. Note the similarity of my LCD monitor to the CS-1000 graph.

So that we might extract some useful information out of this thread, can you expand on your 10 display shootout results and post which displays type head each column and what the Y errors were for each display.

Zoyd,

It seems odd that the I1Pro Physical sampling interval of 3.5 nm is not utilized in the output of data from the instrument and is provided as 36 wavelengths with the associated spectral radiance in 10 nm intervals. Neither the brochure or the SDK advise the number of detector elements in the instrument anywhere! If the piece had a sampling interval of 3.5 nm then they are really doing themselves and the users a huge disservice by not providing this data in the higher resolution which you seem to infer it is capable of from the specifications. The 3.5 nm interval would suggest that the instrument would need to have a detector with no fewer then 1225 elements (3.5 X 350= 1225)which seems to be a rather odd number for a detector to have don't you think? I realize you are only quoting from the printed specifications so this is not a disagreement with you on any level.

Here are the specifications of the instrument as it appears in the SDK.

General (Eye-One Pro, Eye-One Monitor)

Spectral analyzer: holographic diffraction grating with diode array
Spectral range: 380 – 730 nm
Optical bandwidth: 10 nm
Sampling interval: 3.5 nm (100 bands)
Spectral reporting: 10 nm
Analog / digital converter: 16 bit

We are using the manufacturers development tools and I can assure you that there is no way of increasing the number of data points returned to a value higher then 36! If the sampling interval is every 3.5 nm then even if they supplied data in 4.0 nm increments they would supply at least 87 points to log/plot which is not the case.

Here is a typical data return from the i1Pro which we have in our software package. Please notice that the increment is 10 nm!

Wavelength (nm) Spectral Radiance (mw/cm2-sr-um)
380 0.003382836
390 0.002987082
400 0.019512087
410 0.066484712
420 0.285551339
430 1.145173311
440 1.947875738
450 1.676095605
460 1.577716351
470 1.300804257
480 1.520812154
490 2.207514286
500 1.187126756
510 0.381644726
520 0.188358396
530 0.273836195
540 3.266561985
550 3.691952467
560 0.419149816
570 0.157152101
580 0.917529404
590 1.263704419
600 0.662866116
610 2.186607838
620 1.733013988
630 0.67607379
640 0.203153193
650 0.198510468
660 0.158313528
670 0.150739059
680 0.139249682
690 0.117558599
700 0.119688831
710 0.289236009
720 0.075971581
730 0.013550377

Quote:
Originally Posted by umr View Post
Here are some example spectra from my 670. Note the similarity of my LCD monitor to the CS-1000 graph.

So that we might extract some useful information out of this thread, can you expand on your 10 display shootout results and post which display types gave the i1pro the most trouble and what the xyY errors were for each display.

The Plasma sample spectra would be a typical problem for a spectroadiometer such as the i1Pro. Notice that with the PR-670 the two red peaks which are clearly displayed in the graph this is with an instrument with 5 nm spectral resolution and sampling interval of 1.56 nm. If this same display were sampled by the i1Pro the peak wavelength centered at 626 nm would very likely become overlapped with the Red peak adjacent to it which lies at approximately 616 nm.

Zoyd,

It seems odd that the I1Pro Physical sampling interval of 3.5 nm is not utilized in the output of data from the instrument and is provided as 36 wavelengths with the associated spectral radiance in 10 nm intervals. Neither the brochure or the SDK advise the number of detector elements in the instrument anywhere! If the piece had a sampling interval of 3.5 nm then they are really doing themselves and the users a huge disservice by not providing this data in the higher resolution which you seem to infer it is capable of from the specifications.


The brochure (http://www.xrite.com/documents/literature/en/420139_i1System_Brochure_en.pdf) I looked at yesterday clearly states they are using a 128 element photo-diode detector. 128x3.5 nm/pixel= 448 nm coverage. They only report 380nm to 730nm which means they have 100 pixels to work with over that range. Why do they report only 35 values? I don't know but I would bet that each reported point is an average over 3 pixels to increase signal-to-noise. If they are doing that their effective resolution will be degraded to ~3x3.5nm = 11.5 nm. In any case my original point stands, resolution is the response to a delta function in wavelength and what you have been calling resolution is actually sampling interval (10 nm and 3.5 nm respectively in the i1pro's case).


The 3.5 nm interval would suggest that the instrument would need to have a detector with no fewer then 1225 elements (3.5 X 350= 1225)which seems to be a rather odd number for a detector to have don't you think?


Need to revisit your math here, it should be 350 nm/(3.5 nm/pixel)=100 pixels. (It's always a good idea to include and check your units.);)

The Plasma sample spectra would be a typical problem for a spectroadiometer such as the i1Pro. Notice that with the PR-670 the two red peaks which are clearly displayed in the graph this is with an instrument with 5 nm spectral resolution and sampling interval of 1.56 nm. If this same display were sampled by the i1Pro the peak wavelength centered at 626 nm would very likely become overlapped with the Red peak adjacent to it which lies at approximately 616 nm.

Yes, I see where the potential problems are and I have discussed them in previous posts when comparing the i1pro to D2s and spyders, but until you or umr posts data which show the xyY errors on a specific display, looking at spectra is pointless.

The design of a spectrophotometer is much more complex than simply the number of sensors. Here are some articles for those interested in the concepts and limitations of each approach.

http://www.jeti.com/down/basics/basics.pdf

http://www.jeti.com/down/basics/basics1.pdf

http://www.jeti.com/down/basics/basics2.pdf

http://www.jeti.com/down/basics/basics3.pdf

The design of a spectrophotometer is much more complex than simply the number of sensors.

This statement, while obviously true, is an inaccurate representation of the previous discussion. It implies that neither ghbliss nor I know what we our talking about which is not the case. You attempt to set yourself above others instead of engaging in the discussion in an honest manner.

The discussion of the terminology used to express instrument resolution and sampling does not require anything more in-depth than what we covered. If you wish to discuss more complex issues I would be more than willing to. Start another thread on the topic and I will post some of my own designs if there is any interest.

Back to the original intent of this thread, will you or will you not post your full data set so we can evaluate probe errors against various SPD's? The same question goes to ghbliss if he has similar data, especially with respect to the Display 2/LT since that was the OP's original interest, but Tom and I are most interested in the i1pro performance.

Zoyd

Back to the original intent of this thread, will you or will you not post your full data set so we can evaluate probe errors against various SPD's? The same question goes to ghibliss if he has similar data, especially with respect to the Display 2/LT since that was the OP's original interest, but Tom and I are most interested in the i1pro performance.

I just took measurements from my Viewsonic LCD computer monitor and the data which follows shows significant deviation between the i1Pro and the D2 on the y axis whereas the x axis is virtually identical between the two instruments. The brightness difference for Y is almost 10% with the D2 reading higher.

D2

x .3052
y .3337
Y 28.786
K 6,864


i1Pro

x .3053
y .3246
Y 25.909
K 6,962


i!Pro Spectral Data

Wavelength (nm) Spectral Radiance (mw/cm2-sr-um)
380 0.003382836
390 0.002987082
400 0.019512087
410 0.066484712
420 0.285551339
430 1.145173311
440 1.947875738
450 1.676095605
460 1.577716351
470 1.300804257
480 1.520812154
490 2.207514286
500 1.187126756
510 0.381644726
520 0.188358396
530 0.273836195
540 3.266561985
550 3.691952467
560 0.419149816
570 0.157152101
580 0.917529404
590 1.263704419
600 0.662866116
610 2.186607838
620 1.733013988
630 0.67607379
640 0.203153193
650 0.198510468
660 0.158313528
670 0.150739059
680 0.139249682
690 0.117558599
700 0.119688831
710 0.289236009
720 0.075971581
730 0.013550377

While the spectra are interesting - what most casual readers will not know is that it is not what our eyes see. Our ears are spectral analyzers - our eyes are not. What the eyes see has an analogy to SPL for the ear - the total "loudness" of the spectra - called "lightness" for the eye.

Basically it is the area under the spectral curve - weighted with filters. In the ears case SPL(A), SPL(C) are examples. For the eye the filters are X, Y, and Z - with each filter peaking at different wavelengths - and our brain combines these "lightness" functions into colors. Play three notes to the ear and we hear a chord - but chords do not exist to the eye - our brain sees this as only one color. In fact many spectra map to the same color in the brain - whereas our spectral analyzing ears would interpret these as tones of different timbre.

So while it is interesting to brag about fantastic spectral resolution and how you can see the curve better - this is only interesting to the scientest with a need to do spectral analysis. All your eye cares about is the area under the curve. If a lower sampling rate misses a spectral peak that occurs in between samples - it is only significant if the energy contained in the peak is significant portion of the rest of the energy underneath the peak. If subsampling caught that energy and added it into the coarser sampling - that is better than missing it entirely with the coarser sampling.

A tristimulus filter will not miss the peak - it is optical. Its only source of error comes from the coarse approximation of the eye filters - poor quality filters combined with poor quality digital calibration could lead to the peak contributing more/less than it should. A spectro has an accurate digital eye filter - it's limitation is in the coarse spectral sampling impacting the intensity (total curve area) measures. In both cases the GIGO principle applies.

So when the above data shows a spectro was off in Y - then the missing energy it did not see was significant. Or it could mean the spectro saw sufficient energy to be within tolerances - and it is the tristimulus meter that had a digital calibration for the eyes filter that just happened to be over/under the spectral peak causing significant differences in energy. One design is not automatically better than the other - be they cheap or expensive implementations. This is like the age old vinyl vs. CD arguments or graphic EQ vs. tone controls. More precision than the eye can see - is not better - just more expensive.

Zoyd
I just took measurements from my Viewsonic LCD computer monitor and the data which follows shows significant deviation between the i1Pro and the D2 on the y axis whereas the x axis is virtually identical between the two instruments. The brightness difference for Y is almost 10% with the D2 reading higher.


Thanks, your data shows a dE(u*v*) deviation of 7.7 between the i1pro and the D2 on the LCD. I have seen offsets between the two on a plasma (dE(u*v*)=4.4) and posted extensively on it in this (http://www.avsforum.com/avs-vb/showthread.php?t=844188) thread which I will update with RGBYCM comparisons when I get a chance. Do you have any i1pro vs. next tier probe comparisons?

Here are umr's results expressed as dE(u*v*), if the Y results get posted I'll update these numbers.


Display 1 2 3 4 5 6 7 8 9 10
R 1.0 3.1 4.6 1.0 2.4 5.5 1.8 1.7 3.4 1.9
G 5.6 4.2 5.1 3.5 5.4 4.0 6.2 2.4 6.2 7.3
B 4.5 5.4 3.7 1.9 4.7 4.3 5.3 2.4 6.3 4.8
Y 1.8 0.6 1.6 2.1 0.5 0.8 3.0 1.4 3.2 7.6
C 3.1 4.2 2.0 2.2 2.7 3.4 3.0 2.2 3.6 4.8
M 4.4 3.9 4.5 1.5 4.7 4.8 5.4 3.0 6.7 7.8
W 1.3 1.3 1.4 3.0 1.0 3.4 1.1 1.7 6.6 6.1
avg 3.1 3.2 3.3 2.2 3.1 3.7 3.7 2.1 5.1 5.8


Display 9 and 10 gave the i1pro the hardest time, the others don't seem to warrant all the fuss given that state-of-the-art in CMS is an average dE of 4 anyway. Also note the agreement in establishing the white point for displays 1,2,3,5,7,8 is probably within the error bar of each probe.

update: The PR specs for accuracy are +/- 0.0015 in x,y and +/-2% on Y measuring illuminant A type spectral distributions. This translates into a dE(u*v*) accuracy assuming a +x shift and a -y shift of

RGBYCMW: 1.9, 1.0, 1.7, 1.6, 1.8, 2.6, 2.2

I could not find similar specs for the i1pro, only repeatibility of +/- 0.002 (again for illuminant A)

Kraz,

Although your input is valued, unless you have an LMT Berlin tristimulus analyzer then we should not really say that filter based analyzers are just as good as a spectroradiometer! An LMT Berlin costs approximately $50K and I only know of a few manufacturers that own these due to their extremely high cost. This piece is exceptionally accurate due to its hand cut filters which have a tolerance of less then 1% error which is far better then any competitors products provide by a wide margin!

You are correct in saying that filter based analyzers display errors due to the way which they are calibrated however this is a double edged sword so to speak. The calibration is performed using displays which are supposed to be typical to the type which are found in the field and used by the customer purchasing the instrument. In some cases multiple calibrations are saved to the eeprom and may be selected by the user. The problems which the filter based analyzers do not perform well with are from displays which have primaries which lie far outside of the range which the instrument was designed to work within. A recalibration of the probe for that particular display is possible however it will create an error for use with the majority of more "typical" displays which the user may encounter in day to day work.

Manufacturers are attempting to address this issue with filter based analyzers however it is a complex mathematical problem which there is no easy solution to at this time. Until a solution is found which provides accuracy within a very tight tolerance (.002 for x and y) for a filter based analyzer for use with "all display types" the high bandwidth/high resolution spectroradiometer provides the only accurate means of measurement for display analysis for the professional calibrator.

Filter based analyzers aside from the most expensive models made provide varying degrees of accuracy on each and every display type made. The average "Joe" that does not own a spectroradiometer can not possibly know how much deviation his display has after it has been "calibrated" using a filter based analyzer without the display being assessed with a high quality spectroradiometer. The issues are greater with LCD and Plasma displays due to the spectral output of these devices as compared to CRT and DLP based technologies at this time.

I realize that simply looking at spectral data does not tell the entire story of the display being measured. You can however determine a great deal about what is going on from analyzing this information especially in the context of this discussion as to why the filter based products are coming up short when measuring certain display types. My measurements are of just one LCD display however this is quite typical of LCD panels in general and since this is presently the largest display category in the world it seems to be appropriate to me to use as an example. The y axis error is substantial and would not provide an acceptable calibration result for any readers of this forum!

Zoyd,

Display 9 and 10 gave the i1pro the hardest time, the others don't seem to warrant all the fuss given that state-of-the-art in CMS is an average dE of 4 anyways. Also note the agreement in establishing the white point for displays 1,2,3,5,7,8 is probably within the error bar of each probe.

Although the White point deviation is not substantial when comparing these displays the Green and Blue primaries do show substantial error across the board which are noticeable. Half of the ten displays show de of 5.0 or more for Green with the Blue not to far behind! For anyone trying to perform a CMS (color management system) calibration on a display they would likely have a difficult to nearly impossible time of dialing in the primaries and secondaries. The problem would more then likely be compounded on the LCD and Plasma displays.

The point to be made is that each display has its own unique characteristics and generalities do not serve the calibrator very well. You need to be properly equipped to calibrate any display which you may come across. Adding the varying degree of uncertainty to deal with can generally create a major waste of time for both customer and calibrator as they are unable to achieve the level of performance which they expect to.

Zoyd,
Although the White point deviation is not substantial when comparing these displays the Green and Blue primaries do show substantial error across the board which are noticeable. Half of the ten displays show de of 5.0 or more for Green with the Blue not to far behind! For anyone trying to perform a CMS (color management system) calibration on a display they would likely have a difficult to nearly impossible time of dialing in the primaries and secondaries. The problem would more then likely be compounded on the LCD and Plasma displays.
The data set includes LCDs and plasmas but he won't tell us which is which. Other than displays 9 and 10 which you might make a case for (assuming the display even has a CMS), the idea of spending 23K to achieve a dE reduction of 1-2 where the majority of displays don't even have a CMS and those that do achieve average dE=4 is preposterous.


The point to be made is that each display has its own unique characteristics and generalities do not serve the calibrator very well. You need to be properly equipped to calibrate any display which you may come across. Adding the varying degree of uncertainty to deal with can generally create a major waste of time for both customer and calibrator as they are unable to achieve the level of performance which they expect to.
I have no problem with this philosophy but this thread is not about the traveling calibrator biz. The implication of umr's posts, and yours to some extent, was that we mere mortals using an i1pro will by definition not achieve accurate results and that is just silly as umr's own data demonstrates.

I agree with you that $23K is crazy money to spend which is why I have been working on a spectroradiometer which will meet or beat its performance in every respect for less then one third the money! This is obviously not geared towards the casual hobbyist however it will provide the performance matching the best instruments available regardless of cost!

There are other shortcomings of the i1Pro aside from bandwidth etc. such as poor low light level performance which makes it difficult to achieve good Grayscale alignment. Most users of this instrument all agree on this point however the piece does have its place in the calibration arena. It is made for use primarily for calibration of pc monitors for desktop publishing. Expanding the use of this piece for calibration of home theater applications is sometimes a bit of a stretch as it was never intended for this when it was designed.

I agree with you that $23K is crazy money to spend which is why I have been working on a spectroradiometer which will meet or beat its performance in every respect for less then one third the money! This is obviously not geared towards the casual hobbyist however it will provide the performance matching the best instruments available regardless of cost!

There are other shortcomings of the i1Pro aside from bandwidth etc. such as poor low light level performance which makes it difficult to achieve good Grayscale alignment. Most users of this instrument all agree on this point however the piece does have its place in the calibration arena. It is made for use primarily for calibration of pc monitors for desktop publishing. Expanding the use of this piece for calibration of home theater applications is sometimes a bit of a stretch as it was never intended for this when it was designed.

Looking forward to this product. I would seriously consider purchasing this.

I have thought for some time that the market was poorly covered. You have $200 colorimeters, such as the D2, the $800 i1Pro, and then nothing until the $11,000 Minolta CS-200.

Clearly there is a big hole in the market between $800 and $11,000 that I would have thought would have been filled long ago.

Sencore has just stepped into this market with the $7,000 OTC1000, but I haven't heard any field reports about this, so I have no way of knowing whether it offers any significant improvements over the i1Pro.

…the idea of spending 23K to achieve a dE reduction of 1-2 where the majority of displays don't even have a CMS and those that do achieve average dE=4 is preposterous.
If it’s generally accepted that the average HVS perception threshold for short term, non-comparative color discrimination is ~dE 5, then spending even $23 would seem below the level practical, perceptible improvement but I suppose it could be argued that 1-2dE could move it from a just perceptible difference to a not quite perceptible one.

Using a low cost (cheap) filter based sensors that may have a dE inaccuracy on the order of 4-5 seems pretty reasonable considering the initial display colorimietery error is often >20 dE. If it requires $23K (or even $7K) of instrumentation to move a > 20dE error to < 2 dE error then that seems a very high price if a <$100 filter based instrument can move that >20dE to <5dE. Thats way too much a $/dE for me.

Zoyd, as a curios aside- you measured the tri-filter based TAOS sensor as having a dE < 2 compared to your i1 over the D65 grayscale on your plasma. Any guess on how it would compare over the RGBCMY gamut? Just a curiosity…

Dave

There are other shortcomings of the i1Pro aside from bandwidth etc. such as poor low light level performance which makes it difficult to achieve good Grayscale alignment. Most users of this instrument all agree on this point however the piece does have its place in the calibration arena.
I agree this is difficult and a hindrance to the casual user, but not a showstopper if you know what you're doing.


It is made for use primarily for calibration of pc monitors for desktop publishing. Expanding the use of this piece for calibration of home theater applications is sometimes a bit of a stretch as it was never intended for this when it was designed.
It doesn't appear to be a stretch at all for most of the displays above and even for the last two, unless they have color management you're not going to do better than that anyway with the exception of the white point.

Zoyd, as a curios aside- you measured the tri-filter based TAOS sensor as having a dE < 2 compared to your i1 over the D65 grayscale on your plasma. Any guess on how it would compare over the RGBCMY gamut? Just a curiosity…

Dave

That's my next project, update the D2/spyder/TAOS results with RGBCMY comparisons.:)

That's my next project, update the D2/spyder/TAOS results with RGBCMY comparisons.:)

Cool. I'll keep my one good eye out for it. ;)

Dave

Here are umr's results expressed as dE(u*v*), if the Y results get posted I'll update these numbers.


Display 1 2 3 4 5 6 7 8 9 10
R 1.0 3.1 4.6 1.0 2.4 5.5 1.8 1.7 3.4 1.9
G 5.6 4.2 5.1 3.5 5.4 4.0 6.2 2.4 6.2 7.3
B 4.5 5.4 3.7 1.9 4.7 4.3 5.3 2.4 6.3 4.8
Y 1.8 0.6 1.6 2.1 0.5 0.8 3.0 1.4 3.2 7.6
C 3.1 4.2 2.0 2.2 2.7 3.4 3.0 2.2 3.6 4.8
M 4.4 3.9 4.5 1.5 4.7 4.8 5.4 3.0 6.7 7.8
W 1.3 1.3 1.4 3.0 1.0 3.4 1.1 1.7 6.6 6.1
avg 3.1 3.2 3.3 2.2 3.1 3.7 3.7 2.1 5.1 5.8
Display 9 and 10 gave the i1pro the hardest time, the others don't seem to warrant all the fuss given that state-of-the-art in CMS is an average dE of 4 anyway. Also note the agreement in establishing the white point for displays 1,2,3,5,7,8 is probably within the error bar of each probe.
I got similar results using CIELAB with idealized Y data. The biggest dE was 5.6, the smallest 2.2, with the average at 4.1. Errors were larger in blue and green than in red.

Average dE
1----2----3----4----5----6----7----8---9---10
3.8--4.0--3.6--2.2--4.1--4.2--4.8--2.2--5.6--5.6

I'd wager that display 10 is an LCD, 9 is an SXRD, and either 3 or 6 (or both) are plasmas.

I don't know why I waste my time on this. Here is what they were. CMS and wide gamut is where the biggest problems are. It is very likely when you adjust a CMS on a high quality dispaly with a low quality instrument you are making it worse. I see people posting CMS adjustments on this forum all the time and they are never very close to what the PR-670 is recommending.

#10 DLP (new wide gamut)
#9 SXRD (new wide gamut)
#8 Plasma
#7 Plasma
#6 SXRD
#5 LCD
#4 SXRD
#3 Plasma
#2 SXRD
#1 SXRD

...You are correct in saying that filter based analyzers display errors due to the way which they are calibrated however this is a double edged sword so to speak. The calibration is performed using displays which are supposed to be typical to the type which are found in the field and used by the customer purchasing the instrument. In some cases multiple calibrations are saved to the eeprom and may be selected by the user. The problems which the filter based analyzers do not perform well with are from displays which have primaries which lie far outside of the range which the instrument was designed to work within. A recalibration of the probe for that particular display is possible however it will create an error for use with the majority of more "typical" displays which the user may encounter in day to day work.

Manufacturers are attempting to address this issue with filter based analyzers however it is a complex mathematical problem which there is no easy solution to at this time. Until a solution is found which provides accuracy within a very tight tolerance (.002 for x and y) for a filter based analyzer for use with "all display types" the high bandwidth/high resolution spectroradiometer provides the only accurate means of measurement for display analysis for the professional calibrator.

Filter based analyzers aside from the most expensive models made provide varying degrees of accuracy on each and every display type made. The average "Joe" that does not own a spectroradiometer can not possibly know how much deviation his display has after it has been "calibrated" using a filter based analyzer without the display being assessed with a high quality spectroradiometer. ...

I agree. This matches my experience with filter based units. I have owned five different units with one costing $6000 using custom glass filters and none came close to the EyeOne Pro.

My data is only the last 10 displays. The maximum observed white point error has been 0.015 for either x or y between the EyeOne Pro and the PR-670. The error is frequently low for white, but it can be very large. Implying that all displays will exhibit errors within the tolerance shown is foolish.

I agree with you that $23K is crazy money to spend ....


Cliff,

You have spent much more than that for your PR-650 and CS-1000A. I just got sick of the problems with lower cost instruments. $24,000 seemed like a small price to me to take the measurement problems out of the picture.

Hopefully, your new product will be as feature rich as the 670. A big part of the value is in that as well as the measurement quality.

I don't know why I waste my time on this.
I'd give that a 10 on the pompous scale.(NIST traceable of course)

UMR

You have spent much more than that for your PR-650 and CS-1000A. I just got sick of the problems with lower cost instruments. $24,000 seemed like a small price to me to take the measurement problems out of the picture.

Hopefully, your new product will be as feature rich as the 670. A big part of the value is in that as well as the measurement quality.

Fortunately I purchased three PR-650 several years ago and sold two pieces keeping one with a slew of accessories. The cost to me was net $0. My Minolta CS-1000 was purchased used from a display manufacturer for $5K and is in mint "as new" condition! I would never pay the price which these instruments cost new from the factory which I find to be a bit extreme. I also own a PR-880 digital photometer (list price $35K) which is used for light level measurement and is specified by the factory to read to a light level of .0002 fL. I use this to verify our own spectroradiometers. This piece was purchased used for a small amount as well. I feel extremely fortunate to own all of this gear for as little as I have invested.

The equipment is an absolute requirement as far as I am concerned to be in the manufacturing end of this business. I would be unable to verify the accuracy of any display measurements without it. The cost of keeping these pieces in calibration is also considerable!

My new spectroradiometer will be supplied with the current software package which my CA-6X and C-5 customers presently enjoy using with a few enhancements to support the spectroradiometer functions. I do not think the PR-670 will provide anything that my software package doesn't already provide. I have spent the past two years working on making this instrument a match for the best devices in the industry and believe that I have succeeded with this piece. The measurement accuracy, repeatability, low light level sensitivity, short integration times, ease of use, and feature set are all as good or better then the benchmark instruments which are presently available.

My data is only the last 10 displays. The maximum observed white point error has been 0.015 for either x or y between the EyeOne Pro and the PR-670. The error is frequently low for white, but it can be very large. Implying that all displays will exhibit errors within the tolerance shown is foolish.

You posted the table as a representative set of xy errors and a 0.015 deviation is not consistent with your results, why? Was is a bad probe, spectral drift, unusual display gamut? You now argue against paying attention to your own results.

It is very likely when you adjust a CMS on a high quality display with a low quality instrument you are making it worse.This assertion is not supported by the data. Based on the limited data you provided, the difference between the i1Pro and the PR Spectro was on average about 4 dE and under 6 dE in the worst case--not nothing to be sure, but only a modest difference and not one that seems to me cost effective.

Unadjusted displays often employ primaries with dEs over 20. Even an instrument of lower quality than I am comfortable using could easily achieve dramatic improvements. The difference between 20 dE and 6 dE is night-and-day. The difference between 6 dE and 2 dE is close to the threshold of minimal perceivable difference.

One member just used a POS Spyder2 with the new Radiance external CMS to reign in the primaries on his RS1. The results are nothing less than startling.

http://www.avsforum.com/avs-vb/showpost.php?p=11985500&postcount=38

I think you overstate the importance of achieving the last nth in instrument precision and understate the sorry state of current commercial display performance.

If I see any data that indicates otherwise, I'll gladly adjust my opinion on this matter.

I think you overstate the importance of achieving the last nth in instrument precision and understate the sorry state of current commercial display performance.


Especially given the fact the manufacturer's own specifications rate the PR-670 at average dE(Lu*v*) = 1.8 under ideal conditions and illuminant A. It's likely to be higher in the field. One problem with high resolution spectrometers is that spectral registration becomes more important and one needs to constantly recalibrate with line sources to account for drifts as a function of operating temperature and I suspect this will be important when measuring narrow SPD's. Perhaps the 670 already includes this correction, I would hope so for the money.

Additional note on dE: Achieving dE=0 means we've matched the color to the standard observer, but the standard observer is an idealized notion, I'd like to know what the variances are in the color matching functions that the CIE compiled.

Zoyd

One problem with high resolution spectrometers is that spectral registration becomes more important and one needs to constantly recalibrate with line sources to account for drifts as a function of operating temperature and I suspect this will be important when measuring narrow SPD's. Perhaps the 670 already includes this correction, I would hope so for the money.

There are a number of detectors available currently which are extremely stable over a wide temperature range. Many of the new CCD detectors are stable to better then .01 nm/K. I do not believe that many home theater room environments will have a temperature change significant enough to induce an error with what is now available in the industry for high end detectors. As long as one is calibrating in a room temperature below 100 degrees Fahrenheit there should be no real issues.

Zoyd



There are a number of detectors available currently which are extremely stable over a wide temperature range. Many of the new CCD detectors are stable to better then .01 nm/K. I do not believe that many home theater room environments will have a temperature change significant enough to induce an error with what is now available in the industry for high end detectors. As long as one is calibrating in a room temperature below 100 degrees Fahrenheit there should be no real issues.

It's not the CCD that's the problem, it's the optical bench. It depends on the optical layout used, where the electronics are located, and what the temperature gradients are across the instrument. I'm not saying it is or it isn't a problem with any particular spectrometer, but just something to look out for as you improve resolution and try to measure narrower SPD's. btw, the PR-670 spec on spectral registration is +/- 1 nm, that's not very good, they don't mention stability.

On a side note about the PR-670, I can't figure out what makes the damn thing so expensive, it's got a lot of bells and whistles, color LCD screen etc. but performance wise it's ridiculous. You can meet 90% of your accuracy specs with a $90 pod. Getting the last 10% should not cost 250x that. You can get an off-the-shelf fiber optic crossed czerny-turners for ~3K from places like ocean optics, hammatsu, or zeiss with very capable software, take it over to NIST and spend what? another 3-4K to get it calibrated and build your own that will outperform the PR-670.

How much is the PR-670? $10,000+?

How much is the PR-670? $10,000+?

umr said he paid $24,000, don't know what options/accessories that included.

umr said he paid $24,000, don't know what options/accessories that included.
That's alot of calibrations!!

As a hobbyist calibrator simply calibrating my own displays for my own use and enjoyment, the improvements in color reproduction using the i1 Pro have been nothing short of spectacular even if they are not equal to the improvements possible with instruments costing many thousands of dollars. At first, after reading this thread, my thoughts were "Why bother?" - I might as well just adjust color by eye if my instruments are so darned inaccurate. But after reflecting on the dramatic improvements made by using a "cheap" i1 Pro, I decided that I will continue to use this instrument anyway.

As Tom mentioned, the color rendition in displays coming out of the factory are so grossly and horribly inaccurate that even a bad instrument can yield results far more pleasing than the factory calibrations, especially now that extremely oversaturated primaries and goosed up (that's a highly technical term that you guys might not understand :D ) luminance is becoming the norm.

And then there is the matter of the controls available to make corrections. In many displays they are akin to using sledgehammers to make adjustments of microns. I mean, what is the point of having a $50k instrument to measure all the imperfections of a display that has "bass" and "treble" controls to adjust grayscale? And these same displays very rarely have the ability to move the primary coordinates or adjust individual RGB luminance.

The best displays I have owned (the Sharp FPs) at least have full CMS capability that seems to provide controls in the approximate same league as my instrument of measurement (i1 Pro). Basically what I am asking is what is the point of having a measuring instrument that is capable of measuring inaccuracies far greater than the adjustments available to correct them? Sure, like everyone else I would like to have the most accurate instruments available, but at what price? What is the point of diminishing returns where the cost/accuracy ratio is concerned, especially when you consider the prehistoric controls available to correct problems?

And let's not forget, that unlike the professional, most of us hobbyists are only concerned with a single display type (or maybe two) and do not need an instrument that covers all technologies.

I guess I am just trying to cheer myself up here. After reading this thread I found myself pretty depressed about the uselessness of calibration, but after giving it more thought I decided that the improvements I have made in my personal displays were well worth the effort and expense even if lab perfect calibration is not possible with my "cheap" instruments or the controls available on the displays.

Additional note on dE: Achieving dE=0 means we've matched the color to the standard observer, but the standard observer is an idealized notion, I'd like to know what the variances are in the color matching functions that the CIE compiled.

A little quick research turned up this paper (http://www.knt.vein.hu/staff/schandaj/SJCV-Publ-2005/426.pdf) which addresses one aspect of this issue. The authors found the average dE(LAB) variances among observers in a CRT primary color matching experiment was 5-7 with peak values between 10-14(Table I). This is the type of metric one should worry about, not the 4th decimal place in x or y. This is consistent with the observations that even pods provide a significant improvement in getting you from the 20 dE level down to the 10 dE level(or better) and below 5 is in the noise (perceptually).

After reading this thread I found myself pretty depressed about the uselessness of calibration...

And that's my main objection to the attitude of some people on this forum, they have a vested interest in making you feel that way, trying to minimize what can be achieved with lesser equipment, and fostering an impression of superiority. It's often just marketing schtick, so take what they say with a grain of salt.;) Remember, every DIY'er is one less customer.

As a complete novice one thing I question in this accuracy discussion is, how much does screen variation come into play? I know if I move my Display LT around a gray screen on both my A2000 and A3000 SXRDs the RGB levels change. I would have to think that whatever innaccuracies of the instrument are, they wouldn't change based on meter position and so I figure another measurement device would also find the screen to vary based on meter position. So as a professional, do you take this sort of thing into account and choose an average position from which to make readings? I just find some of this absolute accuracy discussion odd if moving the meter is going to read differently. I have no idea though what sort of deltaE changes moving the meter might translate to.

This is actually one good example of what I referred to when I wrote of the poor state of current display technology. What you see is an error in white field uniformity. LCD and LCoS displays are notorious for this. It is one area where DLPs have a huge advantage.

Anyway, what's the point you might ask of demanding extremely accurate instrumentation when the target you seek to measure is non-uniform? The color variations across the screen are much larger than the relatively minor deviations in accuracy between probes that we've been discussing.

As a complete novice one thing I question in this accuracy discussion is, how much does screen variation come into play? I know if I move my Display LT around a gray screen on both my A2000 and A3000 SXRDs the RGB levels change. I would have to think that whatever innaccuracies of the instrument are, they wouldn't change based on meter position and so I figure another measurement device would also find the screen to vary based on meter position. So as a professional, do you take this sort of thing into account and choose an average position from which to make readings? I just find some of this absolute accuracy discussion odd if moving the meter is going to read differently. I have no idea though what sort of deltaE changes moving the meter might translate to.

Screen variation is just another error source, average over an area that fills ~10 deg. from where you view to measure the effect. From Tom's comment I take it this error term for some displays is larger than the interprobe comparisons we have been talking about.

Another error source that hasn't been mentioned is viewing environment, there's a whole lot of "room" for error there.

Zoyd

You can get an off-the-shelf fiber optic crossed czerny-turners for ~3K from places like ocean optics, hammatsu, or zeiss with very capable software, take it over to NIST and spend what? another 3-4K to get it calibrated and build your own that will outperform the PR-670.

You should spend a little time with the pieces that you mention in your post, I have and they are all crap! The pieces typically do not have the low light level capability required and more so have poor color accuracy. The OO piece had accuracy no better then .02 for x/y chromaticity which stinks in everyones book. The Zeiss is simply more expensive and has the same issues I tried it as well. None of them provide a decent radiometric calibration which is a major shortcoming.

Go ahead and spend $3K and try using their terrible software which is very poorly designed (all of them). None of them allow you to perform an automatic Dark Measure or set integration time automatically making home theater calibration a real "pita". Most of the Czerny Turner designs have significant stray light path issues which affects accuracy tremendously. These pieces are typically used for relative measurements rather then absolute so there is no reason for them to have the level of accuracy which we require ie Coca cola wants to see that todays run of "the real thing" is the same color as yesterdays they do a relative measurement and if they are the same great! They do not need to know the exact chromaticity coordinates of the cola!

There is far more to making the spectroradiometer accurate then you might think even using an off the shelf item like you have mentioned. I have already been through this learning curve which is very steep and found that most of what is available is totally unsuitable for our needs do to lack of precision to a huge degree. Most of these manufacturers do not even care when it has been demonstrated to them that their equipment does not perform anywhere near to the level of accuracy that their specifications state. I have the data to back this up and have demonstrated it to three manufacturers already! They simply do not care what their customers are getting and have literally told me "when time and resources permit we will investigate this further".

In my opinion when you know something is wrong with the product it is time to stop the presses and hold everything until it is fixed! These companies care only about the bottom line not the quality of what they are selling.

Regarding the cost of a PR-670 etc. most of it is in the design of the product and the rest is that it is a very limited market for this type of product which necessitates the higher price to recoup the investment. The products which you mentioned which are low cost do not have a shutter, a lens, an LCD color display, compact flash memory card or a host of other features which makes it convenient to use. Granted these items do not total $30K or more however it all adds up and they have to make a profit or in the end they are just a memory ie chapter 11.

As a hobbyist calibrator simply calibrating my own displays for my own use and enjoyment, the improvements in color reproduction using the i1 Pro have been nothing short of spectacular even if they are not equal to the improvements possible with instruments costing many thousands of dollars. At first, after reading this thread, my thoughts were "Why bother?" - I might as well just adjust color by eye if my instruments are so darned inaccurate. But after reflecting on the dramatic improvements made by using a "cheap" i1 Pro, I decided that I will continue to use this instrument anyway.

And then there is the matter of the controls available to make corrections. In many displays they are akin to using sledgehammers to make adjustments of microns. I mean, what is the point of having a $50k instrument to measure all the imperfections of a display that has "bass" and "treble" controls to adjust grayscale? And these same displays very rarely have the ability to move the primary coordinates or adjust individual RGB luminance.

The best displays I have owned (the Sharp FPs) at least have full CMS capability that seems to provide controls in the approximate same league as my instrument of measurement (i1 Pro). Basically what I am asking is what is the point of having a measuring instrument that is capable of measuring inaccuracies far greater than the adjustments available to correct them? Sure, like everyone else I would like to have the most accurate instruments available, but at what price? What is the point of diminishing returns where the cost/accuracy ratio is concerned, especially when you consider the prehistoric controls available to correct problems?

And let's not forget, that unlike the professional, most of us hobbyists are only concerned with a single display type (or maybe two) and do not need an instrument that covers all technologies.

I guess I am just trying to cheer myself up here. After reading this thread I found myself pretty depressed about the uselessness of calibration, but after giving it more thought I decided that the improvements I have made in my personal displays were well worth the effort and expense even if lab perfect calibration is not possible with my "cheap" instruments or the controls available on the displays.

I think you have a good point there about the available controls, Bob. It reminds me of setting G2s on my CRT RPTV. The service manual's preferred method calls for an o'scope. But the pots are so sensitive (and using a regular screwdriver results in such coarse changes) that it's very difficult to get them within tolerance, let alone spot-on. The same goes for its color decoder. The adjustments available just aren't fine enough. I can get very, very close, but not dead on. It can be very frustrating at times.

Zoyd
You should spend a little time with the pieces that you mention in your post, I have and they are all crap!

Oh I have and you exaggerate. Yes, I oversimplified and should have qualified my statement as the comment was not meant for your market where you want to provide an out-of-the-box solution. Of course you don't want to have to deal with another vendor's warts and an in-house (or custom outside) design makes sense so you can add the feature set you want and target the market area you feel most appropriate. I was speaking in purely DIY mode as I have used these instruments in far more demanding applications than color science with very good results. You do however have to have the optical and hardware skills to mod them, and I agree the vendors are of no help. Is it a viable option for everyone? No. For some, maybe. There is no question in my mind that one could improve upon the i1pro performance this way but given the previous discussion on practical accuracy requirements I doubt it would be a worthwhile endeavor other than to say you did it.

(btw, color LCD's are window dressing:p)

And that's my main objection to the attitude of some people on this forum, they have a vested interest in making you feel that way and trying to minimize what can be achieved with lesser equipment, so take what they say with a grain of salt.;) Remember, every DIY'er is one less customer.

I run into the same view with techs who fear DIYers, and this is a faulty notion. It is exactly the DIYers that push the industry and do more to educate the public than the calibration specialist and the technician. In fact, many of the inovators over the past 50 years or so have started as hobbyists. Guys like Henry Kloss and Joe Kane were the ultimate DIYers. Folks like Greg Rogers (Accupel) Jeff Meier (AccuCal) and Bill Blackwell (CalMAN) are from the same mold. Calibration and service techs should not fear the DIY crowd, they should encourage it. If their services and expertise are challenged by someone trying to do it themselves they have no business being in business.

... If their services and expertise are challenged by someone trying to do it themselves they have no business being in business.

There is no threat from DIY. Those people would not be my clients in general. The folks in this thread are mistaken about the value of higher quality tools though. I do not personally consider video even my primary focus, but I do find having $24k tool of very high quality yields better results overall. When it comes to audio I have found that quality tools are available at a much lower price. In the end I buy and develop the tools that deliver the best quality results that I can see and hear. Others I am sure will do the same.

The folks in this thread are mistaken about the value of higher quality tools though.

Obviously that was not the point of what this thread morphed into. The question isn't whether higher end calibration tools "provide value" it's whether or not they are a requirement for an accurate display calibration. Although you never come out with this claim explicitly, many of your posts make this case implicitly. Now that we have some data to explore this question it supports exactly the opposite position, within the limits of human perception it made no difference which probe was used on the standard gamut displays and a marginal difference on the two wide-gamuts. For displays without a CMS and/or course controls an argument can be made that you'll get no further benefit from the higher accuracy probes over using the cheapest pod.

Personally I don't care if you're right or wrong, that's not what this is about, and would happily revise my opinion given new data. Within the framework of the CIELUV or CIELAB systems for evaluating the perceptual errors probes generate, it's a simple matter to show how much one probe outperforms another and disseminate that to the community given your access to multiple display technologies. If, as you say, we can't extrapolate your 10 display data set to like technologies then you must have data to support that conclusion. If you have such data and won't share it you're doing a disservice to this forum.

Well, seems as though everytime I post a question to this forum I start an all out fight... :D

All joking aside, I appreciate everyones input. I think most of the "arguing" is coming from the fact that we have a difference of opinion in what is "accurate enough." I can certainly understand UMR's view. If I was paying him to calibrate my set, I would be happy that he was using the best equipment he could find. I don't think anyone could fault him for that.

I do think that a lot of times UMR looks at an error and says to himself "that really sucks" where most of us would say "it isn't dead on but it is acceptable." It just depends on what is acceptable to each person. I don't think an error of 5 dE is going to kill anybody, especially given the condition most displays are in directly out of the box.

My original question was basically a theoretical one of whether or not the spikes that we see for the new wide gamut displays would affect a colorimeter (with perfect filters). krasmuzik has given a lot of good information indicating that, in theory, the spikes would not matter, which was my original line of thinking. We all know that the story may be different in practice, but some of us are willing to accept the error if for no other reason but it is fun to learn by doing.

In the end, the more expensive tool better be more accurate or I don't think it is going to sell very well... but I also think we can get acceptable results from lower end tools, provided we aren't looking for absolute perfection. At some point you have to realize that it is a TV, not a space shuttle...

All joking aside, I appreciate everyones input. I think most of the "arguing" is coming from the fact that we have a difference of opinion in what is "accurate enough."

My opinion is that the accuracy that gets you below the limit of human perception is as good as one can do, anything more is window dressing and has no value, this is without even considering the display and environmental factors that pollute that response. Would I use a high-end probe if I had one, sure, but it would not lead to any perceptual improvement in my display versus using the i1pro given the results presented in this thread. Would such an instrument be required in some situations, probably for wide-gamuts but without a CMS you'll just get a better measurement of a lousy display.

But I don't agree that this is only about the definition of accuracy, it also highlights a fundamental dichotomy present in this forum. Those who want free and open exchange of information, measurements, and discussion of the real implications of that information as it applies to display calibration, and those that for whatever reason, don't. Those people prefer to dictate rather than participate. And you're right, we aren't calibrating the shuttle, but the reason this kind of detail is important is that it separates out, in an objective way, what you should be looking for and what you should ignore in making decisions about how to improve your viewing experience.

My opinion is that the accuracy that gets you below the limit of human perception is as good as one can do, anything more is window dressing and has no value, this is without even considering the display and environmental factors that pollute that response. Would I use a high-end probe if I had one, sure, but it would not lead to any perceptual improvement in my display versus using the i1pro given the results presented in this thread. Would such an instrument be required in some situations, probably for wide-gamuts but without a CMS you'll just get a better measurement of a lousy display.

But I don't agree that this is only about the definition of accuracy, it also highlights a fundamental dichotomy present in this forum. Those who want free and open exchange of information, measurements, and discussion of the real implications of that information as it applies to display calibration, and those that for whatever reason, don't. Those people prefer to dictate rather than participate. And you're right, we aren't calibrating the shuttle, but the reason this kind of detail is important is that it separates out, in an objective way, what you should be looking for and what you should ignore in making decisions about how to improve your viewing experience.

I don't really want to get into the philosophy of whether or not one should embrace or reject open source projects and open exchange of information, but you can't really expect umr or anyone else who calibrates for a living to come here and spill their guts. Professional calibrators really have no reason to tell us anything, as it just takes up their time and can potentially give valuable information to someone who could become a competitor. By the same token, I can understand the need to react if someone just comes around and states "you must have instrument x to calibrate properly" when that information is not true. I see both sides of the coin in that we need to be thankful for the information pro calibrators take their time to give while at the same time making sure that the information given is correct.

What I'm trying to say is that the ideas/arguments (whatever you want to call them) in this thread are really just the result of several very intelligent, very qualified people "talking past" each other. I don't think anyone's aim is to dictate the forums, but I do think our opinions get thrown out of context from time to time, which can make it look like someone is trying to dictate.

I really didn't mean to start an argument with my original post, and I hope that everyone involved will continue to offer information to the community. The last thing I would want is for qualified people to stop giving information because it always ends up in a 7 page battle to the death.

That is just my 2 cents. I'm a nobody and I'm not taking anybodys side, I just wanted to learn something :D

This forum is not to be used for marketing - neither ones calibration services - nor ones calibration products. Using FUD (Fear, Uncertainty, Doubt) is in fact - marketing.

FUD has a way of backfiring. The chart posed by UMR shows that 8/10 displays he was using to justify his more expensive gear are in fact within perceptual limits with less expensive gear, and certainly the last two displays if they have CMS can be improved on factory defaults. He does not get to backtrack on that data just because someone else understood the implication of that data. He can start his own forum if he wants to market to potential customers that do not understand the data and buy into his FUD. Before it was LED displays it was SXRD displays that you could not calibrate with lesser gear, before that DLP, before that LCD, before that FP CRT..... If you listen to these guys often enough you would think that CRT or LCD or DLP or SXRD or Plasma or LCOS cannot be at all calibrated with less expensive gear. I think this forum is testament enough that there are many people improving their displays to their great satisfaction compared to blown out marketing presets - without having NIST traceability on the sensors against the lamps they are using.

The source I used to verify my gear laughed when I said I wanted to use the PR650 he had. He said it is so expensive to calibrate (which it requires often) that they stopped doing it and bought the KM colorimeter instead because it is much easier to keep NIST traceability on it. They bought it specifically for that feature that a colorimeter needs to be calibrated to your lamp to be considered very accurate - it lets them essentially get lab spectroradiometer results in a portable package - and indeed the portable spectroradiometer was so far off calibration it was not even funny. It was funny that my Spyder2 prototype dithered in the ten thousandth place against the KM colorimeter- even though statements have been made that DLP cannot be calibrated with it - yet there was no need to train it to the more expensive gear. Consider that the only reason I even considered getting it verified was the FUD propagated on this very forum - and my gracious contact thought it might be worth the laugh to see how far off these probes (pucks, pods or whatever derogatory name you want to use) could be. Of course that FUD was based on the simplistic thinking that 256 sensors was better than 3 - even though the Spyder2 has 8 so it was the wrong FUD in the first place.

Hopefully people have learned that there is a fundamental difference in the number of sensors required between spectroradiometers and colorimeters. One needs lots and lots of samples to properly capture a spectra to digitally filter it with the eye function, the other has a few optical sensors intended to match the eye function with proper calibration. Counting the number of sensors between the two as a measure of quality - is and was always FUD! A spectro with all of its sensors might not have enough - or might itself be miscalibrated or have bugs in the filtering math, a colorimeter might itself have cheap filters in a cheap package that got out of factory calibration just sitting in the warehouse, and may never have been checked against certain displays. One is not better than the other just because it has more sensors.

If you are at all concerned about your sensors accuracy - see if your vendor or an associate can arrange a comparison to NIST tracable or certified measurement standards - on the typical light sources you are concerned about. Even better if they can reprogram any inaccuracies in your sensors calibration. Just because their sensor is more expensive means squat - theirs could be worse off than yours simply due to the lack of recent calibration.

I don't really want to get into the philosophy of whether or not one should embrace or reject open source projects and open exchange of information, but you can't really expect umr or anyone else who calibrates for a living to come here and spill their guts. Professional calibrators really have no reason to tell us anything, as it just takes up their time and can potentially give valuable information to someone who could become a competitor.

I appreciate your effort to soft-land this thing but I have no wish or expectation for the procal crowd to "spill their guts". What I do expect is basic fairness. When someone, anyone, is challenged or challenges another on an issue which is not in the realm of opinion, hiding behind the procal tag is not appropriate.

I appreciate your effort to soft-land this thing but I have no wish or expectation for the procal crowd to "spill their guts". What I do expect is basic fairness. When someone, anyone, is challenged or challenges another on an issue which is not in the realm of opinion, hiding behind the procal tag is not appropriate.

I'm pretty sure this one crashed and burned shortly after takeoff ;)

Krasmusik

It was funny that my Spyder2 prototype dithered in the ten thousandth place against the KM colorimeter- even though statements have been made that DLP cannot be calibrated with it - yet there was no need to train it to the more expensive gear.

This statement is misleading to the casual user at it implies that the Spyder probe is accurate for calibration of any display simply because it is in agreement with the reference instrument. The truth is that this simply means it is in agreement on this one display and that others may not necessarily be in agreement at all. The KM colorimeter has been shown to have problems with certain display types even as simple as a Sony BVM 65 Black and White monitor which it was not even close to being accurate in measuring.

You always rant about how your Colorfacts probe is accurate to the .0001 which is a laugh as you have no clue how accurate the reference instrument which you are comparing it with is! Until the people which are interested in this thread understand that trying to take measurement information from other users probes on displays which they have no accurate reference instrument to base the accuracy of the measurements the data is fairly useless. Even with a known reference to verify with there still exists plenty of possibilities for filter based analyzers to have errors due to the many issues already outlined in earlier postings.

The degree of error which filter based probes can have runs anywheres from very little to unacceptable based on the particular display being tested. There is no possible method which you can use to take previously logged measurement data from other displays and come up with a model that tells you the range of error which you may experience on an unknown display model. This simply does not work in the real world and continuing the discussion regarding it is quite pointless for all concerned as it does not provide a solution set for anyone to work with.

This is not about trying to sell more "stuff" and is not a marketing ploy. It is based on facts which I have accumulated working exclusively in this field for the past nine years as well as from manufacturing experts which collectively have experience of many times that. whenever someone has contributed information which has merit regarding this discussion and it involves and instrument which costs more then a pair of Air Jordon's everyone gets upset and feels it is a shot at selling something and that they own an inferior piece of equipment. If you are getting good results with your instrument then be happy and life will continue tomorrow just as it did before. If however you have come across displays which you are not getting credible results with your Air Jordon probe then possibly some merit in the discussions which have transpired thus far have some meat to them and better instrumentation is needed for certain types of displays to obtain any kind of an accurate result.

If you feel getting within 10% is good enough how do you know if you are even this close with your current analyzer without a reference for each and every model of display which is on the market today? Your mileage will vary by model from slim to extreme differences.

ghibliss,

I'm not sure if you are aware of this but your post is a prime example of what Krasmuzik is talking about. You assume "casual users" don't have a brain and won't ask questions about what is and what is not possible or discover on their own what does and doesn't work. So you attempt to define the "universe of possibilities" for them, this is marketing 101. People are not interested in what can't be done. Anyone adept enough to use a probe can read this thread and get a very good, accurate, idea of the limitations of the equipment they are using.

The degree of error which filter based probes can have runs anywheres from very little to unacceptable based on the particular display being tested. There is no possible method which you can use to take previously logged measurement data from other displays and come up with a model that tells you the range of error which you may experience on an unknown display model.I guess what troubles me about this statement is that it seems to insulate claims about probe accuracy from tests against empirical data. If "there is no possible method" employing accumulated data to test a claim about a probe's accuracy, then we are left with little more than testimonials.

I for one am not wedded to the notion that a low-cost probe is necessarily the best approach, but if I were to invest in a higher cost instrument I would want some objective data on which to base such an investment. Also, no one doubts that very expensive laboratory grade instruments can and often do provide more accurate results for a wider range of displays. The question is whether the AMOUNT of increased accuracy justifies the substantially higher investment.

Also, no one doubts that very expensive laboratory grade instruments can and often do provide more accurate results for a wider range of displays. The question is whether the AMOUNT of increased accuracy justifies the substantially higher investment.

Also keep in mind that the PR-670, in my opinion, is not a laboratory grade instrument. Much of the cost of the device has nothing to do with it's performance but rather it's configurability, portability and narrow market. The truth is that the instrument requirements for display calibration within the field of color science are not particularily stringent compared to the kind of spectroscopic measurements that scientists deal with in other fields. That's why you can get something as simple as a pod to work as well as they do in the first place. The tough part isn't even the instruments themselves, they are all based on design principles that have been around for decades, it's in calibrating the instrument and maintaining that calibration. While manufacturers have to pay attention to these requirements it's not always(ever?) the driving force. If someone were to come out with a product akin to the i1pro but with better sampling, improved throughput and leave out the do-dads to keep the cost down I suspect that would be a worthwhile investment.

Except for the calibration, wouldn't a spectroradiometer just be a lens, a diffraction grating, and the ccd chip from a digital camera? With the advances in camera technology, especially with the low noise chips, I'd think it wouldn't be too difficult to put together an accurate meter. The tough part would be calibrating the software that reads the chip....

Except for the calibration, wouldn't a spectroradiometer just be a lens, a diffraction grating, and the ccd chip from a digital camera? With the advances in camera technology, especially with the low noise chips, I'd think it wouldn't be too difficult to put together an accurate meter. The tough part would be calibrating the software that reads the chip....

umr's earlier post with links to the jeti.com material is a good place to start if you are interested in miniature spectrometer design. There are a few issues that need to be overcome to be able to get a well calibrated result. It all depends on what your requirements are. Detector technology has advanced to the point that this is not the driving part. Getting the optics right remains the hardest part, primarily straylight, when dealing with miniature spectrometers. The smaller the box, the harder it is to prevent light from one wavelength contaminating another. It also takes quite a bit of hands-on experience to answer the question "is this going to work" and then evaluate whether your design specs have actually been met.

Except for the calibration, wouldn't a spectroradiometer just be a lens, a diffraction grating, and the ccd chip from a digital camera? With the advances in camera technology, especially with the low noise chips, I'd think it wouldn't be too difficult to put together an accurate meter. The tough part would be calibrating the software that reads the chip....

If it was so easy to produce an accurate spectroradiometer, there'd be a ton of choices and they'd be cheap.

I just spoke to a manufacturer who just finished going through the whole design process for a new spectroradiometer and it was FAR FAR FAR from an easy task.

If it was so easy to produce an accurate spectroradiometer, there'd be a ton of choices and they'd be cheap.

I just spoke to a manufacturer who just finished going through the whole design process for a new spectroradiometer and it was FAR FAR FAR from an easy task.

If there was a market, I'm sure there would be.

Lenses are pretty well understood and an apochromatic triplet shouldn't be all that expensive. Diffraction gratings are pretty straightforward too. With the advances being spurred by the digital camera boom, I'd think a high quality detector shouldn't be too difficult to come by either.

Telescope manufacturers have been dealing with stray reflections for a long time, so materials to absorb them should be out there. So what's so difficult about all this so far? This is all off the shelf stuff.

The tough part is the software that analyzes the pattern read from the detector chip. With the new low noise camera chips that are coming out, that shouldn't be unbeatable either. You'd just have to have known spectral sources to calibrate it against. But that isn't rocket science either, is it?

Lens concentrates the light, grating splits the frequencies, chip measures the intensities of the frequencies. The concept doesn't seem all that complicated to me.

Maybe not - but the lab you need to be able to make sure you are making something accurate - is very expensive. There is a lot more to calibration gear than the BOM.

Sure it is, and I don't want to belittle it. But they do exist and folks have them.

I'm just musing that technological advances in more consumer driven markets, like the digital camera, should make a much more accurate instrument possible without any development time. There's no technical reason the equipment you guys use has to be expensive or hard to find except that the folks with the know-how aren't motivated to improve the situation.

If there was a market, I'm sure there would be.

Lenses are pretty well understood and an apochromatic triplet shouldn't be all that expensive. Diffraction gratings are pretty straightforward too. With the advances being spurred by the digital camera boom, I'd think a high quality detector shouldn't be too difficult to come by either.

Telescope manufacturers have been dealing with stray reflections for a long time, so materials to absorb them should be out there. So what's so difficult about all this so far? This is all off the shelf stuff.

The tough part is the software that analyzes the pattern read from the detector chip. With the new low noise camera chips that are coming out, that shouldn't be unbeatable either. You'd just have to have known spectral sources to calibrate it against. But that isn't rocket science either, is it?

Lens concentrates the light, grating splits the frequencies, chip measures the intensities of the frequencies. The concept doesn't seem all that complicated to me.

You're missing all the difficult issues because you haven't got the level of detail to understand why it is so difficult. EVERYTHING seems like it should be easy when you are sufficiently removed from the hard science and engineering.

Keep this in mind... 30 years ago, people thought we had things like lenses pretty well figured out... telescopes too. What you can achieve today at any given price/cost level was IMPOSSIBLE to achieve 30 years ago. So it is with spectroradiometer design - for someone willing to get to the bottom of what it takes to make a good one.

But that's my point exactly. We've come a long way in the last 30 years. And I'm glad you bring up telescopes. There are these things called computers, and ray tracing techniques, that let you model your optics exactly. This isn't rocket science. We understand refraction and reflection pretty well, I think.

My god, we've got folks doing interferometry of stars and discovering planets revolving around them from the doppler shift they cause in the star's spectrum. The number of photons reaching those detectors is a lot smaller than the number being emitted by your screen or projector and yet somehow they manage to do it.

You're telling me it's impossible to leverage all this technology and make a meter that sits a few inches from a screen or a few feet from a project and get a good readout? Why? Tell me what's new here. What problems haven't already been solved in a much tougher form?

I'm not saying it isn't work, and I'm not saying there aren't engineering tradeoffs to be made. I'm just saying the problems have already been solved, the tools are available, and recent technological advances ought to mean these devices shouldn't be as inaccurate or as expensive as they are.

But of course, I'm too ignorant to understand all that stuff.

It is not impossible at all. It is the start-up costs of doing it right, combined with a limited market, that keeps the picckings limited.

Saying something can be done and making it a reality in the marketplace without losing your shirt (and a couple of years of your life) are completely different. Nothing you have said is far from the truth, but it is naive to think that it can be done on the cheap.

But that's my point exactly. We've come a long way in the last 30 years. And I'm glad you bring up telescopes. There are these things called computers, and ray tracing techniques, that let you model your optics exactly. This isn't rocket science. We understand refraction and reflection pretty well, I think.

My god, we've got folks doing interferometry of stars and discovering planets revolving around them from the doppler shift they cause in the star's spectrum. The number of photons reaching those detectors is a lot smaller than the number being emitted by your screen or projector and yet somehow they manage to do it.

You're telling me it's impossible to leverage all this technology and make a meter that sits a few inches from a screen or a few feet from a project and get a good readout? Why? Tell me what's new here. What problems haven't already been solved in a much tougher form?

I'm not saying it isn't work, and I'm not saying there aren't engineering tradeoffs to be made. I'm just saying the problems have already been solved, the tools are available, and recent technological advances ought to mean these devices shouldn't be as inaccurate or as expensive as they are.

But of course, I'm too ignorant to understand all that stuff.

Well, OK, then. It sounds like you have it all figured out... so go on ahead and make your own spectroradiometer and let us know how it goes.

Except for the calibration, wouldn't a spectroradiometer just be a lens, a diffraction grating, and the ccd chip from a digital camera? With the advances in camera technology, especially with the low noise chips, I'd think it wouldn't be too difficult to put together an accurate meter. The tough part would be calibrating the software that reads the chip....
Some final words about spectroradiometer design:
Several years I was just in this situation - to convert a simple array spectrometer into a spectroradiometer. Of course this is not a world new development but as every developer knows the problems arise immediately after starting the project: misfunction of shutter, problems with diffusor material, temperature and other dependences of parts, non linearity of the "CCD chip", etc. Software development is mainy a question of knowledge of user demands - the calculations themselves are very simple.
If somebody is interested in the result, look for specbos 1201 in Internet (our basics pages were already cited before)

Some final words about spectroradiometer design:
Several years I was just in this situation - to convert a simple array spectrometer into a spectroradiometer. Of course this is not a world new development but as every developer knows the problems arise immediately after starting the project: misfunction of shutter, problems with diffusor material, temperature and other dependences of parts, non linearity of the "CCD chip", etc. Software development is mainy a question of knowledge of user demands - the calculations themselves are very simple.
If somebody is interested in the result, look for specbos 1201 in Internet (our basics pages were already cited before)

What's the pricing on something like that, if you don't mind my asking?

What's the pricing on something like that, if you don't mind my asking?

I was thinking free samples?:D

What's the pricing on something like that, if you don't mind my asking?
It is around $ 8.000.

It is around $ 8.000.Does the luminance sensitivity really only go down to 2 cd/m2?

I just noticed Argyll CMS (argyllcms.com) supported the i1 Pro with its own driver (on Linux in my case), and the docs mentioned "high resolution" 3.33nm support in the latest version. I tried it out on my LCD monitor and I got some plausible results (see attachment).

What have you compared the new 3.33nm data set to for verification aside from the i1Pro which has far less resolution and can not be used for reference?

Is it just interpolating the 3.33nm readings from the hardware's 10nm resolution, or is there something better going on here?

The stated physical sampling interval of the i1Pro is 3.5 nm however the instrument returns only 35 wavelength values with the spectral radiance for those points. I believe that the 128 pixel diode array performs some averaging internally to provide the 35 data points. There is no way to increase the resolution of the data returned via the SDK provided from X-Rite so any values provided by the argyll.cms must use some form of interpolation. You can interpolate with precision when your instrument has greater accuracy then what you are interpolating to however I fail to see how you could arrive with good accuracy interpolating when your instrument has less resolution then your reference device has.

Argyll isn't using the Eye-One SDK. It's accessing the hardware directly and using the 128 raw values from the sensor filtered to 10nm or 3.3nm spacing (based on inspecting the code). The data looks good compared to better devices measuring similar sources (CCFL backlit LCDs). Just from this it does appear that the device does send the more data but the SDK always resamples it to a 10nm interval. The SDK might be generating the XYZ data from the full data set, regardless, but only ever returns the 10nm data to the application.

I see where the code reads the 128 values directly. It sure seems to me to have a native 3.333 resolution. EDIT: After re-reading ghibliss' post, does that mean it takes 128 samples that are 10nm "wide", and if spaced at 3.3 or 3.5, they overlap? Per the docs, it extends down to the UV and up to the IR range, but I'm not sure yet if that data is discarded to generate XYZ. It would be interesting to compare the XYZ values returned from Argyll and the SDK.

What instrument(s) did you (or Argyll) compare the data with?

EDIT: After re-reading ghibliss' post, does that mean it takes 128 samples that are 10nm "wide", and if spaced at 3.3 or 3.5, they overlap?

yes, the optics (slit width and grating dispersion) determine the optical resolution which is 10 nm. This spectrum is sampled by the diode array at 3.5 nm intervals which yields ~100 pixels of information over the relevant wavelengths (380-730 nm). How those 100 pixels are used to form the XYZ values is not known or why the x-rite SDK only send 35 values.

According to the specs on the i1, it's got 10nm bandwidth and a 3.5nm sampling interval. The samples are only 3.5nm wide, but the FWHM is 10nm, so a single spectral line will light up 3-4 sensors at various amplitudes.

I don't have anything handy to compare it to but I can help anyone that has a better spectro to set up the Argyll/i1 combo if they need it.

Does the luminance sensitivity really only go down to 2 cd/m2?

Yes, this is the minimum we can guarantee the precision. More is not reachable with an uncooled photodiode array according to my opinion.

Using the Argyll.cms which does in fact display improved 3.3nm resolution fails to point out that taking the raw data output from the i1Pro does not have the ability to use the calibration tables built into the device to provide the accuracy which one would require to use this instrument. You also will have a substantial reduction in low light level sensitivity using the single pixel data output relative to the binned values from the factory. This information is according to a contact at X-Rite. It would be foolish of X-Rite to not provide users with the added level of accuracy which the device is capable of it truly were capable of delivering it at this level.

It does look like Argyll is reading the calibration data out of the eeprom and using it, but I can't be sure if it's missing anything that the i1 SDK does, obviously.

Regardless, I've been doing some reading and it looks as though there's no significant difference in dE of spectrometer results as long as the sampling interval is equal to or smaller than the bandwidth; so you get no real gain sampling at 1nm intervals over 10nm intervals on an instrument with a 10nm bandwidth. The optical bandwidth here is the true limiting factor, and even with peaky spectra it doesn't look like there's a big difference in dE between 10nm bandwidth instruments and a 5nm or even 1nm instrument. It looks like the choice of the 10nm reported sampling interval is adequate for color measurements and the noise reduction from interpolation is more helpful than having finer sampling.

So in that sense I'm convinced that both the bandwidth and reported sampling interval are well thought-out from an engineering perspective; as the i1 was designed for colorimetry an optical bandwidth better than 10nm wasn't cost effective, and the 3.5nm sampling interval is adequate to cover the 10nm bandwidth and provide a good amount of noise reduction.