EEVblog #967 - Mystery Teardown

[EEVblog]

Mystery Teardown

[ElektronikLabor]

That a nice tear down!
I'm working with similar kind looking photo diodes for spectral measurements.
You can get them for 10$ but if you want a accurate and low light sensitive one you can pay a fortune for one of them.

We are also using filters for different colors.

The left one on the image is a germanium diode that costs 200$.

[mcinque]

What is the spectrometer you're using to test filters?

[thomastheo]

Looks to me like a Vishay BPW21R or something very similar.

http://www.vishay.com/photo-detectors/list/product-81519/





From the datasheet: The device is equipped with a flat glass window with built in
color correction filter, giving an approximation to the
spectral response of the human eye

They aren't that expensive, i've got bunch of them sitting in my parts drawer. In the order of 5 euro a pop.

[jipihorn]

Hello,

I've been working in the colorimetric world for 15 years and this device is a very classic one.
First of all, it is NOT a spectrophotometer. It is a colorimeter.
The principle is the use of  3 filters to obtain a physical space representing the color. But, this has a lots of  inherent limitations. These filters are in fact the standard CIE observer curves (the 1931 2° one). The x one has two humps, related to a weird property of the human vision. All measurement devices use CIE coordinates, which can be XYZ, Lab, LCh... RGB, or CMY(K) are NEVER used in physical measurements. They are useless because they are device related. The link between CIE coordinates and device related coordinates is done using tables (or some formula in very, very specific cases). This is the goal of ICC profiles for example. RGB (CMY) space are not physical and not defined.
All colorimeters use the very principle of filters, but they are NOT RGB or CMY stuff. There is a R,G,B space in CIE definitions using r,g,b observer curves. But this is NOT related in any way to the usual RGB space used in the computer world. RGB or CMY(K) are the worst spaces to work with when you manipulate colors seriously.
The other limitation of a colorimeter is that they are stuck to a given illuminant (or light), often D65 or D50. And there is no way to convert a CIE coordinate in one illuminant to an other. There are some techniques (like the Bradford transform), but they are only crude approximations. The only way to get real colorimetric measurements is from a spectrophotometer. From that point, you can compute any CIE coordinates you want.
Colorimetry is a very rich science subject and it is a shame that it very under-used in the computer world. For example, Adobe software have very archaic color management even in 2016 and there is no way to print accurate colors with usual software, even though everything is available to do so !
Jerome.

[HighVoltage]

Very nice tear down !

I used have a SONY monitor, that came with a similar but much smaller color detector that had to be placed on the front of the monitor from time to time go get it calibrated.

In what episode are you explaining / showing the spectrometer ?
I must have missed that, what brand is it.

 

[EEVblog]

In what episode are you explaining / showing the spectrometer ?
I must have missed that, what brand is it.

[tox3]

Great teardown! 

[Razor512]

That reminds me of a more modern colorimeter teardown.



I just wonder why these companies like to price gouge so much for what is likely $5 worth of hardware.

[bktemp]

I just wonder why these companies like to price gouge so much for what is likely $5 worth of hardware.

Industrial grade colorimeters are really expensive devices (> $10k).
Calibrating the sensors makes them so expensive (see the list of calibration constants shown in the video).

Here is an example of a modern single chip colour sensor:
http://www.mazet.de/en/products/jencolor/true-color-sensors/item/384-mtcsics.html

[PA0PBZ]

So why does it have a petri dish with an Ebola culture on it?

[Razor512]

So why does it have a petri dish with an Ebola culture on it?

The ebola acts as a backup. If you are unsure if the unit is functioning within spec, you can ask the ebola, and see if they can reach a consensus on whether the unit is functioning within spec or not.

[SeanB]

Common to get a fungus growing on optics stored incorrectly, they live off the organic coatings on the glass, or it is a chemical reaction of the coating with water vapour in the air.

[bitwelder]

So why does it have a petri dish with an Ebola culture on it?

Hidden prize for those who are ignore the 'warranty void' sticker.

[eliocor]

It seems to me that the dichroic filters used are not RGB (Red/Green/Blue) but with a response similar to CYM (Cyan/Yellow/Magenta) ones.
If you combine the response from the three CYM filtered sensors, you can reconstruct the RGB response:

R = M + Y - C
G = C + Y - M
B = C + M - Y

For some more details you can take a look at: http://www.egrafton.com/cmyrgb.htm

Unluckily I do not know if such operations are done on the analog level or in the microprocessor.

[amspire]

I think people are forgetting that this device is dedicated to measuring RGB for a CRT. It does not need to model the human eye in any way.

If it is optimised, say, for P22 colour phosphors, the frequencies are
Red:  626nm
Green:  530-535 nm
Blue:  450nm

If you look at Dave's graphs, the Blue filter eliminates all RED and maybe 50% of Green.
The Green eliminates maybe 10% of the red and 60% of the Blue
The Red eliminated probably 50% of the green and 25% of the blue.

With a matrix calculation, they can work out the R, G and B levels.
The reason for using Dichroic filters is they are very stable and predictable.

Once the device is calibrated, it should have good accuracy and stability.

[Someone]

The dichroic filters used are not RGB (Red/Green/Blue) but with a response similar to CYM (Cyan/Yellow/Magenta) ones.

When the display reports in CIE xy format, its a good guess that the filters will be the CIE observers:
https://en.wikipedia.org/wiki/CIE_1931_color_space#Color_matching_functions

I think people are forgetting that this device is dedicated to measuring RGB for a CRT. It does not need to model the human eye in any way.

Since its reporting in the CIE colour space, the filter responses will need to follow that. They could make some simplifications based on a specific phosphor but then it won't work with different phosphors (as would be common when looking at professional monitors compared to consumer units).

For the doubters, attached is an overlay of dave's comedically poor measurements of the filters against the CIE observers. Already it looks like a match for all 3 filters, the cutoffs all match but the relative intensities would need to include the spectral responses of the detectors and the other optical elements in the path.

[amspire]

The dichroic filters used are not RGB (Red/Green/Blue) but with a response similar to CYM (Cyan/Yellow/Magenta) ones.

When the display reports in CIE xy format, its a good guess that the filters will be the CIE observers:
https://en.wikipedia.org/wiki/CIE_1931_color_space#Color_matching_functions

I think people are forgetting that this device is dedicated to measuring RGB for a CRT. It does not need to model the human eye in any way.

Since its reporting in the CIE colour space, the filter responses will need to follow that. They could make some simplifications based on a specific phosphor but then it won't work with different phosphors (as would be common when looking at professional monitors compared to consumer units).

For the doubters, attached is an overlay of dave's comedically poor measurements of the filters against the CIE observers. Already it looks like a match for all 3 filters, the cutoffs all match but the relative intensities would need to include the spectral responses of the detectors and the other optical elements in the path.

The curves are definitely similar, but still a long way off. The fact the probe is only for CRT's and that you need a different probe for LED's makes me think that this probe assumes certain phosphor frequencies.

If it was trying to accurately simulate human eye response, it could be used for any colour calibration - not just CRT calibration.

[Fungus]

The curves are definitely similar, but still a long way off. The fact the probe is only for CRT's and that you need a different probe for LED's makes me think that this probe assumes certain phosphor frequencies.

Either that or the manufacturer simply wants to sell them two probes.   

[Someone]

The curves are definitely similar, but still a long way off. The fact the probe is only for CRT's and that you need a different probe for LED's makes me think that this probe assumes certain phosphor frequencies.

Either that or the manufacturer simply wants to sell them two probes.   

Or that CRTs had almost zero colour shift with angle, but the changes in contrast and colour with viewing angle for an LCD can be extreme. The LCD probe is long and thin to make a high relative aperture, while the CRT probe is a pad up against the screen. Simple explanations from the optical constraints.

[rs20]

My message explains why all the filters have a blue peak, just in case someone is skimming through and thinking that this is just a reply.

For the doubters, attached is an overlay of dave's comedically poor measurements of the filters against the CIE observers. Already it looks like a match for all 3 filters, the cutoffs all match but the relative intensities would need to include the spectral responses of the detectors and the other optical elements in the path.

The curves are definitely similar, but still a long way off. The fact the probe is only for CRT's and that you need a different probe for LED's makes me think that this probe assumes certain phosphor frequencies.

If it was trying to accurately simulate human eye response, it could be used for any colour calibration - not just CRT calibration.

In order for the device to report a theoretically perfect CIE measurement, the three filters don't have to match the CIE observer curves. All that is required is that each filter is a differently weighted mix of the CIE observer curves.

For example, if filter1 was the CIE z observer curve, and filter2 was the CIE x observer curve PLUS 30% of the CIE z observer curve, then it is trivial to subtract 30% of the signal from signal 1 away from the filter 2 signal and hey, you've reconstructed the signal that a CIE x observer filter would see. The more general problem of untangling three arbitrarily mixed CIE observer filters is a simple matrix division.

Hence, it would be absolutely foolish for a colorimeter manufacturer to restrict themselves to trying fruitlessly(/unnecessarily expensively) to construct/acquire filters that match the CIE observer curves one-to-one. As long as the CIE observer curves can be reconstructed as a linear mixture of the three (or more!) filters available, this can be trivially done in software and hugely relaxes the constraints given to the filter designers.

In summary, noting that the filters don't match one-to-one with the CIE observer curves, it is an invalid logical leap to say that they are assuming certain phosphor frequencies. Not saying your conclusion is false, just this attempted argument. I'm thinking the truth might lie somewhere in the middle; they probably surveyed a large number of actual TVs and found the cheapest filter that provided specified performance across all of them. Or, we might discover that if the spectrometer were actually performed properly (i.e., capturing the white torch spectrum and dividing through to get actual transmission), we might find that the measured spectra actually match linear mixes of the observer spectra very closely indeed.