for home lights, not specifically calibrated for flashlight. spectra maybe a little bit different.
They aren’t, they use the same technology and produce the same light. If your sensor reconstructs the spectrum from the eight color channels, the type of LED doesn’t matter.
1 Thank
the key lies in how to reconstruct the spectrum precisely, this is mission impossible. if you measure one specific light source, and the sensor gives you 8 channels values, how could you accurately reconstruct the spectrum? VIS spectrum has 401 values. if you have a formula y = f(x) = x1 + x2, and x is a vector of dimension 2 consisting 2 values. if you already got y value, how could you get x1/x2 values? there are infinite combinations of value x1 and x2. conversely, it’s quite simple and straightforward, you got x1 and x2, y is easy to calculate.
as for channel values, we know for example F1 channel value is the inner product of F1 response curve and light spectrum curve which both have same dimension of 401. we got F1 ~ F8, but you have only 8 formulas ,how can you reconstruct the spectrum?
in the official documents of as7341, they use least squared mean to get the coefficient matrix which try to map channel values to spectra. but this is very sensitive to light source whose Led chip’s excitation wavelength and Phosphor powder proportioning differ to each other, for example home lights and flashlights.
in a word, if you use home lights to calibrate your device, you’d better use the device to measure home lights.
4 Thanks
you got 8 channels values, and you want to reconstruct the VIS spectra which have 401 values, from information theory perspective that’s impossible for all kinds of light sources. for example, if we limit light source to home lights, the coefficient matrix which bridge the channels values and spectra maybe give you a relatively accurate spectrum which you can use to calculate all light metrics.
I heard them saying that even integrating sphere is not absolutely accurate in measuring spectrum., only a German manufacture’s equipment can, but it costs around a million of US dollars.
if precise spectra can be gotten using just several algorithms or calibration methods, why invent an equipment which cost a million dollars?
5 Thanks
This only depends on the spectrophotometer which is connected/installed on the integrating sphere. The sphere itself does not measure anything.
integrating sphere I use can output spectrum and metrics. I see sensor embedded in the shell, and software on PC controls the sensor through wires connected to the integrating sphere. the IS can output spectra which consists of 401 real numbers, but our as7341 can just output 8 values. the resolution ratio is 50 times of as7341.
With an ideal integrating sphere. In reality it won’t have perfect remission and needs calibration.
Of course you won’t get the resolution of a full spectrophotometer. Such a cheap sensor will always be a compromise. But you can use the values to fit a curve. It won’t work with an irregular, spiky spectrum produced from some light sources light fluorescent lamps or some special LEDs with different emitters (like RGB or white with extra red LED to raise R9).
To get better results, you should optimize the processing for different light sources. Add a menu to select the kind of light source (incandescent, fluorescent, white LED, RGB LED, daylight etc). Some will produce better results because they have a smoother spectrum. For those you could output somehow meaningful values like CCT, duv, CRI. For others like fluorescent lamps you need an assumption for the spectrum (which can be different depending on the coating). Try to fit this assumed spectrum to the measurement to get an approximation of the CCT and maybe duv, but don’t show individual CRI measurements as they would be misleading.
4 Thanks
you are right.
now trying to add a classifier which can distinguish between light source, first try to recognize light source types, then try to calibrate and calculate for a specific light source.
3 Thanks
A first iteration could include a menu to select the type of light. And then a future update could try to automatically detect and classify a light.
1 Thank
How does the LM3 work without such a crutch?
Spectrum of Leds is smooth curve with finite support. Moreover usually LED spectrum has 2 maximum and 1 minimum. So you need to find interpolation function. After that you can calculate 100500 point spectrum if needed.
Without the need to disclose the individual R*, I presume linear interpolation is good enough to calculate estimates of CRI, CCT and CIE coordinates. The systematically overestimated CRI tells me that the firmware is not using a realistic model of a LED.
Opple is just starting to develop good software, and I fear this will take a long time.
The thing is: is it even possible to get reliable and precise spectral data from a sensor (which are simply needed to calculate duv and CCT without too much measurement error) with only 8 or even 10-20 channels? A real spectrophotometer uses at least 35 or even more of them. (With interpolation - which is easier with a high amount of channels - the result is even better.) The idea to create a cheap and easy-to-use light measurement device could be great, but it is not as easy as you would think of.
Maybe it would be a better idea to try to create a cheap and reliable spectrophotometer with 25 to 35 channels, but now it is too late I think.
I’m hoping for one of these:
Yes, but the software for analyzing the data has also to be good.
I saw this in my unread threads list today…
538 unread messages out of 531 total (and only 5 actually marked as read). So… based on that, it seems like about 12 messages got deleted more thoroughly than a regular user can do. I guess things got spicy!
After reading through the thread though… my overall impressions are:
- LM4 accuracy is a bit low, but it still provides some useful data. The ripple data in particular is helpful.
- LM4 is apparently only about $30 or $40, which makes the limited accuracy a lot more acceptable.
- An engineer from Opple uses BLF and is trying to make things better.
So I was thinking it might be a good buy. It’s cheap, it gives some useful data, and perhaps most importantly, having an Opple engineer here is a huge benefit.
But – it seems the app requires an account, and doesn’t work offline. And for me, that’s a deal breaker. If I must run a proprietary app, I run it exclusively offline without ever connecting to the internet even once… so if it doesn’t work offline, it may as well not work at all.
12 Thanks
I agree with your point, interpolation doesn’t work because I tried, the accuracy is not comparable to other methods. I think it’s impossible to restore a 401 or 81 sized vector from a vector of size 8, first let’s assume the light source is Led and it has 2 maximum and 1 minimum, this is not enough. as I posted above, two vectors(a and b) of size 401, they do dot product, the result is a fixed real number z, and b is the responsitivity curve which you already have(use monochromator to measure), and a has 2 maxima and 1 minima, “a” still has an infinite numbers of combination, each combination resulting in different Ra value. do you know where the maxima and minima located in the X-axis(380~780)? it’s just mission impossible, nothing else.
it’s kind of like you got the abstract of an article, and you want to restore the article, you can, but you just get an article with same topic, the article would be definitely different word to word with the “hidden” original article.
btw, more channels means more expensive, maybe larger in size too.
the summary is good.
LM3 supports thermal light source/monochromatic color/Led in calculating lux/cct/uv/cx,cy, and supports CFL/Led in measuring Ra/CS/EML, LM4 only calibrated with Led. that’s the difference. LM4 is more accurate in measuring Ra/CS/EML according to test results.
as for why LM3 supports more light sources than LM4, I’d say I’d like LM4 to support other light sources too.
I already convey the “login issue” to PM. she said she would consult with relevant people.
3 Thanks
You say there is not enough spectrum bands to reconstruct the entire spectrum.
However if you cannot reconstruct the spectrum to a reasonable amount then how do you calculate lux? If you have a sensor that can see the entire spectrum but not separate it out i can see this, however with a sensor that can only measure at specific bands then you must interpolate the entire spectrum to get to lux ( believe)
I say this because the LM3 appears to do a very good job at measuring lux with even fewer spectral bands.
I can guarantee that LM3 does not reconstruct the full spectrum. It just support more light source, instead LM4 only supports Led(only home lights, although flashlights are Led, but their phosphor proportion maybe a little different from home lights).
kind of like divide and conquer, algorithm for a whole is not comparable to several algorithms for all parts.