C12880MA，it has 200+ channels, sales person said it’s not evenly distributed channels.
below is its introduction,
Features
My recommendation: Focus on CCT and duv. Those are the most important metrics for many flashlight enthusiasts and to evaluate room lighting. It allows to find light that matches your current lighting.
Flicker should also be possible to measure with the sensor and allows you to troubleshoot issues for sensitive people.
The cause for all the trouble in this thread is the CRI estimation which cannot be done perfectly with this sensor. This should be reflected in the presentation of the data: add a note that is estimated and micht not be accurate. Ask for additional information from the user like type of light to use a proper calibration curve if that helps.
Overall make sure that the CCT and duv measurements work reliable with various types of lights (natural light, incandescent, fluorescent, LED etc).
But it is not only the sensor.
It is mostly the software which gives the correct results. And for this Opple (or whoever develops the software) has a very long way to go.
Its interesting to see in the Application Examples they have “Simple color measurements using smart phones”
This is really exciting, I hope Opple can implement a LM with this device, BLE, and an open protocol.
I think if this actually cost ~$250 (or even more) its business opportunity. Most of the spectrometers we have above LM4 are in the $1000 price range, and they have ridiculous limitations for their costs.
A new colormunki goes for $560. You can get used online but I prefer new.
It’s good to have access to good equipment. I myself did an integrating sphere on a BH1750FVI in 2014 due to lack of money )))). The cheap desktop spheres ($300±) didn’t have the necessary software functionality. In fact, I only used it to determine PCB thermal resistance and LED crystal temperatures.
even only measuring Led spectra, it is kinda difficult for CRI estimation. just like the graph I drew before, the curve with 2 peaks and the curve with 1 peak may result in the same value for a specific channel(approximately, the channel output value is the dot product between this segment of spectral density values with responsitivity curve of this specific channel), and we look in a reverse way, if you got a channel specific channel value, for example 25, how can you know the real density values of varieties of curves behind?
indeed, but the resolution really matters.
for integrating sphere, it’s quite straightforward to calculate CRI.
Of course, but what I wanted to point out: even if you have a good sensor, you also need good software.
And this is not easy, and even if you have the software, it is not guaranteed to get reliable results right from the start.
Basically this thread turns into the direction ’ cheap spectrometer for measuring light sources on-the-fly’, which is really cool, but also really hard to do it the right way.
I confirmed with supervisor that priority should be first with calibration of our own lights. our varieties of lights and this measuring LM device forms a closed loop, when we tell our customers that our specific desk lamp or ceiling lamp is awesome, how to prove that we are telling them truth? we bring out a LM device and measure our lamps in front of our customers, if our customers buy some 10,000 RMB of our lights, we can give them LM for free I guess(10, 000 is just my imagination, but a real number, anyway a big deal or transaction).
I feel regret that LM4 obviously is not a perfect product. It ought to be better. I am starting to learn colorimetry which I really didn’t know before as a CS major graduate.
As a lighting company you sure picked a complicated topic, creating your own colorimeter at a budget. Throwing huge money at something to measure the best possible data, then work with it is not so hard, but getting reliable data out of a cheap device sure is.
Best of luck on your journey, I hope you’ll get the LM4 to measure a bit more reliably. I ordered one, too. Should arrive these days.
here in China, most CS graduates work as computer programmer, there are a varieties of programming languages and platforms, and nowadays machine learning/deep learning or AI is a buzzword, it can be used in computer vision/natural language processing/speech recognition or E-shop product recommendation or somewhere else, I think most CS students either work on programming or work on AI-related algorithms. I never heard of someone with a CS background knowing a little bit of colorimetry before.
Steve I just became interested in the field also. Have you found any good books? I saw Measuring Color 4e, Hunt (Wiley)
I am currently reading “Colorimetry Understanding the CIE System by Janos Schanda” with a very slow speed:-)
Sure, this makes sense. Thanks for the business context. The more accurate your LM is, the more credibility the assessment for your customers will have. If a customer has an OHSP-350B and gets wildly different results, there will be some explaining to do 
hah, this scenario basically will never happen, because we calibrate with our own lights. it’s a balance between overfitting and generalization, we’d give overfitting a slightly more weight. so measuring our lights should not be a concern, while measuring lights from other sources/manufactures maybe is a concern although we are continuously working on it.
seems different Phosphor formulas caused problem, because nowadays LED lights are all blue light excited.
LEDs have pretty much always been, with just a few rare exceptions.
You all are having an inspiring conversation in here. I look forward to what comes from it.
Meanwhile I am just getting started, planning to buy an Opple or similar.
I would appreciate advice on which device to get. I’d prefer to buy something that will last me for a while. I am interested in measuring CCT, CRI, incl R9, lux/intensity, ideally a spectral graph, and flicker measures.
I assume lumen measurement require a separate setup?
Thank you for the detailed answer.
I understand that you can easily calculate various metric if you have the full density. I also understand that you may need the full density to calculate Ra, as it involves 8 different pastel colors (involving spectrum data across the whole range).
However, for R9, I guess you might not need the full density, as it is mostly related to the red part of the spectrum. Also, the red part of the spectrum is usually not as dynamic as the blue region. I have looked at several spectra of well-known LEDs, and I noticed there are not many variations in the red spectrum region. So, guessing the red spectrum regions should be fairly easy because there are not many complex details that are hard to capture with LM4’s coarse channels.
I am okay (less concerned) with LM4’s CCT, DUV, and CRI. Although they are said to be somewhat inaccurate compared to LM3, and many users seem disappointed, I don’t expect high accuracy from sub $40 equipment. If improvements can be made, they are welcome by all means.
However, R9 is where I encounter most problems. LM4 used to give strangely high readings for cheap LEDs. It appeared to be improved in the previous October revision. Now, it is not reporting any R9 numbers for cheap high CCT LEDs in the current December revision. I hope this could be addressed in the future revisions.
on the contrary, I think variables you mentioned is more accurate than LM3 now. CRI and R9 is dependent on spectral density, CRI has a range typically distributed in 80~100, it’s full range is 0~100. R9 is different, it spans a larger range, sometimes it even has negative values, for example -50, in a word R9 is more difficult than CRI to optimize. the October version you mentioned, I notice its R9 estimation error range is within about 5+ for some spectra I tested. so the trend is basically OK, for December version I need to support SDL lights developed by my company, SDL lights has CCT range from 1800K to 12000K, this may distort the original R9 calculation process.
R9 is indeed difficult for my current method, maybe later spectral reconstruction will solve this problem. you can refer to my post above. another way is using a more advanced/higher resolution chip.