thank you for your explanation, I saw 3 matrices in LM3 to map sensor values to XYZ, kind of like divide and conquer, namely M_general, M_thermal and M_color. I don’t know which path those variety of “tints” will go？
in LM4, only 1 matrix is provided, so the sensor values measured whether under white Led with tints or other plain Leds will be multiplied by this single matrix. I guess the other colleague did not use white Led with tints for calibration.
3 Thanks
As I understand LM4 not work with light sourse with hi infrared part in spectrum, isn’t it?
I see some special procidure for this case in documentation.
TimMc, could you give me some advice on how to improve the precision of those “tints” Leds? as I currently know, if you get XYZ, then you could calculate CCT & duv(those formulas can be found in wikipedia or other articles, right?). then the problem lies in how to get accurate XYZ, right? XYZ is inferred by least squared mean with (channel values<F1, … F8>, XYZ) pairs. so if I add more pairs, especially with pairs under a variety of “tints” Leds, then the matrix I inferred is more accurate and more representative???
other metrics are kind of sophisticated, and are dependent on spectrum values(380nm~780nm), color Leds does not need to calculate CRI, those “tints” leds still need to calculate CRI, am I right?
1 Thank
The right way to calculate Ri is use
calibration methods – ‘reconstructed spectrum’ . From the spectrum get all interesteing quantity by standart formulas.
Way you choose is strange and as we see give wrong results.
As7341 could measure right many types of light not only led but daylight, incandescent bulbs and so on. You don’t need thousands reference source for do that.
Cri is valid for near white light source. Light source that is close to black body locus. So you can calculate CRI, CCT for any light, but the more distance from BBL, the more meaningless these values.
1 Thank
thank you for your explaination
this method is workable. giving wrong results is because one parameter is not workable for all types of Led spectra. 730 version would give much better results. It’s kind of equivalent to “reconstructed spectrum”. I know “reconstructed spectrum” is the official method outlined in the as7341 documents.
maybe this method takes less effort because you can easily gather or hand-construct thousands of spectra according to different phosphor composition(maybe pump too).
and “reconstructed spectrum” method is not always workable for all cases of Led spectra, let alone other types of spectra , the “reconstructed spectrum” is not identical to the real spectrum, so errors always exist.
1 Thank
I wish I could but I would need study it to understand. I don’t have a good understanding of the formulas required (CIE 1931?) and what datasets can be used from the AS7341. 
There is another discussion over here:
So you tried spectrum reconstruction method?
The other advantage of spectrum is you can just add standard formulas to calculate some new quantity in future e.g. some Tm-30-15 metrics.
BTW you mention that you calculate Ra by ML, it is not right . Ra is by definition mean of R1-R8 so calculating Ra independent from Ri you get non-self-consistent data.
I didn’t try spectrum reconstruction because even when I got reconstructed spectrum I didn’t know how to calculate Ra and circadian stimulus then. I recently learned how to calculate Ra, CS is much more complicated to calculate using reconstructed spectrum. so it’s easier to regress all the values.
but in the future, I will try spectrum reconstruction.
Ra is the mean of the first 8 Rs by definition, however I found Ra calculation is workable with both methods in this 730 version.
1 Thank
In current application Ra not equal to mean R1-R8 and error sometimes is relatively big. If you already fix it is good news.
@stevechang Any updates/progress to LM4?
In the process of measuring with the Zebralight SC64 series, I measured using the LM3 and LM4. Note the difference - mainly, the LM4 DUV seems wonky.
In discussion on Discord with another user, they quickly put the DUV calculation formula in a spreadsheet, the inputs being the X and Y coordinates. Based on that, can the LM4 not accommodate that?
Somewhat related: the app for the LM4, can you please change it so I don’t have to login? I love that with the LM3 app, you open it and press the “start” button - that is it. Whereas, the LM4 app require lots of user input to get to the device.
| SC65c HI | |||||
|---|---|---|---|---|---|
| OPPLE 3 | |||||
| CCT (K) | CRI | R9 | LUX (@ 1M, CD) | DUV | |
| H1 | 4150 | 93.1 | 9202 | -0.0011 | |
| M1 | 3893 | 96.2 | 972 | -0.0019 | |
| OPPLE 4 | |||||
| H1 | 4245 | 93.8 | 66.1 | 9135 | 0.0007 |
| M1 | 3929 | 96.9 | 77.8 | 918 | 0.0001 |
| SC64c LE 519a 4500K DD | |||||
| OPPLE 3 | |||||
| CCT (K) | CRI | R9 | LUX (@ 1M, CD) | DUV | |
| H1 | 4259 | 97.2 | 5543 | -0.0073 | |
| M1 | 3893 | 96.2 | 517 | -0.0063 | |
| OPPLE 4 | |||||
| H1 | 4387 | 96.2 | 81.8 | 7907 | -0.0045 |
| M1 | 4130 | 96.9 | 72.9 | 735 | -0.0033 |
| SC64w HI Stock | |||||
| OPPLE 3 | |||||
| CCT (K) | CRI | R9 | LUX (@ 1M, CD) | DUV | |
| H1 | 4530 | 79.3 | 10702 | 0.0018 | |
| M1 | 4176 | 81.9 | 960 | 0.0021 | |
| OPPLE 4 | |||||
| H1 | 4733 | 78.4 | 27.0 | 10658 | -0.0010 |
| M1 | 4309 | 84.3 | 22.8 | 966 | 0.0007 |
| SC64c LE Stock w/Lee Minus Green | |||||
| OPPLE 3 | |||||
| CCT (K) | CRI | R9 | LUX (@ 1M, CD) | DUV | |
| H1 | 4018 | 94.8 | 4274 | -0.0005 | |
| M1 | 3966 | 95.9 | 403 | 0.0011 | |
| OPPLE 4 | |||||
| H1 | 4065 | 91.4 | 100.0 | 4473 | -0.0004 |
| M1 | 4016 | 93.7 | 97.1 | 421 | 0.0022 |
| SC64w HI XHP35 Hi 5000K | |||||
| OPPLE 3 | |||||
| CCT (K) | CRI | R9 | LUX (@ 1M, CD) | DUV | |
| H1 | 5188 | 93.5 | 9055 | 0.0008 | |
| M1 | 4858 | 95.6 | 688 | 0.0003 | |
| OPPLE 4 | |||||
| H1 | 5325 | 91.0 | 100.0 | 9189 | 0.0018 |
| M1 | 4961 | 94 | 87.4 | 612 | 0.0030 |
Opple has been saying they were going to give an updated app to fix the DUV error in the LM4. Im hoping they also fix the CIE chart, atm it does not work at all (it does on the LM3).
The new LM4 app is supposed to become available on July 30th…
fwiw, the issue w the LM4, is not the actual DUV calculation… the core issue is that the LM4 produces XY values that are different (more wrong compared to a Sekonic), than the LM3…
3 Thanks
CIE chart bug fixed. XY & duv will not be fixed in July 30th version. It’s the other colleague’s responsibility. this update involves CRI/Ra/CS/EML. I heard from the app team that August 30th version would support firmware update. measurement precision would be elevated especially in low lux environment in the future.
7 Thanks
thanks for the update 
So, CIE chart now works, to show the Wrong DUV… 
I still do not recommend Opple 4… The DUV on Opple 3 is more accurate.
Imagine how it feels to be an early buyer of Opple 4, and month after month a fix has been promised, and Not Delivered.
I no longer believe the DUV will ever be fixed on Opple 4… Plus Opple was a Major Pain when I asked for a refund, even after winning my dispute with Aliex, and Paypal, Opple simply would not refund…
Two thumbs down for Opple 4, I never use mine…
Not Recommended…
(ask me how I really feel sometime)… 
Hopefully some of the other things you said will not be true…as far as fixing things.
Bu, the fact that they did not want to refund your money, even though they know that the device does not work properly, is inexcusable! It means that people need to know before they buy, that they could just be throwing their money away.
3 Thanks
you are correct, CIE chart now works to show the wrong duv.
I personally feel sorry for that, I gave your duv inaccuracy feedback to the other colleague long ago. maybe there does exist some bureaucracy issue.
4 Thanks
It takes time to implement changes, especially in a large company. I appreciate that Opple is taking the time to improve the software. 
1 Thank
I wonder why they do not open ML4 Bluetooth protocol. We all pay only for device, so open source app would be huge advantage for Opple…
There are a lot of features that could be realised using LM4 hardware, but I doubt that there will be anything other than the basic functionality in official app.
2 Thanks
sorry for all of this.
there are two approaches to retrieve sensor data, one is through ble protocol, the other is through serial protocol. I’ll give you the commands that can retrieve a lot of information from the device. first you need to prepare type-c wire(just normal Android charging wire) and a serial port, cut the USB end of the wire, and connect it to the serial port, then you can use software like “sscom.exe” to retrieve info from the device.
as for ble protocol, maybe later.
6 Thanks
instead of writing a mobile app utilizing ble protocol. why not write a Windows app based on Qt or MFC and open source it? maybe this will also give me a lot of help 
4 Thanks
