Opple Light Master 4 discussion thread (new 2023 model)

From the information given, I would say that it appears that Opple Home 3.3.1 may give more accurate R9 measurements, and from the information posted upthread by Koef3 and others, it appears highly likely that the Desktop version gives more accurate Ra measurements.

But while I don’t think it is unreasonable to trust Home 3.3.1’s R9 value slightly more, the data presented don’t seem to give a high degree of confidence one way or the other. Timing of purchase isn’t a very good guarantee your emitters are from the same tint bins as the references. Even if they are within the same bin, various measurements posted over the years by individuals with good gear seem to suggest that individual LED’s don’t always fall within the labeled bin.

To illustrate with the 5000K LH351D, for example, Zeroair’s measurement of R9 = 43 seems low in general for a ~90 CRI emitter, but not unreasonably so. The even lower measurement from the Opple Home 3.3.1 of only R9 = 30.1 would make me scratch my head a bit.

At the same time, there have been samples of the 5000K LH351D where R9 measured close to the 61.5 you measured with the desktop program. Here Maukka had a sample with R9 = 61 to 70 depending how hard he drove it.

With that said, I think it even makes a degree of sense that spectral reconstruction could compromise on the accuracy of saturated color measurements like R9 for the sake of reconstructing a spectrum that is reasonably useful for calculating Ra (assuming the actual spectrum is at least reasonably continuous like an LED, not spiky like a fluorescent light). Trying to fit a curve through 8 points could result in the curve having relatively large local errors in places. If the reconstruction specifically tries to mimic the shape of the black body reference curves, which extend well into the infrared, I could definitely imagine an otherwise very reasonable reconstruction tending to give elevated values in the red range, and therefore report R9 as higher than it really is.

In contrast, training a machine learning algorithm with the data from a sensor that has channels at 630nm and 680nm - both well within the R9 reflection spectrum - could quite credibly converge on a tuning that appropriately weights the values from those channels to produce relatively accurate R9 scores.

In any case, we always have to be careful to remember that the LM3 and LM4 are both built around hardware with significant fundamental limitations, and we’ll have to accept a relatively significant degree of uncertainty in exchange for the low price.

2 Thanks