I don’t see what CCT the tested LEDs are, but it are some interesting numbers. Comparing them to @koef3’s test of the LHP73B, which is also a 7070 LED, it does slightly more lumen at 10A and slightly less lumen at 20A, both with a slightly higher Vf than the LHP73B.
There is one big difference between both; the light emitting surface (LES) of the F150R is a lot smaller. I did a quick/rough calculation (so not super accurate) and came to ~12,5mm²; according to Koef3’s test of the LHP73B, it has a 25mm² LES.
So it looks like a very interesting emitter to put big reflector light; close to 7000lm at 20A, but with (way) more intensity than the LHP73B. LES appears to be slightly larger than the SBT-90.2 and SFT-90, but it also produces more lumen. (~6800lm compared to 5000lm of the SBT/SFT-90, according to Koef3’s tests.)
Very tempted to buy that black camouflage version, as soon as Simon also lists offers the F150R in 5000K or 6500K.
I still have a couple of unused 22mm FET drivers from Convoy, that would nicely combine with a L21B and the F150R. Maybe with some spring bypasses, thicker driver wires and a high drain battery, it could drive the F150R close(r) to the max 30A it is listed at.
It was I early and I wasn’t fully awake, so I didn’t realize that the 150 in F150R is the mil size.
150 mil equals 3,81mm. 3,81 devided by 2 (to get the radius of the LES) is 1,905. π x 1,905² = 11,40mm².
I’ve probably ‘effed up’ the math somewhere, but if the F150R really would have a LES of 11,4mm², it would be less than half that of the LHP73B, with roughly the same output.
It is also just slightly larger than the 9,6mm² of that of the SBT-90.2 and SFT-90, which are hard to get past 5000lm in a flashlight, while the F150R (on paper) can go quite a bit higher.
If the tests are comparable, the higher output at low current suggests that the F150R uses a die+phosphor that is intrinsically more efficacious than the LHP73B. The efficacy only goes down at higher current because the much smaller die induces a much higher power density at the same current.
Amazing eyeballing work! I tried fitting the data to 2 regression models: (1) quadratic, and (2) power law with exponential decay. Modeling the SFH43 up to 20A suggests that the true output at 25A and 30A lie between the predictions of the two models; for the F150R that would indeed be around 7600 and 8000lm!
Just received two T9’s a SFT-25r and a 519a. Seems like the mode switching length of time has been increased to right around a second if not a bit longer. My preference for this is 1/2 a second so it’s easier to go back to 1% with a quick hold. Current muscle memory for this keeps advancing the mode up. Otherwise the lights are great.
I’m not sold on convoy 2xAA efficiency vs the 1xAA lights. A bit brighter, twice the battery needs and half the runtime. 1xAA runs max at 0.5a, 2xAA runs max at 1a. True, you could set the 2xAA lights to the 1/10/50% group and be comparable to 1xAA runtime and brightness, but that still uses two batteries vs the 1. Is my logic sound or am I misunderstanding? Please correct me if I’m wrong.
What’s the benefit to the 2xAA configuration? I’d be much more interested if this driver was in a 2x18650 or 2x21700 light with a battery adapter to take 2xAA. Since it takes 2x14500, it should handle any size lithium ion in 2x configuration. It would be nice to see 1xAA adapters and compatibility on the larger 1x lithium ions too.
I haven’t put batteries to timer yet, just going off the listed max draw on each driver and the mode groups claimed %. As an example and ignoring losses, I’m assuming the 10% mode on 2xAA draws 100ma since max is 1a. Maybe the % refers to only the 6a max and the driver caps at 1a for 2xAA? Would be 600ma draw if that’s the case. The available documentation doesn’t read that way to me, wouldn’t be the first time I was mistaken though
Edit: I got the package last night and stepped onto a plane this morning. Packed the new lights, didn’t pack a NiMH charger though. Will do some tests when I get back.
I think that’s the drive current of the LED. It will be drawing more from the cell because it uses a boost driver.
The 1xAA lights deliver around 500mA to the LED, and the 2xAA around 1000mA. The actual current draw from the cells is the same, but the 2xAA delivers double the voltage to the driver which doesn’t have to boost as much.
I see, a linear forward difference corresponds to a quadratic model, which would predict 7500lm @25A and 7800lm @30A when fitted to the 5 points you listed.
It would be interesting to collect some more data and see which model fits better, quadratic or power law with exponential decay. The quadratic model is certainly simpler, requiring only 2 parameters.
Curve fitting is tricky with LEds because of thermal droop, which isn’t usually measured. The best fit I’ve been able to get so far is with a rational decay curve.
Matches Simon’s numbers pretty closely. It seems to suggest that it will roll over at 49.5A at 9438 lumens. It’s already making 111W at 30A, so not sure how realistic that is.
I agree that it’s very tricky business, with the main contributor being that the rate of droop is not well-understood. The problem at hand is that many models perform similarly well, and it’s unclear which one is the most physically sensible.
A simple quadratic model of the form aI(b-I) performs the worst out of the group. A power law with exponential decay, of the form a * I^b * e^{cI}, performs much better at the expense of adding one more parameter. In fact, the residuals never exceed 9 lumens in absolute value, which is even lower than your model (with max residual at least 18), though it’s not evidence of a better model because real-world measurement noise is on at least the same order as the residuals. Furthermore, a power law with the exponent as a parameter does not seem physically sensible.
Your model incorporates more information by including also the Vf as a predictor, in addition to current. My understanding (please correct me if this is not the case) is that the baseline Vf (bandgap voltage) of an LED is determined by its wavelength; as current rises, the Vf also increases, but all of the excess Vf above the baseline is dissipated as heat rather than as emitted light. In other words: without accounting for droop, the output should be proportional to current (I only) rather than power consumption (I * Vf(I)).
Under this assumption, a sensible model should take the form aI * droop(I, Vf(I)), or just remove the dependence on Vf entirely to get aI * droop(I). Some LED literature suggests that reciprocal droop (as you used) is physically sensible, but it doesn’t seem to do as well empirically as exponential droop, with higher residuals. The physically less sensible power law with exponential decay has by far the lowest and most patternless residuals.
I’ve uploaded some data for the F150R and SFH43, plus a few models, to play around with. Aside from the poorly-fitting quadratic model, every single model mentioned overpredicts the 30A output of the SFH43. LED model fitting | Desmos
I’ve tested two lights with that LED and built one. Basically the numbers are all over the place between the LEDs and you really don’t know what you’re going to get as far as performance. The one in the project light was a NBT160 or one of those 9090 size round die LEDs from Aliexpress did okay. The Fireflies version (a 4500k ffl909mx) had good performance for a warm-neutral tint low CRI LED in the E90. The NBT160 in the Amutorch DM90S wasn’t great (same neutral-warm tint) at all for a big flashlight. It just depends on how you drive them and whatever bin you ene up with. Until we get tests of the Convoy ones (with smaller footprints 7070 vs 9090; there may be a limiting factor there-from experience I highly doubt input currents over 23 amps will produce meaningful output gains), we won’t know what these are capable of, and until people test them professionally in actual lights. Still, they look promising at least.
You’re absolutely right about how Vf works. I had it in there because thermal penalty is a function of power, so it takes Vf, but excluding Vf from the efficiency component makes more sense. Removing it seems to improve the fit a bit. Max residual goes down to 13.5.
One thing I noticed is that you did the fit with emitter #1, whereas I took the average Vf and Lm of the 2. It looks like the exponential decay power law fits the single emitter a bit better, while the updated rational decay with your suggestion fits the averaged values a bit better.
Realistically, there’s like a 50 lumens at 30A. Would be interesting to see the actual test data and see how they hold up
