Personally it starting to step down around 3.5V is what I would consider ideal but thats just me. As it steps down the voltage sag will drop a lot and thus you will get a fair amount of time before it steps down again.
Once it is at ~3.5V it is almost empty afterall, it is time to start thinking about reducing the output anyways.
Well, a 30Q at a 5A discharge has ~22% remaining capacity when it reached 3.3V. At 3.5V, it's much higher - about 39%. That's a pretty big chunk. This is straight off of HKJ's discharge curves, if I'm reading it right.
Sorry, tweaked the #'s.
Also note HKJ tests cells to 2.8V discharge, while the Samsung cell itself is spec'd at a 2.5V discharge cutoff. This probably explains his rated capacities being lower.
I am still with TA on this one.
Pretty sure the voltages on the HKJ curves are measured under load as well. The 5 A curve should line up well with what you would read in a typical single emitter light while in full DD. According to that 5 A curve, if you turn off at 3 V, there is only about 0.125 Ah (4%) left (if you can somehow maintain the 5 A, which we know is not possible and another bucket of worms).
FWIW I step down at 3.4 V (3.2 V for 219Cs) and switch off at 3.0 V (2.8 V for 219Cs). Like TA I prefer to step down and conserve the remaining power. If a spare cell/battery is available I can just swap it out. If not, then so much the better to hang on to the power that is left.
Ok, so you do compensate for the voltage sag, and see that's it's higher on lower Vf LED's. I saw this today comparing a XPL2 light to a XPL. What about triples or quads? Think you will see the same issue. I just tried 4 cells in the 4X BLF Q8 prototype - worked fine, but then tried only one of those cells - the single cell went right into LVP, when still about 0.2V above the threshold value. Then went back to 4 cells, no problems - still runs with no LVP.
Of course I'm not thinking bout this for my own usage. For my own lights, the LVP doesn't matter to me much anyway, like you said. But for documenting and committing to a spec, users will challenge you.
Yeah, the low Vf emitters are tough on the cells. But for general use, in my opinion, a more conservative LVP cutoff is preferred, for three reasons:
The step-down point need to be at least 0.2 V above the LVP trip level to be of any use. Personally I prefer a 0.4 V margin here. It could be a user-programmable parameter (0.0 V - disabled, 0.1 V, 0.2 V, 0.4 V)?
Having the documentation clear on the set-points and the impact of cell impedance and operating current is probably the best you can do. The alternative of having an LVP trip level that scales inversely with the mode can be messy and risky.
I agree, it's complicated. Since we don't know the amp draw or the parallel battery count, we can't do it smartly in firmware. Like in the example I mentioned with the 4X Q8, using 1 cell or 4 cells makes a difference. Not sure, but it might be worse on quality low resistance cells, since they can draw more current at a lower voltage, drained state.
At 3.5V, from what I said above, there's about 39% of the battery's life left for a 30Q discharging at 5 amps, assuming I'm reading this correct, so it's significant.
I think you're right though - there has to be some clarification that the first trip point can be effected by factors causing voltage sag, when a cell is under a high drain state. I think this is a direct effect of us doing this crazy DD/FET thing. Single cell triples or quads of the new low Vf LED's has to be the worse case.
Keep in mind that the battery is not at 3.5V when it trips at 5A. The voltage @ 5A is 3v (or whatever the step down point is set to), which leaves a mere ~150ma charge remaining, or about 5%.
Notice in those HJK plots that the tests that he stopped early, the voltage recovers a lot when the load is cut, you are seeing the same thing in the light. Don’t confuse the reading you get after the load is cut with the voltage the cell was at under load. The under load voltage is what matters when comparing it to the plot.
‘Hi-drain’ cells should actually do better than low quality (or regular cells) at the same current: Lower internal resistance, lower voltage drop.
As TA said (and I tried to say in #143), I think the HKJ voltages are measured under load. So on the 5 A curve, when the cell shows 3.5 V, it is while under 5 A load. And if you then remove the load and measure the voltage again, it is going to be around 3.7 V for 30Qs.
Well, my point why hi drain cells would do worse is the amps will be higher when it reaches the trip point (whether 3.0V, 3.2V, 3.4V, etc.), so the voltage drop is greater because of the higher amps.
Ok, I see it now, comparing a 0.2A curve against a 5A curve on the 30Q.
To optimize battery usage, you really want to be close or below the point voltage starts dropping at a higher rate (curve dips down). 3.2V to 3.4V or below looks good for the 30Q and SANYO GA cells. Interesting the 26650 cells in general have a much steeper drop off.
I think you are overthinking it. Pick a good step down point and run with it. I agree with DEL, Personally I am a fan of a ~3.3v step down with a ~2.8V cutoff. This way I get a good runtime during the stepdown process and I still have a worthwhile amount of runtime left in moon mode should I need it.
You could even allow for a low ~10-15 lumen mode after the final step down instead of moon mode only. 10-15 lumens is enough for most things and yet would still give you several hours of runtime even at those low voltages. Then have it step down to moon mode only once voltage drops below 2.9v.
DEL - any thoughts on building a LDO version of your Y3 driver? Came up in this thread: https://budgetlightforum.com/t/-/45440. someone else look'n to do the same thing I'd like to do - XHP50.2 in a Y3. I got the extra Y3 extenders.
For the Yezl Y3 2S lights, popped an LDO in there:
The new C3 should help if 2S + inductive spikes are killing the 7135. The new R6 is to feed an indicator LED (D1).
Would like to put pads for the LED itself as well, but did not find an elegant way. Any preference/suggestion there Tom?
Dang! You are way too quick on this 
Wait, photobucket and Chrome are not cooperating, once again. Wanted to check my pics. Think I mounted the LED on top of the switch. For these somewhat transparent switch covers, you have to maximize the position of the LED to keep it's power draw relatively low. If it were to lay flat on the driver, it's not an efficient setup.
Ok, here it is:
Electronic art 
That's 30 AWG stranded wire. Got it in a few colors, comes in very handy.
This seems the best way for a Y3, though might be a pain for some working this way.
The resistor value of 4.7K is relatively low because it's a yellow LED, I believe, in this case. Colors like green are pretty bright, so a higher resistor value can work well. In the BLF Q8, I'm using a 15K resistor - still pretty bright even but nice low power draw. The Q8 has two green LED's mounted directly on the switch PCB - best setup, and maybe helped by a more translucent switch cover, and I thnk maybe pretty bright green LED's they are using.
Yes, don’t see a way to point a board-mounted LED to the switch boot, even if side-edge mounted. Top-center of the switch like you did will be optimal.
Should then just try to have two convenient pads around the switch to wire to. Can flip R6 and the D1 pad around and add a mirrored ground pad west of the switch.
That all sounds good!
OK, cleaned it up a little and added pads to air-wire an indicator LED. The share is up:
C2 is the only critical capacitor. It needs to be at least 2.2 uF and good quality. 4.7 uF is a good number for all three to keep it simple. The ‘standard’ 10 uF is a little excessive, but will also work. All pad sizes are 0603, but they are generous and 0805 parts will fit.
C3 is experimental. It should suppress the inductive ringing that finds its way back to the cells when the FET is PWMed. Hopefully it will prevent the 7135 deaths that Tom and TA experienced with 2S DD+1 lights.
LDO regulator is the MIC5235-5 in SOT23-5 package. Pin 1 in, Pin 5 out. Others may work, but the MIC5235 has low parasitic drain, reverse polarity protection and reverse leakage protection. The board can be used for 1S without the LDO, just place a Schottky across LDO pads 1-5.
Nice work DEL! I see on the OSHPark listing you are suggesting 10 ohm for R5 instead of 4.7, and the 47 ohm R4 has been eliminated. I assume the R3 is still 100K, I believe?
Pretty sure I got quality 4.7 caps, so good there. I believe those are the LDO's I have, bought here:
https://www.arrow.com/en/products/mic5235-5.0ym5-tr/microchip-technology
Wow, just noticed I paid $10.71 for qty 10 back on Nov 1st, and now you can get them for $7.50 -- big drop.
For R1 and R2, I believe my tables for setup up for 330K and 47K for 2S cells, least the values I use.
That will work Tom.
With a 2S light we have the luxury of MCU feed voltage to burn, so R5 can go larger. Larger helps the R5/C1 filtering. But whatever you have in 1-22 ohm is workable.
R3 can be 22k to 100k, as usual. Larger is more benign. I always use 100k and never had the inadvertent flash phenomena.
R4 is a nice-to-have, was a little too crowded to fit it and have nice routing.
The MIC is still $1.20 in singles at Digikey. And the price break is only at 25 units. Should try Arrow someday.
K, then. Should be ok for me to place the order now? Just want to be sure you are ok with the design, or want any more time. Are you ordering them?
Dang, you got me think'n bout this issue addressing with C3... Got a Maxtoch Mission M12 want to use a XHP50.2 in, plus couple other 2S lights with a XHP70 or XHP50/XHP50.2.
It should be ready to order. Mechanically it is identical to the 1S version. Electrically it is not tested of course. I do not have a use for it, unfortunately. Only two Y3s and both with single-cell tube only.
The C3 add-on should be easy to test on any driver. It simply sits over bat+ to bat-. Can even tack it on between the spring and the ground ring.