The problem you forgot to take into account is regulation.
You want regulation for brightness to stay the same over the course of the runtime.
Also, keep in mind that at the end of the battery’s charge, 7135s saturate at 0.1V (so a 2.9V LED can still maintain full brightness down to 3.0V at the cell), and a switching driver (buck, boost, or buck-boost) still needs a way to sense current through the LED, usually the small-value chip-resistors, to drop some voltage (that’s proportional to current).
At 3.0V or so, a cell’s voltage will start to drop like a rock anyway very quickly, so that might be a few minutes extra runtime at best. Just watch a cell’s voltage during a NOR test. And that might only be at 700mA or so in the charger.
I did not forget about regulation. It is a separate issue from the problems due to pulsed output though. I choose to focus on a problem I personally consider more important, but the solution was discussed in the first post. Even though the solution is trivial, it comes with additional complexity and cannot implemented as external add-on to existing FET boards.
Firstly, the LC smoothing circuit discussed above does not include current-sensing resistors in its simplest form. Secondly, current can be sensed with a Hall sensor which has negligible internal resistance. Look up ACS712 - that’s a little chip that converts current into linearly scaling voltage 0-5V.
Even the best buck or boost driver at any cost can't match the current? Sorry? You say this because you haven't seen them, don't you? In order to see them design assessment, willpower and funds are required. Buck, boost or buck/boost driver's aim is not to provide raw power but a proper DC output signal in any mode and regulation.
If you aim for amps, ¿how about sticking multiple cells in series feeding 1S emitter(s) with a high current dead short switch? This is what is being done in flashlights with MOSFET “drivers”, only in a more controlled way (matching cell and emitter stages). And I see at least some times (maybe more than that) not the best solutions are taken, like for example the MOSFETs used in the Sofirn C8F series flashlights. According to their datasheets their RDSon values creep much higher below ≈3.8V VGS. I wouldn't choose MOSFETs like that, but ones with good RDSon performance at low VGS values. Maybe they did that because of fear of frying the emitters with very high current batteries.
Where is the progress here?
Hall sensor current sensing? I am all ears dsk3. Glad to hear something new. 
Interesting conversation here, of course, and it’s all relevant to one’s perception to be sure. Me? I have a horrid memory, so the light that stays in regulation til the cell is about dead is somewhat disconcerting to me…. I much prefer seeing the light dim over the life of the cell as a reminder that I will soon have to source a freshly charged cell. To suddenly find oneself with a dead cell and little or no light is not conducive to the way I use a flashlight. At least with the FET driver I KNOW the cell is approaching end of life and I can head towards a fresh cell or a charger. When the light doesn’t change much I tend to forget how long it’s been running, suddenly it steps down and then shuts off and if I am not applying the 2-is-1 1-is-none theory then I am in the dark.
Not to mention that in smaller lights (all lights really but especially single 18650 lights) a regulated circuitry can lead to overheating issues and thermal step-downs with lower output anyway, not much different from a dying FET light once the light steps down on the regulated circuit. And in a Boost circuit, with the light in a constant chase for the optimum brightness level vs. optimum heat level, the output is not consistent due to fluctuations in the step-down-step-up pattern of the thermal regulation. So again, for me, the constantly dimming pattern of the FET light is more to my usage style and preference.
Everyone has their own style of usage of course, and in varying climates the nature of the light also changes. So to each his/her own…
If you want to regulate even smoothed current, you’d probably have to.
And feedback almost necessarily needs an error signal, ie, “wobble” from what’s desired, to regulate it.
One reason yicky PWM is used vs linear regulation is to keep the LED’s output from varying in color. Yeah, color. Its output spectrum will vary whether at low currents (typically greener) and higher currents. PWMing it keeps the output the same spectrum whether at 1% or 10% or 100%.
Nice, but are we still talking about switchers vs linear anymore?
First, it needs 5V, period. No 4.2V down to 2.xV. And it’s an extra chip to save on 0.1V headroom? And another to generate the 5V it needs?
Hey, great if we’re powering a GT or something, but replacing stacked 7135s? 1S cells and “3V” LEDs?
No one’s gonna stack 40 7135s to get “regulated” 14A to an LED. Of course it’d go to a FET solution, whether DD or what.
Even a lowly 8×7135 driver in an S2+ does this. I found out that when I was using one I apparently didn’t charge in a while. Wondered why it was suddenly getting dimmer than I was used to seeing. Sure enough, it started blinking its LV warning.
That’s why I don’t see linears/7135s going away any time soon.
This is getting very tangential to the initial subject. I do realize people have different needs and assign different values to product features, but my interest is almost completely driven by increasing power efficiency with a minimal loss of max current/brightness in a simple, easy to produce circuit. I do not care about color shift or even about regulated current.
Incidentally, the ACS712 does not need exactly 5v (I mentioned 0-5v only as an example in case of regulated Vcc - common in, say, ebike applications). It’s output will linearly scale with current between Vcc and zero. That means that its output signal - voltage - will also change with Vcc(=battery voltage). That sounds bad, but actually that is helful. Remember that the ADC converter in microcontroller can also use Vcc as reference. The net result is that the firmware will see the correct current measured independent of actual Vcc. That’s neat, isn’t it? There is of course an operational range for both microcontroller and hall sensor. In a specific example of ACS712, datasheet calls for Vcc in 3.3V-5V range, but in practice the range is certainly large than that and may be sufficient for 1S whole range of operation. But, again, I am not really interested in the subject of regulation.
Ummm, acto the datasheet at
it’s a 5V-only chip.
(min/typ/max = 4.5V/5.0V/5.5V)
Unless there’s another Allegro ACS712 current-sensor chip…
My bad. I was looking at ACS758 datasheet, not ACS712. The chips are very similar, except for different current ranges they are designed for. It is surprising that one of them is more restrictive about supply voltage.
Why not just design a driver, get PCBs made by OSH Park, build one and test for yourself? If you have an idea which you think can work, why not just go ahead? That’s what I do…
So publish the design — let’s see how the Chinese imitators can cheapen it!
Catches on fire after reaching more than 3A
Is there a reason people here prefer OSH Park over, say, JLC PCB?
If you live in North America, you support a North American company, and always get cheaper shipping in small quantities and about the same shipping times(in Canada anyway)
Otherwise, it depends. For the times I’ve used it, OSH Park has such good customer service compared to JLCPCB.
They are more expensive, but for some reasons: higher wages, they only offer gold-plated ENIG finishes on the boards, smaller production runs.
That said, when ordering PCBs in large quantities, JLCPCB rules because they offer a very high quality service for a very decent price.
I don't know about anyone else, but I feel wasteful putting 7135s behind a Luxeon V. The damn things are often burning over a volt!
I will avoid being judgmental. To me, regulation is great and a smoothed DC output too.
The led tint changes with the amount of driving current, but hardly. If you don't like it, you can get a different led in the first place.
I prefer to have full blast independently of battery state of charge. If you usually end up using your light a lot and fear going out of fuel, how about carrying spare cells?
…
I had written more already, but I'm ruling it out because it is long and I have better duties for my time (i.e. no preaching over deaf ears).
I'll tell you, however, some of you seem to like the simplicity of the “FET” concept. But it is drawback flooded and certainly not for everyone. Now take it to a whole new level, level “zealot”. Zealots are real men and do not even need a microcontroller to PWM the mid modes for them, they do it by pressing the e-switch at ludicrous speed or connecting the MOSFET gate to their own neural 
system. Like me, they can sense the level of battery SoC, so no need for low voltage warnings or shutdowns… :-D
0K, enough already. O:)
I do not think this needs any special design. Just stick an inductor in series with the emitter and some capacitance in parallel, and PWM at the right frequency to obtain a more or less smooth resulting output. Very simple.
What did I just witness?
You answered too fast BlueSwordM, I've edited my message (no preaching O:) over deaf ears).