Thank you very much for explaining! To check my understanding:
LEDs at a particular wavelength (say blue) have some “intrinsic” voltage associated with that wavelength
Whenever an LED operates at a voltage higher than this wavelength-determined voltage, the excess voltage is wasted as heat
The above partially explains why LEDs are less efficient at higher currents: higher drive current corresponds to higher Vf which increases the voltage gap
Linear drivers burn this difference at the driver, FET drivers propagate the voltage difference to be burned at the LED
FET drivers are efficient at transporting power from battery to LED, but paradoxically never achieves higher overall system efficiency compared to other types, assuming ideal implementation
The above alone suffices to give a linear upper bound on an LED’s output as a function of current. This is a nontrivial result because power consumption is super-linear in current. But emitter tests indicate strongly sub-linear growth, which I believe is attributable to the other items you mentioned:
Do you happen to know of some good expository sources on these?
The loss of efficiency shifts from the driver to the emitter.
FET drivers are inherently very bad, while linear drivers are great when you can tune emitter voltages to be as close as possible to the source voltage.
In many lights that run with fully regulated voltages sources like the grid, you can just add enough emitters unless your LED VF gets super close to the wall voltage.
That’s one of the techniques used by Philips’ Ultra Efficient (180-210lm/W) lights: they just string enough LEDs in series until they get super close to the wall voltage. This allows them to make a super compact driver design that is super efficient.
In the context of flashlights, linear drivers are always more efficient until you get to ridiculous power levels.
This is where it gets very interesting in the next example with an LHP531.
Let’s say we have a 350mA linear driver and an emitter that has a forward voltage of 2.65V at that current with a 4.2V source.
That gives it an efficiency of 63%. Assuming the emitter has an efficiency of 200lm/W at 350mA, that’ll give us an adjusted efficiency of 126lm/W.
Driving that same LED at 350mA from a FET direct drive source using PWM at absolute maximum power gives us an efficiency of 68.2lm/W.
System efficiency of the 350mA linear driver with a VF of 2.65V and 200lm/W: 126lm/W.
System efficiency of a FET driver driven at an average of 350mA: 68.2lm/W.
Thank you very, very much for this detailed explanation!
If the source is regulated at a high voltage, it makes perfect sense to keep stringing together emitters in series until the combined voltage is just below source voltage, which makes direct-drive possible and efficient. It comes as a surprise to me that this is an exception rather than the norm–why don’t most bulbs do this?
Thank you for the flashlight-specific explanation! In summary, what I understood is that:
Both linear and FET drivers incur the same systemic efficiency loss via burning of the voltage gap; however, in a PWM scenario, FET incurs additional losses at the emitter due to operation at a higher instantaneous current.
The systemic efficiency gap between them can conceivably narrow if the FET current is lower than the emitter’s absolute maximum current, e.g., if the battery cannot supply enough current or if the Vf is very high. However, the gap never reverses direction, which means a linear driver’s system efficiency is always >= that of a FET driver.
As a result, even though FET drivers are more efficient at transporting power from battery to LED, it is reasonable to informally say that “FET drivers are less efficient” because they always give rise to lights that are less systemically efficient. The distinction between driver and system efficiency rarely comes up in casual discussion, the same way the terms efficiency and efficacy are often used interchangeably.
S21E is my favorite flashlight I’ve found so far. I have it in the SFT40, LHP531, 719A and an SST-40 which I really don’t care for.
Anyone build anything neat with them? I have a spare host and a 3v 8.5a FET with Andruil 1.0 just waiting for a LED or three to shove in it.