Driving a Cree Q5 LED with one NiMH cell - anything much better than an LTC3490?

[0xdeadbeef]

In an attempt to pimp up a cheap LED flashlight with a somewhat crappy driver, I first considered the obvious approach using a PAM2803 but then stumbled over the LTC3490 which seems a bit more efficient and doesn't need a shunt resistor or external diode.
I plan to use a voltage divider at its CTRL/SHDN input to be able to reduce the default current of 350mA to something in the 250mA-300mA range.
Admittedly the LTC3490 is quite a big more expensive than the PAM2803 but as I don't have any commercial interests, this doesn't matter for me. Plus, the LTC is actually easier for me to purchase.
My main concern is that I want to drive 250-300mA through the LED and the energy stored in one (!) 1.2V NiMH battery should be used as fully and efficiently as possible without depleting the battery below 0.9V.
Space is also a consideration as the PCB has a diameter of only 13.8mm and it needs to be single sided (placement wise) due to the battery/case contact on the lower side.
So in a nutshell: is there any much better step-up constant current LED driver for this purpose than the LTC3490?

The attached picture shows my current layout for a 13.8mm diameter PCB. Please ignore that DipTrace can't seem to display plated half-holes properly.

[0xdeadbeef]

Well, OK, I guess this goes to show that nobody really shares my niche addiction
Anyway, some pictures of the final layout with a more realistic render of the inductor used (WE LQS 5040). The model is a scaled down WE LQS 6045 though which doesn't fit 100% but is pretty close.
The voltage divider selects 250mA which I guess is bright enough and should improve efficiency and battery life when using a single AA NiMH battery (Eneloop).
There's also a spice model attached which indicate that an efficiency of ~80% is not totally unrealistic.

[BravoV]

Subbed, keep us updated, especially the final result, thanks. 

EDIT :

Just remembered there was an old discussion at CPF years ago, on adding a resistor at pin 5 and 6 (parallel with internal sense resistor) to increase efficiency, or boost the max current (2 cells configuration).

[0xdeadbeef]

I can somewhat see how bypassing the sense resistor could boost the maximum current, but how would it increase efficiency in single cell configuration?

[EDIT]
Might take a bit longer. PCBWay suck. They let you select castellated holes with no extra price in the original quote but charge 5 times the original quote after the "audit".
I guess I have to remove the castellated holes from the design which is somewhat problematic.

[0xdeadbeef]

The PCBs ordered about three weeks ago at OSH Park arrived today. I'm quite happy with the result, especially considering the price.
I paid $12.80 (free shipping) for 2x12 PCBs (smaller and bigger design for different flashlights).
I hope to populate a few of them this week.

[SiliconWizard]

So in a nutshell: is there any much better step-up constant current LED driver for this purpose than the LTC3490?

Well, I don't know of any similar LED driver that would fit your requirements and have better specs. But if you're willing to compromise and not use a constant current output, you could use the TPS61021A boost converter. It has much lower quiescent current, much smaller package, wider input voltage range and much better efficiency at low input voltages as far as I've seen.

[0xdeadbeef]

Well, I don't know of any similar LED driver that would fit your requirements and have better specs. But if you're willing to compromise and not use a constant current output, you could use the TPS61021A boost converter. It has much lower quiescent current, much smaller package, wider input voltage range and much better efficiency at low input voltages as far as I've seen.

Well, yes, I think a constant current mode is mandatory for a proper LED flashlight. Using only a boost converter would mean that only the LED limits the current - which is usually a bad idea. As far as I can tell, even most ultra-cheap "Cree Q5" LED flashlights use a (more or less functional) simple current control approach but the very few that rely on a step up converter without current control need a resistor to limit the current. Firstly this is bad since setting a certain current (brightness) is a bit of trial and error (varies with LED type, temperature etc. and the forward voltage itself is current dependent). Secondly, the current limiting resistor usually has a much higher resistance than the shunt used for current control. So electrical energy is transformed into heat instead of light. Nothing you want in a flashlight.
Quiescent current is not an issue as the flashlights have a mechanical switch (usually dis-/connects ground from the battery).
Voltage range is not really an issue either as I'm focused on flashlights with 1NiMH cell. Of course the LTC3490 would also support two cells.
Anyway, I made the decision long ago. I have all the parts and now the PCBs - bit too late to reconsider my choices.

[SiliconWizard]

Well, yes, I think a constant current mode is mandatory for a proper LED flashlight. Using only a boost converter would mean that only the LED limits the current - which is usually a bad idea. As far as I can tell, even most ultra-cheap "Cree Q5" LED flashlights use a (more or less functional) simple current control approach but the very few that rely on a step up converter without current control need a resistor to limit the current. Firstly this is bad since setting a certain current (brightness) is a bit of trial and error (varies with LED type, temperature etc. and the forward voltage itself is current dependent). Secondly, the current limiting resistor usually has a much higher resistance than the shunt used for current control. So electrical energy is transformed into heat instead of light. Nothing you want in a flashlight.

Yes it's admittedly not ideal for brightness homogeneity.

As for losing power in a series resistor, you're of course right, although the wasted power depends on the resistor's value, so you can minimize it by setting the output voltage just a notch above the max LED's Vf from its specs, but that makes it all the trickier to tweak, and it's still waste. Not good.

There are means of using a classic boost converter more efficiently though. Attached is an example schematic. The idea is to use a small shunt resistor in series with the LED and use the sense voltage as the feedback voltage. Amplifying the sense voltage allows to use very low-value shunt resistors, so the waste is negligible and you can get much higher efficiency overall, given that again most LED drivers (the one you chose included) are not optimized for very low input voltages, but you have a lot more choice with classic boost converters. The schematic given is just an example (with available models in LTSpice) for a 100mA constant current LED driver. Depending on the selection of boost converter and opamp, it can need some tweaking to avoid oscillations, but that gives some ideas to try.

The shunt value, the gain and the feedback ref voltage will define the current: Iled * Rshunt * gain = Vref or Iled = Vref / (Rshunt * gain)

Quiescent current is not an issue as the flashlights have a mechanical switch (usually dis-/connects ground from the battery).

It adds up to the overall power consumption while on, but I admit at ~300mA output it's going to be negligible as it's stated for under 1mA for the LTC3490 (didn't see any more accurate figure on the DS but ~1mA sounded like a high quiescent current to me for a 300 mA output boost converter at first - it's just that I'm used to track waste when designing low-power battery-operated circuits, but here it doesn't matter much. )

Voltage range is not really an issue either as I'm focused on flashlights with 1NiMH cell. Of course the LTC3490 would also support two cells.

Well, for the higher end, it's not, but for the lower end, it may be. I see the LTC3490 is given for 1V input min, but with NiMH AA cells, average discharge curves I've seen give a usable operating range down to 0.8V-0.9V so you may lose up to maybe 10% of the capacity. That may not matter a lot, but it's worth mentioning IMO.

As for efficiency, though, it's another matter. The LTC3490 has a pretty poor efficiency for input voltages under 1.2V (under 60%) (in conditions similar to yours), which will be the area you're working with when using a NiMH cell. Not that great actually.

The TPS61021A has over 80% efficiency for input voltages down to 0.9V at 300mA output. That's quite significant a difference.

Anyway, I made the decision long ago. I have all the parts and now the PCBs - bit too late to reconsider my choices.

Alright. And yes that should work well enough.
Those were just some thoughts. Could give things to think about for future projects.

[0xdeadbeef]

It adds up to the overall power consumption while on, ... sounded like a high quiescent current to me for a 300 mA output boost converter at first

IMHO, the quiescent current is the current that the LTC3490 consumes if enabled but not driving a load. It doesn't necessarily have to add 1:1 to the current when actually driving a load.
Anyway, at that point, it's part of the efficiency which is pretty high (up to 90%). In my use case (1 cell, 250mA) it will be more like 75% or maybe an optimistic 80% at 200mA.
The typical approach with a PAM2801/2803 should have worse efficiency (but they either only give figures for higher voltages or currents. Plus I guess they don't take the shunt resistor into account).

Well, for the higher end, it's not, but for the lower end, it may be. I see the LTC3490 is given for 1V input min, but with NiMH AA cells, average discharge curves I've seen give a usable operating range down to 0.8V-0.9V so you may lose up to maybe 10% of the capacity. That may not matter a lot, but it's worth mentioning IMO.

I'm not sure that I get your point. This thread is about solutions for a single NiMH cell. NiMH cells have a relative flat discharge voltage curve which breaks down quite rapidly at the end. So it shoudln't really matter much if the low voltage protection kicks in at 0.8V, 0.9V or 1V. As a side note: the only time I see 1V mentioned in the datasheet is the maximum value for the startup voltage - which is not the same as the configurable low battery voltage. Actually the datasheet doesn't give a typical value here for the 1 cell configuration but a range of 0.8V (min) to 1.12V (max). Let's assume the typical value is in the middle, this would mean 0.96V.
Obviously, this range comes from tolerances of the internal reference etc. so getting precisely 0.9V (over the full temperature range and considering aging) might me a bit optimistic. Again, I would think it doesn't matter so much anyway for NiMH cells as long as they are not discharged under 0.8V.

As for efficiency, though, it's another matter. The LTC3490 has a pretty poor efficiency for input voltages under 1.2V (under 60%) (in conditions similar to yours), which will be the area you're working with when using a NiMH cell. Not that great actually.

As said before, the discharge curve of a NiMH is rather flat at least until the last 20% or so. When the voltage drops much under 1.2V, the cell is more or less depleted anyway.
Also keep in mind that the efficiency is the total efficiency. There is not shunt resistor etc.

The TPS61021A has over 80% efficiency for input voltages down to 0.9V at 300mA output. That's quite significant a difference.

It's an interesting device for sure but it would be quite a challenge to fit that extra OpAmp etc. on the very small PCB even though the inductor could be a bit smaller.
Besides, I don't think there's any kind of low battery protection. So what would protect the battery from being discharged below 0.9V? Sure, would be possible to add that somehow but that would cost even more space.
And at the end, it would have to be proven that the efficiency of that whole circuit is really much better than 75% or so.

[0xdeadbeef]

I populated two PCBs and the good news is that they run so far.

There are a few thing though that I didn't really expect.
What concerns me most is that the LED doesn't light up for anything below ~1.26V when I use my bench supply (R&S, should be accurate to 1mV or so).
As soon as it lights up, I can lower the voltage down to ~0.8V before it switches off. So the low battery protection works as expected, but switching on at 1V doesn't.

Then again, and this gives me even more headaches in a way, when I tried to start the lamp with a nearly fully discharged Eneloop (Logitech mouse stopped working, battery measured ~1.11V without load), the LED lit up. It began to flicker after a few seconds, but it started. I did a check cycle with that battery in my charger and it reported 6mAh left in the cell (discharges to 0.9V). So just looking at this second test, the behavior looks OK, but I'm puzzled why I seem to need a higher voltage with my bench supply.

Furthermore, while the current regulation seems to work in principal (i.e. about 1A is drawn at 1.4V and 1.1A at 1.2V), the LED most certainly becomes dimmer for lower voltages (especially under 1V). While it's most probably not a bad idea to lower the current if the battery voltage goes down too much, I can't really find a hint in the datasheet that this is supposed to happen.

Anyway, assuming that the output current is around 260mA at 3.2V or so (didn't measure, I fear neither is not fully true), the efficiency of the whole system (without the LED of course) seems to be more like 63% at 1.2V:

    (3.2V*0.26A)/(1.2V*1.1A)*100% = 63.03%

That's not really totally awesome. But I guess still better than the typical Chinese lamp.
Still, I kinda hoped for something in the >=70% range at 1.2V. Of course my setup is a bit flawed as I trust on output current of my bench supply and there might be a slight voltage drop on the test wires (4mm banana plugs). Without really measuring input /output voltage and current this is a bit of guesswork. But I would think the number is not totally off.

[floobydust]

At low input voltages the (input) ripple current is higher.
Don't forget a weak battery has high internal resistance, so the 10uF input cap you have is probably not big enough. Upsize it and see if that helps.

[0xdeadbeef]

Well, I don't really care for a little ripple. Should I? Actually the capacitor is only 4.7µF but I don't see how that would improve the startup behavior.
I just don't get why the circuit behaves completely different with a bench supply and compared to a real cell and/or the LTSpice simulation.
With the exactly same circuit, the spice simulation starts at 0.9V. Also with a real Eneloop it starts at <1.2V. Just with the bench supply it doesn't.
I checked the voltage supply but it's spot on (down to 1mV). I also measured my wiring but it's below 0.06 Ohm. I'm cluesless.

Anyway, my best guess is that there is a contact resistance involved. Actually when I increase the serial resistance of the voltage supply in LTSpice to something like 0.3Ohm (from 0.025Ohm which should fit an Eneloop), things start to get ugly at 0.9V. I guess my measurements are flawed but it's kinda tricky to get a low ohmic connection to the internals of the lamp without soldering (which I tried to avoid). Probably I have to go that way to get more sensible measurements.