Ideas wanted for low duty cycle PWM generator or similar.

Thread Starter

goingalong

Joined Oct 7, 2015
21
This request relates to creating a torch/flashlight with a very low light output mode - one that has sufficient illumination to be useful over a few feet in total darkness, whilst being "always on". So I am looking at tens of microAmps (9v) battery powered on a small pcb footprint - fixed rate pwm seems the most efficient way to go. Testing so far indicates that I need to drive a LED at a PWM duty rate of < 0.4%.

Thus far I have tried a 555 running at 160kHz driving a 14 bit counter. Using a couple of D-type latches I let one clock pulse through when the counter upper bit ticks over. That way the pwm ratio is clockwidth/totalpulses. But that is 3 ic's plus ancillary passive parts and a significant clock rate, ending up at a low refresh rate at the led. What I would prefer is just the one ic and a clock/update refresh ~ 50Hz.

I haven’t tried yet but I read that a 555 won't get down to this sort of pwm ratio.
There is a LTC6992 pwm generator, but that is 5v only.
There are generators designed for PSU control (also not tried yet) but these tend not to cover the full 0-100% duty range - or so I read.
Would appreciate experienced views on what might be my best approach so I might be able to avoid at least some blind avenues.
 

Papabravo

Joined Feb 24, 2006
12,407
You need to start with a very specific kind of high efficiency LED. Tens of microamps is not going to provide much luminous flux so you will need all the help you can get. A typical LED uses between 10 and 20 milliamps which is about 1000 times more current than you want to expend. A typical High Efficiency LED uses 2-3 milliamps which is still 100 times your budget. An experiment you should try is to power a proposed LED with 20 microamps of continuous current to see if it meets your requirements. If it does not do that, then running it with a PWM is not going to work.
 

Picbuster

Joined Dec 2, 2013
982
Use a PIC ( smallest one) feed from a mcp1701 Low Dropout Positive Voltage Regulator (2.0 μA Typical Quiescent Current)
Any pulse, PWM .
If pic's PWM is not sufficient you build it in software ( 4-6 lines of programming).
As Papabravo said light is energy and 1-2 mA lids are not cheap but you could setup a test and see what happened.

Picbuster
 

Thread Starter

goingalong

Joined Oct 7, 2015
21
Thanks for these replies. I looked briefly at low current LEDs but not long enough to find one in white as reds tend to dominate, also the other mode of the torch is to have a more useful beam - hence I thought that a mid-road 10-20mA led plus pwm would be the way to go. Maybe that choice is wrong and I would be better using two leds, one for each mode. In which case my low-light led could be a more efficient red.

I looked at using a PIC but running purely on my experience (again mid-road) assumed the current demand would be closer toward the mA end, plus the efficiency loss in regulation down to its VDD level. If 2.0 μA is realistic then that is a whole new opportunity.
 

Papabravo

Joined Feb 24, 2006
12,407
Thanks for these replies. I looked briefly at low current LEDs but not long enough to find one in white as reds tend to dominate, also the other mode of the torch is to have a more useful beam - hence I thought that a mid-road 10-20mA led plus pwm would be the way to go. Maybe that choice is wrong and I would be better using two leds, one for each mode. In which case my low-light led could be a more efficient red.

I looked at using a PIC but running purely on my experience (again mid-road) assumed the current demand would be closer toward the mA end, plus the efficiency loss in regulation down to its VDD level. If 2.0 μA is realistic then that is a whole new opportunity.
One way of achieving a duty cycle PWM in a PIC is to let it sleep between pulses. When it sleeps, it's current consumption is measured in microamps. I'm still not clear on how you think 10-20 microamps is going to provide any luminous flux at any wavlength, white, red, green or whatever.
 

Thread Starter

goingalong

Joined Oct 7, 2015
21
Truth is I am trying to replicate a torch I already have but is no longer in production. I measured its demand as 50 μA and best I can tell from the circuit components the led is being driven at 80mA constant current - so I am assuming it is being deliberately overdriven for short intervals to increase the apparent brightness - apparently this is common. What I have not been able to get at is the PWM details - firstly I have no high impedance probe capability and secondly its tight into a closely packed smd board and I have already managed to blow one diode. Should try harder and buy a proper scope - I know.

I cannot measure its luminous output but do know that in total darkness, when my eyes are adapted, it is easily enough to light a room sufficient to avoid walking into furniture.

Anyways, rather than trying to copy the original circuit which is probably a 10 years old design I thought it would be good to start anew from a refreshed component base.
 

Papabravo

Joined Feb 24, 2006
12,407
OK. Now we have something to work with. Question: what is the average current consumed by an 80 mA pulse that is 1 millisecond long every 100 milliseconds. If your answer was 0.8 mA or 800 μA you are correct. If the period is much longer than 100 msec then you'll be able to see the flicker. If it is much shorter then there will not be enough time to turn the LED on at such a low duty cycle. Unfortunately there is no pat answer to this problem. You have to pick an LED and then empirically determine the appropriate period and duty cycle to meet your requirements. What's worse is that every LED, even from the same lot, will be different.

Good Luck.
 

Thread Starter

goingalong

Joined Oct 7, 2015
21
Great answer: I start with the led, not the technique. Also I had not realised how variable each lot of leds can be. No wonder the original is not in production any more.
 

crutschow

Joined Mar 14, 2008
23,377
You should be able to get the PWM duty-cycle you need with a simple circuit.
Here's a reasonably simple one using one IC that is adjustable from 100% down to 0% duty-cycle.
 

Thread Starter

goingalong

Joined Oct 7, 2015
21
I just rechecked my prototype, not the original. Driving a stock high brightness white led at 12mA, 235Hz and a duty cycle of 0.2% I get a full circuit drain of 50 μA (excluding the 555 pulse source running at 120kHz). My cheap moving coil ammeter is probably influencing and reading low. When I put it to measuring the led current alone I get zero reading - even though the led stays lit enough to be useful. I expect the coil is reacting to the pwm impulses but the dominant drain appears to be the circuitry rather than the led.

I excluded the 555 because it is not a low drain cmos type but I have included everything else in the pwm circuit - ie the counter and D-types and my power switch toggling stuff. So I think I may be back to the pwm question. Am I going the right way - with the counter or is there an off-the-shelf generator IC I should be looking at. Obviously that is a question seeking discussion only.;)

And ... obviously I need a much better way of measuring small currents.
 

Papabravo

Joined Feb 24, 2006
12,407
I just rechecked my prototype, not the original. Driving a stock high brightness white led at 12mA, 235Hz and a duty cycle of 0.2% I get a full circuit drain of 50 μA (excluding the 555 pulse source running at 120kHz). My cheap moving coil ammeter is probably influencing and reading low. When I put it to measuring the led current alone I get zero reading - even though the led stays lit enough to be useful. I expect the coil is reacting to the pwm impulses but the dominant drain appears to be the circuitry rather than the led.

I excluded the 555 because it is not a low drain cmos type but I have included everything else in the pwm circuit - ie the counter and D-types and my power switch toggling stuff. So I think I may be back to the pwm question. Am I going the right way - with the counter or is there an off-the-shelf generator IC I should be looking at. Obviously that is a question seeking discussion only.;)

And ... obviously I need a much better way of measuring small currents.
That is 8.5 microseconds out of 4.25 milliseconds, and the average current is 0.2% x 12 mA = 24 μA so there is another 25 or so μA unaccounted for. I'm actually surprised that the LED can respond that fast and that it has enough persistence to give the illusion of being lit.
 

Alec_t

Joined Sep 17, 2013
10,384
Here's my simple-circuit suggestion. The component values shown should give 33uS wide 15mA pulses at a repetition frequency of ~57Hz (i.e a ~0.2% duty cycle), using a 5V supply.
LowDutyCyclePulser.PNG
 

Alec_t

Joined Sep 17, 2013
10,384
If using a 74HC14 (or other CMOS circuit), remember to ground the inputs of any unused gates in the package, and to decouple the IC supply pins with a 100nF capacitor.
 

crutschow

Joined Mar 14, 2008
23,377
Below is a circuit using a CMOS Schmitt trigger that can be adjusted over a wide-range of PWM duty-cycle.
It will also work with the 74HC14.
Discussion here.

upload_2017-10-19_9-59-17.png
 

Thread Starter

goingalong

Joined Oct 7, 2015
21
Some feedback: I tried the circuit from Alec_t, which works well and gets down to the type of duty cycle value I want. The second circuit provided by crutshow is very similar. Implemented using a single schmitt inverter in a SSOP-5 package it will be very compact. I use the schmitt output to directly control a mosfet in a rough and ready manner, which in turn is switching a led and constant current source. The fet load drops the frequency by a few cycles only, which is no bother.

Clock.png

The best I can tell the current consumption is >100 microAmps for the PWM generator alone, which is higher than I would like. The CD4093BE schmitt consumes less than a few microAmps at idle so the bulk must be going through the Capacitor/Resistors and output gate. I can't measure it well but the load does not vary significantly inside the cycle rate range I am interested in.

I have not tried the first circuit from crutshow because providing the -Ve reference adds components.

Unless I can think of a way of reducing or re-using the energy I think is being lost in the schmitt circuit I might give a mcu solution a try, dropping it into sleep mode during the 10mSec off period in each cycle. Alternatively maybe use a crystal source.
 
Top