AVR Freaks

1. Of course I'm not going to scrape the last few μA when I'm dropping about 20mA on the LED. On the other hand some of these measures will have a noticeable effect: according to the site I linked to, they will save on the order of tens of mA, which is about as much as one full LED. Moreover, the LEDs will not be switched on all the time, but only during the night and the nap time: in all, only about 12 to 16h a day (depending on the precise schedule I program). This means that efficient handling of the MCU should about double battery life.

2. And my main point in my original post was indeed not about saving power on the MCU, but on using the MCU to save power on the LEDs themselves. Using current-limiting resistors wastes about 50% of the power (from a 3.7V battery). Surely with PWM we can do better than that.

I need an RTC chip because this is not a brightness-controlled clock, this is a time-controlled clock. (If you really want to know, this is for indicating to the younger ones when to wake up and when to sleep).

I was thinking on the lines of “feed PWM through a low-pass filter”. This has the advantage of requiring only a few passive components.

Of course this depends on the load, and a LED is highly non-linear, which complicates modelization.

Edit: I wrote a short SPICE simulation (using a LED model “found on the Internet”), which says that this should be fine:

.TITLE Filtering PWM through low-pass LC filter + LED
.model LED D (IS=1a RS=3.3 N=1.8)
L1  1 2 100u
C1 2 0 10u
D1 2 0 LED
.OPTION OUT=80
* Make a frequency-domain plot to show cut-off frequency:
Vin 1 0 AC 3.7
.PLOT AC  VDB(2)(-30,0)
.AC 1k 100k decade 3
.DELETE Vin
*PWM signal at ~ 16 kHz with 33% duty cycle:
Vin 1 0 PULSE(IV=0 PV=3.7 PERIOD=60u WIDTH=20u)
.PRINT TRAN V(1) V(2) I(D1)
.TRANSIENT 0 .001 .0000001 > a.dat

I admit that I did not plot the V-I characteristic of my LEDs! The order of magnitude comes from a quick Internet look-up (I use cheap 3mm LEDs bought on Aliexpress; these don't come with a datasheet) and might be outdated. I will measure their current load tomorrow.

The night light will be mains-powered most of the time, but I want it to have as much autonomy as possible in case I need to set it up in a temporary location. Also while writing the code, I realized that there are several ways of achieving the result, and one good way to decide is to select an efficient one (instead of my current prototype, which busy-waits for 1s between each poll of the DS3231 — this works for test purposes, but is about the most inefficient design possible!).

The circuit is described by these lines of code:

D1 2 0 BlueLED
L1 1 2 10m
C1 2 0 1u

Vin 1 0 PULSE(IV=0 PV=3.8 PERIOD=60u WIDTH=50u)

and should look like this (the code above does not have a damping resistor, so that R1=0 in this case):

The voltage source V1 is the PWM output.

(Don't look at the type of the LED, I drew this in Circuitlab and don't know anything about their LED models).

Oh, and 10 mH inductors can be quite cheap.