Another entry in the “The bigger the blob, the better the job” soldering contest:
The four transistors across the bottom row let me test the simulation suggesting there’s no need for the 3× current gain mentioned in the App Note. Spoiler: future LM3909 circuits have the usual two-transistor mirror.
Adding some instrumentation required a bit of unsoldering and clip-lead action: to get the Tek current probe around the LED wiring:
The voltage probe is across the LED, although you’ll also see the voltage across the capacitor and differential voltages measured properly with the common clip leads on the battery negative terminal. I unsoldered two of the mirror transistors after verifying a single mirror transistor can saturate Q3.
Removing the AA cells and feeding it with 3 V from a bench supply:
The yellow trace is the voltage at the collector of Q3 = positive terminal of C1. The purple trace is the voltage at the LED cathode = negative terminal of C1. The fuzzy white trace is the difference of those two, showing C1 charges to about 1 V at the start of the LED flash. The white wedge over on the left marks the 0 V level, confirming the cap doesn’t enter reverse-charge territory during the flash.
The green LED produces a bright flash starting at 30 mA (bottom trace, 10 mA/div) for 15 ms. With 1 V on the cap, the LED + 39 Ω ballast resistor see nearly 4 V at the start of the pulse, because Q3 saturates around 20 mV.
Reducing the supply voltage to 1.5 V flattens the current and lengthens the flash to 35 ms:
The cap still charges to 1 V between on-times, but the lower supply puts barely 2.5 V across the LED + 39 Ω resistor and the current peaks at 10 mA. The increased duration turns the flash into a blink.
It’s good enough, so AA alkalines should last quite a while.