I hadn’t realized the “standards compliant” road design caused the death of so many street lights, but the dead bollard population is definitely under-represented. In round numbers, every traffic circle (“intersection”) always has at least one smashed bollard in addition to the vestigial stumps of those removed rather than being replaced.
The upright bollard is a relic of the earliest installations, back before they realized a bollard with an eye-level light glaring into drivers’ eyes weren’t an effective design, particularly along a road lined with dead-black / non-reflective posts.
Spotted in the Town of Poughkeepsie Highway Department compound.
Having just replaced Rev 1 of the amber running light with Rev 3 (about which, more later) on Mary’s Tour Easy, both the front and rear lights began blinking erratically. Given that they have completely independent circuitry, this strongly suggests a power problem.
Herewith, the headlight circuit voltage:
The voltage should be a constant 6 or 6.3 V, depending on which description you most recently read. That is the case with only one light attached, so the problem occurs only when running both lights.
The four pulses come from the amber LED’s Morse code “b” (dah-dit-dit-dit) with a 85 ms dits; the first dah pulse should be three times longer than the dits and definitely isn’t. The rear light’s red LED stays on continuously, except for two dark dits, so it draws a constant current and does not produce any changes in this trace.
Both lights have 2.0 Ω sense resistors setting the LED current to 400 mA, which corresponds to 250 mA each from the Bafang controller’s 6.3 V headlight circuit. The headlight circuit’s total of 500 mA should work fine, although the “spec” seems to be basically whatever the OEM headlight requires.
The Rev 1 amber light ran the LED at 360 mA with a supply current around 450 mA. That light and the rear light on the back ran fine, so the supply seems to have a hard maximum current limit at (a bit less than?) 500 mA.
The least-awful solution seems to be backing off both LED currents to 360 mA to keep the total supply current well under 500 mA.
A test setup on the bench allows a bit more room for probes:
Some heatsink tape holds the LED to the far side of that oversize heatsink.
The input signal (top trace) arrives from a function generator set to blip the MP1584 regulator’s Enable input at 4 Hz with a 7 ms pulse:
The purple trace is the voltage across the 2 Ω sense resistor. The MP1584 datasheet says the regulator soft-starts for (typically) 1.5 ms, during which the output ramps upward at 600 mV/ms to 800 mV , whereupon the actual regulation commences. The amber LED forward drop adds 2.5 V to the sense voltage, so the regulator produces 3.3 V from the 6.3 V bench supply input.
The cyan trace is the output current through the LED and sense resistor, also ramping up to 800 mV/2 Ω = 400 mA to drive the LED at 1 W.
The furry section shows when the regulator is actively regulating, with the output voltage rising and falling over a small range to maintain the average current (via the sense voltage). Successive Enable pulses may have longer, shorter, or completely missing fur, with no predictable pattern. Increasing the duty cycle doesn’t affect the results, with the fur sometimes extending for the entire pulse and sometimes being completely missing.
I think the regulator can settle in one of two metastable states. The best case has a constant voltage producing a constant LED current, with the sense voltage remaining within whatever deadband keeps the error amplifier happy. When something knocks the sense voltage out of the deadband, the error amp starts the usual regulation cycle, which will stop when the minimum or maximum voltage of a cycle remains within the deadband:
The ripple shows the regulator running at three cycles per 20 µs division = 150 kHz, far lower then the MP1584 datasheet’s maximum 1.5 MHz and the typical 500 kHz in the test circuits. Perhaps a low frequency lets the designers use a cheap PCB and not worry about pesky EMI issues.
In any event, during this pulse the ripple amplitude gradually decreased as the output voltage settled at the point where the error voltage variation stayed within the deadband. The typical amp gain is only 200 V/V, so it’s definitely less fussy than something build around an op amp.
For whatever it’s worth, a 7 ms flash from a 1 W amber LED at 4 Hz is way attention-getting in a dim Basement Laboratory. You wouldn’t need an Arduino to produce that signal, even though I like the Morse capability.
The four traces on the right show the BatMax NP-BX1 lithium batteries (cells, really) originally stored about 3 W·h when they arrived in March 2020. The four solid traces to their left show the capacity dropped to a little over 2 W·h after two riding seasons. Batteries B and C started out above average and are now below, for whatever that means.
The red dotted trace shows the effect of not using the NP-BX1 test holder for that length of time; those homebrew contact pins apparently needed some exercise.
Having replaced the Planet Bike Superflash on Mary’s Tour Easy with a 1 W red LED, testing the eight Panasonic Eneloop AAA cells that have been powering it (and the one on my bike) for the last four years seemed useful:
The sheaf of curves over on the right came from the first full charge, with the untidy collection below them show the current state after a full charge. This is at an unreasonably high 500 mA discharge.
The overall capacity has dropped by 10%, which isn’t all that bad, but the 10% voltage reduction toward the end of the curves is a Bad Thing for an LED flasher intended to run from 1.5 V alkaline cells. In practice, I recharge the batteries once a week while they are still going strong, but the difference between alkalines and NiMH cells is obvious even at full charge.