Skip to main content

You are here

Supercap Solar Results

7 posts / 0 new
Last post
Supercap Solar Results

After reading Jean-Claude's initial information about supercap charging, I became interested in exploring the concept as well. I had done previous tests in which I'd charged a 1F supercapacitor for a few seconds then used it to power a jeenode for up to a minute or two. My conclusion was that the self-discharge rate was incredibly high, as I didn't realize they could falsely appear fully charged. JCW's insight prompted me to undertake a longer set of experiments.

  • Node CHARLIE: A 16MHz JeeNode with AA power board powered by a ~4.5V (open circuit) 100mA (short circuit) solar panel.
  • Node DELTA: An 8MHz custom design powered by a NCP1400 3.3V boost converter powered by a 2.3V (OC) 70mA (SC) solar panel.
  • Both nodes use a 1F 5.5V supercapactor as their energy storage
  • Temperature is provided as a reference as to when the panels are in the sun
  • The sketch transmits a 13 byte payload packet every 5 seconds

Images are Day 1, Day 2, and the start of Day 3 (where it started to rain)

For some reason CHARLIE, the AA board jeenode, always drops off when the supercap voltage hits around 2.0V. When I measure the +3.3V out, there is 3.3V. I can't figure out why it stops transmitting but it drains power pretty quickly at that point. I keep thinking there's something hooked up wrong, but really how complicated is it? Solar panel, diode, AA board's +/-.

Based on the curve in the graphs though, it doesn't look like it would make it through the night anyway. You can definitely see when it gets out of the sun the supercap voltage drops quickly to its "actual" voltage level, so they're not getting fully charged. I'm going to try upping the transmit delay to a minute to see where I end up for comparison with jcw's numbers.


Current draw increases when the input of the AA board gets lower (it has to get to 3.3V and must pull harder and harder).

Consider adding a 100 µF cap on the 3.3V line, as buffer for the brief TX current pulse. Maybe that gets it to last slightly longer.


I'll try adding a 100uF capacitor on the 3.3V, I think it only has 10uF right now. It seems like the AA power board may not like going from 4.2V all the way down to 0V. On a slightly newer setup, the system can make it until about 3am but again stops at 2.3V.

The same system, if completely discharged then charged to only 1.5V for a few minutes will transmit dozens of times before depleting the supercap. I should see at least 1 or two transmits as it dies, sort of like the DELTA setup which takes a couple hours of transmitting to fully deplete the supercap. The DELTA either has a 47uF or 100uF capacitor in it, so you may be right!

In any case I'm going to keep working at it and try a couple different configurations.


@Cpt for the Charlie case, perhaps try a substantial cap in parallel with the supercap and see if the '2.3v threshold' moves lower.


A couple more days of testing and it looks like extra capacitance does help! To save time I put caps on both the power lines, 100uF on the 3.3V and 1000uF on the input power. It looks like now the AA board on Charlie will take the battery down further. After this test I brought it inside and let it run down over the next night, and the AA kept powering it all the way down to 0.65V.

Now I'll try removing one of the caps and see which one did the trick.


@Cpt, the parallel conventional capacitance is acting as a pulse accessible store for the booster chip. The supercap charge storage mechanism can be modelled as a ladder of C's feeding their neighbour through a R. This reflects that charge stored "near" the electrodes is easier to move in and out than the charge stored "deeper" in the structure.
Given time, these charges equalise - this models well the odd macro-charging behaviour where the terminal potential rises faster than expected, only to find that the true stored Q is much less than the C.V implies.

The reverse effect is upsetting the booster chip - it demands a low impedance pulse of current to load up the energy stored in its inductor. Despite being a sawtooth shape, with only the "near" charge available to supply the pulse, the transient voltage collapses and the conversion fails well above the spec sheet 'minimum input voltage'. The charge is there, but can't be extracted so quickly.

Adding the reservoir capacitor supplies the short term need and is "topped up' by the 'Johnny-come-lately' deeper charge out of the supercap. It needs some 'scope traces to optimise the size of C - I'd expect little replenishment goes on during the power conversion process, so C has to take the burden of supplying almost all the charge required. The main topup perhaps occurs during the relatively long 'quiet period'.

I'd speculate that using a reservoir C may also help the overall supercap lifetime. The useable C just fades away over time. Vendors often quote lifetime with shallow discharge rates (e.g. ~10%) implying some interaction with this 'close' and 'distant' storage mechanism.

P.S. Any idea why the Charlie discharge line is more 'staircased' now?


Well that certainly makes sense now as to how the internals of a supercap works. I had assumed they were like amazing electrolytic caps that could produce near instantaneous amps worth of current (based on some articles I had seen about how to charge them without pulling too much current). JCW's post about their slow charging opened my eyes a little to how they work but now I think I'm actually understanding based on your ladder model.

I suspect the capacitance on either side could potentially be a fix, on the 3.3V side the cap there smooths the current draw from the regulator so it doesn't need to pull as much from the supercap. On the supercap side it acts as directly as buffer. It is hard to find the exact moment things stop working on the 'scope because the boost converter is working ok for billions of cycles then pkew something goes wrong (just the right amount of current draw) and oops I missed the trigger.

The staircasing is because the data is being fed into an RRD so I have 3 minute average steps for the past 24 hours, then 20 minute steps for the past week and 60 minute steps for the past month.

Premium Drupal Themes by Adaptivethemes