Those who have been following along in my power bank testing articles will note that the vast majority that I have looked at have hardly been impressive. Some things seem rather fixed – for example, the voltage regulation which is partly determined by the algorithms in the microcontroller.
However, I did have a thought one day – what would it take to improve the ripple performance of a power bank which has already decent voltage regulation (on average)? How much more might the manufacturer have needed to spend?
A Bit of Nostalgia
Remembering back to the early days of my electronics experimentation, when linear plug-pack transformers were more common, a big issue would be the 100Hz ripple encountered from full wave rectification of 50Hz AC mains. A common solution is to stick a few big filter capacitors and you could smooth it out enough for your liking. My methodology back then was to always err on the safe side and throw a little more bulk capacitance – you rather have power that’s smoother than expected rather than ripple/hum-laden power!
Switching converters generally operate through a cyclical process of storing power into an inductor and releasing “bursts” of energy from the inductor which are then smoothed into the required output by controlling the duty cycle (mostly) of the storing process. This can happen at rates from about 150khz up to several megahertz.
Generally there is a tradeoff involved in design when choosing the frequency – with higher frequencies, filtering requires smaller components but losses can be greater in inductors.
Lets Try it Out!
Being a practical sort of guy, I just thought “meh, lets just try it out and see what happens.” The first candidate for modification is the Power on the Go! unit which had decent regulation but a failure to meet ripple specs. It wasn’t too far out, so it was a good candidate to try playing with.
Lets first piggy back a 1200uF Matsushita Low-ESR electrolytic capacitor (overkill) and a polyester greencap on the USB output and see what happens.
Well, what do you know? Ripple improved. Slightly.
Formerly, the ripple at 1A averaged 627.2mV, now it is 248.1mV. Looks like the bulk capacitance did help the ripple somewhat because of the relatively low 300khz switching rate. Unfortunately, the main contributor to ripple is the fast, high frequency rise spikes. How can we deal with this?
I know! Lets try piggybacking a ceramic capacitor onto it.
At this point, the unit failed to start up, and removing all of my modifications made no difference. It seems I have killed it. Damn. Failure is always an option.
It seems I may have inadvertently shorted something out at one stage, which shouldn’t have made contact, and that may have destroyed the switching MOSFET or microcontroller. There wasn’t any heat, smell or visible sign of distress.
Lessons Learned So Far
It seems that I might have been too hasty with my attempt to modify, and now that power bank rests in pieces. Nice cells though.
Using very large bulk electrolytics with low ESR seems to be a bad choice, after some further research, as they often draw a very high current to charge up when the converter is starting up. This “overkill” was okay back in the days of mains transformers as they could tolerate abuse, but it seems switching converters with “closely” specified MOSFETs might not be as happy.
In fact, using electrolytics isn’t a good idea at all at high ripple frequencies. Their apparent capacitance falls quite quickly as a function of frequency, and really aren’t good for applications above 100khz. I sort of knew that, but didn’t know their effectiveness fell off so quickly.
The guides also point to another enemy of the non-ideal capacitor, normally denoted XL, which stands for self inductance. This property affects the capacitor’s “speed” at reacting to changes to input.
As a result, for high frequency operation, ceramic capacitors offer very quick response and low self inductance but very limited capacitance. Electrolytic capacitors offer slow response with limited frequency response, but with much greater capacitance. Tantalum and polyester capacitors sit somewhere in the middle depending on the capacitance. This might explain why so many of the power banks appeared to have a single smaller tantalum capacitor – as it may have a higher effective capacitance at the frequency of interest.
As it seems, there’s a whole plethora of things to consider when choosing the right capacitors during switching converter design. However, it seems a mixture of capacitor types is important for suppressing different sorts (frequency wise) of ripple. Remember, the ripple itself is a complex waveform – the square edges correspond to a wide range of frequencies from low to very high, whereas the “slower” variation corresponds to the lower frequencies. So it seems my idea was … on the right track.
Giving it Another Shot
With the other power bank in the bin, I had to choose another one. In this case, I decided to go for the Lanu Power LP-401B with the counterfeit cells, which I would sacrifice anyway (I don’t need another risk in the house!). I had some “salvaged” capacitors I stole from old boards to play with, so lets give this another go.
At 1A, this power bank originally put out a ripple of 763.6mV which is quite high. Putting one ceramic capacitor across the output, rating 4.7nF …
… increased it to 810.4mV. What? It seems to have an oscillatory effect on the system which increased the spike heights.
Lets try two ceramic capacitors in parallel, for a total of 104.7nF.
It was then, like before, this power bank decided not to power up at all. Not having gotten over the guilt of breaking the last one, did I kill, yet another power bank?
The answer was, no. It turns out that my disassembly and re-assembly had triggered the protection mechanism in the microcontroller, which is reset by applying charger power.
Maybe the last one didn’t die, but I mistook it as dead. Either way, the guilt lingers.
Back to the show …
Now we’re down to 724.9mV but that’s only a marginal improvement. Lets add a 1uF polyester capacitor to add some bulk capacitance to it … for a total of 1.1047uF total capacitance …
… and what a big difference that makes. We are now down to 221.3mV peak to peak, which is starting to reach the territory where it might be comfortable to use. That’s three capacitors, not even 50c worth of components.
But not content with this, lets try super-sizing this. Lets chuck a 680uF Low-ESR electrolytic on the top …
… 228.7mV. Not much change, a slight increase actually. This thus validates that the contribution of electrolytic capacitors at 1Mhz+ is pretty non-existent. But 1Mhz is pretty high for a switching frequency.
Lets try four ceramic and one polyester for a total of 1.3047uF.
This brings us down to 169.4mV peak to peak, on par with wall chargers from something that was out of spec just a moment ago. Lets try adding another polyester cap to it, for a total of 1.3727uF.
A further improvement to 137.8mV is seen, although there is a feeling of diminishing returns. Part of the reason is that, the longer the leads of the capacitor, the more self inductance exists due to the leads themselves. Therefore, it’s best for this to be mounted on the PCB itself (i.e. designed with it in mind) and with as short leads and traces as possible! I have the distinct feeling that it needs another ceramic capacitor.
We get down to 126.7mV from a 763.6mV starting point – an 83.4% reduction in ripple, with a sum total of components that cost no more than $2 but compromise the form factor. It shows that quality needs not be expensive but needs to be designed properly!
Please don’t write about the hackiness of this soldering. I know. It’s hacky. But that’s what happens when I decide to explore. After its noble effort, this power bank was disposed along with several others which were otherwise damaged or very non-compliant (i.e. the majority). I have no room for sub-par equipment.
I think it’s clear that I’m learning as I go, and trying to explain things to the best of my knowledge. I’m rediscovering some things which I had entirely forgotten. Not having dealt with linear power supplies and analog electronics for a very long time meant that I forgot what might even be considered basic for some people. But it was an interesting course of discovery – for example, you don’t need much capacitance for high frequency converters, but you do need effective capacitance. Just throwing capacitors at the problem, especially electrolytics, are unlikely to provide too many benefits unless the switching frequency is low, and the waveform contains larger slow components. A mixture of capacitance is best, allowing each capacitor to contribute where it is able to.
High quality ceramic multilayer capacitors are probably the best pick to augment existing capacitors (if any), given their higher capacitance and high frequency response. The ones I used were traditional ceramic disc capacitors salvaged from old electronics, and so offer much less capacitance for the same space.
So there, it’s possible to fix your own power bank … maybe. Or break it. Whatever.