DC Filtering: best approach here

[ACharnley]

Hi,

My final problem to resolve.

I have a dynamo producing AC, current limited to 0.5A, variable voltage. This is rectified and fed to a large storage capacitor which saturates the current causing the voltage to drop. For now I'm sticking it directly over the load but the real circuit will use a PWM to ensure the saturation doesn't rise over 5V (approx).

Presently this almost works, except that there are some voltage spikes across the storage caps. These could be from the dynamo but I theorise they could also be from the large super-caps as the charge.

Unfortunately these spikes push the voltage about 2.5-4V over the normal, several other circuits use the same net (which can't rise above 5.25v) and I have already damaged a GPS unit.

To see what was going on I attached a data logger. You can see from the graph it's about 0.5us in duration which by my calculation (which could be wrong) is 2MHz. Although my PWM is monitoring voltage both sides of the circuit I doubt the MCU will be fast enough to turn the PWM off for these transients and so I'd like to deal with them via filtering.

However, I am not an analog engineer, and any advice is appreciated.

My thanks, Andrew

 

[MikeMl]

Post the schematic of what you are doing. It may be as simple as bypassing with a few low-series-inductance capacitors.

 

[ACharnley]

Post the schematic of what you are doing. It may be as simple as bypassing with a few low-series-inductance capacitors.

 

[ACharnley]

The diode prevents back-flow, in reality I'll have a MCU controlled PWM. In a perfect world it'll be fast enough to catch the transients and lower the current but I am somewhat sceptical.

I have noticed the following;

1. if there's no load then a small 15pF capacitor in parallel reduces the transients significantly. With a load there's almost no difference.

2. with a Garmin Edge Touring (aka non-resistive load) the inductance of the capacitors at 5V is enough to drop the voltage to 4V very briefly, which upsets the GPS unit. Sticking a ~47uF capacitor in series is enough to deal with it. The transients are still there but have reduced amplitude and haven't caused a problem at any speed. They may however occasionally breach the 5.25V spec but I'm not sure if they are of enough duration to cause damage. I'd rather deal with them via filtering if at all possible.

3. the lower the AC frequency the worse the transient amplitude.

4. an inductor between capacitors and *out reduces the transients but not by too much and the ripple is worse. I fried the Garmin GPS though I didn't manage to capture the data to show how it happened. I guess if the frequency is right then the inductor can potentially act as a boost.

 

[rjenkinsgb]

It appears you are trying to get a stable 5V supply from something like a pedal bike "dynamo" (actually an alternator).

I have built something like that in the past, but using a totally different approach, which may give better results.

I just fed the alternator to a smallish smoothing cap via a bridge rectifier built with schottky diodes to minimise loss.
It had a 100V cap, with a 47V 1W zener across it to remove spikes and prevent excess voltage, plus a 0.1 ceramic for high frequency suppression.
I think it was a 470uF cap, it's a few years back..

Then a high voltage input (60V) version "Simple switcher" 5V regulator to give a stable output.
eg. https://www.ti.com/lit/ds/symlink/lm2576.pdf

(That could feed your supercaps.)

This allows the full output of the alternator to be available & at higher voltages it may be more efficient as the output current is lower for a given power.

 

[ACharnley]

It won't . The super-caps have only 13mOhm ESR from empty which will saturate the switch mode into protection. I have a 7A power supply and they sink all of it.

Your circuit has many losses, you'd be lucky to hit 60% efficiency with that approach and IC.

The problem here is the super-caps are physically changing (probably the electrolyte) as they charge and it's causing the spike.

These spikes are 0.5us in duration, which means a PWM running at 2MHz to catch them, plus time allocated to the MCU processing to capture the voltage and react. I doubt it can be done in software. plus a PWM at that frequency is wasteful in the FET switch.

+ A zener is too slow.
+ The lowest TVS is about 8V (even the 5V SBAJ05CA is 7.5v approx).

I've partially mitigated the problem by fitting a 0.01uF capacitor with a low ESR in parallel with the super-cap. When it spikes it's semi-absorbed. It's actually good enough as the PWM "backs off" after 4V. It's only really in the 3V area where the spikes could go significantly over 5V.

 

[ACharnley]

I'd be interested in hearing ideas such as using an inductor. My understanding is an inductor stops change which sounds ideal. Unfortunately I tried a relatively small one (22mH) without the data logger and my Garmin unit got fried. I theorise it's acting as a flywheel converter so I get a higher voltage when the pwm goes to 0.

 

[rjenkinsgb]

If it takes so long to get to a functioning voltage you have no usable output at all to start with, if everything goes via that. That's the problem with a capacitor, it's working range is from 0V.
A battery has a much smaller voltage "span" between discharged and full charge.

If I'm reading the supercap values right (2 x 50F in series) you need 282 joules to reach 4.75V, with limit of 344 @ 5.25V
That gives you just 62 joules usable stored power.

Rounding to 5V, you have a maximum power transfer rate of 2.5 joules per second at half an amp. Around 25 seconds charge or discharge to the limits of that range, @ 500mA.


OK;
treat the supercap as a battery and have an alternate path around it.

eg. Regulate slightly higher to allow for a couple of schottky diode drops, feed straight from regulator via two diodes to load and also via diode + current limiting resistor to the cap and diode from cap to load.

Limit the charge current to eg. 250mA and you have another 250mA available instantly for the load, while still "topping up" the cap within it's working range in around one minute

You could improve efficiency by using FETs to route power depending on the supercap state of charge & use a switched-mode current controller rather than just a resistor, so the main regulator still operates at the target voltage. You could also split off a second "charge" regulator from the high voltage section.



Re. a series inductor - it would maintain current flow; if your device suddenly takes less current, the voltage will spike.
Use capacitors to damp the spikes and/or add a large reservoir cap plus a ceramic for noise after the inductor, to even out the current through the inductor.