Hi all,

I'm wondering if anyone can lend me some guidance to get this circuit functional. I'm trying to hobble together an output filter for a DC-DC converter with jellybean parts. The noise is roughly in the 500kHz/200mVpp range. So far I've managed to get rid of the noise with an LCL filter (pictured below), only when I connect the output to the load the voltage drops from 5V to 2V.



Can anyone explain how I can remedy this situation? Is the voltage drop happening simply because there's too much current flowing through the inductors? The load itself is somewhat power hungry: three Red Patayas that draw about 3A in total, as well as some smaller filters that draw about 100mA altogether. Unfortunately, I'm uncertain how to calculate the impedance of the load itself... when I stick a multimeter on it it reads 300 ohms, although I realize it's not that simple.

Anyways, any pointers would be greatly appreciated.

 

We don’t know what is to the DC to DC converter box, so it’s a little difficult to help.

Measure the current drawn by the board with a working 5V supply, then measure the voltage and current when powered by your converter.

And postthe converter circuit.

Bob

 

What you consider to be noise is probably normal operation for this DC-DC converter. That amount of output ripple is about 4% (200mV/5V = 4%), and for the intended use may be acceptable. Normally a DC-DC converter will have what it needs internally so hanging external caps and inductors on it is probably a waste of time. If you remove the components (3 capacitors and 2 inductors) do you still get the observed voltage drop? If so you are exceeding the DC-DC converter's output current capability.

DC-DC converters can be designed for low ripple, but it takes some careful engineering and component selection. Some manufacturers cut more than a few corners with their products. I'd be surprised if your filter was effective. To demonstrate that here is the AC response of your filter.


You notice that peak at about 50 kHz? What you have is an unstable oscillator with unity gain and 180° phase shift. The reason you "see" a voltage drop is because V(o) is beating up and down and your instrument is averaging that signal to give you a quasi-DC result. You should look at it with a scope.

A long time ago, in the age of vacuum tubes, a pi-filter would be added to an unregulated power supply, but we seldom do that any more.

 

Thanks for quick response.

Some more context would help: I'm trying to power a fairly complex piece of scientific equipment (I work at a university that's currently suffering from a serious lack of technical expertise) that involves an assortment of Red Patayas, clocks, transducers, transceivers, and filters, all requiring 5V. When we power the system directly through a bench power supply, everything works fine. When we power it through the DC-DC converter (which will be necessary for the system's field application), the data comes out noisy and corrupted.

The converter is rated for 20A, so we're certainly not exceeding its limits. Also, the current draw is consistent between both set-ups, whether from a 5V bench supply or from the converter (roughly 3.5 A).

I agree that the switching noise I'm viewing on the scope may not actually be what's corrupting the system. It really could come down to a whole array of things - all I can say for certain is that it works with the bench supply, but not with the switching DC-DC. The reason for a filter to be absolutely sure it's not the switching noise.

And you're right, this sort of thing does appear to require some careful engineering...

 

Well, the first thing I would do is try another DC-DC converter. Carefully evaluate the specifications and test the samples to see that they meet the specifications that you are expecting. Trying to filter the output of a DC-DC converter is IMHO, a losing proposition. It must be eliminated at the source.

 

The 10 ohm resistor is what causes the voltage do drop with load. Short it out.
Probably one LC will be enough but two is better.

 

The 10 ohm resistor is what causes the voltage do drop with load. Short it out.
Probably one LC will be enough but two is better.
View attachment 268146

I don't know what you are on about. You can't short out the output impedance of the DC-DC converter. To do an AC analysis you need a source impedance. That 10Ω resistor was the output impedance of the DC-DC converter. Most of the output voltage will thus appear across the load at DC which is confirmed by the flat line at 0dB from DC out to just under 1kHz. The TS built an unstable oscillator without realizing it.

 

I don't know what you are on about.

Sorry I mistook your drawing in #3 for the drawing in #1. Some where there must be some serious resistance if the voltage drops to less than 50% with load.

 

Sorry I mistook your drawing in #3 for the drawing in #1. Some where there must be some serious resistance if the voltage drops to less than 50% with load.

I don't know how the measurement was made, but if it was done with a multimeter, averaging the oscillations at the output, that would be consistent with the AC analysis of the combination of inductors and capacitors where the peak rises to the 0dB line and the phase is @ -180°.
To me that says oscillation is possible, then we have to establish that it is in fact being sustained.

 

Measure the resistance of the inductors.
That likely is the source of the voltage drop.
If so, you will need inductors with a lower resistance.
Also the inductors need to be rated for the maximum load current.

If the inductors don't sufficiently suppress the noise, you many need to add a common-mode choke in series with the output and output common.
Switching supplies can have common-mode noise which can otherwise get into the loads.

 

I would suggest separate filters for each output load.
Use chokes with iron powder cores because they tend to absorb and dissipate the interference whereas less lossy inductors reflect it so that it can cause problems elsewhere.
For the light loads try ferrite beads.

 

Hey all, just returning to say that the solution was in fact a simple common mode choke.

It was likely the line noise I was observing (with an oscilloscope of course - not a voltmeter!) wasn't even the cause of the data corruption, but merely a massive distraction. Now what the value of the common mode choke is, I have no clue, it was just a part laying around, which I'll try to identify when there's more time...

Thanks for all the input!