Re: Switcher impulse noise reduction...


Dave wrote:
Does anyone have any experiance removing the switching noise from the
output of a switch mode power supply? I have tried in vain but think
there is a way to actively filter it out somehow. We're talking about
a 20mv peak to peak noise spike at ~ 60 KHz or so. A picture is worth
a thousand words but I cant see a way to attach one in this group. Any
help would be great! I tried this circuit on the problem but it had no
effect...

http://www.wenzel.com/documents/finesse.html

Thanks.

Dave,

What type of SMPS is this (buck, boost, etc)? Did you design and build
it? A schematic would be very helpful. But a lot could also depend on
the PCB layout, even with a "good" schematic design.

I don't call myself an expert, but have learned a few things by reading
manufacturers' application notes (AND http://www.genomerics.org), and
designing and building several medium-power boost-mode SMPS units.
I'll try to generalize some of what I think I know:

First of all: 20 mV p-p is not considered to be too terrible. Many
commercial units have specs with "100mv ripple plus noise". However,
if you're careful, you "should" be able to get it down into the
several-hundred-microvolt p-p range, fairly easily.

To not induce noise, or have noise induced, certain loops should have
as small an enclosed area as possible. Look especially at the loops
with large and/or highly-dynamic currents. Look up Maxwell's Equations
(and/or Faraday's law).

PCB traces should be "wide", and as short as possible, usually. YMMV.

Use a "star" grounding scheme, with no grounds running together,
anywhere. Instead, they should only touch each other at the ground
side of the main input cap (assuming this is an "off line" supply, i.e.
it runs from AC mains, with a transformer, rectifier bridge, and
large-ish input capacitor).

Keep the feedback (and ALSO the ground of any feedback voltage-divider
used) away from everything else, except at the star ground point.
Also, if possible, take the feedback from a point that's AFTER any
output cap and output filtering.

Your inductors should be toroids. Look at the JWMiller 2300 and 2200
series, if your output pushes less than 15 amps or so. (Unless you
like winding your own, you may have to base other parts of the design
on what uH are available at the currents you need.)

If you have a large diode connected to the SW output of a switcher IC,
try designing an RC snubber to go in parallel with the diode; maybe
something like .0022uF in series with 470 Ohms, with the resistor
connected to the diode's cathode and the (film) cap to its anode. The
actual R and C values will depend on the rest of your circuit.

You might also want to consider using a CLC lowpass "Pi" filter on the
output. If there's already a large-ish capacitor from output to
ground, you could add a high-current inductor in series with the
output, and another of the same cap, to ground, after the inductor.
Choose L to make 1/[(2Pi)LC] small; certainly less than 5 or 10 Hz,
while still comfortably carrying your maximum output current.

Look at your big diode(s). Try something faster (or maybe even
slower?), or with lower Vf, etc etc. (And they shouldn't get very
warm, if they're not spending too much time in-between the on and off
states.)

Look at your big electrolytic capacitors. You want them to have low
ESR (Equivalent Series Resistance). If this is through-hole stuff,
maybe look at something similar to the Nichicon "UHE" series (see
www.mouser.com, at al).

You could also try adding a linear regulator to the output (with
appropriate input, output, and feedback Cs and Rs, of course).

You could also try adding a film or ceramic cap in parallel with your
main input cap, say 0.22uF (or whatever helps).

Get LT-Spice (free from www.linear.com), and simulate the circuit!
Remember to add realistic series resistances to all of the capacitors
and inductors. You can get a good idea of what "should" work
well-enough, by trying it in LT-Spice. But your PCB layout and the
types of components used will still need to be correct, to get good
results from your actual circuit.

Good luck.

- Tom Gootee

http://www.fullnet.com/u/tomg

.

Quantcast