Re[2]: [SI-LIST] : power supply filtering and bypassing

Ray Anderson ([email protected])
Fri, 6 Mar 1998 09:05:45 -0800 (PST)

I just wanted to echo Ed's ([email protected]) comments regarding
power supply filtering add a couple amplifiying remarks.

In many cases designers have used the small 3 terminal integrated filters
that combine a couple of capacitors and a ferrite inductor into a single
unit (typically from TDK, Murata and others) and have been suprised that
the addition of the filter exacerbates the problem rather than solving it.

If you look at the frequency response curves in the vendor's catalog you
see a nice, well behaved, low-pass characteristic. The gotcha' is that these
parts are characterized in a 50 ohm system. Indeed, in a 50 ohm system they are
well behaved low pass filters. However, when applied as power supply filters
this assumption falls apart. In that application they are being driven from
a very low impedance source (typically on the order of tens of milliohms or less).
When driven from a very low impedance then nice low pass response degrades to
a very peaky response somewhere near the anticipated roll-off frequency.

I have seen cases with peaking on the order of 10 dB and more. The net result is
that any noise energy on the input of the filter that you are trying to
squash can actually appear on the load side of the filter with a much higher
amplitude than on the input side.

This phenomena can be modeled quite nicely with spice and analyticaly if you
take the time to work through the transfer function of the LPF.

A typical fix is to make sure a small amount of resistance is in series
with the input of the filter to effectively raise the driving impedance
that it sees. One needs to make sure that this doesn't introduce excessive
DC voltage drop when the load is drawing max current. This typically works
well for low current loads such as isolated Vcc pins for PLL's and the like.

And as Ed mentioned in an earlier post, this issue is also directly relevant
to other power supply filtering schemes composed of discrete parts. The same
caveats apply.

-Ray Anderson

Staff SI Engineer
Sun Microsystems Inc.
[email protected]