John Popelish said:
My english was pretty poor. The resistor goes between the rail and
the parallel combination of chip and capacitor (to form a low pass
filter in the supply line and to make the Aground cleaner.
10-220 ohms (a few % DC supply loss)
Noisy Rail --\/\/\/--+--> Chips
|
---
---Bypass
|
###
AGND
This dumps less high frequency current into the Aground than just
using capacitors. It also provides a loss to absorb ringing that is
bouncing around on the board between high Q capacitors and trace
inductance. This approach is only useful on low current loads (where
the resistor can be high enough (relative to the cap ESR) to help the
lower the roll off frequency.
I have also used ferrite beads on leads instead of resistors (works
for higher current loads).
Some analog chips are nonlinear with respect to high frequency noise
and you hear the envelope of the noise pulse density. If the number
of spikes per second varies (because of changes in the operation of
logic), you hear a low frequency AM demodulation of that noise. So
the filter may need to only be effective well above the audio
frequency range to reduce such nonlinear effects and have a noticeable
benefit. This, of course depends on the particular chips used.
Some very good points here.
The general idea is to critically/heavily damp every LC circuit.
Your power supply LC circuit will be the power supply trace inductance and
the decoupling caps (which no doubt completely swamp your trace
capacitance).
An 0V plane with PSU traces will have a moderate inductance, whereas 0V and
PSU planes will have an extremely low inductance (sub 100nH) and therefore
lower characteristic impedance (conversely a psu trace + badly routed 0V
trace will have a very high L and hence high Zo).
Because C is high, Fres is fairly low (even though L can be quite low) and
so it looks like a lumped circuit. This means that your ac-coupled shunt
damping resistor (ie electro/tant cap + ESR) can be pretty much anywhere opn
the PCB. Because it is a shunt damping resistor, a lower value R will
increase damping (to a certain extent).
If you measure a few boards, and do a few calculations, you'll find the
damping resistor tends to work out at a few ohms. One test I like to do is
stuff caps onto an otherwise blank PCB, then do step-response tests and
measure inductance. You can design it like an RC snubber - using no damping
R+C, masure Fres. Add shunt (lossless) capacitance Cx until Fres halves - Cx
= 3Cres, and can now calculate Lres = 1/[(2piFres)^2*Cres] and Rdamp = Zo =
sqrt(Lres/Cres) and Cdamp >= 3Cres. Unitrode have a nice app note on how to
design a snubber.
depending on what you are doing, you may or may not get away with cap ESR to
damp your psu - ESR can increase dramatically at low temperatures, may vary
widely from part-to-part and likely changes significantly with age. When I
damp SMPS filters I tend to use low-ESR caps + series resistors (caveat:
peak pulse power at startup) when looking for long-term reliability. Im not
sure about tantalum ESR behaviour though, every time I consider them I can
do it cheaper with electrolytics/smt ceramics.....
And of course if you dont excite the resonant circuit, it wont ring - but
square waves are rich in harmonics, an edge rise/fall of 10ns has a "knee"
frequency of around 32MHz, and plenty thereafter.....very fast transitions
can excite the resonances between paralleled bypass caps (eg
100nF||3.3nF)...the plot thickens.....and a crappy layout can pick up noise
(outisde or self-generated) that excites PSU resonance......
cheers
Terry