(3) dirty/clean ground again [sdiy]

Neil Johnson nej22 at hermes.cam.ac.uk
Wed Aug 21 00:10:05 CEST 2002


> >So instead of your carefully
> >designed and constructed low-impedance power supply, you now have a 100Ohm
> >(or whatever) power source.
>
> That sounds like very good advice.

In other words, you have 100 ohms isolation between each op-amp.  Good for
reducing feedback paths, inter-op-amp interference, etc?

> >I've had good luck decoupling with ferrite beads. Odd that no one has
> >mentioned this possibility in this thread.
>
> Using inductors, like ferrite beads, is good at reducing RF interrence that
> might be on the power supply lines, but it should be followed by a large
> enough capacitor to provide for sudden demands in current.

Yup, indeed.  But even on our audio-related stuff you can get fast edges,
e.g. reset pulse on a sawtooth VCO, or the output of a comparator for
making square or triangle waves from sawtooths, etc.  And its these fast
edges that most easily couple into anything near them.

> >Think of each rail as a low-impedance path to the PSU ground.  Sure, they
> >have some DC on them, but at AC they look like a path to ground.  Except
> >they only join ground back at the PSU, via the output decoupling caps.
>
> So the signal would flow to ground, not the other rail.

True, but that would be the ground point *at the PSU*, and not the local
signal ground.  And anyway, that common ground point should, hopefully
(!), be the lowest impedance ground in the circuit, so that, to continue
the pipe analogy, everything flows downhill, and not back up :-(

> >Bzzt!
> You have obviously chosen making rude noises over thinking.

Ha ha!  Thats "Bzzt!" like in game show buzzer :-)

> >Wrong---now you have two sets of lead inductance in series, so
> >you've doubled the total inductance, upping the impedance even more.
>
> If you use only one cap, the shortest possible trace length is the
> diagonal line between the supply pins minus the width of the cap itself.

Which, if the cap is on the opposite side of the board to the op-amp can
be made pretty short indeed (but only tends to be done with surface-mount
stuff).

> In practice, the traces would actually be much longer since both the cap
> and the op-amp would be mounted on the same side of the board.  If you
> use two caps, you can place each one right next to each supply pin.

Or use two caps in parallel, one at each end of the chip, assuming the
PCB layout uses the standard pattern of parallel supply tracks.  That way
(a) we get plenty of capacitance close to the op-amp supply pins, and (b)
we don't contaminate the ground in the region of the op-amp, which is
generally where we also tend to use it for a reference point too.  At
worst the noise will travel along the supply rails back to the large
electrolytic at the board power connector (which should be shunted with a
smaller capacitor for HF decoupling).

Ultimately, each circuit is going to have its own peculiar needs.  The
engineering solution is to implement only what is required---any more and
it is over engineered and too costly, any less and it won't satisfy the
specification.

The decoupling needs of a low-noise preamp are considerably different to
that of a power amp, which would be different from the needs of a digital
board.  It also depends on the operating environment---if its all digital
then we can use switched-mode power supplies and to hell with switching
noise.  If, however, we're sharing the box (and maybe PSU too) with
analogue circuits then the digital designer has to be more careful with
keeping his noise to himself, and not letting it leak out all over the
place.  Just your usual EMC design rules of thumb (or as a lecturer once
used to say "rules of tum", I guess he meant some sort of gut feeling??!)

This has certainly turned into some interesting discussion!

Night all,
Neil

--
Neil Johnson :: Computer Laboratory :: University of Cambridge ::
http://www.njohnson.co.uk          http://www.cl.cam.ac.uk/~nej22
----  IEE Cambridge Branch: http://www.iee-cambridge.org.uk  ----




More information about the Synth-diy mailing list