Re: [SI-LIST] : return current distribution in diff pairs

D. C. Sessions ([email protected])
Thu, 23 Sep 1999 11:26:01 -0700

Eric Bogatin wrote:

> There has been some discussion recently about the current distribution of
> the return path in differential pair lines. I think it is a common
> misconception that the other line "carries" the return current of the first
> line. This may be true when the off diagonal elements of the characteristic
> impedance matrix are very small compared to the diagonal elements, as in
> shielded twisted pair, but not in typical board geometries.
> In the classes I teach, I show an example of the current distribution in the
> case of two 50 ohm coupled microstrips, 5 mil line and space, coupling of
> about 10%. There is less than 10% overlap of the return currents in the
> planes. This is ultimately a "skin depth related" effect. I have appended a
> copy of one of my slides showing the current distribution at 100 MHz sine
> wave freq for the current in the signal lines and the return path in the
> plane below. This was done using the Ansoft Maxwell 2D Extractor field
> solver, assuming copper for all the conductors. Unfortunately, I can only
> plot the magnitude of the current, not the sign. So, I plot in the top
> example, the current in the plane when only one conductor has current, +1A,
> showing that most of the return current is directly under the signal line.
> Then I plot the current when one has +1A and the other has -1A. You can see
> there is clearly a lot of return current in the plane.

I think the essential question is not whether the return currents overlap at
the wavefront, but how large the loops are. This is inherently a 3D problem
and not a 2D one. I admit I don't have any publishable research to back my
own view up, but experience has so far been consistent with the theory that
the parallel components of the return current are small except in the immediate
neighborhood of the wavefornt, with the result that EMI from plane crossings
tends to attenuate very rapidly.

We can, however, certainly place a lower bound on the mischief caused by
differential signals crossing plane splits, since (as has been pointed out)
differential signals in PWBs often aren't very. Any common-mode currents
will cause the familiar problems as they hit the plane split.

Aside: wasn't there a paper presented recently (DAC?) about using nonuniform
planes to reject common-mode signals?

> The lesson here is to always treat the return currents with as much care and
> respect as the signal currents, even in differential pairs, unless you know
> for sure the return currents are cancelled in the planes. Of course, the
> actual current distribution in the planes will depend on the precise cross
> section and spacings.

Always a good idea. Engineering consists of institutionalized paranoia, so
even _if_ you have reason to believe that the parallel current vectors are
negligible, it usually doesn't hurt to take precautions anyway.

> If the traces go over a split in the return path, the currents will probably
> mix, and may go to zero at one spot, but the impedance of the two modes will
> be radically changed in this region and you will generate common mode
> voltages where there were none before- causing discontinuity problems,
> termination problems, switching noise problems and EMI problems (did I leave
> any out?). Of course, you need to simulate the magnitude of the problem to
> evaluate whether for the given split, the noise is still under an acceptable
> limit. But the defensive strategy is treat return paths in differential
> pair, like you would for single ended lines.

PS to EB: I never did see that article. Did it get published?

> > -----Original Message-----
> > From: [email protected]
> > [mailto:[email protected]]On Behalf Of D. C. Sessions
> > Sent: Wednesday, September 22, 1999 7:35 PM
> > To: [email protected]
> > Subject: Re: [SI-LIST] : Q: Plane-jumping return currents
> >
> >
> > Eric Goodill wrote:
> > >
> > > Mike Jenkins wrote:
> > > >
> > > > Eric,
> > > >
> > > > One line of your question, "My system is running pretty fast
> > > > (> 1 Gbps)", caught my eye. At that speed, which I assume might
> > > > be Fibre Channel or Gigabit Ethernet, you may well be running
> > > > differential. (If not, good luck to you.) But if your lines
> > > > are dif'l, they carry their own return current. Depending on
> > > > geometry, there is some discontinuity, but MUCH less than
> > > > single-ended. If your lines are, in fact, differential, and
> > > > if you wish me to elaborate, I will.
> > >
> > > Mike,
> > >
> > > Yes, differential. However, we're using edge-coupled pairs, and it's my
> > > understanding, though I've done no analysis, that about 10% -
> > 15% is about
> > > as much coupling as you can get between edge-coupled lines.
> > Thus, there is
> > > still a strong coupling between the trace and it's reference place.
> > > Therefore, I suspect that there's non-ignorable amount of
> > return current in
> > > the reference planes. I'd be interested to see a
> > > return-current-distribution plot for a diff pair both in the reference
> > > planes and the coupled traces.
> >
> > I don't think so. Sure, there's a fair bit of capacitive current between
> > each trace and the adjacent plane, but since they're equal and opposite
> > the loop is very small and entirely lateral. Cross a plane boundary and
> > there's no need for any current across the break.
> >
> > --
> > D. C. Sessions
> > [email protected]
> >
> > **** To unsubscribe from si-list: send e-mail to
> > [email protected] In the BODY of message put:
> > UNSUBSCRIBE si-list, for more help, put HELP. si-list archives
> > are accessible at ****
> >
> ------------------------------------------------------------------------------------------------------------------------------------------------
> Name: Current dist.PDF
> Current dist.PDF Type: Portable Document Format (application/pdf)
> Encoding: base64

D. C. Sessions
[email protected]

**** To unsubscribe from si-list: send e-mail to [email protected] In the BODY of message put: UNSUBSCRIBE si-list, for more help, put HELP. si-list archives are accessible at ****