FreeLists / si-list / [SI-LIST] Re: Diff. Bal. to Unbal. Transition

["Loyer, Jeff" <jeff.loyer@xxxxxxxxx>]

Hi Scott,
While I strongly agree with your first point, I'm not clear on your =
second point.  Some clarification, please...

If I understand you correctly, you're saying that a "long" transition =
(relative to the signal risetime) represents a problem?  I.E., when =
going from differential to single-ended, if I slooooowly transition the =
traces, that will induce an impedance discontinuity due to the modal =
conversion effects.  So, if you take your 2 traces that are very close =
(say 5mils) together, and then decouple them (say 20mils), if you make =
that transition between tightly coupled to loosely coupled (so they're =
each effectively single-ended) very gradual, that can be a problem.

The alternative is to use a short transition section - separate them =
very quickly.  This avoids the issue.

Is that what you're saying, or am I misunderstanding it?

_____________________________

What I would have said is that, when you transition a tightly-coupled =
differential pair to 2 single-ended traces, you:

1) May have to compensate for any impedance change you induce.  If a =
tightly coupled diffential pair has a differential impedance of 85ohms, =
each single-ended trace should have an impedance of 85/2, or 42.5ohms.  =
That probably means your traces will be a different width when they're =
single-ended (a bit wider) than when they were differential.  This only =
applies if the single-ended length is significant, relative to the =
rise-time (for instance, if you're separating your traces to go around a =
via, the discontinuity may be short enough to be ignored).
2) Lose the advantages of routing a differential pair tightly-coupled =
for the length they are routed single-ended.  The advantages I know of =
(in no particular order) are :
   a) less susceptibility to noise - noise that's injected on 1/2 of the =
pair gets injected on the other half (hopefully, equally), and that =
"common mode" noise has less effect on the receiver than on a =
single-ended signal.
   b) less EMI - radiated waves of the 2 halves cancel each other.
   c) less noise induced on other traces - radiated waves of the 2 =
halves cancel each other
   d) differential signals are much less susceptible to problems caused =
by reference plane changes
   e) differential pairs can be AC-coupled - allowing different DC =
levels for the transmitter and receiver.
3) May induce more subtle problems.  If 2 halves of a differential pair =
are routed over dielectric with significant differences in their =
respective properties, you might inadvertently induce =
differential-to-common mode conversion.  For instance, the 2 halves =
might be perfectly differential when they transition to 2 single-ended =
signals.  Then, one trace might be routed over a portion of the board =
with a different effective dielectric constant than the other (due to =
processing variation).  (Or, you might do something very silly and route =
them on different layers).  When the 2 halves came back together, they =
might now be out of alignment - the purely differential signal now has =
some common-mode component.  That would be a bad thing for the =
differential receiver.  But, I haven't heard of this being a significant =
problem.  Anyone who has information showing otherwise, please correct =
me.  In the case of routing the two halves on different layers, that =
would be asking for trouble.

So, I would have said that, theoretically, it is acceptable for a =
differential pair to be routed both tightly-coupled and loosely-coupled, =
as long as no significant impedance discontinuity is introduced.  The =
caveat is that the advantages of routing tightly-coupled are lost for =
the length of the single-ended portion, and you must be careful to not =
route the two halves in significantly different media.

And finally, I'm not sure how a balun is applied in this situation.  A =
single balun would transition the differential pair to a single =
single-ended signal (say that 3 times quickly!), and then another one of =
the far end would convert that back to a differential pair?  That may be =
done, but I haven't heard of it in PCB applications.  But, I wouldn't be =
surprised to find out it's done for applications I'm not familiar with.

Jeff Loyer

-----Original Message-----
From: Scott McMorrow [mailto:scott@xxxxxxxxxxxxx]
Sent: Wednesday, September 10, 2003 7:27 AM
Cc: si-list@xxxxxxxxxxxxx
Subject: [SI-LIST] Re: Diff. Bal. to Unbal. Transition


Brad,

There are two issues that need to be considered in a differntial to=20
single ended transition.

1) Differential impedance must be matched on both sides of the=20
transition.  For differential wire to single ended trace transitions,=20
this can be accomplished easily with loosely coupled 50 ohm traces,=20
where the differential impedance is approximately 100 ohms.

2) Modal converstion control. (Yes, you did use the correct term.) In a=20
balanced pure differential wire line the field distribution is=20
significantly different than that in PCB traces.  Maximum field strength =

is in the area between the lines.  For single ended traces, the maximum=20
field strength is contained between the trace and the plane.  So, a=20
differential to single ended transition will facilitate the transfer of=20
energy from one field pattern to the other.  At most edge rates, this=20
happens naturally as the fields find a plane in proximity to the=20
traces.   The fields transfer first from the wire line to the traces,=20
and then they are redistributed between the traces and the plane.  At=20
high edge rates, this constitutes a discontinuity.  (Actually, it is=20
always a discontinuity.  At low edge rates the discontinuity is small=20
with respect to the edge rate, and so is generally invisible.)  To=20
smooth the transition, the traces can be designed to be highly=20
differential coupled in the region near the two-wire line, along with=20
some ground removal underneath the traces, and then smoothly vary until=20
they become decoupled 50 ohm single ended traces.  I'd recommend this=20
for signals switching in the sub 100 ps region.

In general, if the modal conversion region is << risetime, then little=20
has to be done.  But when the region approaches a reasonable fraction of =

the risetime, adjustments should be made. In reality, differential=20
transitions are a bit easier to accomplish than single ended=20
transitions, such as with coax cable to board transitions.  In the=20
single-ended coax case, the field patterns are significantly different.  =

Here the transition must be accomplished in 3 dimensions.

These transition optimizations are not easily done by hand, although it=20
is possible to get quite close by a Zen-like "being the fields" and=20
imagining where they might want to go.  Unfortunately, we humans tend to =

miss some of the places where electromagnetic fields like to travel.  In =

that case, I'd recommend a full wave 3-dimensional field solver like CST =

Microwave Studio, Ansoft HFSS, or many of the others, that will=20
correctly solve (and allow us to visualize) these sorts of problems.


best regards and happy transitioning,

scott

--=20
Scott McMorrow
Electromagnetic Field Wrangler
Teraspeed Consulting Group LLC
2926 SE Yamhill St.
Portland, OR 97214
(503) 239-5536
http://www.teraspeed.com

Bradley S Henson wrote:

>
>
>For years I've encountered systems with balanced differential signals =
that
>transition to 2 single ended lines in a PWB. For example, an RS422 =
signal
>on a 100 ohm diff. pair wire that connects to a PWB that then routes it =
as
>a loosely-coupled pair of 50 ohm striplines. This seems to work fine =
for
>these slower edge rate systems.
>
>I'm starting to see attempts to do the same thing with sub ns edge rate
>signals in the 100s of MHz and GHz range. This bothers me seeing a
>transition between the 2 modes (probably poor term to select) without
>seeing a transformer or other balun-like device. Should I be concerned? =
If
>this is a problem, how do you simulate it (tools)?
>
>Thanks,
>Brad Henson,Raytheon
>
>
>------------------------------------------------------------------
>To unsubscribe from si-list:
>si-list-request@xxxxxxxxxxxxx with 'unsubscribe' in the Subject field
>
>or to administer your membership from a web page, go to:
>http://www.freelists.org/webpage/si-list
>
>For help:
>si-list-request@xxxxxxxxxxxxx with 'help' in the Subject field
>
>List archives are viewable at:    =20
>               http://www.freelists.org/archives/si-list
>or at our remote archives:
>               http://groups.yahoo.com/group/si-list/messages=20
>Old (prior to June 6, 2001) list archives are viewable at:
>               http://www.qsl.net/wb6tpu
> =20
>
> =20
>

--=20
Scott McMorrow
Teraspeed Consulting Group LLC
2926 SE Yamhill St.
Portland, OR 97214
(503) 239-5536
http://www.teraspeed.com




------------------------------------------------------------------
To unsubscribe from si-list:
si-list-request@xxxxxxxxxxxxx with 'unsubscribe' in the Subject field

or to administer your membership from a web page, go to:
http://www.freelists.org/webpage/si-list

For help:
si-list-request@xxxxxxxxxxxxx with 'help' in the Subject field

List archives are viewable at:    =20
                http://www.freelists.org/archives/si-list
or at our remote archives:
                http://groups.yahoo.com/group/si-list/messages=20
Old (prior to June 6, 2001) list archives are viewable at:
                http://www.qsl.net/wb6tpu
 =20

------------------------------------------------------------------
To unsubscribe from si-list:
si-list-request@xxxxxxxxxxxxx with 'unsubscribe' in the Subject field

or to administer your membership from a web page, go to:
http://www.freelists.org/webpage/si-list

For help:
si-list-request@xxxxxxxxxxxxx with 'help' in the Subject field

List archives are viewable at:     
                http://www.freelists.org/archives/si-list
or at our remote archives:
                http://groups.yahoo.com/group/si-list/messages 
Old (prior to June 6, 2001) list archives are viewable at:
                http://www.qsl.net/wb6tpu