90 degree bends in microstrip

[Bill Sloman]

I was digging through some old reports from 1989 and found this text

Microstrip Discontinuity Capacitances and Inductances



A paper of this title was written by Peter Anders and Fritz Arndt

and published in the IEEE Transactions on Microwave Theory and Techniques

MTT-28 (11) pages 1213-17 (November 1980).



Amongst other things it discusses 45 and 90 degree junctions in microstrips.

My reading of the conclusions is that



1. The problem is worst for 50R microstrip; the reflection

from a junction will be roughly at factor of three less from
the same junction in 75R track and another factor of three

down again for l00R track.



2. The reflection from a 45 degree junction is about a third of that
from

a 90 degree junction, so there is a small advantage in making a

90 degree change in direction as two 45 degree junctions.



3. The discontinuity can be reduced by about an order of
magnitude by bevelling (they use the term mitering) the
outside edge of the junction. For 90 degree junctions, the
scale of the trim decreases with decreasing impedance, while
for 45 degree junctions there in a smooth maximum in the trim
around 75R.



Gigabit Logic's Application note 2 at Fig. 5, gives a figure of 1.8 times
the track

width as the optimum length for the trimming cut for 90 degree junctions.



Applying the Anders and Arndt results, I get closer to 1.55 for a 50R track
on PTFE,

1.77 for a 75R track, and about 2.2 for a l00R track, all for 90 degree
junctions.



For 45 degree junctions in 50R track the equivalent figure is 0.91 of the
width of the

track. For 75R track this increases to 1.5 of the track width, and for l00R
track it goes

down a bit to 1.46 track widths.



Note that while our CAD system puts a radius on the outside corner of all
track corners,

the area trimmed off by this feature is at least an order of magnitude less
than that

removed by any of the trimming cuts specified above.



It is also worth noting that a via or a test pad inserted into a "constant
impedance"

microstrip will introduce much larger reflections than an untrimmed 90
degree junction.





l l

l l

l l

\ --------------

\

\----------------



The "trimming cut" is just ths 45 degree cut off the outside of the cormer -
for 90 degree

bends, and 22.5 degrees for 45 degree bends.

 

[[email protected]]

Below several GHz, risetimes above maybe 100 ps, square corners don't
matter.

John- Hide quoted text -

- Show quoted text -

correct.
everything in Bill's post may be true at 10 GHz but irrelevant below a
few GHz.

Mark

 

[[email protected]]

But at least he posted something about electronics, for which we can
be grateful.

And John did express his gratitude with just the sort of graceful
courtesy I've come to expect from him -

"Below several GHz, risetimes above maybe 100 ps, square corners
don't
matter."

In fact the whole point of the post was the reference to published
paper on the subject

Microstrip Discontinuity Capacitances and Inductances
by Peter Anders and Fritz Arndt
IEEE Transactions on Microwave Theory and Techniques
MTT-28 (11) pages 1213-17 (November 1980).

and the salvaged text was thrown in to give an indication of the
content. But John Larkin doesn't seem to read the electronics
literature, and that aspect of the post would not have got his
attention.

 

[Bill Sloman]

You expected gratitude? You believe in courtesy? Funny and funnier.

You did claim to be grateful.

And all I stated in that post were facts. You find facts to be
discourteous? Funnier yet.

All I was saying that your initial reaction doesn't read like a grateful
response.
I did find the contrast funny.

Your post didn't mention frequency at all.

The reference to Gigabit Logic's application notes would make it
clear enough to anybody who could remember Gigabit Logic.

I wouldn't want people
going to the trouble of clipping trace corners and worrying about vias
when it makes no sense... which it usually doesn't. [1]

And they aren't going to do it if the application notes for the logic
that they are using don't make a fuss about track impedance.

Right, I don't read all those 30-year-old old journals, but I do have
a few pretty good books about such stuff. More important, I *do*
occasionally add to my boards test traces and SMA connectors into
power/ground planes, so I can measure capacitances and resistances and
TDR things. That's reality.

It is, if you've got a time domain reflectometer. Of the places where I've
worked, only EMI Central Research and the Nijmegen University electronics
workshop had their own, and even there they weren't easy to get at.

Look:

ftp://jjlarkin.lmi.net/TDR1.jpg

ftp://jjlarkin.lmi.net/TDR2.jpg

On the TDR display, the junk at cm 2-3 is the SMA/pcb transition. The
two right angles start at cm 4 and the capacitive bump at cm 5 is the
via. On regular boards, right angles are down in the noise, buried in
trace width variations and fiberglass weave.

This is on a scope with a 30 ps reflected TDR risetime. At 100 ps,
things would be a lot flatter.

TDRing power pour/ground plane structures is fascinating. It blows
away all sorts of popular dogma about planes and bypassing.

John


[1] Corner clipping is a legacy of hand-taped layouts anyhow; CAD
makes nice smooth curves, which lots of olden-days layout people
forbade because it contributed to tape creep.

Even a smooth curve adds extra capacitance to the track.
A 90 degree corner with a right-angle on the inside and a quadrant
of a circle on the outside has 0.79 track widths squared in the
quadrant. A 1.4142 long mitre has only 0.5 trackwidths squared,
and the rule of thumb 1.8 track widths long 45 degree mitre gets rid
of half of that.

In an ideal world, you'd want to narrow the track in
according to the tightness of the bend.