Means of determining microstrip impedance

[Joel Kolstad]

I have some microstrip traces that were designed to be 50 ohms, but I'd like
to find out what they really are. I have a network analyzer good to 6GHz at
my disposal. Does anyone have a favorite means of doing this? This was my
thought:

-- Terminate the trace in 50 ohm loads
-- Have the network analyzer plot S11 up to whatever frequency corresponds
to at least half a wavelength; this will ideally display a circle that has
one edge at the center of the Smith chart (Zin=50 ohms) and the other edge
either greater than Z=50 if Z0 of the microstrip is >50 ohms or less than
Z=50 if Z0 of the microstrip is <50 ohms.
-- With some straightforward calculations, one can show that Z0 of the
microstrip is the geometric mean of these left and right 'edges' of the
circle. E.g., if the circle intersects Z=32 ohms and Z=50 ohms, Z0 is 30
ohms.

My problem is that my microstrips start behaving 'badly' well before the
trace is half a wavelength long (this occurs around 2GHz on my test board),
so it's difficult to decide what the edge of the 'circle' (which is really
more of an arbitrary curve) should be. I am planning on making some longer
test microstrips, but I'm curious if the 'practical' people out there have a
well-known method for performing this measurement.

---Joel Kolstad

(...who hasn't quite gotten around to reading Rober Witte's book yet...)

 

[Joel Kolstad]

E.g., if the circle intersects Z=32 ohms and Z=50 ohms, Z0 is 30 ohms.

Typo. This should be "Z0 is 40 ohms."

 

[Joerg]

Hello Joel,

My problem is that my microstrips start behaving 'badly' well before the
trace is half a wavelength long (this occurs around 2GHz on my test board),

Strange. If it was me I'd find out the material, measure the thickness,
the trace width and calculate the impedance. The math is in the old Moto
MECL databooks or the Fairchild version.

Else I'd probably "cheat" and change the termination resistor until no
resonance is seen anymore. That would be the line's Z.

Regards, Joerg

 

[Joel Kolstad]

Hi Joerg,

Strange. If it was me I'd find out the material, measure the thickness,
the trace width and calculate the impedance.

Well, that's how I designed them -- assuming a geometry, dielectric
constant, etc. However, the material I'm doing this on is "generic FR-4" so
the dielectric constant is a little iffy (I assumed 4.6, but I believe
people tend to use anywhere from 4.4-4.8); hence the desire to see what
actually happened and then back-calculate a few parameters.

I don't necessarily expect such a generic (read: cheap) dielectric to be
good past a couple of GHz, but it would be nice.

Else I'd probably "cheat" and change the termination resistor until no
resonance is seen anymore. That would be the line's Z.

Hmm... just get myself a bunch of 1% resistors and play, eh? Not a bad
idea; thanks!

---Joel

 

[John Larkin]

I have some microstrip traces that were designed to be 50 ohms, but I'd like
to find out what they really are. I have a network analyzer good to 6GHz at
my disposal. Does anyone have a favorite means of doing this? This was my
thought:

-- Terminate the trace in 50 ohm loads
-- Have the network analyzer plot S11 up to whatever frequency corresponds
to at least half a wavelength; this will ideally display a circle that has
one edge at the center of the Smith chart (Zin=50 ohms) and the other edge
either greater than Z=50 if Z0 of the microstrip is >50 ohms or less than
Z=50 if Z0 of the microstrip is <50 ohms.
-- With some straightforward calculations, one can show that Z0 of the
microstrip is the geometric mean of these left and right 'edges' of the
circle. E.g., if the circle intersects Z=32 ohms and Z=50 ohms, Z0 is 30
ohms.

My problem is that my microstrips start behaving 'badly' well before the
trace is half a wavelength long (this occurs around 2GHz on my test board),
so it's difficult to decide what the edge of the 'circle' (which is really
more of an arbitrary curve) should be. I am planning on making some longer
test microstrips, but I'm curious if the 'practical' people out there have a
well-known method for performing this measurement.

I use TDR, a Tek 11801 scope with an SD-24 sampling head. That
time-resolves the whole path, so you can separate out things like
connector bumps and termination mismatches, and see the impedance at
any location along the trace. To see where things are, you just run
your finger along the trace and follow the bump that makes; great fun.

I think one can FFT the S11 data and theoretically reconstruct a
TDR-like plot of impedance vs distance.

Send me a board with a test trace and I'll TDR it for you. It would be
interesting to compare to a frequency-domain measurement.

John

 

[Joerg]

Hello Joel,

I don't necessarily expect such a generic (read: cheap) dielectric to be
good past a couple of GHz, but it would be nice.

I have seen a 5GHz antenna done on FR4. It may not be perfect but cheap.
Maybe you could find the dielectric constant by taking a really large
sheet and measuring the capacitance. On a multi-layer there can be a
substantial difference between pre-preg and the other layers, likely due
to the different number of fibers per unit.

Hmm... just get myself a bunch of 1% resistors and play, eh? Not a bad
idea; thanks!

No need for 1%. Just take SMT resistors where you are sure they are low
inductance. Probably it's safe to start high and stack up to three. When
things jibe and there is no resonance just ohm it out.

As a consultant I am trying to stay low on the food chain when asking
for equipment access at a client's site. The big network analyzer would
be nice, but.... So, another trick I have used was to place a lower
value resistor and then gently taking the dremel to it until "Z=Ideal"
happened, then measure. But be careful with respect to the dust. Some
might even be a bit unhealthy.

Regards, Joerg

 

[Mac]

I have some microstrip traces that were designed to be 50 ohms, but I'd like
to find out what they really are. I have a network analyzer good to 6GHz at
my disposal. Does anyone have a favorite means of doing this? This was my
thought:

-- Terminate the trace in 50 ohm loads
-- Have the network analyzer plot S11 up to whatever frequency corresponds
to at least half a wavelength; this will ideally display a circle that has
one edge at the center of the Smith chart (Zin=50 ohms) and the other edge
either greater than Z=50 if Z0 of the microstrip is >50 ohms or less than
Z=50 if Z0 of the microstrip is <50 ohms.
-- With some straightforward calculations, one can show that Z0 of the
microstrip is the geometric mean of these left and right 'edges' of the
circle. E.g., if the circle intersects Z=32 ohms and Z=50 ohms, Z0 is 30
ohms.

My problem is that my microstrips start behaving 'badly' well before the
trace is half a wavelength long (this occurs around 2GHz on my test board),
so it's difficult to decide what the edge of the 'circle' (which is really
more of an arbitrary curve) should be. I am planning on making some longer
test microstrips, but I'm curious if the 'practical' people out there have a
well-known method for performing this measurement.

---Joel Kolstad

(...who hasn't quite gotten around to reading Rober Witte's book yet...)

If the trace is long enough, and you have a scope fast enough, and if you
can put a fast edge on it, you can measure Zo using the voltage divider
approach.

The idea is that you would put a fast-edge in series with, say, a 47 Ohm
resistor, then measure the ground referenced voltages (using an
oscilloscope with active probes) on both sides of the resistor.

You want to use the first plateau voltage, and ignore the step caused by
the reflection from the end of the line.

For some period of time after the edge starts to rise, what you have is a
Voltage divider formed by Rseries and Zo. So the rise time of the edge
must be significantly shorter than the two-way flight time down the trace
and back for this to work.

Note, I've never actually done this with a circuit board, although I have
done it with coax, just for fun.

--Mac

 

[Robert Baer]

Hello Joel,



Strange. If it was me I'd find out the material, measure the thickness,
the trace width and calculate the impedance. The math is in the old Moto
MECL databooks or the Fairchild version.

Else I'd probably "cheat" and change the termination resistor until no
resonance is seen anymore. That would be the line's Z.

Regards, Joerg

...and have care with the termination resistor(s).
For 50 ohms, use two 100 ohm 1/8 watt carbon composition, minimum
lead length, spread in a "Y" from the microstrip to the ground plane (i
think30 degrees may be optimim).
With chip resistors, one might find that two 100 ohm SMD placed in a
similar "Y" configuration may be better than one 49.9 ohm SMD.
..and beware of ground plane under the resistors...
For the carbon comps, you want that; for the SMDs, you may want the
ground plane to go no further than half the length (at most).

 

[Robert Baer]

Hi Joerg,




Well, that's how I designed them -- assuming a geometry, dielectric
constant, etc. However, the material I'm doing this on is "generic FR-4" so
the dielectric constant is a little iffy (I assumed 4.6, but I believe
people tend to use anywhere from 4.4-4.8); hence the desire to see what
actually happened and then back-calculate a few parameters.

I don't necessarily expect such a generic (read: cheap) dielectric to be
good past a couple of GHz, but it would be nice.




Hmm... just get myself a bunch of 1% resistors and play, eh? Not a bad
idea; thanks!

---Joel

Perhaps a PCB material that is more uniform would be better.
Megtron 5 is price competitive with FR-4:
http://oil4lessllc.com/HTPCB.pdf

 

[mike]

I have some microstrip traces that were designed to be 50 ohms, but I'd like
to find out what they really are. I have a network analyzer good to 6GHz at
my disposal. Does anyone have a favorite means of doing this? This was my
thought:

-- Terminate the trace in 50 ohm loads
-- Have the network analyzer plot S11 up to whatever frequency corresponds
to at least half a wavelength; this will ideally display a circle that has
one edge at the center of the Smith chart (Zin=50 ohms) and the other edge
either greater than Z=50 if Z0 of the microstrip is >50 ohms or less than
Z=50 if Z0 of the microstrip is <50 ohms.
-- With some straightforward calculations, one can show that Z0 of the
microstrip is the geometric mean of these left and right 'edges' of the
circle. E.g., if the circle intersects Z=32 ohms and Z=50 ohms, Z0 is 30
ohms.

My problem is that my microstrips start behaving 'badly' well before the
trace is half a wavelength long (this occurs around 2GHz on my test board),
so it's difficult to decide what the edge of the 'circle' (which is really
more of an arbitrary curve) should be. I am planning on making some longer
test microstrips, but I'm curious if the 'practical' people out there have a
well-known method for performing this measurement.

---Joel Kolstad

(...who hasn't quite gotten around to reading Rober Witte's book yet...)

Theory is great. But as a practical matter, if you want to know the Z0,
it's best to just measure the Z0.
I always used a 100 ohm 1/4" pot at the end of the trace and a TDR at
the beginning. Tweek the pot till the flat part after the glitch caused
by the pot is equal to the flat part before the part, the transmission
line. Measure the resistor when you're done.

In a perfect world, you could do the same with a network analzyer, but
I've never been able to get frequency domain measurements to yield
results in a less than perfect world.

As another practical matter, if you need accurate Z0, you've got a
severe production problem using FR-4.
Your vendor probably doesn't have any idea what the dielectric constant
is for each batch. You MUST specify the exact layup you want. That
won't keep 'em from changing it on you, but you can get your boards
rebuilt when they do. I've seen problems caused by too much or too
little pressure when they mashed the layers together. When there's only
a few mils of dielectric, it don't take much squishing to change it
radically. I found it better to make sure that critical traces and the
associated ground plane were on opposite sides of the same piece of
dielectric with thickness specified very tightly. Sometimes it takes a
non-sandard layup to make that happen.

Don't get me started on what happens two years down the road when
some bean counter decides he can save a buck by changing vendors.
Make sure EVERYTHING is written into the board spec. and that incoming
inspection checks for it.
mike

--
Return address is VALID but some sites block emails
with links. Delete this sig when replying.
..
Wanted, PCMCIA SCSI Card for HP m820 CDRW.
FS 500MHz Tek DSOscilloscope TDS540 Make Offer
Wanted, 12.1" LCD for Gateway Solo 5300. Samsung LT121SU-121
Wanted 12" LCD for Compaq Armada 7770MT.
Bunch of stuff For Sale and Wanted at the link below.
MAKE THE OBVIOUS CHANGES TO THE LINK
ht<removethis>tp://www.geocities.com/SiliconValley/Monitor/4710/

 

[Joerg]

Hello Robert,

At those frequencies you really have to use SMT resistors.

Regards, Joerg

 

[[email protected]]

Couldn't you terminate in 50 ohms, use network analyzer to get mag &
phase of reflection coeff & calculate from there?

 

[Joel Kolstad]

Couldn't you terminate in 50 ohms, use network analyzer to get mag &
phase of reflection coeff & calculate from there?

That's really what I'm doing now (just that I'm using a method that doesn't
require knowing the actual length of the tracee). I was hoping there was
some way that you could visually get an indication of relative impedance
(relative to the termination, which is the same impedance as the reference
for the Smith chart) as a function of frequency... similar to how you can
estimate Q of a tuned circuit from the size of the 'loop' it makes on a
Smith chart.

---Joel