coplanar waveguide

schreef in bericht news: snipped-for-privacy@j39g2000yqn.googlegroups.com...

For what it is worth, strip-line - a track routed over ground-plane - is intrinsically dispersive. Microstrip - a track routed on an inner layer between two ground planes - is not.

--
Bill Sloman, Nijmegen

snipped-for-privacy@j39g2000yqn.googlegroups.com...

I think you have your terms backwards there. I use microstrip all the time (a track routed over the groundplane , but exposed). You can combine lumped elements to the circuits. All power transistors that I have used are matched into microstrip circuits.

It is a legitimate , bonified, transmission line, like a coax or strip line.

A skinny trace has more losses than a wide one. Adding the cpw grounds will, for a given dielectric and board thickness, force the 50 ohm trace width down, increasing losses. I think you may get more edge-of-trace crowding losses, too.

I'd suggest a thicker dielectric and a wider trace. Also have your board shop tell you about the underside of the copper (or peel it up); many disgusting adhesion-enhancement treatments like "black oxide" greatly increase skin losses.

Right here I've got a 4" cpw trace on crap FR4, 52.1 ohms, 62 mil diel, 100 mil wide trace, 100 mil gaps (about), Tr = 88 ps. I have a similar thing on teflon, if I could just find it.

John

berichtnews: snipped-for-privacy@j39g2000yqn.googlegroups.com...

He's right that microstrip is dispersive, but that's not the major risetime-loss contributor in the op's situation. On a given board stackup, stripline is generally lossier... the trace will be skinnier, the dielectric layers thinner, and any pcb dielectric is lossier than air. Not to mention the vias.

John

berichtnews: snipped-for-privacy@j39g2000yqn.googlegroups.com...

Thanks, man. I figured as much on a gut level, but now I have to figure out the why of the terrible loss. Then maybe hit up some simulators.

berichtnews: snipped-for-privacy@j39g2000yqn.googlegroups.com...

No mystery. The trace is too skinny.

John

x-no-archive

50ps corresponds to about 20 GHz.

Sweep your PWB out to 20 GHz and see how much it rolls off at 20 GHz.

This problem may be easier to understand in the frequency domain instead of the time domain.

Figure out how to reduce losses at 20 GHz, the other posters already gave some clues.

Mark

Yes the initial simulation (which I now can't find to see who and with what it was done) showed that. We were going for 90ps out of the line, which was acceptable for the application, and I needed the trace width because of

1) density 2) stackup (hands are tied) ... but we now measure 150ps and I confirmed that to be sure. This is bad. We are over spec...

So I'll be attacking the problem in a few ways

1) Find out about the adhesion, how is it done? foil type, etc 2) The formulas often used for impedance are empirical approximations and never give attenuation/inch, 3) Again with the soldermask, I never got a good answer to the question of removing it: does it change anything? Most of our fast circuits expose the microstrip to be gold plated. All RF boards I've seen (eval boards for the chips, etc use this approach as well, but it looks like hard gold which is a problem in itself) 4) Stackup (thicknesses) can't change since it would be a major re- spin, so I'm trying to see if re-arranging ground planes will allow wider traces and not screw up the rest of the design 5) Run away screaming like a girl.

Rip up a trace (heat it with a soldering iron and peel) and look at the underside.

TXLINE does, but I don't know that I believe the numbers. And it's still loss at frequency, which is not risetime.

Plugging in your numbers, it claims 51.8 ohms and 12 dB/m loss at 3 GHz. The loss number sounds unlikely; maybe they're ignoring skin effect.

No.

That's the thing to do. Possibly slot out the plane on layer 2... barely possible.

Are you sure the lab's risetime measurement is accurate? I'm guessing your 150 ps with that skinny trace is real, but you never know. How are they doing it?

Oh, why does the risetime matter? Does it have to be 50 ohms?

No, run and scream like a MAN.

Or replace the trace with a piece of coax. Or fab a little transmission line kluge board and glue it on top; that would be fun.

John

A risetime of 50 ps corresponds to a system bandwidth of about 7.5 GHz... depending on the system.

BW * Tr = 0.35 for a gaussian-response system, like the typical oscilloscope. It's similar for a simple 1st order rolloff, like an RC.

John

Since there isn't much I can do about it, I've elected to email Rogers and ask their opinion.

I'll look into it. You always have the right free tool it seems.

Yes that's what I'm looking at but it interferes with the transition from the inner (short) microstrip. So I'll have to look at the tradeoffs here.

Don't know. The engineer is overworked so I stepped into the lab this afternoon and will test myself. It's chaos, there's only one 13GHz infiniium and only one 12GHz probe, all the other stuff is 6GHz stuff.

It's the spec and yes, it's the spec too.

approximations

John's number of 12 dB/m at 3 GHz seems awfully high to me. Could it be for FR-4, which as a loss tangent of about 0.015? I'm assuming your material is similar to Rogers 4350B, which has a dielectric constant of 3.5 and a loss tangent of only 0.003

I was going to ask you what your transition was between coax and your microstrip. But now it seems you have another transistion on the board? I suspect all these transistions. Can you look at the whole path with a fast (30 ps or 50ps) TDR? I've been surprised at how it doesn't seem to take much of a=20 glitch in impedance to slow down the risetime.

--=20 Regards, Howard snipped-for-privacy@ix.netcom.com

I use a Tek 11801/SD24 tdr sampling scope, 20 GHz, with around a 28 ps effective risetime as a combined pulse generator/scope. Under $2K on Ebay these days.

John

berichtnews: snipped-for-privacy@j39g2000yqn.googlegroups.com...

Why is microstrip dispersive, and how much? (excluding poor dielectrics with frequency-dependent dielectric constant of course)

Chris

It's dispersive because some of the dielectric is air and some is substrate. It's sort of like parts have different propagation speeds.

John

Thanks! I need to think about that a bit more...

Chris

Hmmm... I should make a multi-section L-C lumped model and simulate it. As far as I can see, it must have something to do with the component of the electric field that is not at right angles to the direction of the microstrip trace, otherwise I would have expected the magnitude of the electric field to be independent of frequency. I need to do some more thinking.

Chris