microstrip capacitances

John, if you'd describe the pair of traces (I can't access dropbox.com directly) I'll see if I can answer your question.

Hul

John Lark> Given a pair of parallel run traces on a board, FR4, like this,

Cool. Two parallel microstrip traces, each 6 mils wide, with a 6 mil gap between. Ground plane below.

_______ C3 _______ C1 C2 ___________________________________________

The dielectric is 10 mils thick FR4, between the traces and the ground plane.

(mils, 0.001 inches!)

If you can do that easily, it would be interesting to increase the gap to 12 mils. Eventually at large gap C3 goes away and each trace goes to about 1.8 pF per inch.

Thanks

John, maybe... the capacitence between two 6 mil traces 6 mil apart on the top of a pc board w/ dielectric const. of 4.8 is calculated at 7.155e-13 farads per inch. This calculation has not yet been verified so the result need be considered a ballpark estimate. The capacitance of a single 6 mil trace between 2 ground planes 20 mils apart comes to 2.698 pf/inch. This sould be equivilent to the sum of 2 distant traces on the top of a board. That I'll check. The calculation came, I think, from Analog Device's "Microstrip & Stripline Design", "MT-094 tutorial". The sum of these 2 should approximate your answer.

Hul

John Lark> > >John, if you'd describe the pair of traces (I can't access dropbox.com directly)

I was hiking up to my car, over the freeway after work, and WAG'd 0.6 pf/inch! I probably assumed a lower Er. That makes the minimum feedback C of my 80 MHz opamp about 1 pF, which is probably OK. We'll spread the traces a bit anyhow.

That's embedded stripline. Saturn does that, calculates 3.3 pF/in.

Did you do the calc by hand? It would be good to have a little program. I might do one.

Thanks for the help.

John Larkin <jlarkin@highland_atwork_technology.com> wrote in news: snipped-for-privacy@4ax.com:

I find it hard to believe that you do not have any magnetics software. Or if you do, are unable to use it right.

How long you been doin' this?

You have analysis software? You must just be missing something in trying to use it then.

Also, you could try the tried and true ARRL Handbook.

John - putting such calculations into code for the first use takes twice as long as doing it "by hand" but any further use pays nicely.

Hul

John Lark> > >John, maybe... the capacitence between two 6 mil traces 6 mil apart on the top of a pc board w/ dielectric const. of 4.8

For a stripline, you can just do a two-D solution; instead of Q = charge, it's K = charge_per_inch, and the C_per_inch = d(K)/dV with boundary condition being three conductors (two of 'em set to ground?). That doesn't give three two-wire capacities, though: you have to do something akin to the delta-wye inversion, I suppose.

One could define a 2d matrix, where each element is the voltage on a node.

Then define the conductors, with initial voltages. Force ground in a box around the space. Iterate and see what voltages you get. ATLC does that but computes txline impedances.

Maybe this will do what you want?

they don't make it easy to work out how much it would cost for a commercial license. This one is free, but it might take a bit more effort to define the problem. are lots of other free finite element modelling packages around that might also be suitable.

John

Their example is almost my exact problem. I might try the limited student edition, maybe buy the real thing if it looks OK.

The real thing costs 1K to 8K BP's.

You don't need the detailed voltage in space, that's easily approximated. All you need is to know how much CHARGE is in the electrode to make that voltage, i.e. you need to know the integral of the field divergence around the electrode. Yeah, a two-d grid of points is good enough. I'd be surprised if PC epsilon was controlled to better than 10%.

Seems to me that you can work it out from the even and odd mode impedances and velocities. Works fine for isolated traces.

Cheers

Phil Hobbs

Quickfield quoted me over $13,000 for the single-seat Electro module.

I assume it would require the usual FlexLM sort of horror to run.

I might lay out a proto board with a few cases, measure, and graph. Good enough.

I have a little Basic program for single traces, but one input is effective Er.

Saturn gives c and l for single microstrips.

I have never seen a calculator for my three-capacitor case.

I see a lot of formulas online for trace-trace capacitance, but they assume that the facing edges of the traces are simple parallel-plate capacitors in a uniform dielectric. That doesn't feel right to me.

I believe Clayton Paul's book Analysis of Multiconductor Transmission Lines describes several methods of calculating these capacitances. IIRC Fortran code was provided. I don't have my copy to hand but no doubt it'll be available for download somewhere.

It's a 2-D Laplace problem, which you can code up pretty fast--set up a uniform rectangular grid, draw equipotentials, and sweep through the array setting the potential at each cell to be the average of its nearest neighbors. With N conductors, I expect that it would needs N*(N-1)/2 runs with different boundary conditions to get all the mutual capacitances, but in complicated situations you could probably prune that some, based on the configuration.

It converges reasonably fast, especially on modern hardware. You can do a fair bit better by using successive over-relaxation (SOR). For a small 2D problem it's hardly worth it.

Cheers

Phil Hobbs

(Who had to code something like this for a second-year undergraduate lab, using an original Data General Nova.)

You might try FEMM

Or make a simple PCB and measure the C ?

boB

It would be fun, but the big lift is defining the geometry. ATLC uses Paint and inputs a colored bitmap.

Bumpfiles are super simple, so that's not difficult to reproduce. One of your Python brethren probably has a library that'll do that part right off.

For programs that put out pictures, I always just generate bumpfiles and then call 'convert' from ImageMagick to make whatever more complicated file type I actually want.

Cheers

Phil Hobbs