Multiple Grounds

Yes, that's precisely the problem but eliminating crossovers only solves part of the problem. Ground currents can still "want" to go across the slot. Slots (discontinuities) are to be avoided for all but the simplest (low performance) boards.

I make a couple of prototype boards with a microstrip trace on top and a ground plane on the opposite side, and measured the signal integrity with 30 ps TDR/TDT. Cutting a slice out of the ground plane had no visible effect. I've also added test traces to 4 and 6 layer boards, with various ground plane insults. Again, small width slits in the ground plane have no visible affect on microstrip integrity.

So, we don't worry about traces crossing small plane cuts, like for instance crossing regions of a mixed-voltage power plane.

I didn't test for radiated emission. I suspect that nanosecond-risetime edges will weakly excite a multi-GHz slotline resonance.

--

John Larkin         Highland Technology, Inc 

lunatic fringe electronics

On 2016-02-18 03:34, John Larkin wrote: [...]

I differ. I've just made a picture of a 30ps step on a microstrip line going over a 3cm long transverse cut in the ground plane under the strip. I see a 20% reflection!

I suppose you still have other planes that continue under the cut.

See here: .

Jeroen Belleman

You have practically cut the board in half! A TDT would probably show a decent logic level at the other end.

A mixed-voltage power plane might have 8 or 10 mil gaps between the pours. That will have no effect on an adjacent-plane signal. On a multilayer board, all the various layer pours are almost equipotential.

Yes. The various ground and power planes stich one another together.

--

John Larkin         Highland Technology, Inc 

lunatic fringe electronics

Agree, mostly it is a bad idea to split grounds. Exceptions are audio and analog video where external ground loops could occur. With audio one should go differential then (XLR connectors) but with video there is often no choice.

The customer is always king. I give them a fully isolated circuit as well if they insist but not before alerting them to the additional cost and board space.

--
Regards, Joerg 

http://www.analogconsultants.com/

I did a balanced differential video feed once. Worked fine - to my surprise, and that of the junior engineer who ended up testing it. I threw in a balun to reference the differential output back to local ground.

Once bitten, twice shy ... my 1996 Peltier junction thermostat circuit drives the MOSFET switches in the switching output via opto-isolators, just so that I could keep the ampere-level gate drive currents tightly isolated within the output stage.

When I last looked the paper had 18 citations, and nobody had bitched about the opto-couplers.

--
Bill Sloman, Sydney

1000Ghz?? I must be reading that wrong. It's early in my part of the world and I haven't had a single coffee yet.

We talk picoseconds whenever we get below 1ns. We'll talk femtoseconds when we'll get below 1ps.

My fastest pulse generator says it will make 15ps edges, but my fastest measurement instrument will only do 30ps. That's acceptable for the fastest gadgets I build, like this 70ps risetime particle beam current transformer.

Jeroen Belleman

I didn't say *one* picosecond. The fastest edges we are currently generating are around 40 ps, with pulse widths of 100 ps and timing (edge placement) resolution of 1 ps. That works on multilayer FR4, but it takes some care.

The trickiest part of that board was the transition from the tiny final amp chip to the SMA connectors on the right. We did a bunch of actual e/m simulation to get that right.

It is tricky to post before coffee.

--

John Larkin         Highland Technology, Inc 
picosecond timing   precision measurement  

jlarkin att highlandtechnology dott com 
http://www.highlandtechnology.com

Every kid should be required to learn how to use machine tools, and to solder and weld, and to cook.

--

John Larkin         Highland Technology, Inc 
picosecond timing   precision measurement  

jlarkin att highlandtechnology dott com 
http://www.highlandtechnology.com

I'd settle for making something, anything - rather than just buying.

Cooking comes naturally once they leave the nest. If not then it is evolution in action.

No harm in teaching them that you can create good edible meals without being a professional cook, though.

I'm surprised that SMAs were adequate, regardless how you fed them. I thought you were a fan of SSMB's?

No, we use a lot of SMBs. They are good to at least 4 GHz. SMBs are fast to mate/unmate and can be placed a lot closer together than SMAs.

SMAs work fine at 20 GHz or so. Even BNCs are pretty good to several GHz.

--

John Larkin         Highland Technology, Inc 

lunatic fringe electronics

Perhaps you misunderstood and he meant "Devilishly Cunning"?

SSMB is not SMB.

Ok, good to know. I thought 6GHz was more realistic.