Does this sound a little fishy? (Noise currents, split planes, etc.)

There is little point in splitting a plane between two clattery digital circuits.

You don't want large AC currents flowing into the bypass capacitors of the low level analog circuits. It makes sense to split or partly split the power plane so that you prevent the big currents from going through the area.

You never want to run a trace of any kind over the split between planes. It is often better to put a separate plane on a different layer as the analog reference plane and continue the power plane and ground under it. On a mixed board you usually have extra layers to work with in the analog area.

For things like DC-DC converters, you also want to give it its own plane layer that hooks to the overall ground in one place. This point should be to the degree you can, the return side of all of the filter capacitors. Ideally all of the connections other than this ground should have some small impedance in them so that all the AC current tends to go to ground in the capacitors.

If what your making is going to be part of a larger system and has an exposed metal chassis, never-ever use the chassis as the ground connection of the supplies. If you are trying to integrate a system where two people have used chassis as supply ground, you often end up with nightmares.

You always want to provide some sort of lossy device to eat up the RF power crawling along your power planes. In most cases, the capacitors are enough but if you use inductors to isolate power supplies, you can end up with a tuned circuit. This can lead to very hard to find problems that only show when the code in the micro runs in a loop past

07FFFH and 08000H.

from

rd" --

elf was

re "when

Vcc]

ten you

be

is

ice.

islands

hurt

c

ted

I don't really disagree but would like to emphasize some points. You want to "know where your currents are going at night". That is, you don't want 5A stepping motor currents flowing under the address bus, or high frequency digital spike currents flowing under your 16 bit converter. In cases like this actual physical separation (split planes) is appropriate rather than multiple parallel ground planes. It becomes an exercise in city planning; keeping the industrial areas separate from the boutique shops separate from the residential areas. The guy from Silent Solutions (an EMC company I would recommend) pointed out that if you have to do this and also have to communicate across the splits; make the joins bridges and pass all traffic over the bridges. If you can't afford optoisolation. My old adage is: over 50% of mixed signal performance is determined by the physical implementation. Take it as a serious part to the design! There is no black magic, just thorough control.

Ray

--

was

"when

you

islands

hurt

Another point to make is that making your analog signals differential or even quasi-differential helps too.

If you have many (lets use 10) analog signal to take from one domain to another, it makes sense to create one resistor divider that feeds the non-inverting input of 10 op-amps used as inverting amplifiers at the receive end. The divider between the two grounds can have a low resistance.

Interesting idea; thanks.

On chassis grounds... you're suggesting that a chassis should generally not be the same "ground" as your various power supply grounds? At all? Or just at DC? (I.e., still AC couple the chassis ground back to your power supply grounds.)

---Joel

ot be

just at

y

Chassis grounds and others If you have a current probe (or perhaps a scope), attach (or look differentially) a jumper between almost any two "grounds" to see the problems. "ground" is a much overused term. With uncontrolled connections current will flow (in all varieties), and unless controlled will wander over to the weakest part of your circuit; pessimistic viewpoint. This attitude subsumes "ground loops" and ilk. You can manage IO connections and grounds but never let the currents loose on your board; even a perfect lead/connection can pick up voltage from stray magnetic fields. Perhaps it "should be"a common ground but don't count on it. Even solving the difficulties of tying power supplies and transformers down there is always UL safety issues to side track intents. My favorite story is from the 60's where a janitor plugging in a buffer electrocuted a patient by means of equipment that was plugged into separate electrical outlets.

Ray

Do it lousy: then fix it and be a hero.

RL

I've noticed that happens a lot in software development... the guys who just write good, solid code that works on schedule aren't usually noticed, but those who write buggy code that endangers the schedule get plenty of kudos when they "go the extra mile" to fix the bugs just in time...

Projects where they split planes generate part of my income 8-D

That's it, inadequate bypassing at the CPU will do that. Maybe they goofed up the proper placement to 0.1uF caps or used lengthy traces stubs to their vias.

1uF? Best is to strategically sprinkle good 0.1uF caps, preferably in 0603 packages. And use the plane-to-plane capacitance to your advantage.

Keeping the planes intact is better, but shhht, don't tell anyone. I haven't reached retirement age yet.

Splitting up any plane in a system with fast signals greatly increases the chance of serious egg in the face at the EMC lab. Essentially you are creating a dipole antenna that happily radiates off any spikes that can be measured across the split. To add insult to injury it also receives any kind of EM spike or RF that might saunter through the ether. "Dang, why did the CPU lock up while we ran the same code as before when it didn't lock up?"

--
Regards, Joerg

http://www.analogconsultants.com/

"gmail" domain blocked because of excessive spam.
Use another domain or send PM.

The ideal situation is that the electronics has no connection at all to any exposed metal and that all shielding and ground be done inside the product. Sometimes you have to put a capacitor between the chassis and the electronics ground to meet the EMI restrictions.

Real nightmares can get started when an offline switcher uses its chassis as the electrical ground in its circuits. I have seen voltages as large a 1V between corners of the chassis. Needless to say, applying a 1V signal to the chassis of something with small signals was not a good thing.

The folk like UL want all exposed metal to have its own ground wire that runs all the way back to a good solid connection to the planet earth. They don't want any path that may bring dangerous voltages to the metal. This is a safety issue not a performance issue.