Re: gEDA-user: basic anti-EMI design q

[John Doty <jpd@xxxxxxxxxxxxx>]

On Mar 24, 2006, at 7:41 AM, Steve Meier wrote:

> Ideally true but in reality not true. Put you oscilloscope probe  
> tip to
> the ground plane at a point further then the probes ground. More often
> then not you will see some sort of noise. If the ground had zero
> resistence and an infinite supply of available electrons this noise
> wouldn't be there.

Not true. This usually isn't due to resistance. Taking skin effect  
into account, at 100 MHz a copper ground plane has a surface  
resistance of ~3 milliohms/square. On the other hand, a 3 cm diameter  
loop has an impedance of ~50 ohms. Now this is just dimensional  
analysis, but the mutual impedances that cause the crosstalk tend to  
be proportional to these numbers. So at this frequency you can expect  
crosstalk due to mutual resistance to be something like 80 dB below  
the mutual inductance crosstalk for circuits of typical dimensions.

The reason you see this on the scope is that the ground lead -> probe  
-> probe tip -> ground lead loop picks up the induction from loops in  
the circuit.

> Since ground is the return path of most circuits
> (obviously not differential circuits) and two or more circuits can  
> have
> return paths that cross each other on the ground there will be cross
> talk on the ground plane. In the case of digital circuits which often
> have high frequency clocks that clock itself can propigate through the
> ground and power and into the analog circuits where it can then be
> amplified causing distortion.

Propagate through ground? No! The fields in a copper ground plane are  
very low. The interference propagates through fields surrounding the  
circuitry. The most common problem is coupled loops, as in your probe  
experiment.

I wish EE profs would teach their students that the user of  
Kirchoff's Voltage Law is pledging to account for all of the  
induction in the circuit.

> So it is often recomended using two grounds which are connected at a
> single point. One for analog one for digital. The question is often
> where and how to connect the two grounds.

Wherever you have net current flow between the circuitry on the  
separate grounds, you want them connected at that point to provide a  
return current path. Wherever wish to accurately carry a single ended  
voltage between the ground systems, you want them connected at that  
point to provide a return reference. If you're serious about EMI,  
only differential signals may be allowed to cross between the ground  
systems without a return connection between the grounds at the point  
where they cross. Even there, you should be careful about common mode  
excitation and CMRR: it may be better to provide a common mode return  
connection.

The purpose of this is to collapse the loops that are the source of  
the trouble. The return current has a strong tendency to flow as  
close to the signal current as it can, so giving it a way to be  
really close to the signal can make the effective loop area very  
small. By reciprocity, similar considerations apply to voltage  
transport.

Alexander Graham Bell attempted to patent the "return next to signal"  
idea in 1878, but in 1881 the US Patent Office ruled that David  
Brooks had beaten him to it (although Bell was successful at  
obtaining a patent in England).

If coupling is through mutual resistance, the "single point ground"  
idea is a good one, because in this case the coupling must occur  
through a common conductor. For precision measurements below ~1kHz,  
this may be the main story. But for mutual inductance coupling, no  
contact between circuits is required. It's usually better then to  
"let ground abound": you're much more likely to get into trouble by  
being too clever.

John Doty              Noqsi Aerospace, Ltd.
jpd@xxxxxxxxxxxxx