I must confess, this is a an issue on which I don't have a comfortable grasp. If the noise voltage is common to both signal ground and chassis/safety ground, where are we shunting the common-mode current to? In other words, how can any current be shunted between two voltages which are common to each other? What am I missing here?
Perhaps you're only noticing the smaller loop - that between the signal and its accompanying ground/shield. But common mode currents come in over one pair of wires and exit via another. So the shunting that's needed is around internal sensitive circuits so no noise voltages are imposed across their 0V reference points.
If everything in your device including the metal box is at the same (common mode) potential, then there is no problem. The circuits inside don't care, everything is relative.
Problems begin when for example two cables carry a different common mode voltage (and therefore, common mode current will flow).
Those cables end up connected to a board inside your device.
Common mode between these cables will cause a current to flow, across whatever non-zero impedance is between your two connectors shields (maybe a bit of track, a ground plane, etc).
If those cables are connected say, to BNC screwed to an aluminium plate, very low impedance, thus very low voltage.
If they are connected via thin tracks or wires to your star ground, higher impedance, higher noise voltage injected into sensitive spot = asking for trouble.
It works both ways : if both cable shields are connected to points in your device which are at different potentials (like an analog and digital ground plane) then the cables will carry different voltage on their shields and act as an emitting antenna. Check out the video on the first page of LearnEMC - EMC Training for Engineers and Technicians
The same happens if your box capacitively couples to something. Draw the current loops, you'll see what I mean.
Now, transformers.
Ethernet unshielded twisted pair : the TP is connected to a transformer. However when the device transmits, some current will leak thru the interwinding capacitance. Thus a capacitor connects the transformer center tap (cable side) to chassis to provide a local return path.
SPDIF uses a coax. Coax shield must be terminated to the chassis as we've seen. Here we have a problem if we want isolation... A solution can be to connect the shield to chassis via capacitors (again, provide a return path for transformer leakage current). It doesn't work as well as twisted pair, though.
Peufeu, Richard,
Thanks, for helping to clarify this topic for me. So, if I correctly understand, the shunt impedance half of an effective common-mode filter shunts any shield/chassis/safety ground currents around the signal handling circuit. To achieve this, we need to ensure that all shield/chassis/safety grounds that are to be connected together have a low impedance path provided between them, one which totally bypasses the signal circuit. Does that understanding sound correct?
Thanks, for helping to clarify this topic for me. So, if I correctly understand, the shunt impedance half of an effective common-mode filter shunts any shield/chassis/safety ground currents around the signal handling circuit. To achieve this, we need to ensure that all shield/chassis/safety grounds that are to be connected together have a low impedance path provided between them, one which totally bypasses the signal circuit. Does that understanding sound correct?
Sounds spot on to me at least 
Solutions are either
a) Put all I/O and power connectors close together on only one side of the PCB and use the traditional groundfill with a break to a point so that 'lateral' currents (between I/Os) don't impinge on the rest
or
b) Adopt rigorous star earthing so that all CM currents converge to and diverge from an infinitesmal point and don't use a global groundplane/fill. Local 'island' planes are still compatible with this.
I prefer (b) myself as with (a) there are still common ground impedances (albeit reduced).
Solutions are either
a) Put all I/O and power connectors close together on only one side of the PCB and use the traditional groundfill with a break to a point so that 'lateral' currents (between I/Os) don't impinge on the rest
or
b) Adopt rigorous star earthing so that all CM currents converge to and diverge from an infinitesmal point and don't use a global groundplane/fill. Local 'island' planes are still compatible with this.
I prefer (b) myself as with (a) there are still common ground impedances (albeit reduced).
Yes.
If there is a current, then there is a loop.
If it is common mode current, you only control the part of the loop inside your device, not outside (cables, other gear...)
If this current should not go somewhere (for exemple the ground reference of your input stage) then the solutions are :
1- Increase the loop impedance by adding common mode filters, transformer, etc so the current decreases
Transformers isolate well at low freqs, not at RF. Common mode chokes provide some impedance at RF.
2- provide a low impedance path to short a portion of the loop
For example an aluminium plate at the back of the chassis. This works well at RF but creates a ground loop inside your gear if it is not balanced. A solution can be to bind the connector grounds to the chassis only at RF (using small capacitors).
If there is a current, then there is a loop.
If it is common mode current, you only control the part of the loop inside your device, not outside (cables, other gear...)
If this current should not go somewhere (for exemple the ground reference of your input stage) then the solutions are :
1- Increase the loop impedance by adding common mode filters, transformer, etc so the current decreases
Transformers isolate well at low freqs, not at RF. Common mode chokes provide some impedance at RF.
2- provide a low impedance path to short a portion of the loop
For example an aluminium plate at the back of the chassis. This works well at RF but creates a ground loop inside your gear if it is not balanced. A solution can be to bind the connector grounds to the chassis only at RF (using small capacitors).
When the source of the CM current is an SMPSU increasing the source's impedance is tricky because its already very high - just stray capacitances between trafo windings which might amount to of the order of 10-20pF - but from a high potential (mains). Thus its in effect a noise current source..Common mode chokes can provide a high enough impedance to make a dent in this but usually only over a fairly narrow band near their self-resonance. I wind my own segmented CM chokes to get the highest SRF (lowest strays).
Sounds spot on to me at least
Solutions are either
a) Put all I/O and power connectors close together on only one side of the PCB and use the traditional groundfill with a break to a point so that 'lateral' currents (between I/Os) don't impinge on the rest...
Yes, I agree, that is the logical implication. Conventional wisdom calls for separating power and signal connectors as far as possible, so to minimize normal-mode noise coupling. Yet, I can now see how such physical separation can adversely affect common-mode noise coupling. Interesting.
Engineering a decent 75R digital output is not rocket science. Fairly easy to do with solid-state; possible, but much much harder, with valves. Only a fool would deliberately introduce a valve at this point. The difference between the two plots is LF response (not audio LF response!) and almost entirely irrelevant.
Yes! Finally someone who sees it. These to plots are different, one apparently not triggered well, so seem to be made with a purpose.
Jan
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.