Consider an unshielded length of cable. Say you have twinax (twisted pair in coax shield), and there's a gap where the shield is removed, replaced with a shorting wire instead.
Consider a source of common mode energy. It might be coupled onto the cable with a transformer, or by this cable forming a loop through an RF field, or...
The common mode energy tends to travel over the outside surface of the shield. Remember that the currents on the inside and outside of the shield are independent (it acts as a Faraday cage). And the inside shield carries the image current of the signal lines.
So, in the unshielded length, there are three wires (two signal and one shield), which have equivalent inductance*. Normally these are well coupled (in effect, a wraparound shield ensures 100% coupling between the shield and what's inside), but the coupling coefficient is less than 1 for individual wires (twisted closely or otherwise). So we get a nonideal transformer, and different voltage drops for a given current flow.
(*Again, inductance is an approximation, but that's sufficient for our purposes here.
)
So we have a voltage drop across the shield wire, which is not shared by the other wires. Looking at the signal voltages at the far end of the cable, it looks like a voltage source has been added in series with the shield, and therefore with the signals as well. Or to think about it a different way: the current that was supposed to stay 100% outside the shield, is now part of a current divider (i.e., the parallel equivalent of a voltage divider), mingling with the signal current(s) which were supposed to be separate!
Or if there's voltage drop between shield and circuit ground, at the connector, same thing: the signals are pushed up by that amount.
The fundamental point is: if you want to measure a voltage, you must do so at the point of origin. Whether it's an oscilloscope probe or digital logic input, you don't want to have stray wiring picking up outside signals (or, causing loss of the intended signal, since it works both ways).
My most dramatic applied example of this was a client's application, which used USB signaling through ungrounded connectors. The equipment was failing EFT immunity even on the machine's lowest setting (300V). Ferrite beads did nothing. What I had to do was replace all the connectors with shielded metallic connectors (no small cost, sadly..) and string up all the shields (at least they started with shielded cable!) to get all the grounds together. After that, plus some grounding strips and a few well-placed ferrite beads, I had the EFT immunity beyond 2500V: at which point my crude setup was arcing at the (non-HV BNC) connectors. But that was 8dB beyond what we needed, so that's plenty!
Consider that a 2500V EFT pulse has to be attenuated over 60dB (to 2.5V) before it will stop wrecking logic thresholds (USB's immunity range is about half the supply, or ~1.6V). That's a good shield! Just 1cm of missing shielding has ~10nH of [uncoupled] inductance, and if the average frequency content of an EFT pulse is 100MHz (probably not unrealistic, for being such a wideband signal), then the reactance is 6.3 ohms, which acts as a voltage divider with the ~150 ohm common mode source impedance**, putting a mere 100V on your logic signals!
**Common mode impedance is rather arbitrary, but it's rarely less than 50 ohms, and rarely very far into the 100s. The impedance of free space is 377 ohms, and most transmission line geometries have an impedance some ratio to this value (usually a lower ratio, except at rather extreme distances). The common mode is measured between all conductors of a cable, and ground. Which is to say: only the shield matters, in a proper shielded type. But, unshielded multiconductor will be very similar, taking the dimension of the 'shield' as the outline of the bundle of wires. So you can still have well-defined common mode impedances, with, say, CAT-5 (UTP) cable running through a duct or raceway (= ground plane!). The impedance of a 6mm round cable about 25mm above ground plane is around 150 ohms, which is a very typical condition for immunity testing.
This, by the way, is also why ferrite beads don't tend to do much, or at least much more than a few dB (say 3 or 6, but not much more than 10). They act as a common mode impedance divider with respect to the transmission line impedance. At frequencies of 100MHz or so, ferrite beads are most commonly 50-300 ohms (more for very large or multi-turn affairs, though multi-turn cores also peak at a lower frequency). The primary application of them is actually to dampen resonances: where you might've had a resonant cable before (where the
worst case amplitude might be +20dB, causing problems to your circuit), now it can be dampened, so that the wideband might be only a few dB better (or worse, as the case might be!), but the worst offending peaks (or dips!) will be shorter and wider (lower Q).
To maximize the value of a ferrite bead, try to arrange them in a voltage divider against even lower impedances (e.g., grounding clips -- maybe you can't fully shield the connection point, but you can get a ground in there. Effectively, you're using the ferrite bead to increase the coupling coefficient of the unshielded wires -- you aren't reducing the leakage inductance (which is the difference that's causing problems), but if you can drop most of that voltage across the ferrite bead rather than the wiring (reducing the ground current), you can have it shared evenly, and thus improve signal quality.
Tim
Home
Help
Search
Login
Register
