When I teach my "Grounding and Interfacing" seminars and deal with cables in balanced line level interconnections, I find that students frequently ask which end they should ground. It is a rather controversial issue that divides industry experts into one of three camps - ground at the driver only, at the receiver only or at both ends. Proponents of each usually argue simply that their choice works fine for them. I have learned over the years that people continue to do wrong things because most of the time, they can get away with it. In this column, I will offer two factual arguments to show that grounding only at the receive end is a bad idea.
First, which end is grounded will affect how much ground noise is converted to signal by driver and cable capacitance imbalances. As I wrote in a paper that appeared in the AES Journal (Ref1), the most significant effects of a shielded twisted pair cable are caused by the capacitances of its inner conductors to its shield. The total capacitance is about 67 pF per foot (221 pF/m) for a typical, widely used cable. What is more important is that this capacitance is not evenly divided between the two conductors. Measurements on several brands of cable show that imbalances of 4% to 6% are common. These capacitance differences are due to normal tolerances in the thickness of the wire's insulation. Ultimately, it is because wires of different colors are extruded from different machines.
If the cable's shield is grounded only at the receive end, then these capacitances and the output impedances (R superscript S1 and R superscript S2 in the Figure 1 and Figure 2) of the driver form two low-pass filters. Remember that driver output impedances are not well matched either. Unless these two filters match exactly, which requires an exact match of both driver output impedances and cable capacitances, mode conversion will take place. Conversion makes some of the ground noise (V superscript cm in the figures) appear as differential signal on the cable. This conversion is, of course, worse with longer cables and large driver impedance imbalances. Examination of Figure 1 shows how the common-mode noise is low-pass filtered. Remember that our reference point is the receiver ground.
This conversion, however, can be completely avoided if we simply ground the cable shield at the driver end instead, as shown in Figure 2. Now, no ground noise voltage appears across the cable capacitances, and no filters are formed. Because the shield now is at the same voltage as the driver ground, there is no ground noise voltage across the cable capacitances, and they effectively disappear. This virtually eliminates unbalancing effects of mismatched cable capacitances.
Grounding of shields at both driver and receiver creates some tradeoffs. The conversion effects, predictably, fall between the schemes of Figure 1 and Figure 2. The benefit is that such a connection can reduce the ground noise voltage difference between the boxes even though it degrades the rejection of it by the receiver. If there is no other ground path between the boxes, then using the cable shield to connect them will likely reduce the common-mode (noise) voltage. It would be far better, of course, to use some other means, such as a dedicated grounding system or the AC safety ground to limit this voltage. Devices with two-pin AC plugs are the most troublesome, with their chassis voltage often well over 50 VAC above utility safety ground. The current available is small, posing no safety hazard, but it creates a large common-mode voltage unless there is a grounding path.
Second, which end is grounded will affect how high-frequency signal currents in the shield get back to the driver. These currents are caused by the same imbalances in cable capacitance and imbalances in the symmetry of the two signals from the line driver. If the capacitances were exactly equal and the signal swings were exactly equal and opposite (symmetrical), then there would be no signal current flow in the shield at all. In the real world, however, neither is exactly matched, resulting in signal currents in the shield, which must return to the line driver. If the shield is grounded to the driver only, there is no problem. If, on the other hand, the shield is grounded only at the receiver, this signal current will return via an unknown path. This path could include sensitive circuits that couple it back into the signal path. If this occurs downstream, it can cause crosstalk. If it occurs upstream, it can cause oscillation. Remember, this coupling is capacitive, making its symptoms worse at higher frequencies. I have seen newly wired consoles break into ultrasonic oscillation (silently pegging the VU meters) as the faders were opened up. This problem was caused by cables with shields grounded only at the receive end.
The only possible benefit of grounding at the receiver end is that it sometimes helps to suppress RF interference, but a lack of immunity to RF interference at an input is really a symptom of inadequate design. The real solution is an RFI (lowpass) filter on the input inside the equipment. Good equipment has this already built in.
A good compromise solution is to ground shields at both ends. As I described in my AES paper, the CMRR degradation is trivial in most systems. The only possible downside to grounding at both ends is caused by an equipment design problem which has been dubbed "the pin-1 problem" by Neil Muncy (Ref 2). This defect essentially makes the shield connection (pin 1 on an XLR connector) behave as an audio input. Because a shield connected between two boxes will always carry some small power-line noise currents, this defect will cause them to be amplified and added to the signal. Obviously, a shield connected at only one end will not carry current from box to box and would prevent this problem. As you can see, the issues are complex, and there are plenty of tradeoffs, but, in any case, grounding a cable's shield only at the receive end is a generally bad idea. I invite any reasonable arguments to the contrary.