- electric cabinet that contains power supplies and a main board. It has a proper connection to ground earth.
- several boxes placed around the cabinet, few meters far. They are connected to the main board (inside the cabinet) using a shielded cable.
- each one of these peripheral boards has another shielded cable that connect to an external sensor.
I'm wondering how manage the cables shields. My thoughts:
- main board (inside the cabinet): provide a large connection to the earth of the cabinet itself. Connect the local earth to the shields of outgoing cables.
- peripheral boards (inside external boxes): provide two pins to connect together the shields of both cables (the one from the main board and the one from the sensor).
- sensor: leave the shield floating.
Questions:
a) is it correct? b) even if the sensor is placed very close to a metallic item already grounded, I must not connect the shields there. Otherwise I'll run into a ground loop, won't I?
The goals are, as usual, reducing EMI and induced noise, protect the boards from high voltage spikes...
I'm doing a system just like that now. The main box connects to 32 external boxes, each of which connects to a sensor.
No. Ground the shields as often as possible.
Yes. Ground loops are good. Of course, avoid getting any ground loop potentials into your signals.
Exactly. That's why the shields should be grounded, and why the PC board ground planes should be grounded.
An ungrounded shield is an antenna, and can have large RF and ESD potentials relative to local ground, and it will couple those potentials to the wires inside it. The result is nasty common-mode signals.
There could be exceptions, like a coax running to a photodetector that's in a metal enclosure. Grounding the enclosure would create a ground loop that gets into the signal. So connect the shield to the PD box, but float that box. I'm doing that in some cases. Still, neither end sees a big RF common-mode signal.
How to avoid differences of potential along ground? We're talking about an heterogeneous structure, with different kind of ground earth: e.g. main supply, physical grounding of buildings, etc...
Ok I've got it.
Yeah, but AFAIK the shield of a cable should be connected to ground on at one end. I understand your explanation so I'm wondering when it's correct my idea.
Uh, you're just answered to my last question :)
Is there a way to measure the quality of grounding in such a system?
The situation is complex, so generalizations must be questioned.
Single-ended signals, like audio between chassis over RCA cables, are tricky. Using single-ended signals and grounding both ends will inject ground loop potentials into the signal; and floating one end of the shield will be even worse.
In general, the best way to run analog or digital electrical signals is as balanced differential pairs inside a shield that's grounded on both ends. But every specific situation needs to be thought about.
It is rare, very rare, when it would be advantageous to shield a cable but not ground the shield on both ends.
In the event that a shield winds up carrying a serious amount of ground-loop current, more thinking is required.
Strong recommendation: read Jim Brown's writeups at
formatting link
--
Dave Platt AE6EO
Friends of Jade Warrior home page: http://www.radagast.org/jade-warrior
I do _not_ wish to receive unsolicited commercial email, and I will
boycott any company which has the gall to send me such ads!
Well, assuming that the ends of these chains of boxes are not connected to anything, then you are right. But, if these sensors at the ends are electrically connected to noisy equipment, then there will be significant currents flowing through the shields, and possibly grounds, too, throughout the system. This can be quite troublesome. Also, if the cables from sensors to central box don't follow the same route, they form loops with included area. If there is significant magnetic radiation, such as from large transformers, magnets, big DC power supplies and especially welders, then you pick up large induced voltages around the loop.
There are high-priced gurus who travel the country fixing problems like these at several thousand $ per day.
Finding the optimum scheme to achieve shielding and yet not create troublesome ground loops can be quite a challenge, and you may end up breaking most of the general advice.
I work on nuclear instrumentation systems at national labs, and they have all the ingredients for havoc. Extremely small signals with high impedances, very high gain amplifiers, huge magnets with SCR or transistor switching supplies (up to 2500 A @ 100 V is commmon) located hundreds of feet from the magnets, and signal cables routed several hundred feet away to control rooms for observation while experiments are running. Combine all this, and the potential for noise to get into sensitive systems is huge.
And, this is where it gets really messy! You want to avoid ground currents flowing in your shield. But, if you need to have the remote box or sensor at the same ground potential as the central box, then you HAVE to have a heavy ground conductor between the boxes. This creates a compromise. One way is to have coaxial shields, an inner one for the signal return, and an outer one for shielding against external influences.
And, of course, the shields only exclude electric fields, not magnetic.
In many cases you have to give up, and either use optical connections or fully differential signals to get rid of the noise affecting single-ended signals.
On a sunny day (Thu, 01 Dec 2011 14:06:56 -0600) it happened Jon Elson wrote in :
The TV studio I worked is located next to a railway, the railway has a special stop for that place. There are hundreds of km (really) of coax running in that place.
1 Vpp signals mostly.. I have heard that the (all electric) trains induced strange signals... Say a hundred meters or so away. Dunno how they solved it, different department (installation).
An other company I worked was next to a big railway station in Amsterdam, there was a strong 400Hz magnetic pickup from the trains in the whole building.
Get hold of a copy of R Morrison's "Grounding and Shielding Techniques in Instrumentation"
It's old now, and was originally written when mains interference was the biggest problem, but it's great virtue was that Ralph Morrison not only had a lot of practical experience in the area, but also wrote a very clear description of what was going on when external interference induced noise in instrument circuits.
I got my employers to buy copies everywhere I worked, and finally bought one of my own - the fourth edition "Grounding and Shielding Techniques" ISBN 0-471-24518-6.
Except in situations where a cable runs for longer distances to something on another power circuit or connects to modules where ground can have a different potential but is capable of delivering lots of current. In that case it can be better to have a parallel RC connection at one end.
If you think of a shielded cable as a transmission-line transformer, you *want* ground loop current in the shield. Or, to be more precise, you want the shield ends to be at the potentials of the chassis on both ends, so the ground voltage difference will be induced into the signal wires. The current in the shield is a side effect.
One way reducing the ground loop currents is to wrap the coax around a toroidal core, making a balun.From the point of view of the ground loop, the screen lead wrapped around the toroid is a simple (if rather low value) inductor and inserts some impedance into the ground loop.
Interestingly shield current INDUCES noise into poorly made cabling. And, all cabling is 'poorly' made, by its very nature.
You want to see some noise problems, try to reduce noise in a system around the neodynium laser fusion experiments when teramegawatts fire off. Or, during a good EMP wave at 50,000 V/m with spectrum beyond
I made some measurements of coax cable leakage a few years ago. See .
In short, the amount of signal leaking into a coaxial cable is given by its --frequency dependent-- transfer impedance, specified in ohms per meter of cable. At DC and low frequencies (
It induces ground difference voltage into the center conductor such as to cancel noise. Delta-V at the transducer becomes delta-V at the other end. As Bill points out, adding common-mode inductance enhances the effect by improving the idealness of the effect, specifically by increasing the L/R ratio of the shield.
One manifestation of this isolation effect is the coaxial pulse inverter, just a hank of coax that you cut at midpoint and resplice, but with the inner and outer conductors swapped. If you connect it to a grounded pulse generator and a grounded scope, it's technically a short, but it transmits and inverts pulses like a charm. Running it through a toroid extends the low frequency response.
formatting link
The LT Spice transmission line model is the ultimate example of this idea. The generator end can be grounded to anything, and the receive end grounded to anything else, and it still transmits the differential signal perfectly.
There are limited scenarios where this approach is optimal. The coupling will never be quite perfect, the induced voltage on the signal conductors will differ with any differences in termination impedances, and shield to conductor coupling drops with frequency below the shield "cut-off frequency" where skin depth is greater than shield thickness, typically anything under around 2 MHz, with skin depth not having anything like a sharp cutoff. (Calculated "skin depth" is the depth at which, if the current was constant up to that depth, the same resistance would be seen as for the actual gradually reducing current with depth profile. Something like 30% of the current actually flows below the "skin depth" where conductor thickness allows.) See Henry W. Ott, 'Noise Reduction Techniques in Electronic Systems', for a really nice treatment of shield effectiveness and coupling ratio vs frequency (among other things).
I once fixed a system designed by someone else where a 3 wire shielded tachometer cable picked up better than 5 volts of noise from a VFD, causing an overspeed trip at about 5 RPM actual, over 4000 RPM indicated. The 3 wire tachometer gear tooth presence sensor switched its output between +5 and COM with under 1 ohm impedance, isolated from the fully sealed grounded metal case (except for parasitic capacitance of course), with cable shield connected to the case of the sensor and to chassis ground at the tachometer end. Top quality cable with heavy copper foil plus braid shield. After estimating the value of the significant parasitic circuits, something which is easier after looking at a lot of examples (Morrison, 'Grounding and Shielding' is good), I determined that for this case the biggest contributor was VFD switching edges injecting current to ground through the largest capacitance to ground in the circuit, the motor. The tach cable shield had more surface area than the safety ground wire that shared the return current from the motor frame, and not much more loop area due to routing. Shield current partially coupled to all 3 tach wires (frequency below shield cut-off), no effect on power or ground low impedance connections but injected enough charge into the diode clamped RC low pass filter at the tachometer signal input to get over the trigger threshold.
While there were many ways to fix the problem, the best being to get rid of the interference by placing the VFD very close to it's reconstruction filter (motor), which would have been free at design time but expensive after, or add an output filter to the VFD which would have cost about $1200 and 3 days of my time waiting for it to arrive. Because the system was a turbo-generator control set maintenance trainer (with VFD replacing turbine) no changes to the tachometer or it's wiring were permitted per the contract, but as a quick test I lifted the cable shield to ground connection at the tachometer end, immediately knocking the crosstalk down to under half a volt - insignificant for this application.
While I was resigned to hanging around Black's Beach patiently for a few days while waiting for the permanent fix VFD output filter to show up, the customer immediately accepted my test fix as good enough, signed off on system acceptance, and I caught the return flight of the same non-stop I arrived on a few hours earlier. 15 hours travel and 1 hour work, a good day of troubleshooting :-).
I would have liked to explain the problem to the original designer, but the customer was overly fond of secrecy and wouldn't tell me who it was, to the extent of providing me with drawings having blank title blocks. I expect the designer got my markups showing the single change to the control set wiring and concluded that it was a control set problem, not a turbo-generator simulator problem, and thus not due to his incompetent design :-).
I have only found one universally applicable rule of thumb for EMI problems: Never apply any rule of thumb without understanding its basis and being sure it applies to the situation. I think that proper application of this rule of thumb will solve all EMI problems - sometimes with a shield connection at one end, sometimes at both, and sometimes at neither (semi-shielded CAT-6 being the only example of neither I can think of offhand), and sometimes without any shield at all (UTP, fiber).
On Dec 2, 11:54=A0am, Glen Walpert wrote: ...snip...
ng
...snip...
Worth mentioning to people who are going to use the skin depth equation: The value of attenuation predicted by the standard skin depth formula is usually higher than reality, because the standard formula's derivation is based upon the assumption that the fields are 'planar'. A planar field is a good representation a long distance away from the source where the radius of curvature is large and the sphere looks almost identical to a planar wave. In actual applications noise sources are much closer where the radius of curvature is much smaller. The result is that the fields will 'punch' through a shielding conductor and penetrate much further than skin depth equation predicts.
Our NDE Eddy Current Instrument takes advantage of this 'punch through' by using very small dipole magnetic fields. The punch-through allows us to routinely 'see' to depths of more than 5 skin depths and resolve small details clearlly at depths of more than 3 to 4 skin depths. In addition, the 'punch through' makes it exceptionally easy to see broad separations in material of less than 0.1 mil, measurable at depths more than half inch below the surface. Not bad when you consider the received field is inversely proportional to the distance into the material to the sixth power!
Anyway, the point is if the noise source is close to a shield, the shield won't attenuate the field as much as the skin depth equation would predict.
However, it *IS* a good starting point and the formula is super easy to remember, square root of quantity 2 over conductivity, permeability, frequency with conductivity in Seymans per meter, permeability in MKS, and frequency in radians per second and the result is in meters.
Oh, yeah! I was once in a major department store headquarters. They were in a historic old building, and it had electric service coming from three different substations. Two were from one utility, the other one came from Ontario Hydro, from a 25 Hz to 60 Hz motor-generator station. Talk about insane electrical problems! They had a huge sheet of copper installed below the raised computer-room floor to try to create a single-node ground, but it didn't work. There was a separate neutral and safety ground for each service entry. You could get a significant shock just putting your hand on two pieces of machinery walking through the computer room! They had RS-232 cables burn out the ground pin due to the ground potential difference. They eventually ended up moving the whole data center out of that God-awful building.
ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here.
All logos and trade names are the property of their respective owners.