Common-mode surge rejection of power supplies

Is the 48 volt supply powered by the AC line? If so, does it have an IEC connector or equivalent, L-N-G?

Is the supply inside the same chassis as its load?

What, ultimately, "eventually", gets grounded to the box?

--

John Larkin         Highland Technology, Inc 
picosecond timing   precision measurement  

jlarkin att highlandtechnology dott com 
http://www.highlandtechnology.com

e is a 48 VDC supply, and the output floats (by itself). The negative of t his supply is eventually taken to "ground" at some location in the system. This "local ground" is returned to another ground in the system, but the i mmunity testing is exposing the fact that low resistance|impedance is not z ero resistance|impedance.

n-mode," or at least pseudo-common-mode. The aforementioned 48 VDC supply seems to pass this common-mode surge to some extent. I am very confident o f this fact, if not the magnitudes involved. (I am 1500 miles away from the compliance lab, and so have relied on a few measurements made by other peo ple.) Whatever the magnitude is, it is locking some stuff up, that then nee ds a power cycle to return to normal.

Normally when this happens, it is because you do not have a good low impeda nce clamping path from line to earth and your chassis/earth has not suffici ent low impedance or is connected to other parts of the circuit (this creat es ground bounce internally, which reset microcontrollers, etc)

at the lack of specifications for power supplies for common-mode transmissi on (attenuation). I didn't see any information regarding standards for tes ting such attenuation.

It really does not make sense, EMC is not quantitative in that way. You nee d to increase your robustness, for surges with kA currents, it is hard to p redict the robustness level, so you need to do your tricks in the lab

ion versus frequency. (Although details, regarding impedances, for example, are not always obvious. Yet I have read that 50 ohms is an industry stand ard, an "interesting" selection of impedance.) Is anyone aware of standard s or common practice with regard to common-mode attenuation specifications and standards for testing it in DC power supplies?

nd usually unipolar. That means they have DC content and EMI filters only help so much, as they pass DC. This may mean that it is helpful to include a time domain based surge attenuation measurement, although that might be tricky. But, the isolation XFMR in AC-DC supplies does block DC. Common mo de RF attenuation seems easier to evaluate than common mode surge attenuati on. What do you think?

Sound strange to me about your statement of DC. Surge pulses are superimpos ed on the mains voltage, normally a 0.5us/20us pulse, and that's not DC

Cheers

Klaus

re is a 48 VDC supply, and the output floats (by itself). The negative of this supply is eventually taken to "ground" at some location in the system. This "local ground" is returned to another ground in the system, but the immunity testing is exposing the fact that low resistance|impedance is not zero resistance|impedance.

It is a molex that has LNG.

Not really. But they are attached together. (In the sense the load clamps onto the main chassis.) There is no real coherent scheme of low-R/I chassi s ground in the main chassis itself, and a low-R/I bond is not available to the load. It will never be available. The "system" I mentioned is the mai n chassis, its internal elements, and the load.

What "box"?

here is a 48 VDC supply, and the output floats (by itself). The negative o f this supply is eventually taken to "ground" at some location in the syste m. This "local ground" is returned to another ground in the system, but th e immunity testing is exposing the fact that low resistance|impedance is no t zero resistance|impedance.

ps onto the main chassis.) There is no real coherent scheme of low-R/I chas sis ground in the main chassis itself, and a low-R/I bond is not available to the load. It will never be available. The "system" I mentioned is the m ain chassis, its internal elements, and the load.

When you connect the load to chassis, then the currents are going to flow i n chassis, and that's really not good. Why do you connect the load to chass es and not to an individual/seperate wire?

Cheers

Klaus

ere is a 48 VDC supply, and the output floats (by itself). The negative of this supply is eventually taken to "ground" at some location in the system . This "local ground" is returned to another ground in the system, but the immunity testing is exposing the fact that low resistance|impedance is not zero resistance|impedance.

mon-mode," or at least pseudo-common-mode. The aforementioned 48 VDC suppl y seems to pass this common-mode surge to some extent. I am very confident of this fact, if not the magnitudes involved. (I am 1500 miles away from t he compliance lab, and so have relied on a few measurements made by other p eople.) Whatever the magnitude is, it is locking some stuff up, that then n eeds a power cycle to return to normal.

dance clamping path from line to earth and your chassis/earth has not suffi cient low impedance or is connected to other parts of the circuit (this cre ates ground bounce internally, which reset microcontrollers, etc)

Yep. That is the problem.

d at the lack of specifications for power supplies for common-mode transmis sion (attenuation). I didn't see any information regarding standards for t esting such attenuation.

eed to increase your robustness, for surges with kA currents, it is hard to predict the robustness level, so you need to do your tricks in the lab

I agree robustness is hard to predict. And yes, of course, the design shoul d be robust.

I don't think I agree that no meaningful method of quantifying transmission of surge energy via common mode is possible. I think it is possible and d esirable, if non-trivial.

ation versus frequency. (Although details, regarding impedances, for exampl e, are not always obvious. Yet I have read that 50 ohms is an industry sta ndard, an "interesting" selection of impedance.) Is anyone aware of standa rds or common practice with regard to common-mode attenuation specification s and standards for testing it in DC power supplies?

and usually unipolar. That means they have DC content and EMI filters onl y help so much, as they pass DC. This may mean that it is helpful to inclu de a time domain based surge attenuation measurement, although that might b e tricky. But, the isolation XFMR in AC-DC supplies does block DC. Common mode RF attenuation seems easier to evaluate than common mode surge attenua tion. What do you think?

osed on the mains voltage, normally a 0.5us/20us pulse, and that's not DC

So if I do a Fourier Transform of a unipolar energy signal, you are saying there will be zero content at DC (w=0)?

You don't know the trigger level for fault, so what do you gain by knowing the frequency spectrum of the pulse?

:

rgy, and usually unipolar. That means they have DC content and EMI filters only help so much, as they pass DC. This may mean that it is helpful to i nclude a time domain based surge attenuation measurement, although that mig ht be tricky. But, the isolation XFMR in AC-DC supplies does block DC. Com mon mode RF attenuation seems easier to evaluate than common mode surge att enuation. What do you think?

rimposed on the mains voltage, normally a 0.5us/20us pulse, and that's not DC

ing there will be zero content at DC (w=0)?

Just to be clear, I am not saying it is a DC problem. Like I said, the tra nsformer blocks that anyway. I mean that the pulse has some "low" frequenc y content some "mid" frequency content, to say it qualitatively.

g the frequency spectrum of the pulse?

Well, I didn't say I was looking for that when I questioned your contention about the DC content of the surge. But since you ask, it does bring up a p oint. If I can know where the energy is in the frequency domain, and I know what the response of a filter + supply is, then I can know if a reasonable amount of energy will be rejected by the filter + supply. If a "new" filte r manages to reject some energy, and smear the remaining (passed) energy ou t over a longer time, then perhaps the peak can be suppressed below the so- called trigger level. But this is not guaranteed and the corner may need to be lower than practical.

It makes me wonder if a Fourier XF of the 1.2/50us voltage surge and 8/20us current surge have been done by someone. But I don't know that those test waveforms are the same as the actually imposed on the DUT.

There is a 48 VDC supply, and the output floats (by itself). The negative of this supply is eventually taken to "ground" at some location in the sys tem. This "local ground" is returned to another ground in the system, but the immunity testing is exposing the fact that low resistance|impedance is not zero resistance|impedance.

amps onto the main chassis.) There is no real coherent scheme of low-R/I ch assis ground in the main chassis itself, and a low-R/I bond is not availabl e to the load. It will never be available. The "system" I mentioned is the main chassis, its internal elements, and the load.

in chassis, and that's really not good. Why do you connect the load to cha sses and not to an individual/seperate wire?

Currents are not flowing from the load chassis to main chassis at some poin t of contact. The return current is flowing through wires and the outside of RF coax.

Actually, if I had a low (enough) impedance bond available chassis-to-chass is, or on wire(s), that would do it. But I don't have that available, and I won't have that available. Again, and if I understand the system people c orrectly, I will not ever have a definitive and low R/I bond to ground/refe rence available in the current class of the main system.

Incidentally, when bonding the negative of the 48 VDC supply to main chassi s ground as it exits the 48 VDC supply, that "fixes" the problem. They don 't want to do this because it decreases the "general usefulness" of the 48 VDC supply---other loads may want to use a positive ground, for example.

I was the first person to use this supply, and also go through compliance t esting.

I wasn't really asking how to fix the "ground bounce" problem, but am inter ested, of course. I was thinking that if someone knew a good way to charac terize the common mode transmission of supplies, then perhaps I could evalu ate a few and report to the system people regarding the supply they had cho sen. If it had "poor" attenuation, then maybe they would be interested in a better relative performer, as they do "source" it from an outside vendor. I have a feeling it "sucks," or could at least be improved, but I can't g et probes on it and I am not exactly sure of measurement method. I may be able to test it the same way as an EMI filter, for all its caveats. Hopefu lly, it will be above the LISN/CDN corner frequency. It is also conceivabl e that very aggressive low pass filtering and surge suppression on the AC i nput will "do it."

We have experienced some immunity issues during compliance testing. There is a 48 VDC supply, and the output floats (by itself). The negative of this supply is eventually taken to "ground" at some location in the system. This "local ground" is returned to another ground in the system, but the immunity testing is exposing the fact that low resistance|impedance is not zero resistance|impedance.

The issues are related to G-L and G-N generated surges. These are "common-mode," or at least pseudo-common-mode. The aforementioned 48 VDC supply seems to pass this common-mode surge to some extent. I am very confident of this fact, if not the magnitudes involved. (I am 1500 miles away from the compliance lab, and so have relied on a few measurements made by other people.) Whatever the magnitude is, it is locking some stuff up, that then needs a power cycle to return to normal.

In looking for information on this sort of thing, I was rather surprised at the lack of specifications for power supplies for common-mode transmission (attenuation). I didn't see any information regarding standards for testing such attenuation. As a counterpoint, EMI filter folks certainly publish common-mode attenuation versus frequency. (Although details, regarding impedances, for example, are not always obvious. Yet I have read that 50 ohms is an industry standard, an "interesting" selection of impedance.) Is anyone aware of standards or common practice with regard to common-mode attenuation specifications and standards for testing it in DC power supplies?

The common mode surges imposed during immunity testing are high energy, and usually unipolar. That means they have DC content and EMI filters only help so much, as they pass DC. This may mean that it is helpful to include a time domain based surge attenuation measurement, although that might be tricky. But, the isolation XFMR in AC-DC supplies does block DC. Common mode RF attenuation seems easier to evaluate than common mode surge attenuation. What do you think?

I have a VNA that goes down to 5 Hz, but I do not have a LISN or CDN. Of course, the LISN/CDN are useless that low in frequency. After all, they pass 50-60 Hz. My point is that I have a low-f analyzer available. I think I need a current probe VNA. lol

We have experienced some immunity issues during compliance testing. There is a 48 VDC supply, and the output floats (by itself). The negative of this supply is eventually taken to "ground" at some location in the system. This "local ground" is returned to another ground in the system, but the immunity testing is exposing the fact that low resistance|impedance is not zero resistance|impedance.

The issues are related to G-L and G-N generated surges. These are "common-mode," or at least pseudo-common-mode. The aforementioned 48 VDC supply seems to pass this common-mode surge to some extent. I am very confident of this fact, if not the magnitudes involved. (I am 1500 miles away from the compliance lab, and so have relied on a few measurements made by other people.) Whatever the magnitude is, it is locking some stuff up, that then needs a power cycle to return to normal.

In looking for information on this sort of thing, I was rather surprised at the lack of specifications for power supplies for common-mode transmission (attenuation). I didn't see any information regarding standards for testing such attenuation. As a counterpoint, EMI filter folks certainly publish common-mode attenuation versus frequency. (Although details, regarding impedances, for example, are not always obvious. Yet I have read that 50 ohms is an industry standard, an "interesting" selection of impedance.) Is anyone aware of standards or common practice with regard to common-mode attenuation specifications and standards for testing it in DC power supplies?

The common mode surges imposed during immunity testing are high energy, and usually unipolar. That means they have DC content and EMI filters only help so much, as they pass DC. This may mean that it is helpful to include a time domain based surge attenuation measurement, although that might be tricky. But, the isolation XFMR in AC-DC supplies does block DC. Common mode RF attenuation seems easier to evaluate than common mode surge attenuation. What do you think?

I have a VNA that goes down to 5 Hz, but I do not have a LISN or CDN. Of course, the LISN/CDN are useless that low in frequency. After all, they pass 50-60 Hz. My point is that I have a low-f analyzer available. I think I need a current probe VNA. lol

Simon, When the power supply has an isolated output, that is just DC isolation, not AC. Most CM noise is caused by the primary switcher, coupling noise thru the transformers parasitic capacity to the output ground. To negate this, a cap (10nF/2KV) is placed from the output ground to the input ground to return this noise to it's source. Do you have such a cap in place??

Cheers,

Harry

e

his

his

ity

de

at

n

ing

Is

nd

elp

ink

hru

a

Very good point. But, no.... Or at least, not that I know of. I am not in control of the 48 VDC supply, or the main chassis where it resides, or the wiring up to the "load board" I am responsible for. It is designed by a d ifferent group in my company. I asked the "system group" to try a common m ode filter on the output, that has the caps you mention and a common mode c hoke. They won't even bother because they don't want to retrofit the exist ing, and very expensive, main chassis' that are out in the field. I told t hem they have a lurking problem, but they shrug it off.

I don't know much about the supply other than what the very basic data shee t says. It is a module purchased from an outside vendor.

Just to reiterate, the problem is not noise generated by the supply. It is some (uncertain) level of common-mode transmission of a common-mode (L-G & N-G) surge applied on the AC input. It is part of the CE compliance testin g. The level of this ground referenced surge is a contrived requirement, b ut that fact does not help me because it is a letter-of-the-law matter, not a practical one.

[...some immunity issues during compliance testing... related to G-L and G- N generated surges ... it is locking some stuff up, that then needs a power cycle to return to normal.]

eet says. It is a module purchased from an outside vendor.

is some (uncertain) level of common-mode transmission of a common-mode (L-G & N-G) surge applied on the AC input. It is part of the CE compliance test ing. The level of this ground referenced surge is a contrived requirement, but that fact does not help me because it is a letter-of-the-law matter, n ot a practical one.

If the lockup doesn't happen at the power supply, the problem is in the (p resumably capacitive) coupling to the ground at the power supply, for which the 'return' current is from the OTHER ground, at your remote box. Find a way to make that return curr ent happen without making the reset happen. One approach: common mode choke on L and N, will increase the current risetime. That might be enough to eliminate ind uctive coupling to a sensitive conductor, by deflecting current to stray L+N versu s GND capacitance in the connecting wiring.

coupling to the ground at the power supply, for which the 'return' current is from the OTHER ground, at your remote box. Find a way to make that return current happen without making the reset happen. One approach: common mode choke on L and N, will increase the current risetime. That might be enough to eliminate inductive coupling to a sensitive conductor, by deflecting current to stray L+N versus GND capacitance in the connecting wiring.

CMCs can help, but the (as yet undefined) surge can saturate that in a fraction of a microsecond (or just plow right on over it). In general, it's preposterously difficult to filter any kind of common mode transients (ESD, EFT, surge) by ferrite alone.

As an example, I was once playing around with an EFT tester; at 2kV (5/50 ns pulse rise/fall time, 50 ohm system impedance), the pulse is essentially unstoppable by simply adding inductance. I had a 3" high-mu toroid that I put as many turns on as possible (8 or so); it softened the pulses, somewhat taking the edge off, but the majority was simply passed on through unaffected.

What's needed to handle this sort of stimulus is either shunting it to ground, or shielding *everything* in the system so that it washes over all cables and boxes without interfering with signals. (Same thing, but defining local ground inside the Faraday cage, and only bringing in signals that are matched to that ground via solid shields.)

Suppose your test system has a pretty generous amount of cable (>3m), maybe a few boxes, and the cables tee apart from the boxes in various places. Everything will be arranged on top of styrofoam, above a ground plane. So that the common mode impedance of the cables is in the vicinity of 100 ohms or so.

Pulse comes in from the machine, flies along the power cable (or whatever), hits the first box; a little reflects (the box has more capacitance / a lower impedance to ground), most transmits. A little later, the reflection sloshes around the source end and eventually comes back (it's something like an E&M tsunami). Wave hits a box with multiple cables, more transmission and reflection. More reflected waves from the tips of those cables. Etc. etc.

Stopping a pulse assumes you can provide much more than 100 ohms in series with the cable. And at lower frequencies (once those reflections start building up potential everywhere else), considerably higher than that still. I would guess my ferrite bead was upwards of 1000 ohms at some frequencies; unfortunately, those frequencies will be on the low side (~100kHz), due to the mu(f) characteristic of the material. So like I said, the pulses pass over with little change.

You might have some luck with a board mounted CMC with many turns on it, but if it's too small, it won't handle nearly enough flux to do the job. It will also have a lot of parasitic capacitance (depending on design) which makes it at least partially self-defeating. And leakage inductance, which may make things worse, or impossible (say by destroying high-speed signals).

If you have some impedance to ground (be it galvanic or only at RF with a Y-type cap), the performance of the CMC is vastly improved. With low enough shunting impedance, you can do a good job with the cable's impedance over space alone! But you may also invite ground loop (at RF or otherwise) by doing so. Which means more ferrite beads to break those loops, and so on. At least ground loop is a smaller signal thing (depending on the scope of conducted susceptibility, and the magnitude of radiated, of course).

Needless to say, dealing with slow surges (1.5/50 or 8/20us, or worse) will be entirely intractable; your only choice there is to ride it out (not terrible as it's generally slow), and shunt what you can (GDTs, MOVs or thyristor TVSs?).

Tim

--
Seven Transistor Labs, LLC 
Electrical Engineering Consultation and Contract Design 
Website: http://seventransistorlabs.com

Yes, because the V= L dI/dt thing only works with some large-ish I and that's probably intended to be small, here.

Interesting possibility: I was recalling an AC line test, and MOVs to ground weren't allowable from line or neutral. Maybe with a 48V floating supply, those ARE allowed?

With my system integrator hat on, chaining groundings is asking for trouble.

Everything should be connected directly to a single point, which may then be connected to a real grounding electrode.

In a big factory wide control system, you typically have three bars close to the mains service entry. One bar is the Technical Earth (TE) into which all DC supply returns, signal cable shields are connected with separate conductors to the TE bar. Then there is the Protective Earth (PE) bar which is connected to all metallic structures and mains EMC filter grounds, And of course there is the Neutral (N) for any three phase feed (which is also connected to the real grounding electrode in any TN-x grounding connection).

All three bars are then connected together with 10 cm long jumpers with thumb wide jumpers. The reason for these jumpers is that by removing these jumpers, it is possible to make sure that the network wirings are really separated from each other.

My request to board and subsystem manufacturers is that to keep the TE, PE and N wiring internally separated and bring them available to the external connector. The installer can then select TE, PE, N separated (preferred) or TE+PE and N in TN-S convention or TE+PE+N together in TN-C environment.

Connecting the EMC filter ground directly to the chassis ground makes it impossible to connect any DC grounds and signal return to the chassis, without completely polluting the TE grounding network. Using separate outside terminals makes it possible to use separate TE and PE separated or combined TE+PE at the equipment level.

Of course, by default there should be a TE+PE connection between the terminals and possibly also with N in case TN-C convention is used in some countries.

I've never heard of triple grounded systems. Where is that even allowed?

Tim

--
Seven Transistor Labs, LLC 
Electrical Engineering Consultation and Contract Design 
Website: http://seventransistorlabs.com

Why would that be illegal, after all, it ends up to a single point equipotential bonding

The separate TE ground is handy in audio/video systems as well as in large industrial sensor networks.

It makes some slight amount of sense with isolated power subsystems (like allowed in hospitals) or emergency-power subsystems. Nomenclature, in this case, is impenetrable, you have to sketch a diagram on a whiteboard and point at it a LOT.

Big-motor factories use isolated power because it allows deferred maintenance when something shorts.

Otherwise, if it's not a power-ground safety system, but a laboratory/measurement ground and shield system, that lies outside the concern of the electrical safety code.

You are referring to the IT convention

In an operating theater IT is often used so it will function even after a single phase to ground fault.

At least in Europe big motors are delta connected (either 400 V or 690 V) so there are no neutral/ground.

The three bar system that I was talking about is just an extension of the TN-S grounding