Radiated vs conducted noise with flyback PS module.
in [Design]
|
From: dbvanhorn on 14 Apr 2010 14:17 Sanity check please: :) I had some fun recently with a design that used a small isolated flyback module, with large EMI issues. The device in question will pass conducted and radiated emissions at class B, if it is operated by itself with a load resistor, but when placed in a real board, emits all over the place, conducted and radiated, up through the 4000th harmonic (!) of the switching frequency. I suggested to the vendor that they slow down the turn-on and turn-off of the FET if possible, re-select core material, and apply an inter- winding shield connected to the high side of the primary, but they don't like those options. They wanted to work with snubber values on primary and secondary, and on the rectifier diode, and they did try some shielding options, but nothing seemed to help much when the supply was attached to our system. Boiling down a bunch of testing and experimenting, the supply has two capacitors in series, connected from the high side of the primary to the high side of the secondary. Adjusting these values seems to help a fair bit, but only in specific ranges, and of course this has implications for stability which limit the values that are usable. One pair of observations were particularly interesting: When we put the supply in the system with the output leads disconnected and loaded with a resistor, everything was fine. When we attached ONLY the ground lead to the system, conducted and radiated emissions went WAY up. Also, attaching cables pushed up the conducted emissions. I came up with a model for the problem that seems to work, considering the supply as a two-terminal noise source, with one terminal connected to the AC line cord, and the other connected to the rest of the system. I'm theorizing that the drain of the fet is pushing energy capacitavely through the transformer. The PCB, enclosure, and cables on the output end act as antennas at high frequencies, and some value of series capacitance to earth ground at lower frequencies. Adding any metal mass to the output end of the supply would then push up the conducted and radiated emissions, by lowering the impedance between the output terminals and earth ground. Applying this idea, I suppressed the radiated emissions with a pair of 600 ohm ferrites in series with both output leads (one ferrite each), and the conducted by adding capacitance from the input terminals to the output using UL rated caps. Radiated emissions went WAY down, well below class B limits when the ferrites were applied, but of course they don't do much at <1MHz. Conducted emissions took a similar stomping when the "end run" caps were applied, but I didn't have enough time left to optimize the values. So, it seems to me that my model is valid, since the predictions it makes are testable, and they seem to be correct. What else am I missing here?
From: John Larkin on 14 Apr 2010 15:25 On Wed, 14 Apr 2010 11:17:57 -0700 (PDT), dbvanhorn <microbrix(a)gmail.com> wrote: > >Sanity check please: :) > > >I had some fun recently with a design that used a small isolated >flyback module, with large EMI issues. > >The device in question will pass conducted and radiated emissions at >class B, if it is operated by itself with a load resistor, but when >placed in a real board, emits all over the place, conducted and >radiated, up through the 4000th harmonic (!) of the switching >frequency. > >I suggested to the vendor that they slow down the turn-on and turn-off >of the FET if possible, re-select core material, and apply an inter- >winding shield connected to the high side of the primary, but they >don't like those options. They wanted to work with snubber values on >primary and secondary, and on the rectifier diode, and they did try >some shielding options, but nothing seemed to help much when the >supply was attached to our system. > >Boiling down a bunch of testing and experimenting, the supply has two >capacitors in series, connected from the high side of the primary to >the high side of the secondary. Adjusting these values seems to help >a fair bit, but only in specific ranges, and of course this has >implications for stability which limit the values that are usable. > >One pair of observations were particularly interesting: When we put >the supply in the system with the output leads disconnected and loaded >with a resistor, everything was fine. When we attached ONLY the >ground lead to the system, conducted and radiated emissions went WAY >up. Also, attaching cables pushed up the conducted emissions. > >I came up with a model for the problem that seems to work, considering >the supply as a two-terminal noise source, with one terminal connected >to the AC line cord, and the other connected to the rest of the >system. >I'm theorizing that the drain of the fet is pushing energy >capacitavely through the transformer. The PCB, enclosure, and cables >on the output end act as antennas at high frequencies, and some value >of series capacitance to earth ground at lower frequencies. Adding >any metal mass to the output end of the supply would then push up the >conducted and radiated emissions, by lowering the impedance between >the output terminals and earth ground. > > >Applying this idea, I suppressed the radiated emissions with a pair of >600 ohm ferrites in series with both output leads (one ferrite each), >and the conducted by adding capacitance from the input terminals to >the output using UL rated caps. Radiated emissions went WAY down, well >below class B limits when the ferrites were applied, but of course >they don't do much at <1MHz. Conducted emissions took a similar >stomping when the "end run" caps were applied, but I didn't have >enough time left to optimize the values. > >So, it seems to me that my model is valid, since the predictions it >makes are testable, and they seem to be correct. What else am I >missing here? > There is one type of component that excels at producing insane harmonics like the 4000th: a step-recovery diode. Some fast-recovery rectifiers (not schottkies) can act like SRDs. And some synchronous switchers (like the LM3102) can snap their own substrate diode and blast RF into everything nearby. We've had opamps fail (huge offsets) just being on the same board as an LM3102, with no obvious connection. Here's the 3102 waveform: ftp://jjlarkin.lmi.net/SwitcherRise.JPG The risetime is likely that of the 500 MHz scope. Poke around with a fast (500 MHz at least) scope, waving a fet probe around as an antenna. Look for very fast spikes. John
From: Nico Coesel on 14 Apr 2010 15:46 dbvanhorn <microbrix(a)gmail.com> wrote: > >Sanity check please: :) > > >I had some fun recently with a design that used a small isolated >flyback module, with large EMI issues. > >The device in question will pass conducted and radiated emissions at >class B, if it is operated by itself with a load resistor, but when >placed in a real board, emits all over the place, conducted and >radiated, up through the 4000th harmonic (!) of the switching >frequency. > >I suggested to the vendor that they slow down the turn-on and turn-off >of the FET if possible, re-select core material, and apply an inter- >winding shield connected to the high side of the primary, but they > In some cases a 10 Ohm series resistor with the boost capacitor will also help. Its not always the diode that causes the problem. -- Failure does not prove something is impossible, failure simply indicates you are not using the right tools... nico(a)nctdevpuntnl (punt=.) --------------------------------------------------------------
From: Wimpie on 14 Apr 2010 16:19 On 14 abr, 20:17, dbvanhorn <microb...(a)gmail.com> wrote: > Sanity check please: :) > > I had some fun recently with a design that used a small isolated > flyback module, with large EMI issues. > > The device in question will pass conducted and radiated emissions at > class B, if it is operated by itself with a load resistor, but when > placed in a real board, emits all over the place, conducted and > radiated, up through the 4000th harmonic (!) of the switching > frequency. > > I suggested to the vendor that they slow down the turn-on and turn-off > of the FET if possible, re-select core material, and apply an inter- > winding shield connected to the high side of the primary, but they > don't like those options. They wanted to work with snubber values on > primary and secondary, and on the rectifier diode, and they did try > some shielding options, but nothing seemed to help much when the > supply was attached to our system. > > Boiling down a bunch of testing and experimenting, the supply has two > capacitors in series, connected from the high side of the primary to > the high side of the secondary. Adjusting these values seems to help > a fair bit, but only in specific ranges, and of course this has > implications for stability which limit the values that are usable. > > One pair of observations were particularly interesting: When we put > the supply in the system with the output leads disconnected and loaded > with a resistor, everything was fine. When we attached ONLY the > ground lead to the system, conducted and radiated emissions went WAY > up. Also, attaching cables pushed up the conducted emissions. > > I came up with a model for the problem that seems to work, considering > the supply as a two-terminal noise source, with one terminal connected > to the AC line cord, and the other connected to the rest of the > system. > I'm theorizing that the drain of the fet is pushing energy > capacitavely through the transformer. The PCB, enclosure, and cables > on the output end act as antennas at high frequencies, and some value > of series capacitance to earth ground at lower frequencies. Adding > any metal mass to the output end of the supply would then push up the > conducted and radiated emissions, by lowering the impedance between > the output terminals and earth ground. > > Applying this idea, I suppressed the radiated emissions with a pair of > 600 ohm ferrites in series with both output leads (one ferrite each), > and the conducted by adding capacitance from the input terminals to > the output using UL rated caps. Radiated emissions went WAY down, well > below class B limits when the ferrites were applied, but of course > they don't do much at <1MHz. Conducted emissions took a similar > stomping when the "end run" caps were applied, but I didn't have > enough time left to optimize the values. > > So, it seems to me that my model is valid, since the predictions it > makes are testable, and they seem to be correct. What else am I > missing here? Hello, I think you are right, the main problem is the capacitance between the primary and the secondary (as there is no ground shield in between). It is like there is a source in between. The result is a noise source between the complete primary circuit and secondary circuit. As you also have problems below 1 MHz, that part will not be solved with other rectifiers. As at low frequency, the output impedance of the noise source is rather high (capacitive coupling), standard EMI ferrites will not have sufficient impedance at low frequency Probably you need filters in the low voltage output and mains input with common ground (but you will probably not like this as this may require safety grounding and it is probably more expensive then an internally shielded transformer). The same thing also happens with some isolated DC-DC converters. When only very short wires are on the secondary DC-side, everything is OK, but with a real circuit, lots of noise is emitted. Good luck with keeping the noise inside, Wim PA3DJS www.tetech.nl remove abc first in case of using PM.
From: dbvanhorn on 14 Apr 2010 18:44
> Some fast-recovery rectifiers (not schottkies) can act like SRDs. And > some synchronous switchers (like the LM3102) can snap their own > substrate diode and blast RF into everything nearby. We've had opamps > fail (huge offsets) just being on the same board as an LM3102, with no > obvious connection. Makes sense.. They aren't being too receptive to any significant changes to the internals. The fet they are using is promoted for it's insanely fast on/off times, which I'm sure is helping their thermal issues, but it's sure not helping EMI I don't know the output diode specifically, they are not forthcoming with component data. They do have a series RC network across the output diode, which they've adjusted a bit. Maybe it's not a schottky, I don't know for sure. Agreed on the pix, the scope may be the dominant factor in that shape. :) Looks almost like a pulsed laser driver, except for the overshoot. |