How to perform EMI filtering on a ready made SMPS?

[dzseki]

Hello,

Abstract:
There is an SMPS module for a projector. I built an improvement board for that projector which otherwise working fine except for that it picks up the noise radiated by the poorly designed SMPS, the radiated noise is easy to measure so I want to find a solution to reject it.

Details:
I circled down the source of the problem to the stand-by section of the Power Supply, it generates about 400mV pp RF noise on its output, at the fundamental switching frequency (~80kHz).
I attached here the schematic diagram of the power supply in question (sorry it is upside down...), the active PFC and mains input section can be found on pages 5-6, and the standby section is on page 1, the rest is not important at the moment.

So TX8 is the standby transformer, I tried the following things so far:
- Added copper foil shielding outside the transformer: no help
- Played with ground loops on PCB: no help
- Disabled the PFC circuit to make sure it is not making the interference: no help
- Added L-C filtering (220uH + 1000uF) to the standby output lines: this heled somewhat but the noise is still there it is just rejected somewhat.

I'm not an expert in SMPS designs but I know that for low radiation snubber circuits are essential to the rectifiers, now in this section seemingly there are 3 secondary windings, but only on one there is an R-C snubber. Although I tried to add an other before CR11 with 39 Ohm+500pF and later tried 100 Ohm+500pF but this did not make any difference.

I just want to add that however with the stock board configuration for this projector the problem is not very apparent but it isl there (had been checcked with other power supplies as well), with my modified board it gets more pronounced. Although I don't address this to my design flaw because my board does not even use any of these voltages, it only passes the 6,35V line (for 3mm trace length) but this is also surrounded by ground plane, also with added buffer capacitor, so I really can't do more from my side, although it is clear that the power supply itself radiates a lot of EM noise regardless...

Can anyone help how to move on?

[MagicSmoker]

The culprit is likely Q27 or D61. The network formed by D61, et al., is a peak clamp which does not dampen ringing and snappy turn-off of D61 can spray noise around like you wouldn't believe. You can try using a *slow* diode for D61 (like try a 1N4007 here; seriously) or you can put an RC damper across Q27. Maybe both.

[dzseki]

I changed D61 to BY133 recently and did not cahnge anything, do you have a suggestion for the Q27 RC damper, what should be the rule of thumb?

[MagicSmoker]

I just noticed that you are actually complaining about noise at the switching frequency, rather than higher frequency ringing (as I had ASSUMED from not reading your original post carefully...). If this is the case then a snubber is not going to help you. All you can do to fix that is either/both a grounded metal shield around the entire power supply or/and additional LC filtering on every output.

Otherwise, there are numerous resources on the web on designing snubbers, both for the switch in a flyback, as well as the rectifiers. For a really comprehensive explanation search for a document on snubbers by Rudy Severns (it's huge), while a much more focused/concise explanation of snubbers for flyback converters is Fairchild AN-4147.

[dzseki]

Well I think it is a bit of both... There is some noticeable ringing at the switching frequency. See the attached scope shot.
The power supply is in already in an aluminium box that is in a complete different compartment of the projector, the noise is radiated by the wires, since if I simply disconnect the +6.3V line off my board the noise goes away.

[MagicSmoker]

Well I think it is a bit of both... There is some noticeable ringing at the switching frequency. See the attached scope shot.
The power supply is in already in an aluminium box that is in a complete different compartment of the projector, the noise is radiated by the wires, since if I simply disconnect the +6.3V line off my board the noise goes away.

You are misinterpreting what the scope is showing you: that is definitely high frequency ringing which is repeating at the (much lower) switching frequency. Select a shorter timebase period to zoom in on the two spikes.

Generally speaking, ringing across the switch at the beginning of turn-off is from the lumped capacitance present at the switch drain and the primary side leakage inductance. Ringing at the end of conduction of all secondary diodes, while the switch is still off, is comprised of the same lumped capacitance and the primary magnetizing inductance. Damping the former is easy as it tends to be very high frequency; damping the latter is usually impractical as it tends to be very close to the switching frequency (maybe 4-8x higher) so losses in the damping network will be high.

Also note that you must use the ground spring that goes on right at the probe tip - and not the pigtail + alligator clip - to connect to the ground/common reference point when probing for noise like this.

[dzseki]

Thanks for the help, I really appreciate it. The power supply itself is constructed in an aluminium box that can be opened in to halves, one board on each side, the primary drive and PFC is one side the secondary is the other PCB. It is not very pleasing to work with, and is essentially only possible to test when the box is closed, then I only can rely on the output connector which is constructed off the PCB itself (like an old ISA plug for PCs). I may be able to run the stand by section and taking in-circuit measurements with using an other mains input section.

I indeed used aligator clip for grounding, because of the above mentioned problems, and also because the SB- output and their ground return pins are quite far from eachother. I also have a high impedance FET probe if that helps.

[T3sl4co1l]

Oh good, you have a location.  You can try filtering that.  Best case: two filters, one for each wire, grounded to the aluminum box.  Cutoff frequency well below the ringing frequency.

If the two wires are ringing in phase, you can use a common mode choke.  It may be enough to use a ferrite bead, but more likely you'll need to loop the wires around a core many times, and add some caps (to the aluminum box).

Tim

[dzseki]

A little advancement:
I put the PSU on the bench by making an alternate input rectifying and filtering, also added an isolation transofrem to the mix, so I can safely measure the primary side as well.
So, there is a lot of ringing on the Drain of the FET (Q27) with 833kHz frequency.
I found the attached snubber design guide pretty clear to understand and straightforward.
As a first step I should add RC snubbing to Q27. BUT calulacting with the hand written leakage inductance on the schematic of the transformer and with the measured ringing frequency would give ~53Ohm impedance for R, then the C calculation give about 3nF. These seem way too "strong" for the first sight. Is the leakage inductance probably off?

[dzseki]

With trial and error I mnaged to find a suitable RC filtering fo the primary side.
But as it seems the main problem is the missing RC snubbing from the rectifier diodes, and this is a though one. As it was suggested I measured at the puffer capacitor with minimal grounding length for the probe, no riplle at all. Measuring this with the aligator clip gives severe ringing. But as I approach towards the output the ringing is getting there no matter how I ground the probe, in fact the output + and - are phisically quite far from eachother on the output connector 

[dzseki]

Not that it is so relevant now, but I managed to solve this issue a few days later than the previous post I just did not care about updating so here we are:
Based on the known informations the calculations of the attached pdf suggested 53Ohm+4nF for R-C snubbing at the FET's Drain, this sounded fishy rightaway, so I put a guesstimated 470pF+2k7 combo instead. This rejected the oscillation greatly however at the output the improvement was subtle still.
Further examination revealed that the Gate drive signal is not so nice either, so by trial and error I replaced the original 47Ohm || reverse diode with 680 Ohm || 47Ohm + reverse diode.
By doing this the noise on the output droppped to it's half.
I also made a loop antenna from my scope probe's crocodile clip and examined the noise sources "touchless" further on and found four major sources: The FET itself, the transformer and the D54 and CR10.
The FET was mounted on a small heatsink but without isolation (= antenna), so I applied isolation between the Drain and the heatsink, and then tied the heatsink to the ground. Also put 220pF capacitors parallel with those diodes. By doing all these the result was satisfying and the goal was achieved.

Also I attach a scope shot that was done with the same settings as above.

[Someone]

Further examination revealed that the Gate drive signal is not so nice either, so by trial and error I replaced the original 47Ohm || reverse diode with 680 Ohm || 47Ohm + reverse diode.
By doing this the noise on the output droppped to it's half.

Watch out for increasing the dissipation in the FET after you have slowed down its transitions, a current probe would be convenient to check the SOA under different loads.

[dzseki]

Since it is a relativel low power supply there the switching currents are not too much. I have modified several power suplies since then, and none of them developed any problems.