Problems snubbing HF noise on flyback transistor

[Jdo300]

Hello,

I am working on the design of a flyback converter based on the LT3751 capacitor charging circuit. My design is similar to the 42 A cap charger example on pg. 25 of the LT3751 datasheet. In my case, the circuit runs on 12V and steps up the output to 430 V, but I am having some major noise issues with the MOSFET driving the primary of the transformer. The flyback powers some other high voltage circuitry and the noise from the FET during turn-off is radiating into the other parts of the design, both through the power and ground lines and through the air.

My first attack plan was to add an RC/RCD snubber across the primary of the transformer to snub the high frequency oscillations on the switch. But for some reason, none of the values I calculated for it, (or just plan experimented with) had any real impact on the waveform—in particular, the first high voltage spike, which is rising to as high as 200V. I have attached scope shots of the turn-off waveform, including a closeup of the first spike, which is around 12 ns wide. One challenge is that this waveform's spike is so narrow, and also the waveform seems to be rich in harmonics.

Has anyone run into problems like this before or seen a waveform such as this which seems to defy normal RC/RCD snubbers? I'm not sure what to try next to remove this noise.

- Jason O

 

[berkeman]

Can you upload the schematic and describe where you are putting the snubber? Can you post the winding stackup and winding information for the transformer? Do you have the flexibility to wind the transformer differently, or are you using a stock unit that you can't change? There is a trick in the winding stackup that can help to reduce that ringing (when used with a good snubber circuit).

 

[Jdo300]

Hello Berkeman,

Thank you for your response. I have attached a basic schematic of the flyback circuit showing the parameters of the power stage and position of the RCD snubber I've been playing with. The transformer itself was a custom-wound part from Sumida based on a recommended part number from the LT3751 datasheet (Sumida PS07-300) so I'm not sure how they wound it internally. The only difference between the part number listed above and the one we're using is that it was wound with higher voltage wire. All other parameters are the same.

 

[berkeman]

So, a few thoughts:

• There does not seem to be much in the way of high-frequency decoupling at your power input. The HF impedance of that 470uF electrolytic cap is probably not going to let the snubber circuit damp the high-frequency oscillation that your first 'scope pictures shows. Consider adding more HF decoupling at the input, and see if that helps the snubber to work better
• I didn't see the RC values that you are using for the snubber (I could have missed them). In a recent power supply we used 0.01uF and 499k. I was inclined to use smaller values of the resistor, but there is a tradeoff in efficiency loss if you have to absorb too much energy in the snubber resistor
• The input inductance of the primary seems a bit low for your switching frequency, but I guess if the IC manufacturer recommends that transformer design, it must be about right.
• I'd recommend that you ask the transformer house to show you the layer stackup for the transformer. Often they wind the primary 1/2 against the core and the other 1/2 as the outer layer outside of the secondary. They do this to reduce the leakage inductance Lk. But this puts 1/2 of the noisy primary on the outside of the transformer which can lead to more noise coupling and EMI. We recently were working with Wurth on a power supply design, and they re-wound our transformer with all of the primary as the innermost layer, then a grounded shield layer, then the secondary layer. This helped a lot in reducing our noise and EMI (along with improved decoupling).

Maybe give some of those ideas a try to see if they help. Let us know how it goes please.

 

[Tom.G]

Note that for RFI radiation, the source impedance is around ¼Ω. Since this is a rather poor match to the 377Ω of free space, it is likely magnetic field radiation that is causing external interference, if any.

Now on to the snubber.
Energy stored in transformer inductance at no load (use the Primary inductance):
J = ½ x L x I²
. = ½ x 2.5E-3 x 42²
. = ½ x 2.5E-3 x 1764
. = 1.25E-3 x 1764
J = 2.2Joules

Capacitor sizing for 100V spike at no load:
J = ½ x C x V²
2C = J / V²
2C = 2.2 / 1E4
2C = 220E-6
C = 110uF

For full load, use the leakage inductance which is about 1/8 the primary inductance. This gives a snubber capacitance of 14uF.

Edit
Since the above capacitor values are typically Electrolytics, you will have to parallel them with several high frequency caps. Considering the frequencies involved, probably a few different cap constructions will be needed. Check the impedance versus frequency curves to obtain sufficient capacitance at all needed frequencies.
End Edit

That full load approximation for C is for a resistive AC load directly on the secondary. With a filtered rectifier load, the numbers will be between the above values.

Note that with this flyback configuration, the Primary side voltage spike will be the output voltage divided by the transformer ratio. If the Primary spike is clamped below that value, there will be insufficient Secondary voltage to charge the output capacitor.

 

[jim hardy]

There's another inductance present. Did you minimize area of this loop. created by your interconnections?

 

[berkeman]

I have attached a basic schematic of the flyback circuit showing the parameters of the power stage and position of the RCD snubber I've been playing with. The transformer itself was a custom-wound part from Sumida based on a recommended part number from the LT3751 datasheet (Sumida PS07-300) so I'm not sure how they wound it internally. The only difference between the part number listed above and the one we're using is that it was wound with higher voltage wire. All other parameters are the same.

Your schematic shows a single ground shared between the primary and secondary. Is the transformer constructed as a Safety Barrier? The input voltage of 12V is human-safe SELV, but the output voltage is high enough it is not considered SELV and is subject to certain safety considerations.

One reason I ask is because in our power supplies, where we have AC Mains coming in and use transformers that provide a Safety Barrier so that the output voltages are SELV, we have found that minimizing the area of the input ground near the switching circuit helps to lower the radiated EMI and self-interference with the secondary circuits. Have you tried breaking the two grounds (primary and secondary)? You don't show any feedback for the output high voltage back to the switching circuit. Does this DC-DC converter circuit just run open loop?

 

[Jdo300]

Hi Berkeman,

Sorry for my delayed response. Thank you for getting back with me so quickly. Here's my feedback below.

There does not seem to be much in the way of high-frequency decoupling at your power input. The HF impedance of that 470uF electrolytic cap is probably not going to let the snubber circuit damp the high-frequency oscillation that your first 'scope pictures shows. Consider adding more HF decoupling at the input, and see if that helps the snubber to work better

This is a great idea! I'll be working with the circuit more today so I'll try it and see if it helps.

I didn't see the RC values that you are using for the snubber (I could have missed them). In a recent power supply we used 0.01uF and 499k. I was inclined to use smaller values of the resistor, but there is a tradeoff in efficiency loss if you have to absorb too much energy in the snubber resistor

Yes, we experimented with several different snubber values. Based on the calculated ringing frequency (we estimated a period of roughly 15.47 ns from the scope shot), we tried R values ranging from about 10Ω to 50Ω (we also tried a few higher values), and C values from 100 pF to 10 nF. It seemed that the initial spike was unaffected by any particular combination of RC values, though I did notice the other oscillations/harmonics reorganize a bit (no super significant change though). At the largest C values, I did see a larger low-frequency ripple develop, superimposed on the higher frequency ringing (which never seemed to go away), but the circuit also drew a lot more power too, so I didn't use values in that range for long.

The input inductance of the primary seems a bit low for your switching frequency, but I guess if the IC manufacturer recommends that transformer design, it must be about right.

Yeah, I thought so too. But it looks like the IC is designed to operate in the boundary mode between CCM and DCM, so as the loading on the transformer increases, the switching frequency also increases. For the most part, it seems to operate in the 200 kHz-ish range during charging, but the frequency does vary a bit. The other reason for the low inductance, I believe, is because of the high current draw needed to quickly charge the secondary (42 A peak).

I'd recommend that you ask the transformer house to show you the layer stackup for the transformer. Often they wind the primary 1/2 against the core and the other 1/2 as the outer layer outside of the secondary. They do this to reduce the leakage inductance Lk. But this puts 1/2 of the noisy primary on the outside of the transformer which can lead to more noise coupling and EMI. We recently were working with Wurth on a power supply design, and they re-wound our transformer with all of the primary as the innermost layer, then a grounded shield layer, then the secondary layer. This helped a lot in reducing our noise and EMI (along with improved decoupling).

This is a good question. I actually did speak with them a couple of days ago to ask more about how they wound the transformer, and the engineer told me he thought they did the split primary/secondary design to reduce capacitive coupling, as you mentioned, though, he didn't say whether the primaries were on the inside or outside of the layer stack-up. I'm currently waiting for them to send me a drawing showing how it is constructed, but we should be getting some more samples with the electrostatic shielding between the two layers.

Maybe give some of those ideas a try to see if they help. Let us know how it goes please.

Will do, thank you for all the suggestions!

- Jason O

 

[Jdo300]

Your schematic shows a single ground shared between the primary and secondary. Is the transformer constructed as a Safety Barrier? The input voltage of 12V is human-safe SELV, but the output voltage is high enough it is not considered SELV and is subject to certain safety considerations.

One reason I ask is because in our power supplies, where we have AC Mains coming in and use transformers that provide a Safety Barrier so that the output voltages are SELV, we have found that minimizing the area of the input ground near the switching circuit helps to lower the radiated EMI and self-interference with the secondary circuits. Have you tried breaking the two grounds (primary and secondary)? You don't show any feedback for the output high voltage back to the switching circuit. Does this DC-DC converter circuit just run open loop?

Ok, this is a really good question too and one that I've been considering a bit more lately. This boost converter is part of a battery operated HV power supply for charging and discharging a capacitive load through an H-Bridge, and also has an embedded processor and some analog circuitry implementing a HV CC limit function. One thing I'm wondering is if it is necessary to isolate the HV output/power plane from the analog and digital sections, or keep everything unisolated. It makes to total sense to isolate if it is powered directly from the mains, but I'm not sure what is customary in this case here.

 

[Jdo300]

@Tom.G,

Thank you for the feedback on the snubber design. I haven't tried capacitors anywhere close to that size. It seemed that when I used any values above about 10 nF, the input current to the circuit started to increase substantially. However, it's good to know how to calculate the values for full vs, no load. So far, my calculations have been based on the full-load leakage inductance. I'll take a second look at the no-load calculation (since this will be the case the majority of the time).

@jim hardy,

HI Jim. Yes you pointed out an important point, which I think is one of the flaws of my current layout. Though, I did have the circuit designed onto a PCB, the distance between the transformer and the controller and MOSFET is larger than it could have been, and I suspect that the di/dt and dV/dt from the input and output current loops could be a major source of the noise I'm seeing. A much more compact layout is definitely on the list of improvements for the next version.

 

[berkeman]

One thing I'm wondering is if it is necessary to isolate the HV output/power plane from the analog and digital sections, or keep everything unisolated. It makes to total sense to isolate if it is powered directly from the mains, but I'm not sure what is customary in this case here.

Separating the two grounds can help to isolate the output HV noise (leave a spot for a capacitor tie between the grounds as a contingency), but it can be required for safety reasons in some situations.

You mention that this whole system is battery powered, so that may mitigate the safety issue some, depending on what pieces of metal are user-accessible (if any). The 12V stuff is safe for a user to touch, but the HV stuff definitely isn't. And when the two grounds are tied, that could make it easier for a single fault in the device to couple HV back to the primary side.

Did you ask the transformer folks if the transformer is constructed as a double-insulated safety barrier? What metal pieces (switches, connectors, etc) are accessible to the user on this device?

 

[Tom.G]

Thank you for the feedback on the snubber design. I haven't tried capacitors anywhere close to that size. It seemed that when I used any values above about 10 nF, the input current to the circuit started to increase substantially. However, it's good to know how to calculate the values for full vs, no load. So far, my calculations have been based on the full-load leakage inductance. I'll take a second look at the no-load calculation (since this will be the case the majority of the time).

Now that the problem has had a chance to brew in my subconscious, here are some more details.

The 12nS spike you are seeing is probably from the distributed capacitance of the transformer. Using the 300nH leakage inductance, to resonate at 41.7MHz takes 49pF. A 100V Zener closely and directly wired across the primary may clamp the spike, easy enough to try. Watch out for power dissipation at different load conditions. A different transformer configuration may help... so may a board layout change. For instance a board pad under the transformer connected to one end of the primary may help. Try it on both primary leads (not at the same time! ).

The "conventional" (cheap) cure for inductive kickback is R and C in series directly across the inductor. In this case your R would have to be above 2 Ohms to get the needed 100V spike on the primary. Unfortunately a 110uF cap with that transformer resonates around 10kHz, or 50uS for a half cycle. So that wasn't as good an initial approach as I had hoped. To get the resonant frequency high enough to not interfere with the next pulse, the cap must be less than 500nF. That would also be the upper limit for the present snubber circuit. You could try a few ohms of resistance in series with the present snubber capacitor to limit the current increase at turn-on.

 

[berkeman]

Has anyone run into problems like this before or seen a waveform such as this

BTW, how did you probe this signal? If you used a conventional 'scope probe with ground lead and clip, some of that ringing can be in the 'scope probe inductance itself. For probing high-frequency waveforms like that with a 'scope, it's best to use a "Z-Lead" probe tip like in the picture below.

Also, some of that ringing can be due to common-mode to differential-mode conversion in the scope input. One way to tell how much this is adding to the ringing is to try a "ground minus ground" check, where you leave your conventional probe ground clip connected, and probe that same ground with the probe tip. If the common-mode spike is getting converted into what looks like a differential signal, you will still see much of it during this ground minus ground sanity check.

Z-lead accessory is the small gold-colored ground ring + tip just below the end of the probe in this picture. It slips over the bare end of the probe, with the ring part making contact with the ground ring of the probe tip, and projects a short, sharp ground tip to make contact with your circuit right next to the signal you want to probe. This can reduce the probe's ground loop inductance by a couple orders of magnitude. Of course you need to have a ground probe point next to your signal probe point on the PCB in order to use this type of probe...

http://cdn7.bigcommerce.com/s-9x745...s/105/276/5904RA-lg__28096.1433344387.jpg?c=2

EDIT -- Fixed a typo in one of the sentences...

 

[Tom.G]

Another trick with the 'scope probe is to use a Ground lead long enough to wrap around the probe several times, space the wraps over the length of the probe. Of course you still need to have a nearby Ground point.

 

[Jdo300]

Separating the two grounds can help to isolate the output HV noise (leave a spot for a capacitor tie between the grounds as a contingency), but it can be required for safety reasons in some situations.

You mention that this whole system is battery powered, so that may mitigate the safety issue some, depending on what pieces of metal are user-accessible (if any). The 12V stuff is safe for a user to touch, but the HV stuff definitely isn't. And when the two grounds are tied, that could make it easier for a single fault in the device to couple HV back to the primary side.

Did you ask the transformer folks if the transformer is constructed as a double-insulated safety barrier? What metal pieces (switches, connectors, etc) are accessible to the user on this device?

The ground separation question I am still pondering. From looking at the secondary vs, the primary side of the circuit, it actually seems like most of the noise is from the primary side. The noise on the secondary is coming from the diode junction capacitance, but, thankfully, I was able to suppress this with an RC snubber across the diode.

As for interfacing. This power supply doesn't have a direct UI. It will be designed into a larger application which has a separate external UI that will communicate with it through a wired-in header. And as for the transformer, that's a great question. I'm currently waiting for the manufacturer to get back with me. They should be sending me a datasheet with more details of how the transformer was wound.

Now that the problem has had a chance to brew in my subconscious, here are some more details.

The 12nS spike you are seeing is probably from the distributed capacitance of the transformer. Using the 300nH leakage inductance, to resonate at 41.7MHz takes 49pF. A 100V Zener closely and directly wired across the primary may clamp the spike, easy enough to try. Watch out for power dissipation at different load conditions. A different transformer configuration may help... so may a board layout change. For instance a board pad under the transformer connected to one end of the primary may help. Try it on both primary leads (not at the same time! ).

The "conventional" (cheap) cure for inductive kickback is R and C in series directly across the inductor. In this case your R would have to be above 2 Ohms to get the needed 100V spike on the primary. Unfortunately a 110uF cap with that transformer resonates around 10kHz, or 50uS for a half cycle. So that wasn't as good an initial approach as I had hoped. To get the resonant frequency high enough to not interfere with the next pulse, the cap must be less than 500nF. That would also be the upper limit for the present snubber circuit. You could try a few ohms of resistance in series with the present snubber capacitor to limit the current increase at turn-on.

Thanks for the tip on the Zener Tom. I am currently making a list of changes to make the the current PCB for the next version. So I will make sure to include pads for a HV zener across the primary for testing (don't have any 100V zeners to test with at the moment)

BTW, how did you probe this signal? If you used a conventional 'scope probe with ground lead and clip, some of that ringing can be in the 'scope probe inductance itself. For probing high-frequency waveforms like that with a 'scope, it's best to use a "Z-Lead" probe tip like in the picture below.

Also, some of that ringing can be due to common-mode to differential-mode conversion in the scope input. One way to tell how much this is adding to the ringing is to try a "ground minus ground" check, where you leave your conventional probe ground clip connected, and probe that same ground with the probe tip. If the common-mode spike is getting converted into what looks like a differential signal, you will still see much of it during this ground minus ground sanity check.

Z-lead accessory is the small gold-colored ground ring + tip just below the end of the probe in this picture. It slips over the bare end of the probe, with the ring part making contact with the ground ring of the probe tip, and projects a short, sharp ground tip to make contact with your circuit right next to the signal you want to probe. This can reduce the probe's ground loop inductance by a couple orders of magnitude. Of course you need to have a ground probe point next to your signal probe point on the PCB in order to use this type of probe...

http://cdn7.bigcommerce.com/s-9x745...s/105/276/5904RA-lg__28096.1433344387.jpg?c=2
View attachment 223565

EDIT -- Fixed a typo in one of the sentences...

This is a REALLY good point. I literally just found out a few days ago that I shouldn't have been using the normal grounding clip with the scope probes for HF measurements! Thankfully, the probes I have did come with the spring-tip contacts so I'll be using those for testing moving forward. However, unfortunately, the way I laid out the current PCB, there are no close ground points near the drain or primary connections where I can take good measurements. I'll make sure to include some test point pads (any recommendations for test points?), for the next version to make this easier.

Another trick with the 'scope probe is to use a Ground lead long enough to wrap around the probe several times, space the wraps over the length of the probe. Of course you still need to have a nearby Ground point.

This is a good idea here. Thankfully I do have the spring tip ground connections that berkeman showed in the picture above, though I don't have any nearby ground points to measure it with directly.

 

[berkeman]

I don't have any nearby ground points to measure it with directly.

How far away is the ground side of Rsense from the Drain of the MOSFET?

 

[Jdo300]

How far away is the ground side of Rsense from the Drain of the MOSFET?

The distance between the ground of the Rsense resistor and the MOSFET is about 0.8 inches. Thats roughly the length of the MOSFET package + the length of the resistor itself. This is too long for me to stretch or bend the ground probe. However, there is a bypass capacitor GND near the drain so I could spring for that as a close alternative...

 

[Tom.G]

(any recommendations for test points?)

Test points are very handy, especially in the early product versions. If periodic field calibrations or field repair, are needed they are also useful. They can always be left unpopulated if the above do not apply.

Here is an example of one you can hook a scope probe or alligator clip to. Product selector at:
https://www.digikey.com/en/supplier-centers/k/keystone-electronics/test-point-product-selector

Image from: https://www.digikey.com/product-detail/en/5009/5009K-ND/362671

 

[Jdo300]

Test points are very handy, especially in the early product versions. If periodic field calibrations or field repair, are needed they are also useful. They can always be left unpopulated if the above do not apply.

Here is an example of one you can hook a scope probe or alligator clip to. Product selector at:
https://www.digikey.com/en/supplier-centers/k/keystone-electronics/test-point-product-selector

View attachment 223683
Image from: https://www.digikey.com/product-detail/en/5009/5009K-ND/362671

Thanks for the info Tom. Yes, I regularly use these types of test points in my design. I was just wondering if there were any suggestions for high frequency test point measurements with the probe and spring clip combo. At the moment, I'm planning to provision for some SMT pads on the board near the spots that need measurements so I can just touch the probe to them. But if there are any other good approaches, I'm interest too :-)

 

[Svein]

Another thought: 12ns in an electrical lead where c=20cm/ns adds up to 2.4m or 1.2m back and forth. Is that the length of your oscilloscope lead?

 

[Jdo300]

Another thought: 12ns in an electrical lead where c=20cm/ns adds up to 2.4m or 1.2m back and forth. Is that the length of your oscilloscope lead?

Actually, that's a really good question. The scope probe I used for the original measurement had a lead length of 1.98 Meters from the end of the cable to the tip of the probe head. Not sure if this directly correlates?

 

[Tom.G]

Actually, that's a really good question. The scope probe I used for the original measurement had a lead length of 1.98 Meters from the end of the cable to the tip of the probe head. Not sure if this directly correlates?

It could. The OEM probes have a resistive center conductor (400Ω in the one I just checked) to suppress such things, I haven't found that in the one or two cheap replacements I've checked.

At one job I did, they couldn't figure out why some 'scope readings were not reproducible. They had saved money with some cheap imported probes. The probe designer saw the 10MΩ input impedance spec and dutifully put a 10MΩ resistor in the probe! Oops.

Cheers,
Tom