Mains Transient protection circuit looks very bad.....should we bin it?

[ocset]

it would be a way much better way to open your circuit using a high enough voltage rated MOSFET that is normally on. Just turn it off while the input voltage is too high.

Thanks, due to our circuit situation, I don’t believe we can do that. Our    LED driver is basically several successively switched linear regulators, and we need to limit the current through the leds when a transient comes, so we do really need the current clamp.
I mean, we could put the turn-OFF fet in there aswell as the current clamp, but  there simply is not room on the current PCB size.......so we have to have the current clamp and just make do.

[dmills]

I was not suggesting you USE the thing, I was giving it as an example of why you should not play in the low end, because down there volume and buying power really matters.

Seems to me that you thought you could play in the low end at an acceptable margin and are now finding that meeting both COG and product reliability targets is way harder then you thought.

The way to look at warranties on things like garden lights is to realise that your retail chain can just refund the sale as long as the failure rate is low enough, which takes the fact that the OEM has vanished after 5 years rather out of the picture (Not that many will still be able to find the purchase paperwork for something like that after 5 years anyway, I know I never can).

[dmills]

Thanks, due to our circuit situation, I don’t believe we can do that. Our    LED driver is basically several successively switched linear regulators, and we need to limit the current through the leds when a transient comes, so we do really need the current clamp.

Just switch off the current clamp mosfet completely when the voltage gets too high, not hard.

Regards, Dan.

[ocset]

Just switch off the current clamp mosfet completely when the voltage gets too high, not hard.

Thanks, however, that  can’t exactly be applied to our particular unusual  type of offline LED driver….we use PFC’d successively  switched linear regulators, each with their own LED string……when the mains is at the low part of the cycle, only the first led bank is switched in, …..at this point a serious overvoltage would be a DC bus voltage of > 70V……as the mains cycle goes higher, the value of an “overvoltage” gets higher, up to the mains peak, where the mains would  need to peak above 400V to represent an “overvoltage”.
I understand this is a very unusual way we are doing it, The chips we use are little known about.
But yes, there is simply not an  single  “overvoltage level” that we can use as the “overvoltage threshold”….it depends on which particular linear regulator is on at the time, and what part of the mains cycle it is at. This is why we absolutely  need the current clamp and cannot use an overvoltage triggered FET switch off.

....Uuuurrr, but Having said that, you are right in that an overvoltage switch off, set at say 413V or so, would be better than not having one. And it would have the advantage of protecting the Vdd of the LED driver which is 450V max. So i will look into it, but will struggle to find the room for it.

Seems to me that you thought you could play in the low end at an acceptable margin and are now finding that meeting both COG and product reliability targets is way harder then you thought.

If im honest, ive no idea if our own built product  is profitable  or not.......maybe we just survive because of the direct-from-china stuff which comes straight from china to the customer, middle-manned through us.

[dmills]

So just have the bus exceeding 400V or so cause a transistor to pull the current clamps gate voltage, job done.

Then you can probably remove that unreliable shunt thing and use the space saved for an input usable resistor. 

Regards, Dan.

[ocset]

Thanks very much DMills. Much appreciated.

The attached shows what i believe you meant. There will be space for this.
(schem in pdf and ltspice attached, plus ZR431 .asy file if needed.)

[dmills]

Why not just replace Q1 with the ZR431 directly, then you can eliminate R1,R2,R25,R24,R28,D20,C2, Q4 & Q1?

You have ~200uA down the R20 chain, and the ZR431 is good down to 50uA so that should work?

I would be a little concerned about the open loop gain, particularly in the 8us rise time case, it is rather unimpressive by the time you hit ~100kHz, could possibly improve that with a small cap across R16 or something? I cannot be bothered to work the numbers now. Possibly put Q1 back in but make it PNP and do the obvious. If going there I comment to your attentions such parts as the BC847BVN and related things.

I still say you want that 10 ohm fusable resistor in there ahead of the MOV well worth the cost.

Who do I make the invoice out to?

Regards, Dan.

[CopperCone]

i think the 'board of investors' is going to send a hit man after you if you mail anyone an invoice from the sound of things

[dmills]

Seems likely. 

Well that  is around half the parts from that input section gone, let me sleep on it and I will try to get us down the the fuse, MOV, Class X and pair of class Y plus CM choke that we all know is what the universe intended for a reliable mains input section.

Weird company setup, having investor types sticking fingers in Engineerings turf.

Regards, Dan.

[CopperCone]

yes, most people don't realize that transients of high frequency are attenuated by filter components, I see them as 'taking the edge off' for further components down stream.

Protecting capacitors with MOV or protecting MOV with capacitors though, which is better? I would at least put it behind the magnetic though, but the fuse in front of everything. I guess capacitors don't degrade like a MOV does?

[T3sl4co1l]

Mind that filters can saturate, depending on materials used.

CMCs saturate in common mode, but diff mode is air cored, so that helps.

Y caps are Y5P ceramic and saturate, but at fairly high voltages (low kV).

If both of these combine, you can end up with significantly higher peak common mode voltage than you were expecting.  This can be a good reason to add spark gaps or GDTs around the CMC, or to ground, as is sometimes seen.

X type film caps don't saturate, but do self-heal.  Which actually helps, since it's a current sink acting in parallel with the MOV -- yay, at least until the cap is self-healed in half, which does happen over time!

There are a number of reasons the IEC 61000-4-5 surge shape was chosen.  The mains network itself has a filtering effect, due to stray inductances (power lines are typically a higher characteristic impedance than the power they transfer, thus appear inductive), filter capacitances (typically for PFC), lightning arrestors (stacks of MOVs), transformers (a ball of wire, having parasitic capacitance and leakage inductance), and other loads (what with SMPSs being the norm, it's not uncommon for a few thousand uF to be clamping a household's mains line under transient conditions).  Lightning itself is not terrifically fast; it may be a spark discharge, but due to its length, the 100kA+ is delivered at a fairly relaxed rate (many microseconds).

Overall, especially to differential mode -- a typical mains input DC rectifier circuit has good filtering effect, which greatly helps the MOV.  The increased source impedance (mainly due to the CMC's LL and DCR) absorbs a lot of surge voltage, when the MOV is placed after it.

Tim

[ocset]

Thanks yes i should be paying up for this.
I've just realised what will be said about this.
The original use of a DPAK NFET (M3) to shunt the HV DC Bus was  said (not in this thread)  to be beneficial because it conducts the transient and takes the energy out of it, whereas now, since the rail is just being switched off when the transient comes, ...well this means that we are no longer de-energising the transient, and therefore it will not be de-energised, and therefore it will likely overvoltage the current clamp FET which gets switched off.

For this reason, i know they are not going to like me getting rid of the shunt NFET M3.

The space for the current clamp FET means it has to be DPAK, and the highest voltage we can find in low enough RDSon is 900V....as such, i think we will need the shunt FET (M3) back in to de-energise the transient.

..Yes, the let-thru voltage of the LSP05 MOV is 1100v,  and so we need to de-energise this let-thru.
LSP05 MOV
http://m.littelfuse.com/~/media/electronics/datasheets/varistors/littelfuse_varistor_lsp05_datasheet.pdf.pdf

(...I  confess that some of the units dont even   have the MOV fitted...to save cost.)

I tried this years ago with a zener based transistor shunt regulator, which is much simpler than your circuit, as a substitute for a high power TVS or zener.  I will not say that it cannot work when combined with a current limit as you have done but I am not convinced that it is the best way.

Thanks, do you think a power BJT with a zener fed base would make a better active-TVS than the M3 based circuit of the first post?..ie would it be more robust because of using a BJT?

[David Hess]

I tried this years ago with a zener based transistor shunt regulator, which is much simpler than your circuit, as a substitute for a high power TVS or zener.  I will not say that it cannot work when combined with a current limit as you have done but I am not convinced that it is the best way.

Thanks, do you think a power BJT with a zener fed base would make a better active-TVS than the M3 based circuit of the first post?..ie would it be more robust because of using a BJT?

It amounted to the same thing but was simpler and it could be implemented with a power MOSFET just as easily.  Using a 431 shunt regulator strikes me as unnecessarily complicated.

The only virtue to the circuit was that we had bipolar power transistors which were much larger than available zener diodes.  But failure was still possible if the power transistor's power ratings were exceeded and with a bipolar part, secondary breakdown becomes a major limitation.  The only reason it worked was because we could throw a large area of silicon at the problem.

[hermit]

I can understand the reluctance to share a proprietary design but has the OP ever reasonably demonstrated they are even chasing the right problem?  I know this stuff is spread over two or three  (or more?) threads so the information gets needlessly diffuse.   Seriously though how propriety can an LED driver on the low end market be?  I think the OP could get some excellent help here if they were more open.

[ocset]

secondary breakdown becomes a major limitation.

Thanks yes i appreciate that.
IGBTs  are known for their robustness, -certainly more so than mosfets, and i wonder if an IGBT with a zener fed gate woudl be the most robust way to make an active TVS.
I also remember that MOSFETs, whilst not suffering from secondary breakdown, do have paraisitic BJTs etc which suffer things similar to secondary breakdown.

[T3sl4co1l]

secondary breakdown becomes a major limitation.

Thanks yes i appreciate that.
IGBTs  are known for their robustness, -certainly more so than mosfets

If by "more" you mean "more important to protect".  IGBTs have, uh, ballpark 5 times or so higher current density, and have never been optimized for linear operation as far as I know.

Tim

[dmills]

Also a 10kHz IGBT is considered to be really fast.
Any sort of active thing is really going to be fighting against the miller effect, so you will need stiff gate drive.

I still say a few ohms or so of series R is what you want in there, and that that shunt clamp cannot do much for the energy in a transient because it is in series with a current clamp which limits the shunts ability to reduce the line impedance seen by a transient (In fact once the current clamp actually clamps the line impedance becomes very high which makes the over voltage actually worse at the input to the circuit). 

Regards, Dan.

[ocset]

Thanks,
We are wondering of the attached Mains Transient absorber would be   effective? 
(its to go ahead of the product, in the mains line)
Its just basically a 1500V, 1.8uF  film capacitor in the Mains line leading up to the product.
185PSB152K6R film capacitor, 1500V, 1.8uF
http://www.illinoiscapacitor.com/pdf/seriesDocuments/PSB_RSB%20series.pdf


We were thinking that if we used this, then it gives some absorbtion of a transient, and then we can do DMills kindly given  idea of switching off the current clamp FET when the voltage goes above aprox 430V.
 

[splin]

If you don't have enough space for a discrete resistor, how about a PCB trace resistor? 5 ohms would require a .127mm (5mil) trace on 1oz copper and 18mm x 18mm board area and should be good for 500mA. Not sure what it would do for EMI though.

You could make it into an 13uH inductor to slow things down a little too.

[T3sl4co1l]

Surge is tested with a 2 ohm source impedance. What's the 4.7 ohm resistor for?  It will have a small effect, but between the series resistance and the small capacitor, it will be maybe 10%, an insignificant reduction.

Consider that the surge tester (the real thing, not your toy circuit) uses 20uF charged to 1.5kV or more.  To get that under 800V, you need more like 50uF or more.

Commercial surge protectors of this sort use electrolytic capacitors around 470uF, and a large series inductor (100uH or so, air core).

Tim