help about SMPS compenstion network.

how to get rid of high voltage spikes in smps

Ok VVV later today i am going to build a fixed freq SMPS for the controler board. I took some pictures that you can take a look at please.
I attached comments to the waveforms
I disconnected the snubber just to make sure that was not the cause.
It was not causeing the weird bump
Anyways take a look please and let me know what you think

This second board in the photo was made by a tech here and is really good as far as singals go. It is really clean and he said he has done this before because it made two really nice ground planes and you can see where they are seperated on the photo with the line under the transformer and opto isloater. The first board i made i will not picture because it was garbage and a waste of time with noise being the only thing i could measure!
Thanks again VVV

uc3845 pin3

this is the layout used for the cramer coil transformer
the ground plane is seperated by the green line. This has basically no noise as you can see from the waveforms.

I will soon have the layout done for an acutal manufactured board and i can send it to you to see what you think about it

This new board pictured works really good as far as singal clarity goes.
I hope to figure out what the strange pulse is and where it is coming from.

smps loop compensation

OK, I got it.

The ringing you are seeing on the primary is actually a normal occurrence for discontinuous mode flybacks. It comes about because at the end of the flyback time, or diode conduction time, the MOSFET's D-S capacitance is charged to a voltage higher than the input voltage and it now has to discharge to the input voltage. The voltage will not simply decrease to Vin, because Coss forms an LC resonant tank with the transformer inductance. To that you need to add the reflected capacitance of the secondary diode (which has just turned off, so it looks like a variable capacitor), reflected by the square of the turns ratio. (Even if Coss were zero, the diode's capacitance, reflected to the primary would still cause ringing, resonating with the transformer's inductance.) Anyway, that is why it happens: you have a charged capacitance (part of an LC tank) which needs to discharge to a lower voltage. The discharging process will cause the LC tank to resonate, creating the ringing you see.
The waveform will decay exponentially, but if there are low losses in the circuit, then the exponential time constant will be fairly long.

You can connect a snubber across the secondary diode (say a 10Ω/2W in series with a 2.2~10nF cap). That will reduce the frequency of the ringing but it will also make it decay faster, due to the resistor, which is a lossy component.

The ringing is usually harmless, but sometimes can cause EMI issues. The cure is usually provided by the snubber across the secondary diode. The ringing does not reduce the duty-cycle, since the ringing can only occur AFTER the diode has turned off and the diode turns off only AFTER the MOSFET has turned off. The gate drive waveform is clearly in sync with the primary waveform.
(The primary waveform is normally viewed across the transistor, because that allows you co connect the scope's ground to the primary ground, not Vcc. Just keep in mind for future measurements.)

I do not see any evidence of oscillation in your waveforms. That would manifest itself as a "jitter" on the drain waveform, which I do not see (perhaps it is there, but it's hard to tell from only two cycles; the second one might seem shorter, although to my eyes, the both look OK).
To make sure, you should measure the voltage at pin 1 of the chip, with respect to primary ground. The voltage should be a DC voltage that changes with the load, but if you keep the load constant it shoulbe pretty much DC. If you see something that looks like a sinewave (a few kHz) superimposed on it, then you know the loop oscillates.


Note that in SMPS you do not measure voltages with the probe tips and ground leads attached, as they will pick up lots of noise. Instead, you remove the probe's tip and attach a short ground adaptor. If you do not have one, just use some bus wire that you wind (4-5 turns) on the ground sleeve of the probe, close to the probe's inner tip (pin). Leave only about 0.5" of bus wire attached, which you bend parallel with the probe's tip. Touch the ground wire to GND and the probe's pin to the point you need to measure. When you measure the error amp output (pin 1) you touch the ground wire to pin 5 and the tip to pin 1. That way you have the minimum loop area to pick up noise. But be very careful, a slip of the hand and... the short is sure to happen (I hope you are using some sort of isolating transformer while you are working on this P/S). You can solder several of these ground "coils" at key points and only insert the probe when you need to measure.

offline smps high voltage 50 600vac

so your saying that this is normal in DCM flyback so should i have a snubber across secondary diode or not and there is no real way to remedy that waveform? Does that cause the circuits because of that LC tank as far as regulating?

Since this is normal then something else must be wrong since the circuit is not regulating to 12 at the moment.
I have ordered some more OnSemi uc3845 and they will be here late today. I have been getting nothing buttin strange results from the MicroSemi IPG controller.

I read over and over again what you want for the probes but i am still a little confused. So you want short distances so make a short GND connector for the probe and remove the tip so its just a point? Somethings a picture is worth a million words do you possibly have a picture of what your describing.

Today i am also construcing that other board for testing purposes.


As always i apperciate your help and time!

Added after 16 minutes:

Another side note as you were say that i should possibly snub the secondary diode.

I was looking at other flyback schematics and noticed that some of them snub on the primary like i did but also snubbed across the mosfet drain to ground. Are there any advantages disadvantages to having both a snubber on primary and across the mosfet?

Thanks

Added after 53 minutes:

Yeah we have a nice instruement from california instruments for isolation that is feeding the AC to the SMPS for testing.

flyback smps using uc3845 bipolar switch

That ringing is normal and it is not easy to get rid of. In new circuits, they clamp it with another switch (Maxim has such a device, for DC-DC). Anyway, this is not something you want to do. You are using the low-cost controller, you wil just live with the ringing, like millions of power supplies out there.
The ringing will not affect the regulation.

Yes, the probe ground lead has to be very short. I do not have a picture, but maybe I will find an appnote that describes better than me what is needed. Perhaps you received such an adaptor with your scope. It's just a plastic sleeve with one hole through which the pin of the probe can protrude and one short ground wire, made out of steel.
Anyway, to make something, imagine you are winding an air-core inductor and you are using the ground of the probe as the coil former. You wind 4-5 turns and you stop (the GND sleeve is not too long anyway). Then the lead that remains, bend it so that it is parallel with the probe tip. Now you have a probe with two "tips" that are parallel and about 0.5" apart. One of them is the GND, the other the probe tip.

Adding a snubber to the primary was done in the bipolar transistor era to prevent secondary breakdown of the transistor. I can explain more, but we are not concerned with that now. Your RCD clamp should be sufficient.

You are saying the controller is not regulating at all. In yesterday's pictures it looked like the chip was outputtting maximum duty-cycle (close to 50%). What was the input voltage to the circuit? If it was not the minimum input voltage, then I beleive it is not regulating, it just keeps the duty-cycle constant.

Again, please measure the voltage at pin 1 of the chip. At full load it should be about 3.5V and it should drop with lower output currents. If the voltage is higher, then that is causing the problem. The circuit will just stay at full duty-cycle. I would measure that voltage with the scope, like I described.

Did you have a chance to remove all those additional components and do a test at fixed-frequency, and using just a pullup resistor to Vref from pin 1? I would have been interested in seeing those results.

Can you post one more picture showing the D-S voltage of the power transistor, together with the voltage at the anode of the secondary diode? (the two grounds will be connected together, even if only through the scope).

I am not too fond of the ON Semi UC384x series, but I do not understand what you mean by strange results with the Microsemi.

ringing problem in smps

Ok first things first. I checked the control votlage at Pin 1 and it was 5V. I swept input AC from 60VAC to about 130 VAC with a .1A load and a 1A load and the control voltage stayed at 5V. So like you said its just banging along at 50% duty cycle because if i increase input AC the output Increases proportionally. I now think i need to check and make sure the tech hooked up all the wires correctly. Other than that do you have any ideas as to why its at a constant 5V. When i made the first prototype on vectorboard (it was very noisy and ill never build a SMPS on it again) but the control voltage did change.

Next i am working on as second board to do the fixed freqeuncy tests. I got about half way done today.

Thridly i got onsemi chips coming tommorrow so i can test those as well. I will get you a Drain to GND waveform tommorrow so that you can look at it.

Thanks for your comments. I really apperciate it!

Added after 51 minutes:

Alright first props to you VVV for all your help. The tech wired Vref pin 8 to the bottom of VR2 so the frequency was fixed and Pin 1 was not hooked up at all.
So i fixed that and walla! The power supply worked from 70VAC to 270VAC with out put of 0A to 3A at 12V beatuifully!!!!
Tommorrow i will be testing at higher input voltages (270VAC to 600VAC)
Thanks for you help and i will have more questions soon i promise!
One thing is that at low input line 70VAC and 3A output at 12V the line sags to about 11.6Volts. So thats within 3.3% of 12.
Can i increase Cout so that possibly the output line does not sag that much or is this just a fact of load and flybacks?
Thanks again for you input and intuition its been great corresponding with you!

OK, I like hearing the good news.

The fact that the converter goes out of regulation at 70V can be because "you run out of duty-cycle", so to speak. Basically, the duty-cycle gets to its maximum (about 50%) and it can no longer compensate the output. That is something that cannot really be fixed, unless you decrease somewhat the current sense resistor. At the same time the inductance of the transformer may need to be lowered a little. As you can see, that is no so easy.
But since the minimum specified is 90VAC, you have sufficient margin, so I would not worry and I would not do anything at this point.

The output capacitor does not help keep the output voltage constant. It is there to simply keep the ripple low, but the DC voltage is maintained by the loop and the controller.

Let me know how it goes at high line.

VVV i have downloaded and printed the Loop Compensation cook book from dixon. I need to obtain a beter grasp though on how to correctly graph it and what gain equations to use. Every example i see uses slighty different dc and ac gain equations and plots different load lines and input levels.
One example i plotted for high and low lines with high and low loads giving me 4 lines on the bode plot. Then your supposed to combined them to make a control line or soemthing. Anyways i am looking for a better grasp on loop compensation so maybe i can get a tigher regulation.
Ill post output waveforms for you tommorrow so you can have a look and give me suggestions on tweaks on stuff of that nature.
Thanks for you time and input!

Unfortunately, the gain will vary with line voltage and load current, so that is why you get more than one graph. The loop has to be compensated in such a way as to maintain stability (with reasonable margin) under all conditions of line and load.

The compensation will not affect the regulation, just the transient response (how fast the voltage goes back to the steady-state value after a sudden change in load current). The regulation (how stable the output voltage stays when the load current or line voltage change slowly) is only dependent on the DC gain of the loop. That means you can do your line/load regulation tests with a loop that is excessively slow. In cases where you have a stubborn loop, it is normal practice to slow it down until it becomes stable, do the rest of the tests and then deal with the loop.
Once you have completed the tests we can assess the loop compensation, and possibly try to improve the transient response.

The regulation should be good in your case, since the TL431 will provide sufficient gain (it always has for me) and you have no cross-regulation problems, since it's a single-output supply.

I have one more question for you: what type (manufacturer and series) of output caps did you use?

As far as capacitors i havent deicided on the ones for final board
i am using ones the techs had with a little M CE NHG +105C 3300uF 25V

so i have to adjust DC gain of the loop to have it regulate tigher over input ranges and outpuit loads?

Then after that i adjust the loop compensation in order to take care of transients?

I have been looking at loop compenesation and it seems as if ever one use a different set of equatiosn for DC gain, and somepoeple cacluate the esr zero of output cap and somepeople do not calculate the esr zero of the output cap.

I will get DS voltage weaveforms today and also output waveforms for you.

One last question when looking ad DS of the MOSFET when does it break down at when you have a 1000V mosfet? Does it break down at forward 1000V reverse 1000v or both?
THanks

OK, so you have 3300uF. I thought they were 330uF and I was really getting concerned. I will try to find the datasheet on the one you mentioned.

Yes, the DC gain is the one that maintains tight regulation. But before you start changing things, please check the following: is the TL431 (actually its K and the divider) sensing the true output voltage? I mean, the sensing should be at the actual output connections, otherwise you may believe the regulation is poor, because you lose voltages across traces. You want to use almost Kelvin connections for the error amp.

Yes, the loop compensation is usually the last thing to adjust. I do not know why some people do not calculate the ESR zero, perhaps because they rely on measurements, rather than calculations. The typical ESR zero is around 2kHz for Al electrolytics, but it will vary.

The breakdown voltage is for reverese bias.

Ok i tested it up to 500vac so the DC was about 700V. The mosfet was only a 1000V mosfet and at a 1 amp load for 30 secs something went bye bye. Thing is i cannot tell what part to replace ebcause the fuse blew before andthign was noticably damaged. I replaced the mosfet thinking i need to use higher voltage mosfet cause with the ring its easily hitting 1000V accross the mosfet. I replaced with a 1500V one and still when i apply only 40vac to the circuit the fuse blows right away. Any thoughts on what else went bad. I removed the controller chip tried again and the fuse blew imediatly. Can the EMI filter burn up? Is the diode bridge bad now? I am not sure what to replace. Any thoughts or answers VVV?
Thanks again

Added after 5 hours 44 minutes:

well i tested everything to find it was the mosfet , so i replaced it and the fuse still blows . That made me thought something else was wrong so i rechecked everything to find that the mosfet is causing the fuse to blow. How did the mosfet go bad i just replaced it with a new one and it doesn't work. How could a new mosfet already be bad. Right now i am lost at what to do

The chip most likely is gone, simply because when the MOSFET gets shorted the chip sees a high voltage at pin 3. The current sense resistor usually goes, too.
And you are right, the bridge can get shorted. The EMI filter should not fail short, so I don't think that is the problem.

I suppose the transformer was built properly and it did not fail. It is not easy to check it, if you do not have the right equipment. A shorted primary could potentially cause the MOSFET to fail, in spite of the current mode controller.

But I would not start with the transformer.
I would just check everything, replace the MOSFET and the UC3845, the sense resistor and try again.

Perhaps it is a good idea to first remove both the MOSFET and the chip and apply power, without any load connected. If the fuse blows, you need to find the defective component first.

If it is OK, carefully discharge the input caps with a resistor, install the chip and MOSFET and apply a low voltage from a current-limited lab supply. Watch the drain voltage with a scope and see if that provides any clues, especially about a shorted transformer, or even output diode. I do not see the type of the output diode, nor do I know the turns ratio of the transformer to tell if the diode may have failed at high line. Although this may be difficult, check the operating frequency, too, perhaps after the failure it became too low.
Note that the startup resistors may be too high for this test, so you may have to lower them. Alternatively, you can just apply 12V from a separate supply to the Vcc point.

Just curious, what voltage rating have the two input caps? (No, I do not think they failed, because you would have heard that, loud and clear. You would have smelled it, too.)

Just curious, what voltage rating have the two input caps? (No, I do not think they failed, because you would have heard that, loud and clear. You would have smelled it, too.)

Click to expand...

The input caps are 450V in series 100uF

I found the problem! Arcing! The mosfet tab was arcing to the heatsink through the hole in the mosfet because the heatsink is tied to CGND. I removed heatsink and used a tabbed heatsink! I also upped the insuloation rating.

This solved the fuse blowing problem and i am back to a working supply!
I just got up to 600VAC (850DC) with no problems. The output is circa 12 votls with various loads.

Next question is heat. What is max allowable heatsink tempature (heatsink is only thing i can measure atm). I was getting up to about 106 F on the heatsink. Is that allowable as i have never had to deal directly with heat issues.

Question 2 is can i leave the voltage sensing where it is and add and LC network to the output so it cleans up the output waveform.? If i leave the voltage sensing where it is and add a LC on the output it should nt mess with compensation network correct?
Or is it better to votlage sense after an LC filter and adjust compensation accrodingly?

Thanks for all you help!

Added after 1 minutes:

What sorta mosfets are do you know about the only one i can find over 1000V atm is a 1500V mosfet from STmicroelectronics and its about 3.31 per 1k. I am looking to possibly reduce that price and find a 1200v mosfet. if you have any suggestions [please let me know

Added after 8 minutes:

i just wanted to add that reason i am thinking about the LC filter is because the output is 12 volts with a load but its sorta noisy and has some like possibly transients!
If you would like i can get some output graphs for different input votlages and loads?
Let me know thanks!

OK, good.

The input caps could use a little more derating. I mean, at the maximum input voltage of 600VAC you get 425VDC across each capacitor. That does not allow for any increase in the input voltage whatsoever. and perfect balance of the resistors across them. Even without considering any input voltage variation I think they are marginal.
Normally, AL electrolytic caps are derated to about 70% their working voltage. That means that the maximum DC voltage across each should be only about 340-350V under normal conditions. Anyway, it's your call, especially that the only solution I see is using 3 caps in series, in case you cannot find something better.

The usual junction temperature of semiconductors is limited to about 100~110C at the maximum ambient. The junction temperature is measured with good approximation on the tab of the transistor, closest to the case. It is only a few degrees lower than the actual junction temperature. Attach a thermocouple to the tab and measure with teh P/S in a thermal chamber, at the maximum ambient. Make sure you allow the temperature to stabilize, you will probably have to run the test for about one hour and a half.
Also, since the transistor dissipation varies with load and line, you need to measure at the worst combination.
Don't forget the output diode and the transformer. Put a thermocouple on each. The transformer usually gets two, one for the core, the other for the winding.

The output filter only reduces the noise, the output spikes, and it is the usual feature on flyback supplies. It is usually placed outside the loop. The voltage drop across the inductor will affect somewhat the load regulation, but you should select an inductor with low enough resistance.

Unfortunately high-voltage MOSFETs are both espensive and hard to get. Sorry, I am not aware of other parts. Have you tried Infineon, IXYS, Philips, IR, and ON Semi? With the 1200V device you have to be really careful and make sure the RCD clamp really clamps. Don't forget testing under overload conditions. A short at the output will produce a huge leakage spike.

ok i have 330uF on the output being sensed then i have an LC Filter with a capacitance of 3300uF. The output looks good with a couple spikes.

I have some real boards being made ATM and when i get a chance i will let you see the layout. The boards will take about a week for turnaround. I am going to be doing some surge testing on friday with the proto board the techs made. I hear its really hard to pass sruge testing so i hope that my MOV's hold up and that the EMI filter knocks the clamp votlage down low enough to not destroy anything after the filter.

should i have a larger output cap? say 1000uF then same LC with a cap of 3300uF ? how should i do that? also i see lots of schematics with a lets say 100nF ceramic on the output. what is the purpose of that cap?

thanks again for all you help!

Usually the cap right after the diode is the large 3300uF cap and the one in the filter is much smaller, in the hundreds of uF range. Have you done a clculation of the cap's RMS current? I am almost certain that you are close to the rating, if not exceeding it for the 330uF cap. Does it get hot?

The ceramic caps help remove the spikes, because the spikes have a high-frequency content that is not removed well by the electrolytics, due to their high parasitic inductacne. Ceramics on the other hand have small inductance and therefore offer better rejection of spikes than the larger Al electrolytics, despite the much smaller capacitance.

i have not calculated my rms current for output cap. is that so i dont ruin cap because of ripple curernt? how do you calculate it corectly and arey ou talking about cap after diode or part of the LC filter.

I haev some various LC filters setup here and i dont see much of a difference from between the LC filters so what is a good design method for the LC filter.
On a side note some ps schematics have a common mode choke and cap on the out put. what is the pro cons of using L versus a choke.

For the output ceramic cap common value is about 100nF. Is this suitable? or is there some calculation behind the ceramic caps value?

thanks again

i just want to note that the problem was with the uc3845 from onsemi. It turns on at a higher voltage than the uc3845 for microsemi IPG. I am getting way better results from the IPG vs onsemi. I assume i could make the on semi 3845 turn on sooner if i lower the string of resistors charing that cap. Strange though casuse i would assume that both chips would turn on really close to each other. Also for some reason the IPG has better regulation. When i go from light load to heavy load on the IPG there is no dip in output votlage but when i use the 3845 from onsemi there is a dip to 0 then comes back up to 12.

Strange results. Also the uc3845 from onsemi seems to have a cleaner output, so i might not even need an LC output filter but i am baffled beacuse of its strange behaviour vs the IPG 3845

Added after 3 hours 15 minutes:

BANG wow that was fun! just blew up the output capactiance of 3300uF!!! it was rated at 25V. Heres what happened. I discovered that i had an bad uc3845 chip so i put in a new uc3845 from onsemi. Turns out this one worked fine at all input voltages for no load. at about 400VDC on the rail i went from no load to 1A load and the output went from 12 to 40 volts and BAM the output cap went up in smoke. Is the reason the output jumped because i have not properly found the correct compensation network yet?

Thanks any input is greatly apperciated!

i was just wondering how come in the app note they set Ve as 1 and did not discuss why they set it to 1 volt in the Adc gain equation.

another question i am using a tantalum cap in the loop compensation. Does it make a difference wether its ceramic or tantalum? I do have the cap correctly polarized.

Thanks