Forward SMPS - unexpected behaviour

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
Can anyone voice an opinion on this:

I'm building a single-switch forward SMPS, 50W, 100KHz and although I've got it working, I'm now trying to optimize the design for lower heat output.

Strange thing though:

If I run it with no inductor in the output ( just a diode and capacitor filter, like flybacks), the MOSFET runs cool, while providing full power (18V into a 7 ohm dummy load). The secondary side rectifier (MUR1560) and filter capacitor run a bit hot, but manageable. I expect that, because of high ripple current.
However, when I insert an inductor between the secondary rectifiers and bulk cap, the MOSFET switch gets very hot. (secondary side components run cooler, since less ripple current). My calculated inductor value was 12 uH. I've tried inductor values from 3uH to 68uH, and the higher the inductance, the hotter the MOSFET gets. The only thing the MOSFET seems to like is no inductor at all.
Plus the free-wheeling diode doesn't ever get warm (with the inductor present), like it's not conducting any current. With or without it, there's no difference in output power or waveform.
Forward topologies are supposed to use a series inductor on the output, but the MOSFET seems to be rejecting it. Given the choice, I'd rather keep the MOSFET happy, but isn't this an odd behaviour for a forward SMPS?

Andy
 
I've chacked the free-wheeling diode, and it's OK. Ive repeatedly interchanged it with the forward diode just to make sure, and every time, the diode in the forward position gets warm, the shunt stays cold. If the inductor and capcitor are correctly sized, these two diodes should conduct equal loads.

But in any case, adding a series inductor only reduces rms current through the secondary, not increases it. No reason for the MOSFET to get hotter.

Trouble with SMPS's is that, unlike a 60Hz transformer, they're not really isolating. Anything you do on the secondary, gets reflected to the primary, sometimes in a big way.

Sometimes controller behaviour such as leading-edge-blanking, and other intrinsic features, can take decisions based on waveforms reflected into the primary, and produce unexpected behaviour.

This is particularly the case with highly integrated "solutions" like "Top-switch", "Tiny-switch", "FPS-Fairchild Power-Switch", etc, etc, which take too many decisions (some of them questionable) without your knowledge.

This controller is not one of those, though, it's a Motorola NCP1200.

I'm scanning schematic, will post shortly

Andy
 
Here's the schematic of the forward converter.

An externally hosted image should be here but it was not working when we last tested it.


I've tried reversing the polarity of the transformer secondary with the following results:
MOSFET much hotter, output power reduced to about a quarter of previous, waveform at switching node very chaotic, unstable and frequency dropped from 100KHz (fixed frequecy controller) to 50 or 40KHz. Doesn't seem the converter likes the reversed secondary.

Andy
 
Vector,

I think your duty cycle is going over 50%, this is not allowing enough time for your reset winding to clear the core. When you saturate the core current in the switchFET sky rocket.


Limit you output your duty cycle to 50%... just for example

165 * (8/45) * 50% = 14.66

Just to test... change your output sense divider to yield 12 Volts. This should get your duty cycle under 50%. Then see if the thing behaves like a forward convertor.

If it does... then thats it. Then rework your secondary turns including voltage losses in the output diodes...

Ns = [(Vo + Vdoide) * NP] / [DutyCycle *(Vin - Vfet - Vsense)]

Use a maximum value for Vin to arrive at a minimum for Ns. Use a maximum value of DutyCycle at say 0.45 to assure reset time.

It looks like jacking your secondary turns to 11 or 12 will do the trick.

Let me know if this works... haven't done switchers for awhile... little rusty obviously.


:D
 
Well, I'm not a forward kinda guy but I'll try to give some helpfull comments... What you just decribed with reversing the windings sounds like it was correct originaly. The freq change was probably skip cycle mode taking over...
Why are you not using a strait flyback (easier) and if you do you should go register at ON Semi and download the design tool "ON Power Designer" it works pretty good... I don't remeber if it simulates a foward topology or not...
First... don't you need a snubber on that primary??? That could be a problem... Without one destruction is eminant...
It sounds like your inductor in the output is blocking most of your output current... speaking of output current, the 2.5 Amps or so D.C. you are expecting is probably generating 6 or more amps peak (probably more) wich can't make it to that cap very well through the L... And, I don't know of any 500uF cap that would servive that kind of ripple current... Your output ripple Voltage probably looks terrible if it is loaded that much??? You probably need 3x 1500uFs on there...
Since the L is blocking the energy transfer to the output, the switch is trying to operate in continuos mode and that will generate a bunch of heat... Also killing effeciency if you care...
You are using a fast/ultraFast diode in the output, I would suggest a shotkey(spelling?) of high enough voltage because it will be hot when it works. I usually play with lower Voltages but there is something like an MBRS4525 or something and that series are real low Vf (less Heat)...
Sorry if my comments may only apply to flybacks "I'm not a forward kinda guy".
If that did not help I'll try again latter.
Happy Holidays:D :D :D
 
Umm... I assumed that you understood the basis of a forward converter. The overheating problem is clearly due to transformer saturation. Your control IC has a maximum duty cycle of 80% and it's fixed internally, it can't be adjusted externally, so your frewheeling winding should theoretically have 1/4th the turn count of the main primary winding (not practical due to excessive high voltage buildup at MOSFET drain).

In other words: In any forward converter the maximum duty cycle must be limited according to the turns ratio between the primary winding and the freewheeling winding. For a 1:1 ratio 50% duty cycle is the absolute maximum. This duty cycle limitation should be also taken into account when selecting a proper turn count for each secondary winding.

max_DC=n_pri/(n_frewheel+n_pri)

This is 50% for 45T+45T


n_sec=Vout*n_pri/(max_DC*min_Vin)

This is 17T sec. for 45T pri., 50% DC max. , 19V out (18 + diode) and 100V min. input

You may try a 1000V switching device and a 11T frewheeling winding if you dont feel like finding a new control IC. That would allow for 80% duty cycle.

Note that the circuit will be asked to work at the maximum allowed duty cycle during brief periods of time, both during startup and during load transients, so inherent duty cycle limitation is mandatory.
 
A Primary to Reset winding of ratio is 1 to 1 is popular because it allows for a bifilar winding, primary and reset wound together, which minimizes leakage inductance and also reduces spikes on the drain.

The goal is not change the reset turns to accomodate the duty cycle. The goal is to correct the secondary turns to reduce the duty cycle.

Schottky diodes on the output would be a good idea to reduce waste heat.

The inductor size is balancing act. The larger the inductor the less ripple current you have in the output caps. But the power can be slow to respond to changes in load.

Eva makes a very good point, you should also calculate at low Vin to be sure you have enough windings on the secondary... 11 or 12 is minimum maybe 15 or 16 would be safer... you have to choose your value for Vin (low).

:D
 
But duty cycle limitation is still mandatory because the control circuit will be asked for maximum duty cycle during startup, during load transients, and also during low line conditions. No matter how much secondary turns are employed, transformer saturation and possible latch-up or oscillation may still occur if duty cycle is not properly limited. I Just wanted to make that point clear :)
 
A few comments - Eva is right about the possibility of core saturation., especially since you are using a simple 1:1 reset winding. Also, the NCP1200 has an inherently high resistance gate driver structure, if you take a look at the spec sheet. The chip is really optimized for very low power flyback applications. The series resistor and diode you have on the gate of your MOSFET are probably not needed.

Since you are dealing with only 50W output at relatively high output voltage, is there any reason not to use a flyback? This will get around a lot of the problems you are facing, at the expense of having a couple of output capacitors to handle the ripple current.

If you stick with the forward, you will need a more flexible reset scheme to accomodate duty cycle > 50%, or a different controller. At the 50W level, something like an RCD snubber/reset network would work better than a 1:1 reset winding. A controller that has a sliding duty cycle limit vs. input voltage would also be a good idea. Otherwise, you could use one of the old indusry standard Unitrode 384X series that has a 50% duty cycle limit. These chips require a startup resistor, but designers have dealt with this for years.
 
Thanks
poobah, eva, wrenchone

Ouch! that's true, it's an 80% DC controller!
It was stupid of me not to have looked closely at the spec sheet.

I substituted this chip for an NCP1216, which was unavailable, and assumed everything was the same.

When the switching waveforms never went past about 50%, I found confirmation that it must be the controller limiting the DC.

It was actually the transformer saturating at maybe 55% and not letting the DC go any higher.

Thanks again
Andy
 
I do have room for more turns on sec.

I'm actually recycling an existing transformer from a dead SMPS.

It was a push-pull, so that's why I've got 1:1 on the primary.

I wouldn't want to mod the primaries 'cuse they're buried deep and I like the trafo the way it is. But I'll certainly be able to add turns to the secondary, since there is about 1mm of space still remaining between coil and core.

This is only a temporary fix, since it doesn't address the duty cycle surge mentioned by Eva at startup ( I hadn't thought about that ). I'll wait a couple of weeks to get the NCP1216 that's backordered, and plug that in when it comes.


Andy
 
You'll still need to increase your Ns turns for keeps... use EVA's equation Mine had small mistake... or did it??


If your power supply will have relatively constant load & you have the DC limit chip you could easily set your Ns turns for a DC of 40% at low Vin. This wont push your Ip up too much and the supply will just come up. and react to load changes, a little slower.

Cool!
 
Yes, I'll increase n(s) permanently.

The big pitfall was having trimming potentiometers on both the voltage and current sense loops. I had intended to experiment with different settings for different MOSFETS, different output powers etc.
But the adjustment ranges of these pots go into some unsafe territoty. There is a reason no PS are made with w-i-d-e adjustment ranges on the sense dividers.

Valuable lesson learned: if I'm going to experiment, I'd better not get lazy about reading the specs line-by-line.
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.