Flyback smps problems (Now forward converter) / General Science and Electronics / Forums

Forums

4hv.org :: Forums :: General Science and Electronics

« Previous topic | Next topic »

Flyback smps problems (Now forward converter)

previous 1 2 3 4 next

Move Thread

LAN_403

uzzors2k

Thu Feb 14 2008, 08:33PM

Registered Member #95 Joined: Thu Feb 09 2006, 04:57PM
Location: Norway
Posts: 1308

Here's a good app note on RCD clamps. Apparently you can run more than 50% with a properly desinged clamp, can a reset winding also do so? Since I'll be winding my own transformer I'll use a reset winding since they're more efficient and simple, it's just that I'm running out of pins on the bobbin!

Actually I forgot all about the output inductor for the test run! I'll use a multi-winding one like you say once I rewind the xfrmr.

About the minimum load thing, browse down to Fig. 5 here.

GeordieBoy

Thu Feb 14 2008, 11:15PM

Registered Member #1232 Joined: Wed Jan 16 2008, 10:53PM
Location: Doon tha Toon!
Posts: 881

You can't just remove the reset winding in a forward converter, without providing an alternative means to reset the transformer core. Also flyback transformers with their cores gapped in the centre limb do not make good forward converter transformers. Transformers in forward converters should have no air-gap, (except rarely in the assymmetric half-bridge arrangement where 2 switches and 2 diodes are used and a tiny air gap is sometimes introduced to stabilise the inductance and reduce flux remnance. The much greater magnetising energy here is recycled to the bus capacitor.)

Normally switch duty is limited to less than 50%. The volt-seconds applied across the primary when the switch is closed must equal the volt-seconds developed across it when the switch is open for the core to successfully reset after each switching cycle. So if the DC bus voltage is 320volts, the MOSFET drain voltage will need to reach 640 volts + some margin for the leakage inductance spike.

The duty ratio of the switch can sometimes be pushed beyond 50% with low input voltages (in USA for universal-voltage supplies) but this means that the reset voltage now needs to be even higher than the DC bus voltage because it has less than 50% of the total time to achieve volt-second balance and reset the core. At anything other than low input voltages the Vds burden on the MOSFET is too high.

The best reset arrangement for your power level is to use a reset winding and reset diode of equal number of turns to the main primary winding (but much thinner wire), and to limit the duty ratio to less than 50%. Above 50% duty you need slope-compensation etc anyway so its really not worth getting into that unless the supply needs to be universal input rated 80V-270VAC etc.

Hope this helps,

-Richie,

uzzors2k

Sat Feb 16 2008, 11:57AM

Registered Member #95 Joined: Thu Feb 09 2006, 04:57PM
Location: Norway
Posts: 1308

It did, thanks.

For the transformer do 58 primary turns and 24 secondary turns sound reasonable? The core is still 70mm^2. I used a bunch of equations from an app note, but I'm not sure if I understood them properly.

Marko

Sun Feb 17 2008, 12:24PM

Registered Member #89 Joined: Thu Feb 09 2006, 02:40PM
Location: Zadar, Croatia
Posts: 3145

Normally switch duty is limited to less than 50%. The volt-seconds applied across the primary when the switch is closed must equal the volt-seconds developed across it when the switch is open for the core to successfully reset after each switching cycle. So if the DC bus voltage is 320volts, the MOSFET drain voltage will need to reach 640 volts + some margin for the leakage inductance spike.

I was thinking about this, so wouldn't this kind of topology seriously limit the max duty cycle?

Since reset voltage will add to primary voltage on switch drain with common 500V devices we have around, it would have to be much less than 50% and I'd have really poor use of the transformer?

Could I simply use larger reset winding/primary ratio to keep that voltage down, why keep it 1:1 if that is unfavorable for most available switching devices?

GeordieBoy

Mon Feb 18 2008, 11:22AM

Registered Member #1232 Joined: Wed Jan 16 2008, 10:53PM
Location: Doon tha Toon!
Posts: 881

so wouldn't this kind of topology seriously limit the max duty cycle?

Marko for a universal-input rated power supply, you only need maximum duty ratio when the input is at it's minimum voltage. In practice this value is the minimum that the DC bus voltage sags to in between mains peaks with like 90VAC input and full load. So even with 50% duty ratio the peak MOSFET voltage doesn't need to be that high because the bus voltage is low at this time.

At the other end of the input voltage scale (260VAC) the peak DC link voltage can be 370 volts, but the duty ratio is probably around 12% now to keep the output in regulation. So at least with an RCD reset circuit the drain voltage doesn't need to go that high because there is a lot more time to acheive core reset at a lower voltage. That is why most universal-input rated single-transistor forward-converters use a carefully optimised RCD clamp.

Commercial single-transistor forward converters to operate up to 240VAC often use 800 or 900v MOSFETs. These are standard devices. The higher Rds(on) of the high-voltage MOSFET is less of a problem in the forward converter because it doesn't have to support the very high peak currents that plague the flyback supply.

As a side note, output current in the forward converter is usually continuous so components on the output side also are not plagued by such high peak currents. The output capacitors see many times less ripple current in a forward converter than a similarly rated flyback. Output voltage ripple, EMI and capacitor lifetime are all more favourable.

-Richie,

Marko

Mon Feb 18 2008, 10:29PM

Registered Member #89 Joined: Thu Feb 09 2006, 02:40PM
Location: Zadar, Croatia
Posts: 3145

t the other end of the input voltage scale (260VAC) the peak DC link voltage can be 370 volts, but the duty ratio is probably around 12% now to keep the output in regulation.

That looks like a poor use of a core compared to a halfbridge forward - I guess that's the trait of using one low side switch. Could it be helped by using a different reset winding/primary ratio?

Right now I'm trying to understand what factors determine minimum load to keep continuous mode on multi-output converter... (that article is after all this time still losing me a bit :( )

I can take feedback only from one output, and if I loaded it too lightly to enter continuous mode but to keep controller working, voltage on unregulated output rises even up to output voltage of the transformer...

From equation 11 of the article, higher the frequency, inductance, and lower the input voltage less load is required to keep continuous current?

I don't understand one thing though, it appears impossible from the discontinuous mode equation to get this 'delta' to zero as it would be required to get an equation for continuous mode, ir the equation just becomes invalid at that point?

Marko

Marko

Sun Apr 27 2008, 08:19PM

Registered Member #89 Joined: Thu Feb 09 2006, 02:40PM
Location: Zadar, Croatia
Posts: 3145

Hey guys,

looking at this again and I have some questions on forward converter:

1. - regarding clamps: If I use 500V devices, and decide my max drain voltage to be 425V (Ucc+100V), my maximum duty cycle would be 100/325 = 0.3 in any case, right?

So, with 500V switch it would be impossible to use 1:1 reset winding, but rather, I would need to increase the number of turns in reset winding (it is not just about volt-seconds anymore, but rather volt-turn-seconds since I have 2 windings).

Using 3x the number primary windings for reset winding only 108V would add to switch drain voltage, right?

And if I had a switch rated for high enough voltages, I could make reset winding with less turns than I have on primary and thus go beyond 50% duty cycle, is that right?

Am I understanding some of this?

Other thing I'm wondering about, could I simply use for example a 100V zener or TVS with reverse diode to clamp the primary voltage directly to the specified value?
That appears easier than RCD clamp, but why don't I see it anywhere?

2. Optocoupler feedback: I guess this was always the most troubling part of SMPS's to me.

It is easy to use UC3842 error amplifier directly, as I have everything more-or-less ready made and just need to use a resistor and compensation capacitor between output and negative input of the error amp...

I could directly stabilize the auxiliary voltage and keep output voltage regulated that way (as long as there's enough minimum load?), but for truly good regulation, especially if I have just one output, I need isolated feedback...

That is where I fall into trouble.
What exactly is required in order to get this typical TL431+optocoupler feedback to work?

I first thought that, since TL431 acts as an error amplifier of sorts, I must *disable* the UC3842's own error amplifier by giving it strong negative feedback, so optocoupler can more or less linearily adjust duty cycle.

But, looking at these two circuits:

(second is flyback converter, but I don't see how feedback should differ?)

Both of them seem to have pretty closely same configuration of error amp on UC3842.

So how can I have 2 error amps doing the same thing?

Can anyone explain me what exactly do I need to do for this opto-coupler feedback to work?
In fairchild app-note phototransistor forms a network with R6, R7 and R8, with output from R8 being fed directly into feedback voltage input (pin 2.)

How can this be stable, I have 2 error amplifiers and 2 reference voltages? What am I missing there?

Then there comes feedback compensation... where is exactly the compensation network in this TL431+OC circuit? R15, 16, 17 and C18 in fairchild circuit?

Edit; one more:

- I tried defining max duty cycle to about 30% as I need, but I'm running into trouble due to very large capacitance and small resistance needed. Oscillator becomes unstable and shaky at that point, I guess because of potentiometer I used.

The fairchild forward converter circuit has duty cycle set to nearly 100% and seems to be fine with it. What am I missing there? Wouldn't this allow possibility of transformer saturation in some cases?

Marko

GeordieBoy

Mon Apr 28 2008, 02:37PM

Registered Member #1232 Joined: Wed Jan 16 2008, 10:53PM
Location: Doon tha Toon!
Posts: 881

Okay Marko, this is a big subject area but i'll try to answer your points very quickly...

1. - regarding clamps: If I use 500V devices, and decide my max drain voltage to be 425V (Ucc+100V), my maximum duty cycle would be 100/325 = 0.3 in any case, right?

500V devices aren't usually used in single-ended forward converters for 220-240VAC countries. The voltage rating is too low. 800V or 900V devices are used to allow for the voltage swing during reset. 500V devices are however used in the assymmetric half-bridge arrangement (2-switch forward converter) because the reset voltage is clamped to the DC bus voltage in this topology.

So, with 500V switch it would be impossible to use 1:1 reset winding, but rather, I would need to increase the number of turns in reset winding (it is not just about volt-seconds anymore, but rather volt-turn-seconds since I have 2 windings).

I think you are a little confused here. Altering the number of turns on the reset winding doesn't do anything to mitigate the voltage requirements of the single MOSFET switch. It is the volt-seconds on *every* winding of the trasnformer that must balance during the forward and reset periods. You can alter the number of turns on the reset winding to clamp to a different voltage, or remove the reset winding completely and use an RCD clamp but the volt-seconds across the primary during MOSFET conduction must be equaled by the volt-seconds across the primary during reset. Only altering the on-time, the DC bus voltage or the reset time can allow the clamp voltage to be changed.

Using 3x the number primary windings for reset winding only 108V would add to switch drain voltage, right?

Nope. Volt-second balance still has to apply to the primary winding and dictates the voltage seen by the switch during reset.

Other thing I'm wondering about, could I simply use for example a 100V zener or TVS with reverse diode to clamp the primary voltage directly to the specified value?

I think you can, but losses in the zener might be high during start up of the PSU. There are also times when the sliding clamp voltage of the RCD can minimise losses. If you use a zener clamp you will need to make sure there is absolutely no air-gap in the transformer or dirt on the mating surfaces and keep magnetising inductance high. Use lots of turns on the primary!

The normal arrangement for a universal voltage forward converter is the RCD reset-voltage clamp because it doesn't need an additional reset winding on the core and the reset voltage can float with the dc bus voltage as I mentioned before. It is one of the few things that increases efficiency at low line.

That is where I fall into trouble.
What exactly is required in order to get this typical TL431+optocoupler feedback to work?

Stable loop compensation with sufficient sufficient DC gain to maintain good regulation and sufficiently high crossover frequency to obtain good transient response.

I first thought that, since TL431 acts as an error amplifier of sorts, I must *disable* the UC3842's own error amplifier by giving it strong negative feedback, so optocoupler can more or less linearily adjust duty cycle.

Often the internal EA is disabled in PWM controllers when an amplified reference like the 431 is used on the secondary side. But there is nothing to stop some of the loop compensation being applied around the error amplifier on the secondary side and the remainder of the loop compensation to be applied on the line side. The transfer functions just multiply together to give a single combined closed loop response. Usually the tolerance on the optocoupler CTR is the biggest concern here!

But, looking at these two circuits: , (second is flyback converter, but I don't see how feedback should differ?)

Be careful. Forward and flyback converters have totally different transfer functions so their requirements for closed loop compensation are completely different. The voltage programmed forward converter has a two-pole resonant open-loop response which is quite complex to compensate. In contrast the commonly used current-mode discontinuous flyback has a simple single pole open loop response that is inherently stable with simple closed loop feedback. That is what those two Error Amplifier circuits in the app note are trying to show you. A discontinuous current mode flyback is the simplest thing there is to compensate because it just looks like a current source charging the output capacitor. Things get way more complicated if the flyback enters continuous current mode though, and a right-half-plane zero makes feedback compensation very challanging.

Then there comes feedback compensation... where is exactly the compensation network in this TL431+OC circuit? R15, 16, 17 and C18 in fairchild circuit?

The R's and C's around the TL431 set the mid-band gain, and the placement of the poles and zeros that shape the frequency response of the compensator. As I said this is a huge topic, but they are basically chosen to get high DC gain, wide bandwidth with good phase marging, a high loop crossover frequency, and a graceful rolloff in gain with a -1 slope so the transient response doesn't ring. That's it in a nutshell.

If you want to know more google for terms like "buck type II compensation" and get out some graph paper

- I tried defining max duty cycle to about 30% as I need, but I'm running into trouble due to very large capacitance and small resistance needed.

Unless you *HAVE* to design a single switch forward converter for universal voltage applications I would suggest building a half-bridge or assymmertric (2-switch) forward converter. If you mess up the reset calculations for the single-transistor converter you will saturate the transformer or burn out the clamp. Either way the switch will die quickly. The high voltages developed in an off-line forward converter can also be tricky to measure safely without the right probes, so best to opt for something more robust if you are unsure of how it works and haven't got 6 months spare to plod through it all.

I hope this info helps,

-Richie,

Marko

Tue Apr 29 2008, 11:51AM

Registered Member #89 Joined: Thu Feb 09 2006, 02:40PM
Location: Zadar, Croatia
Posts: 3145

Thanks Richie for the long reply,

, I'll try to get through it:

500V devices aren't usually used in single-ended forward converters for 220-240VAC countries. The voltage rating is too low. 800V or 900V devices are used to allow for the voltage swing during reset. 500V devices are however used in the assymmetric half-bridge arrangement (2-switch forward converter) because the reset voltage is clamped to the DC bus voltage in this topology.

I thought 500V devices are more common, the mini 80W ATX supply I had here used a single IRFP450.
800V devices are hard to acquire in my country and quite expensive, while I already own loads of 500V mosfets.

I think you are a little confused here. Altering the number of turns on the reset winding doesn't do anything to mitigate the voltage requirements of the single MOSFET switch. It is the volt-seconds on *every* winding of the trasnformer that must balance during the forward and reset periods. You can alter the number of turns on the reset winding to clamp to a different voltage, or remove the reset winding completely and use an RCD clamp but the volt-seconds across the primary during MOSFET conduction must be equaled by the volt-seconds across the primary during reset. Only altering the on-time, the DC bus voltage or the reset time can allow the clamp voltage to be changed.

Nope. Volt-second balance still has to apply to the primary winding and dictates the voltage seen by the switch during reset.

Apparently we are in misunderstanding here. At beginning of my post I implied that I *would* decrese the duty cycle, that is, on time, for use of 500V switch, and that also implies for reset winding of higher ratio idea. Sorry for being unclear on that.

Maximum duty cycle I can ever use is *directly* limited by the maximum voltage the switch may see, agree?
I know it will be low with 500V switch, like 20..25% or so practically for 230V mains, but I accept that.

Take an example:

I have a switch I can only allow to see 400V D-S, and 300V DC supply.
This would yield maximum reverse primary voltage of 100V, which automatically implies my off time would need to be 3x on time to balance Vs, right?

In order to achieve that with reset winding, to my logic, it would have to be 3x the primary turns;
During off time, the reset winding would see supply voltage on it, 300V, which translates by transformer action to 100V on primary, and satisfies my max allowed switch voltage;

Maximum duty cycle would be low;

say my on time is 1us, this would equate 300uVs, which would with 100V reset voltage need at least 3us off time, right?
This would equate max duty cycle of 25%.

This is what I meant, hope I'm not failing to see fallacies in there?

I think you can, but losses in the zener might be high during start up of the PSU. There are also times when the sliding clamp voltage of the RCD can minimise losses. If you use a zener clamp you will need to make sure there is absolutely no air-gap in the transformer or dirt on the mating surfaces and keep magnetising inductance high. Use lots of turns on the primary!

The normal arrangement for a universal voltage forward converter is the RCD reset-voltage clamp because it doesn't need an additional reset winding on the core and the reset voltage can float with the dc bus voltage as I mentioned before. It is one of the few things that increases efficiency at low line.

I'm using a toroidal ferrite core, and zener clamp seems to work pretty well, although I can't seem to be able to make feedback to work properly. Optocoupler feedback was always my direst problem.

Often the internal EA is disabled in PWM controllers when an amplified reference like the 431 is used on the secondary side. But there is nothing to stop some of the loop compensation being applied around the error amplifier on the secondary side and the remainder of the loop compensation to be applied on the line side. The transfer functions just multiply together to give a single combined closed loop response. Usually the tolerance on the optocoupler CTR is the biggest concern here!

What is puzzling me is, how can I actually ''disable'' the error amplifier inside UC3842, which has it's positive input permanently fixed to 2.5V reference.

The two-error-amplifier ocnfiguration just doesn't work to me.

I've made the supply work well with load, no feedback and duty cycle manually set...

Should I just try to tie it's negative input to Uref and somehow use the optocoupler directly on PWM comparator input? It looks like I could do that since amp has diodes on it's output.

Would something like that work?

Be careful. Forward and flyback converters have totally different transfer functions so their requirements for closed loop compensation are completely different. The voltage programmed forward converter has a two-pole resonant open-loop response which is quite complex to compensate. In contrast the commonly used current-mode discontinuous flyback has a simple single pole open loop response that is inherently stable with simple closed loop feedback. That is what those two Error Amplifier circuits in the app note are trying to show you. A discontinuous current mode flyback is the simplest thing there is to compensate because it just looks like a current source charging the output capacitor. Things get way more complicated if the flyback enters continuous current mode though, and a right-half-plane zero makes feedback compensation very challanging.

Uh, I see.

More I learn this just looks more hopeless to me, my previous ideas of reasons for feedback oscillation apparently just don't make sense.

Can you tell me just this:

If I have nothing but series zener diode and optocoupler for feedback... and optocoupler controls the voltage into PWM comparator with one more resistor.

How do I compensate such a kind of 'error amplifier'? I only know I need to use a resistor in series with the zener.

Unless you *HAVE* to design a single switch forward converter for universal voltage applications I would suggest building a half-bridge or assymmertric (2-switch) forward converter. If you mess up the reset calculations for the single-transistor converter you will saturate the transformer or burn out the clamp. Either way the switch will die quickly. The high voltages developed in an off-line forward converter can also be tricky to measure safely without the right probes, so best to opt for something more robust if you are unsure of how it works and haven't got 6 months spare to plod through it all.

I hope this info helps,

-Richie,

Well I fixed the duty cycle issue now, I'm keeping it <25%... it was just choice of the right oscillator cap and resistor.

I wanted a small simple SMPS that would not use any specially critical components.
I need no more than like 15W of power, and half bridge converter would be an overkill.

Also, I don't seem to have any current mode halfbridge controller IC's. I only have common ones like SG3525 and TL494.

The big pitfall of SG3525 and mosfet halfbridge would be impossibility of startup without an auxiliary supply, and it would make little sense of using SG3525 on high side rather than IC that is designed for it.

TL494 and bipolar tranisstors have a way of self starting, by using the proportional-current drive transformer as a saturable transformer, working as an blocking oscillator for short period of time to power up the controller.

Still I thought of that as too much work in comparison to simply using UC3842.

Big posts.. uh!

thanks for answers,

Marko

Steve Conner

Tue Apr 29 2008, 12:49PM

Registered Member #30 Joined: Fri Feb 03 2006, 10:52AM
Location: Glasgow, Scotland
Posts: 6706

About disabling the error amplifier:

One technique I've seen is to tie the + and - inputs to some fixed voltages. So the - to ground and the + to Vref or whatever. If you get it the right way round, the error amp will actually turn itself off. They usually only have a single-ended output, so if it tries to pull its output all the way high, it ends up disabling itself. This is what people do to use the TL494 as a SSTC driver.

You can now feed the signal from your optocoupler into the compensation pin, since this is the error amp's output and also the input to the comparator.

Another way to "disable" the error amp would be to use heavy negative feedback with a resistor from the compensation pin to the - input. In this case, you're just lowering its gain. As Richie mentioned, a little gain in the error amp can be helpful, in case the optoisolator isn't powerful enough to drive the compensation pin directly.

The Fairchild datasheet even shows how to do that, see Figure 1:

previous 1 2 3 4 next

Moderator(s): Chris Russell, Noelle, Alex, Tesladownunder, Dave Marshall, Dave Billington, Bjørn, Steve Conner, Wolfram, Kizmo, Mads Barnkob