Dual Rail SMPS / General Science and Electronics / Forums

Forums

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

« Previous topic | Next topic »

Dual Rail SMPS

previous 1 2 3 next

Move Thread

LAN_403

Steve Conner

Fri Mar 11 2011, 08:13AM

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

Oh, if the transformer has an air gap and the load current was low, I guess it's OK.

Can't really tell from weight. Usually Powdered iron is yellow/White, or green/blue for the nicer grade. MPP, very nice if you can find it, is painted grey. I've also seen black ones used as output filters, but I don't know what they were made of.

Really your best bet is to see how it was used in the thing you tore apart. Active PFC boost choke, output filter choke, OK. But if it was a common mode filter (2 windings, connected in the AC input) it'll be ferrite, no good.

GeordieBoy

Fri Mar 11 2011, 03:02PM

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

As Steve C said, you really need that cross-coupled buck choke in there. If you wound the buck chokes for the +ve and -ve rails on different cores the cross-regulation will be poor.

With seperate +ve and -ve chokes each rail can transition independently between continuous and discontinuous current depending on it's load current. If one rail has a heavy continuous current and the other has a light discontinuous current then they obey different duty cycle equations and the output voltages will be different for the same duty ratio on the primary side.

If you tightly couple the two windings on a single core it's like making a single load current loop between the +ve and -ve outputs now with a virtual single inductor. Then there is only one current to be either continuous or discontinuous. At least then both output voltage rails obey the same duty ratio equation, and the feedback loop can work on minimising the combined error.

A cross-coupled choke is wound like a differential-mode choke, not like a common-mode choke. Remember it's job is to "choke" the HF ripple in the load current, so it needs to be wound so that the core sees the load current, not so it cancels out. Have a look at the cross-coupled choke in any PC SMPSU rated over a couple of hundred watts.

Iron-powder or gapped ferrite will both work. Ferrite with a closed magnetic path like that used for common-mode EMI chokes is not suitable at all. (A neat thing about iron-powder for SMPSUs is that it's DC current dependent permeability ("progressive saturation") can be used to delay the onset of discontinuous current operation under light load. This action improves regulation down to lower load currents, so I usually go with something like Type-52 Iron powder.)

-Richie,

Turkey9

Fri Mar 11 2011, 09:11PM

Registered Member #1451 Joined: Wed Apr 23 2008, 03:48AM
Location: Boulder, Co
Posts: 661

I'm pretty sure that the core I pulled from the supply is the choke. This supply was a dc/dc type with 24V input and variable output. The core was wound with 2 strands of 12awg wire, but I'm not sure if they were different windings as the whole thing was potted in silicon and wired point to point. I had to tear it apart to get at the parts.

Here's a picture of the core next to a quarter... now that I think about it I should have used a to-247 MOSFET as a size gauge as not everyone on 4hv knows the size of a quarter!!! the numbers on the side are A-254168-2 2G6+5

The load I was using when I took those primary waveforms pics was just the highest watt resistors I had on hand. Now that I think back, when I tested the supply using a saltwater load, it was a nice square wave without those abnormalities.

Steve Conner

Sat Mar 12 2011, 09:04AM

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

Gah, never seen a blue one before.

Never mind, a quick google of "a-254168-2 core" told me it is some sort of powder core from Arnold Magnetics.

Richie: I didn't realise that the coupled choke was as big a win as that. Must try retrofitting one to the remains of the Coldamp.

One subtlety: in an ATX power supply the choke windings are all in the same direction, like a common mode filter, but that is because all the rails are positive. You have one positive, one negative, so you have to reverse one winding.

Turkey9

Sun Mar 13 2011, 05:20AM

Registered Member #1451 Joined: Wed Apr 23 2008, 03:48AM
Location: Boulder, Co
Posts: 661

I wound the inductor and it works great! Powered up my amp and watched the primary waveform, which changes as more power is needed for the music! The change is not what I thought it would be though. The strange spike to 0V gets wider and shifts slightly to the beginning of the square wave. Is this normal? It seems that there should at least be a little dead time in the primary waveform between the pulses as the tl494 outputs a max of around 95% duty cycle. When looking at the signal at the base of the IGBT, the voltage at the dead time which should be zero is actually around 3V. Could this be turning on the IGBTs too soon? I don't have a cap in series with the GDT primary, is this needed? Is there any way that I could compare the gate drive waveform and the primary waveform with only a two channel scope? Possibly using a current transformer on the primary?

Also, I'm getting a lot of heating from the half bridge. More than I've gotten from a small SSTC that I built. The heat sinks I'm using are 6in by 2.5in with 3in fins running the length; they're fairly large. I have the bridge and the output transistors from one channel mounted to one heat sink and the other output transistors and the rectification diodes mounted on the other.

Thanks for all the help!

Steve Conner

Sun Mar 13 2011, 08:57AM

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

Glad to hear you got it working. Building a SMPS with regulation isn't exactly a picnic.

The spike to 0v is doing what volt-second balance would predict. When the deadtime starts, the bridge output shoots over to the opposite rail, returning magnetising current to the DC bus, until the core has reset, then it collapses to zero, usually with assorted ringing. Because of volt-second balance, it must spend as long on the opposite rail as it spent switched to the previous rail. So, the spike to 0v appears at duty cycles below 50%.

But normally you don't see it because it is small compared to the load current, so the freewheel diodes clamp it down, and the transformer voltage just collapses to zero, and doesn't reset until the next switching cycle. If your iggbits are getting too hot, and you have that funny waveform, again it makes me think maybe you have too much magnetising current. Can you put a CT on the primary and measure it?

Or then again maybe you have some slow IGBTs and are just going too fast for them, what switching frequency are you running? This also feeds into the magnetising current issue: the number of turns you quoted would be OK for over 100kHz. But I think the little transformer in an ATX PS has nearer 25 primary turns, for half bridge drive off a 320v DC bus at 30-50kHz. Are you using 320 or 160?

Since you have a gapped core, it won't get hot even if the flux density is excessive.

Turkey9

Sun Mar 13 2011, 07:51PM

Registered Member #1451 Joined: Wed Apr 23 2008, 03:48AM
Location: Boulder, Co
Posts: 661

Here's a pic of the current waveform. It peaks at about +/-3.25A or 6.5A p-p. I'm running at 50kHz 160V and using these IGBTs: STGW40N120KD

I have others that say specifically in the data sheet that they are intended to be used in SMPS. that I could use if needed. It is easier for me to get at the transformer than the switches however, if it needs to be rewound.

Thanks!

Steve Conner

Sun Mar 13 2011, 08:20PM

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

OK, you have way too much magnetizing current there. The load current is just a pimple on it, whereas it should be the other way around.

I suggest trying more turns, and/or a core with no gap.

Turkey9

Tue Mar 15 2011, 06:57AM

Registered Member #1451 Joined: Wed Apr 23 2008, 03:48AM
Location: Boulder, Co
Posts: 661

Well I changed to a smaller core and increased the primary turns from 18 to 30 and my problem didn't quite go away. It got close when I changed the frequency to 80kHz. At that point, the spike was still there but only dropped down a couple volts.

I was thinking about your post about this strange thing only happening when the duty cycle was below 50%. I also read that the ripple would go way down when used at a much higher minimum duty cycle. So I decided to remove a secondary turn so that at the minimum power draw of my amp, the duty cycle would be much higher. This worked as far as I can tell. There is no spike and a wonderfully shaped square wave. I don't know if this fixed the problem though... did it? The current through the primary is still 3A. This is at 50kHz. Do I just need to pile on the primary turns? Increase the frequency also?

GeordieBoy

Tue Mar 15 2011, 05:39PM

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

The magnetising current for a half- or full-bridge forward converter should only be a few percent of the full-load primary current! IGBTs will heat up a bit if the magnetising current is large because they are being asked to interrupt the current right at it's peaks! This causes current tailing and turn-off losses.

In contast, SSTCs typically operate with roughly sinusoidal current waveforms so IGBTs interrupt less current and suffer reduced tail current losses at turn-off.

Use a big power grade Mn/Zn ferrite core with the core halves firmly clamped together. There shouldn't be any air gap in the magnetic path for a half-bridge or full-bridge converter. You want the transformer to transfer energy from primary to secondary during the device's on-time, not store up energy like a flyback transformer.

Flux density, core loss and ultimately temperature rise will dictate how many primary turns you need. You should be able to find details in a magnetics textbook.

-Richie,

previous 1 2 3 next

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