Dual Rail SMPS / General Science and Electronics / Forums

Forums

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

« Previous topic | Next topic »

Dual Rail SMPS

1 2 3 next

Move Thread

LAN_403

Turkey9

Thu Mar 10 2011, 12:33AM

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

I've decided to try and build a dual rail SMPS based on the TL494. I've got it working pretty good with the + 35V rail. It's really quiet and can supply plenty of power when needed. I'm using the two rectifier and center tap ground setup to rectify the output. I have a 33uH and 1uF filter network on the output.

I was running the supply with a 2.5K ohm load on the positive rail and it was working great. For the negative rail, I used the same center tapped windings but with the diodes backwards for the negative rail. With no load resistor on the neg rail, it sits at about -60V, but when loaded with a similar value resistor, the output stays at the desired -35V.

When I scope the output, the positive rail is very smooth and filtered, but the negative rail has a fairly large amount of noise on it. Why is this? I have the same filter network on the negative rail, inductor in series with the output and a 1uF cap from the other side of the inductor to ground.

Also, why does the unloaded voltage not get regulated on the negative side? I'd think that since they are the same windings, they would be at the same voltage.

Thanks!

Adam Munich

Thu Mar 10 2011, 02:09AM

Registered Member #2893 Joined: Tue Jun 01 2010, 09:25PM
Location: Cali-forn. i. a.
Posts: 2242

Well I'm not too sure about the noise, but an SMPS usually requires some loading to maintain regulation, even something fancy like an ATX PSU.

Marko

Thu Mar 10 2011, 02:19AM

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

Hello,

If you're taking feedback from the positive supply rail, it's natural for it to expect it to be more stable than the negative one. It's stability is only assured by choice of matching components and hopes that it will be loaded about equally as the positive rail. Also, to see the same output voltage as on the positive rail, you need to load the negative rail enough to make it go into continuous current mode. Only there you'll get real "buck" action (the voltage being proportional to the duty cycle). Otherwise the voltage will just tend to the maximum output voltage of your transformer; the choke won't be able to buck anything with no current going through it. Try a few tens of watts of load instead of a kiloohm resistor.

If properly designed and loaded this approach is fine for most applications (including PC power supplies), although it might not look pretty. If you really need both rails regulated with direct feedback, your best choice is to make two single-sided supplies and wire them in series, and use a double potentiometer to adjust the voltage.

Marko

ConKbot of Doom

Thu Mar 10 2011, 05:21AM

Registered Member #509 Joined: Sat Feb 10 2007, 07:02AM
Location:
Posts: 329

While this may not help with the minimum load aspect, but for tight regulation I just saw in an appnote I was digging through recently, that your filter inductors should be wound on the same core, with the same turns ratio as the transformer to help maintain tight regulation between them despite feedback only being on one rail.

Steve Conner

Thu Mar 10 2011, 09:26AM

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

Yes, winding both inductors on the same core will help.

But the fundamental problem here is that the negative side is going into discontinuous mode when unloaded, as Marko points out. I have a Coldamp SMPS kicking around here, and their solution was to sense the difference between the + and - rails for regulation. That makes the action of the feedback a compromise between what the two rails need. If you do that, use a decent sized inductor so the DCM boundary occurs at a low current, and put a bleed resistor on there to keep it in CCM, it should work out fine. Oh and wire your amps out of phase so bass transients load both rails evenly. (the bass content of stereo recordings is usually the same on both channels)

OTOH, Coldamp designed theirs to work in discontinuous mode all the time. I blew up two of them without ever achieving the rated output.

Turkey9

Thu Mar 10 2011, 04:51PM

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

I was wondering about sensing the difference and using that as feedback. The only problem I foresaw was that if I average the two voltages into a single feedback, the individual voltages could get out of whack. For example, you could get +40 and -30 to give the right average.

I think I'll try winding the inductors on the same core just for the piece of mind. I've gotten it to work really well loaded with about 24W total, 12 on each rail. The voltages are the same down to the .01V as far as my meter can tell. The main problem was that I had calculated the inductors for an average current of 10A! That meant a minimum current on 1A or about 60W total for continuous mode according to the book I'm using. I changed the values and now it's working pretty well.

Below is a picture of the primary waveform at 24W. It looks strange to me with those spiked to ground in the middle of the square wave, what is causing this? My GDT doesn't have very much ringing at all. BTW this is a half bridge topology.

And a GDT waveform for good measure. This was taken at the base of one of the IGBTs.

@Steve, if I wire the channels out of phase, won't there be cancellation if the speakers get too close? Will this not mess with the sound somehow?

Steve Conner

Thu Mar 10 2011, 06:10PM

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

The spikes to ground are probably just once the transformer core is finished resetting and dumping its magnetizing current back into the DC bus. They are not the middle, rather the end of the cycle and the beginning of the next one. The waveforms all look fine, nice project!

You invert the input to one channel using an op-amp or similar, and also reverse the speaker wires to avoid cancellation. On this channel, the red speaker post will be grounded and the black one driven. That way, when you get a kick drum thump in your recording, both cones shoot out in the same direction, but one amp channel draws from the negative rail and the other from the positive, keeping the rails balanced. This is an old trick used by PA guys, that long predates SMPSs in power amps.

I think allowing the individual voltages to get "out of whack" is an acceptable compromise. The total out-of-whackness is still less than if you sensed one rail only. Think of it this way: You only have one point of control, so you can't regulate both rails. Therefore, the best compromise is to split the error evenly between the two rails, and that's what sensing the difference (average) does.

Turkey9

Thu Mar 10 2011, 10:50PM

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

Oooooohhhh that makes much more sense.

As for the inductor, how should it be wound? Should the windings be in phase or out of phase like a common mode line filter? I imagine I should use powered metal since it is functioning as a choke and will have large DC bias current. The yellow cores from ATX supplies are powdered metal, right?

Steve Conner

Fri Mar 11 2011, 07:54AM

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

I'm not completely sure, but I think for the coupled inductor to work, you have to connect it so that the fields of the two windings add, not cancel. So it's not like a common mode choke between your rectifier and your caps: one winding is reversed.

Yes, the big yellow toroid from an ATX will do. That's what its function is in the PS anyway, one of these coupled filter inductors. ATXs are cost reduced to the bone, so if you can get two toroids and stack them, that would be easier.

I thought about that waveform some more. I think it means your transformer has too much magnetising current: greater than the load current. Are you sure your transformer design is ok, you didn't get a 2x4 caught between the core halves or anything?

Turkey9

Fri Mar 11 2011, 08:06AM

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

Hahaha I'm not completely sure on the transformer design. It's larger than most ATX transformers with a cross section area of around 2cm^2 through the center. I didn't break the core apart so it still has the gap from the factory. I have 18 primary turns and 9 secondary turns (well 18 center tapped for the secondary to get the dual rails). I used Uzzor's spreadsheet for smps to determine the turns on the transformer. I have a bunch of ATX transformers too if I need to redo the transformer. Also, that waveform was taken with .35A on the output, not much at all.

I just recently tore apart a fairly large dead buck converter or something of the sort that has a couple large toroids in it. Two of them are the same size, but one is considerably heavier than the other. Is the heavier one the powdered metal?

Thanks!

1 2 3 next

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