Linear PSU design

[Phaedrus]

Ironically, though I work at a company that makes PSUs, I've never been able to sit down and design one from the ground up and actually have it made, for a number of reasons. Welp, I need a bipolar supply in the lab, so I figure it's a chance to make one.

I've set my specs as +/- 1.5 - 32V, 10A (5A per channel), accuracy +/-1%. I probably won't need to run the negative rail over 1-2A, but I will need to run the positive one that high, so might as well make it symmetrical.

Here's my basic schematic for the power stage. I'll add a PIC16 and a 7 segment display for output readings. Probably add some comparator based protection, etc. This is just the basic outlines.





Cx, Cy, and L are maybe not strictly necessary; but I have an entire bin of those parts pulled out of SMPS, and there's no harm in filtering your input. I also have a dozen bridge rectifiers to choose from. Protections diodes will probably be 1N4148s. C3 and C6 will be an array of 2200-4700uF Rubycons or some such. The LM338s can be heatsinked on a spare CPU heatsink I have.


My main concern is the transformer. Commercial center taps in the power range I need run around $75-$80. How much would it cost if I were to wind my own instead? Also let me know if I've made any rookie mistakes.

[retrolefty]

C6 and C5 will most likely blow up if you don't correct the polarity symbols.

And lower regulator needs to be a negative voltage regulator.

[Phaedrus]

Thanks! Derp.

[T3sl4co1l]

Your schematic doesn't show connections to:
Any capacitor, P1, P2 to GND
C4 (+)
D2 K
D3 A,K
D4 K

D3, D4 are backwards.

You can use filtering on the input, doesn't really matter.  The common mode choke won't make any difference; the primary capacitance to the core will be much greater than anything a CMC will do.  The core should be grounded.  You can use a pair of R+C (say, 10 ohms + 0.01u) across the secondary (one for each end to CT) to help suppress diode recovery transients.  Likewise, some high frequency filtering at the output might serve to keep this noise off the output (you'll need a three winding CMC to filter the outputs and ground, however).

Note there is no way to get 10A out of the thing, and 5A will only be available at the highest output voltage.  You will need massive heatsinks to cover the full range safely, even at reduced current (1-2A).

Not worth winding transformers; the parts, if you can find them, aren't much cheaper, and you won't have a vacuum impregnated, UL rated, fully specified part.  I have seen toroid kits before, which might be good for those cases where you need just a mess of outputs, but for a single CT winding?  Meh.

Edit: yes, C4, C5 and C6 are backwards ^^ and the FWB VV

Tim

[Andy Watson]

The bridge rectifier is not correct!  You've got the A.C. where the D.C. should be. You might also want to take a look at the negative supply - it appears to be the same as the positive supply - but drawn upside-down. I suspect this is not what you intend.

Here's my basic schematic for the power stage. I'll add a PIC16 and a 7 segment display for output readings.

Consider an off the shelf DVM module solution - unless you specifically want to gain experience in using PICs.

The LM338s can be heatsinked

It's not optional - they will need to be heatsinked (heatsunk ?)

My main concern is the transformer. Commercial center taps in the power range I need run around $75-$80. How much would it cost if I were to wind my own instead?

It depends how much you value your time!

[Phaedrus]

Ouch. Draw schematic in haste, regret at leisure. Teach me to sketch something and post it without double checking everything. 

[Phaedrus]

I've updated the schematic with the dumb polarity mistakes corrected. Note, the capacitors, etc. are connected to the center ground, it's just not super visible in this EDA software (ExpressSCH).

The concerns T3sl4co1l raised are probably more important in the big picture, though. If this topology won't provide the rated current, what topology would? I don't want to go SMPS. I could maybe drop the current requirement to 2-3A, but I suspect the LM338s still wouldn't cut it...

[Bored@Work]

I'll add a PIC16 and a 7 segment display for output readings.

Digital instruments got so cheap on ebay and other places that it is simpler and easier to just buy them. Many use some MCU internally.

Protections diodes will probably be 1N4148s.

Too weak. Use at least 1N400x or better.

Edit:
Oh, and the bridge is was wrong.

And, well, the second LM338 will not magically become a negative regulator. 

And some fuses on the secondary side. See also crowbar below.

And some reverse protection diodes  parallel to the outputs.

C3 and C6 will be an array of 2200-4700uF Rubycons or some such.

Skip them if you can. Big caps at the output means trouble.

My main concern is the transformer. Commercial center taps in the power range I need run around $75-$80.

Try surplus shops.

For extra fun:

Crowbar?

Negative supply tracking positive one?

Transformer tap switching to reduce power?

Some power  switch?

Output on/off switch?

Adjustable current limit?

[Phaedrus]

And, well, the second LM338 will not magically become a negative regulator.

True. I can drop in an LM337 and get 1.5A out of it though. But the topology concerns are the bigger worry...




Heatsinking is no concern for me, mind. I work for a company that manufactures heatsinks for Intel, AMD, Via, Nvidia, and a dozen other companies. Worse comes to worst I can slap on some CPU heatsink rated for 250-300W TDP, just a matter of pulling a refurb from the warehouse. But, if I can avoid going to those lengths that would be good.

[mariush]

You could try something more modern instead of LM337, LM338 .. for example see LM1084, LT1084  etc for up to 5A with just 1.3v drop , or  *1085 for the 3A version with same specs. LT1083 can do 7.5A but it's even more expensive.

For negative voltages  lm337 is good enough, if you want more amps you can go with LT1033 or LT1185 for example, both do 3A ... downside is they're somewhat expensive.

Be careful if you go with LM337 or similar... they can only dissipate about 15w. This means you can't just give it 30v input and configure it to output 5v at 2A ... that would mean you're dissipating 50 watts [ (30-5) x 2 = 50w). Get a center tap transformer and switch between taps when you need to go above a particular voltage, for example 0-16v , 16-32v

Transformers aren't that expensive.

Here's some transformers (long digikey url) : http://goo.gl/vtLMw3

When picking one (or two, see below), keep in mind that you need to rectify that AC voltage to DC using bridge rectifier, which means your peak DC voltage will be Vac x 1.414 - 2 x Vdiode drop ... so a 24v AC transformer will give you ~ 32v peak dc voltage.
Then you need to size up the capacitors accordingly, so that the input voltage will always be above a particular voltage, for the regulators to work.
Example.. for 3A of current and just 2v drop between peak dc voltage of 32v you'd need   C = current / (2 x ac freq x vripple) = 3 / 2x50x2 = 3 / 200 = 0.015 farads or 15.000 uF

So it would make more sense to pick a 28-30v AC transformer, like this one for example that's 28v AC, because then you'd have  about 37v peak DC, so you could size the capacitors for minimum 34v, therefore you could use C = 3 / 300 = 10.000 uF (or about 3 x 3300uF 50v electrolytics in parallel)

Then also keep in mind the VA rating.. for example a 24Vac 100Va (4.16A) will be 32v peak dc at about 0.68x4.16 = 2.8 A

My advice would be to use TWO transformers with center tap, for example this one: http://www.digikey.com/product-detail/en/VPS28-6250/237-1282-ND/666168  this one will give you ~ 37v DC @ 4.25A 
Connect all four secondary windings in series, the link between transformers becomes your ground, and  then you have two windings for positive, two for negative..

If you really want 5A, either go with a larger transformer (but expensive, 40$+ as far as I see on Digikey), or make a compromise and make it 5a+ just on 0-16v by placing the secondary windings in parallel.

[Rerouter]

Once you get the above issues fixed up, you can make it able to go down to 0V by biasing the bottom of the set pin with 2 diode drops to the opposite rail, many lm317/lm318 supply schematics online showing how its done,

[liquibyte]

My advice would be to use TWO transformers with center tap, for example this one: http://www.digikey.com/product-detail/en/VPS28-6250/237-1282-ND/666168  this one will give you ~ 37v DC @ 4.25A

I just bought two of these to do a dual 0-30V 3A setup.  I know I'll get flak for this but you can see some of the work I did over at Electronics Lab.  The circuit isn't perfect and I can only get down to 31-33mV but it works fairly well and there's a 5A version also.  I haven't started hacking on the circuit yet but I have four spare boards to play with in the future.  First board I ever had made so don't be too harsh, it works but the transistor footprint is backwards.