New Improved +/-45V PSU

... Quote: "... Features 100µA Typical Starting Supply Current 500µA Typical Operating Supply Current Operation to 1 MHz Internal Soft Start Internal Fault Soft Start Internal Leading-Edge Blanking of the Current Sense Signal 1 Amp Totem-Pole Output 70ns Typical Response from Current-Sense to Gate Drive Output 1.5% Tolerance Voltage Reference Same Pinout as UC3842 and UC3842A ..."

You can not ask for more?

Glenn

Ah... how about "not being 5x the price?" :)

Some of those features, like the leading-edge blanking and soft start, eliminate external parts. The tradeoff depends on your purchasing and placement costs. If purchasing/placement costs 25 cents per part, it doesn't take many extra parts to sway the balance.

I don't mind the cost of some of these switcher parts as much as the complexity and the number of discretes they often need.

They're about five times the price--what's better about them that's enough to outweigh that?

Cheers

Phil Hobbs

If you're building a multi-kilobuck instrument in low volume, with high purchasing and placement cost, the book price of a small part doesn't much matter. What really matters is design time and risk. EMI is a big part of that risk.

All true. However, when you're trying to figure out where the bodies are buried, inquiring minds want to know. ;)

Cheers

Phil

You folks are correct, it is not the cost of the parts but design time, risk, heat and EMI just to name a few. I believe that you are taking a risk in your design without protection features that are normally needed. Such as under voltage shutdown and poor short circuit protection. If the 16V supply come up low or if the output is shorted you have a good chance of letting the smoke out. In normal operation your design has a large charge dumping spike at R1 at turn on. This screams for a 1nF cap at Q1-B to ground to mitigate the spike but will greatly effect short circuit protection. Needs lots of tweaking. I used the LT1242 only because it was the only model available in LTSpice. You are correct about LT controllers. They have too many hooks and features, to allow easy integration. Designers using LT parts normally copy the published designs to the letter and not think on their own. LT never learned that a great IC controller has less than 10 pins. If I were task with this design I would probably look at the LM3475 Hysteretic controller from TI. Your design works because this is a simple requirement but most PSU are not this simple and you need to be conformable with IC controllers.

Highest regards, Harry

The thing I posted was just a concept, a starting point for playing with. Like I said, I had to run to Safeway.

I did a production board of something similar, a schmitt ac-coupled boost converter. It uses a comparator for the source current sensing and it does have an RC in that path. The breadboard works fine and I haven't blown up a single IC so far. The first PCB will be here on Friday.

This circuit doesn't mind a shorted load, because the mosfet peak current thing always works.

Here's a schmitt-based voltage flipper. It's actually used on a couple of products.

There *is* a schmitt trigger model too!

Your sketch is a good case on point. If you replace the LM5112 with a LTC3801 (6 pins) with your feedback scheme, you now have a solid PSU and not a kludge. I have used that topology often, with an "off the shelf" coupled shielded inductor, to get tracking +/-V outputs. Safeway?? Have you ever eaten at "Brown Sugar" in Oakland. We should meet there some time an have grits an eggs.

Cheers, Harry

Why would you call a perfectly fine, works-as-designed, cute-as-a-button production circuit a kluge? Because it doesn't cost $3 and isn't blessed by some reference design?

Besides, the LTC3801 isn't good to 12 volts. Would it even work as an inverter?

I have used that topology often, with an "off the shelf"

The food in San Francisco is fabulous, except for two glaring exceptions: no decent Cajun/Creole, and no good BBQ. Oakland has both.

Sure, any time. I love grits. Or come to SF and I'll give you the plant tour and take you to Zuni. They have grits, except that they call it polenta and charge 4 times as much.

Zuni is where I discovered cranberry beans.

We should all meet at some time. I actually have met Joerg, great guy with his heart the right place (give my regards to your wife, if you read this)

But, we would need bouncers, to keep JT and JL apart, although it seems to be better the last couple of days ;-)

Cheers

Klaus

I've met Joerg too, took him to Zeitgeist for beers and burgers.

Let's leave JT out. I shipped him a six-pack of our house beer, and he thanked me by saying truly vile things about my wife. Not that he's ever met her... just to be nasty.

There you go again, tipping over the edge of your manic-depressive self... what a sick puppy :-( ...Jim Thompson

There's nothing manic-depressive about calling you the coarse asshole that you are. And nothing wrong with defending my wife; she's worth a hundred of you.

ng

ed

ing

d I

ouple

h a

SU

ld

h.

with > >>>his heart the right place (give my regards to your wife, if you r ead this)

s to > >>>be better the last couple of days ;-)

t you

ou.

A curiously quantitative assessment from somebody who can write dozens of l ines on electronics without getting closer to numbers than saying that some thing didn't include a $3 part.

Also a bit silly. Jim Thompson's net social worth may well be negative - an ybody who publicly speculates about shooting his neighbours after the count ry falls apart can't be considered as an asset to his neighbourhood - and i f John Larkin's wife worth was a 100 times more negative than Jim Thompson' s it would be a very negative assessment.

Maybe I am extremely confused, but I don't understand why Q1 is connected the way it is.

I think R1/R2/C1/U1 makes an oscillator when you turn on the juice. That turns Q2 on and off through R3. OK.

When Q2 is on, a magnetic field builds up in L1. When Q2 turns off, the field collapses, inducing a voltage in L1. This is rectified by D1 and filtered by the capacitor that must not be named to become the -12 V out. OK.

C2 is a power supply bypass capacitor, probably to deal with the current peaks when Q2 is on. OK.

R4/R5/Q1 seem to be part of the feedback. I think the idea is that when the -12 V goes too far away from ground, you want the oscillator to have Q2 turned on for shorter amounts of time. I further think that the idea is to "steal" some of the current that would normally charge C1 and dump that current to ground instead, which changes how fast C1 charges.

If that is true, I would sort of think that the base of Q1 would be towards the R4/R5 junction, the emitter would be grounded, and the collector would go to the R1/R2 junction.

I'm not saying it doesn't work (it *is* in production); I am just trying to understand how it works.

Matt Roberds

Right. The duty cycle, ignoring Q1, is about right to ramp up the load.

VOLDECAP!

Bypass caps are good, like wearing garlic to keep vampires away. They both work.

OK.

Ideally yes, but the feedback polarity would be wrong.

When the output voltage is too low (that is, not negative enough, like -6 maybe) the voltage divider R4-R5 will keep the emitter of Q1 positive, which keeps Q1 off. The RC Schmitt oscillator runs at the "powerup" duty cycle and pumps a bunch of current into Voldecap. When the output finally gets close to -12, the emitter of Q1 hits about -0.6, and Q1 starts to conduct. It's a common-base, non-inverting amplifier. As the emitter goes negative, Q1 collector current flows, pulling C1 down, reducing the oscillator duty cycle. It stabilizes with the output voltage around -12.

It's not super accurate, but I'm just using it to power opamps.

Certain Engineers Who Must Not Be Named think I should use Cuk converters or something complicated like that instead.

What if I put a bypass cap on the clove of garlic before wearing? Will that keep twice as many vampires away? (Or a clove of garlic across the bypass cap in a circuit. Hmm.)

If it was hooked up like I said, as the base goes more and more negative, the transistor conducts less and less, and steals less current from C1? That isn't what you want, yeah.

Aha. That, and the fact that negative voltages are involved, makes it make sense to me.

I am used to seeing either common-emitter or common-collector circuits with a single supply, where the signal into the base tells the transistor what to do. From that perspective, having a grounded base seemed weird at first - "it's never gonna do anything". The fact that it isn't one of those topologies, plus the fact that there is a negative supply, means the emitter can get down to a voltage where the transistor will start turning on.

I have seen common-base circuits before, but drawn with the whole transistor symbol rotated 90 degrees counterclockwise, mostly to emphasise that something different is happening.

The current "stolen" from C1 via Q1 ends up in the -12 V out, right? Along with the < 1 mA from the +5 V reference that comes through R4 and R5?

If you needed more accuracy, and as long as the opamps need between a few and few dozen mA or less, maybe you could follow your circuit with a 79L12 and be happy. On the other hand, if the opamps only need a couple of mA total, the current that the 79L12 needs to run becomes a big part of the load. If you're running on a watch battery or something, you're counting every mA.

PIC, AVR or equal with a software PID loop. :)

Matt Roberds