Linear Regulators in parallel - Playing in LTSpice

[intabits]

I want to make a lab/bench "workhorse" power supply. Not too noisy, so linear rather than switching, and high power so
as to utilize a 2x25V 500VA toroid transformer that I have. Precision would be nice, but initially the emphasis is on power.

So I've been investigating putting linear regulators (eg LM350) in parallel for increased current output, and according to LTSpice, looks to be quite doable.
In practice, there will be a pre-regulator ahead of this, but here I'm just playing with the main regulator part.

The schematic I've come up with (using Lx317's):-

The output voltage is controlled in the usual way via the value of resistor RVAdj, with the usual lower limit of 1.25V.
The ballast resistors can also double as current sensors.

Sweeping the adjustment resistor to cause an output range of 2V to 30V, with a 5R load.



Here the output voltage is set to 12V, while the load ranges from 2R to 20R.
The start of the simulation is confused with the heavy starting load, but the load regulation is pretty good once it gets underway.



But I messed around some more to try for current limiting, and output down to 0V, and both operated via control voltages.
Adding some underground opamps, resistors, diodes and voltage sources, we have:-


Sweeping the Voltage control with a 5R load and current limit set to 5A:-


At 24V output, sweeping the load from 2R to 20R, with the current limit set to 4A:-


With 18V output into a 6R load, sweeping the current limit from 1A to 6A:-



This seems to be a cheap and cheerful way to get a reasonable high current, linear regulator with voltage driven controls.

Or not?

I'm going to knock this up with LM350s soon, and see how the reality pans out.

[Kleinstein]

This combines 2 bad ideas:

Paralleling regulator chips is tricky, as they are not exactly the same voltage and thus normally only one will be in normal operation. The other are either off (because they think the voltage is a little too high) or in current limiting mode, because they think the voltage is too small. So the power will not share equal, but some will get really hot and some will stay cold.

Adding an extra current limiting to a 3 pin regulator is a pain in the ass and very difficult to get stable with variable loads. One problem here is that the models for the regulators may not be that accurate and the standard parts are made by different sources. The Ti version may behave different from the ST or Motorola version.

So it is usually easier and have an lab supply build from scratch - so OPs+reference and transistors, maybe LM723 + transistors.

[mariush]

Some regulators have notes in the datasheets saying this is possible, by having some balancing resistors on the output (or some amount of wire that acts as a resistor).

For example, LT1083 / LT 1084 / LT1085 has such application example, see page 12: http://www.analog.com/media/en/technical-documentation/data-sheets/1083ffe.pdf

[intabits]

I know it's not recommended to parallel regulators, and I take your points on this.

But I just want to see this for myself, and since it's easy to do, I'll try it out anyway.

Going the "proper" OPs+reference+transistors route was always my ultimate intention.  I was just looking at this as an interim/fun/educational option.

I have some LT1084s coming, so I'll be trying them as well as LM350s

[MosherIV]

Hi

I agree with Kleinstein. It is not a good idea to parallel Linear regulator.
Most data sheets for linear regulators show how to use a 'bypass' transistor in emitter follower mode to increase the current capacity for the regulator.

I am not familiar with the LM350, I assume it is a variable one like the LM317.

For that much power/current, it is not advisable to build a linear psu.
At 500VA 25V, you can get 20A. The worst case power dissipation for the psu is when your output is say 1V. That means  your psu must drop 24V and at 20A output (worst case), you are talking near 500Watts to dissipate in heat!

Like Kleinstein says, it is hard to add current limit to a linear regulator. It is actually easier to build a psu with your own voltage and current regulation based around opamps. See this design
http://www.microsyl.com/index.php/2010/03/31/bench-power-supply-0-25v-0-5amp/comment-page-2/

[Siwastaja]

Most 3-legged power regulators have so poor output voltage tolerance that you tend to need to waste a lot of power in the balancing resistors, and then you'll load regulation will severely suffer from these resistors. This is why you cannot parallel most 3-legged linear regulators.

There's nothing wrong in switch mode design. I'm always intrigued about where do all the beginners get this idea that they need a linear lab supply to give their projects "clean" power a switcher "cannot" provide.

[intabits]

....talking near 500Watts to dissipate in heat!

I know all that. That's why I said a pre-regulator would be needed.

It is actually easier to build a psu with your own voltage and current regulation based around opamps.

Yes, I said I intended to do that. But I'm happy to spend some time exploring possibilities and experimenting, even if the conventional wisdom says it's a bad idea.
On paper, it works. But I want to see the practical issues and limitations involved for myself.

Most 3-legged power regulators have so poor output voltage tolerance that you tend to need to waste a lot of power in the balancing resistors, and then you'll load regulation will severely suffer from these resistors. This is why you cannot parallel most 3-legged linear regulators.

That's what I read also. I want to see how bad it is.
However, the load regulation is much better with the {RF} feedback resistors before the ballast resistors than after.
Since they are not in the feedback loop, the only effect they have on load regulation is the IR voltage drop.
So it's a tradeoff between keeping them small due to that, that and making them large enough to balance out the tolerance variations of the regs.
(the stated aim of this experiment was more for power capability than precision...)

...beginners get this idea ...

Electronics has been my hobby for well over 50 years - hardly a beginner!

[ocset]

i think its fine to parallel linear regs with the current share  resistoring that you describe.

another way to do it is  to have one linear volt reg as the  master, and then lots of parallel slave current regs which copy the masters output current......so the masters output current is sensed, then you use opamps etc to get the pllel lin currnt regs to copy the master current.

so its like in the attached, but for  lin regs, not smps's

[David Hess]

Linear Technology has a warning about not paralleling LM series regulators because their greater voltage tolerance requires too much ballast resistance to yield good load regulation.

Current sharing can also be enforced using operational amplifiers and current shunts as shown in the National LM338 datasheet.  Note that the LM107/LM307 is an LM301A with internal compensation so it has a common mode input range which includes the positive supply.

For those who are not familiar with them, the LM350 is a 3 amp LM317 and the LM338 is a 5 amp LM317.

However, the load regulation is much better with the {RF} feedback resistors before the ballast resistors than after.
Since they are not in the feedback loop, the only effect they have on load regulation is the IR voltage drop.

This is not to be underestimated.  The top of the feedback network should be attached as close to the regulator as possible.  The bottom may be located at the load.

[duak]

Maybe another way to think of this is to consider the 3 terminal adjustable regulator as a high current gain transistor with integral thermal current limiting.  Vbe is about the same as a darlington and is almost insensitive  to temperature.  The "base" current is also pretty much constant and flows the other way.  Current division could be performed by ballasting resistors.  It would need an opamp or two and maybe a reference to compensate for the ballasting resistors.

I've been working with hp lab supplies where the regulator is floating and the output voltage is determined by a reference current passing thru a variable resistance to the common terminal - very much like adjustable 3 terminal regulators.  These supplies are far easier to parallel than a standard non-floating series regulator.  Food for thought, eh?

[Kleinstein]

Maybe another way to think of this is to consider the 3 terminal adjustable regulator as a high current gain transistor with integral thermal current limiting.  Vbe is about the same as a darlington and is almost insensitive  to temperature.  The "base" current is also pretty much constant and flows the other way.  Current division could be performed by ballasting resistors.  It would need an opamp or two and maybe a reference to compensate for the ballasting resistors.

I've been working with hp lab supplies where the regulator is floating and the output voltage is determined by a reference current passing thru a variable resistance to the common terminal - very much like adjustable 3 terminal regulators.  These supplies are far easier to parallel than a standard non-floating series regulator.  Food for thought, eh?

The sensible idea of using regulators like LM317 or similar as the output stage of an lab supply is to treat it as a kind of special transistor.

Adjusting the output voltage in the scheme of reference current and variable resistor may not be the best solution. In many such old implementations changing the resistor also effects the loop gain and this way makes has a negative effect on regulation (often slow at high voltage and possibly oscillating at low voltage).  So this is a detail that is better not copied. Otherwise the floating regulator is a good choice, as it allows for a low dropout and an easy way to use 2 or more transformer taps to lower the worst case power loss.

[David Hess]

Maybe another way to think of this is to consider the 3 terminal adjustable regulator as a high current gain transistor with integral thermal current limiting.  Vbe is about the same as a darlington and is almost insensitive  to temperature.  The "base" current is also pretty much constant and flows the other way.  Current division could be performed by ballasting resistors.  It would need an opamp or two and maybe a reference to compensate for the ballasting resistors.

An LM395 "Ultra Reliable Power Transistor" would be more suitable but they are also more expensive.

I've been working with hp lab supplies where the regulator is floating and the output voltage is determined by a reference current passing thru a variable resistance to the common terminal - very much like adjustable 3 terminal regulators.  These supplies are far easier to parallel than a standard non-floating series regulator.

LM317s are floating regulators which can be used at high voltages but be careful that their maximum input to output voltage is not exceeded during current limiting or thermal limiting which will destroy them.

Adjusting the output voltage in the scheme of reference current and variable resistor may not be the best solution. In many such old implementations changing the resistor also effects the loop gain and this way makes has a negative effect on regulation (often slow at high voltage and possibly oscillating at low voltage).  So this is a detail that is better not copied.

The bottom resistor can be bypassed to ground decreasing closed loop gain at high frequencies so the frequency compensation does not change with output voltage.

[intabits]

Made a test jig with 3x LM350:-



The schematic:-



Testing with 30V input, using halogen bulbs as a dummy load. The bulbs also function as illumination for the shitty displays on HP3468As.
Current was 7.5A (on analog meter), Top HP shows input voltage (29.9654V), bottom HP is output voltage.


I tried to get the lowest I/O differential to keep dissipation down, but 4.3V is the best I could do, though that does include 1.3V across the ballast resistors (0.51R).
The voltage across these resistors was within 5mV on all three, indicating very good current sharing.
But 0.51R is way too large - dissipation is about 1.3W on each, and they got very hot. The regulators were also quite hot (maybe a little came from the resistors), so better heatsinking is required.

With the output resistors disconnected, the regulator outputs (now isolated) were within a few mV of each other (no load), rising to within 20mV at 800mA load. So the tolerances on these seem to be very tight.

Next things to try:-
* Lower value ballast resistors
* Check on scope for any oscillations or other unwanted behavior
* Other loads:
  capacitive & inductive (motor?)
  pulsed: to see transient handling & recovery   
* Try 5 regulators for 12-15A capacity
* Look at a better heatsink than a bit of angle ally (and use thermal grease)
* Try using LT1084 regulators

So at first sniff, it looks like this can be done after all, though more testing is required.
Efficiency might be an issue though, but with lower ballast resistors, the drop across the regulators could come down to more like 3v, which is not much more than what any pre-regulated supply would be aiming for.

[Siwastaja]

The issue with this is that you are depending too much on luck.

Most 3-legged regs specify the output tolerance at +/- 2% or worse, some even +/-5% (such as certain brands of 78xx).

Of course, by luck, a batch of 3 regs might be very closely matched, and work with fairly small ballast resistors, at least as long as you thermally couple the devices, which is not difficult to do.

But, you'd need to handpick the regulators if you want to keep the ballast resistors low-value.

I fell in this trap once; did a prototype with 3 regulators which actually looked just like yours. Worked fine. Then I added a fourth device, and one of the regulators was never taking almost any current! I then needed to increase the ballast resistors too much for it to be usable, and still one of the regulators wasn't taking as much load as the others, which is a waste of silicon.

This doesn't mean this is impossible to do. I can see it works in some cases. But, most often, it's a disappointment.

[james_s]

When I was a kid I built a power supply with 3x 7812 regulators in parallel, it worked fine and I used it for many years. YMMV though, this is definitely not recommended practice but if it's for a personal project and not a commercial product and you can make it work, go for it.

[ocset]

what you could also try would be making your own linear regulators out of transistors....then drive all their gates/bases with a single error amplifier output...that error amplifier "looking" at the single shared output voltage....this way, you dont get feedback  loop problems (feedback loops fighting each other) because you only have one feedback loop.........and you  have the transistor regulators in parallel with each  other....as you know you still need the sharing resistors.......unless you do the "master and slave" thing that i describe above in my previous post....

You can of course just put a current clamp on each of them so they can't send more than a certain current...........this way none of them can hog all the current.

[IDEngineer]

In addition to everything that's been said above: Remember that an ideal constant voltage source has infinitely low output impedance. Yet here we are intentionally inserting series resistors to increase the output impedance....

[james_s]

I didn't use any resistors when I made my power supply, I just connected the regulators in parallel and it worked for years. I never tried it again though so I don't know how repeatable that result is. Lots of things you can get away with when it's a one-off hobby project.

[intabits]

Yes, the tight tolerance is probably mostly down to luck. But then, selecting matching parts is no big problem.

As I've said all along, this is just for fun, and to see what's possible.
I do intend to develop a "proper" regulator using op-amps and transistors - In fact, that's how I got here.

Before working on a proper post-regulator, I first need a pre-regulator for it, and I want to experiment with various ideas to do that.
But to develop the pre-regulator, I would also need a "dummy" post-regulator to test it with.
Which is what this is really for, if I can get it to work well enough for that purpose.
Just a cheap and cheerful aid to testing, so all the possible drawbacks that it may have, are not really an issue.
Efficiency is not very important, it can just use fat resistors, as long as it works half reasonably and doesn't explode.

That lovely 500VA toroid transformer is too good for this thing, it's reserved for the use in the "proper" setup.
And if this does turn out to be a half decent method, I have an older transformer/rectifier/filter front end that I could use it with as an interim measure while I continue working on the "good" one.

[james_s]

A few years ago I built a power supply using a circuit from a late 80s Radio Electronics (or Popular Electronics?) magazine, I even wound a custom power transformer for it. Anyway it uses the HV version of the LM317 along with a pair of big pass transistors on a common heatsink for each channel. 0-50V, 0-5A, the circuit is fairly similar to a lot of lab power supplies of the era.

[intabits]

... LM317 along with a pair of big pass transistors ...

Many circuits like that out there, and often in the data sheets.
That would be my next choice for a testing/dummy post regulator, but I think this parallel thing will do for now.

[Zero999]

This is not something I've tried to play with, as I've been told it's not good design.

As far as the topic of this thread is concerned: relying on simulator models for this kind of thing is silly, because it will produce better results than real life.

Linear Technology has a warning about not paralleling LM series regulators because their greater voltage tolerance requires too much ballast resistance to yield good load regulation.

Current sharing can also be enforced using operational amplifiers and current shunts as shown in the National LM338 datasheet.  Note that the LM107/LM307 is an LM301A with internal compensation so it has a common mode input range which includes the positive supply.

Is current sharing important? As long as the end result of a greater total output current, than with one regulator, is achieved, the goal will be accomplished.

What will happen of two 1A linear regulators are connected in parallel: one with and output voltage of 10.1V and the other with an output voltage of 9.9V?

Theoretically, the 10.1V one will pass all the current until it goes into current limiting, then the 9.9V regulator will start providing the extra current when the voltage drops to 9.9V. Surely the end result will be a 2A regulator.

Or will it cause instability in the feedback loop, when the 9.9V regulator sees a higher output voltage, than expected?

[Siwastaja]

Theoretically, the 10.1V one will pass all the current until it goes into current limiting, then the 9.9V regulator will start providing the extra current when the voltage drops to 9.9V. Surely the end result will be a 2A regulator.

Or will it cause instability in the feedback loop, when the 9.9V regulator sees a higher output voltage, than expected?

Even though non-optimal, this works without issues as long as all the regulators are actually designed to provide unlimited time current limit, in "true" CC mode. With most 3-legged regulators, this is not the case: they are designed to work in CV only, and their current limit is very crude, and is set far higher than the actual specified safe maximum continuous current, so even with proper cooling, they tend to start throttling into thermal protection. People have also widely reported reliability issues when running against thermal protection continuously.

[David Hess]

What will happen of two 1A linear regulators are connected in parallel: one with and output voltage of 10.1V and the other with an output voltage of 9.9V?

The output current from each regulator will depend on the resistance between their outputs.

Is current sharing important? As long as the end result of a greater total output current, than with one regulator, is achieved, the goal will be accomplished.

Or will it cause instability in the feedback loop, when the 9.9V regulator sees a higher output voltage, than expected?

I would not expect it to be unstable exactly but reliability will be compromised if one is current limiting due to excessive junction temperature.

[bson]

It works in simulation because all the individual regulators are exactly identical, and all the Rf/Rc's are exactly identical.  It's a special condition that won't occur in reality.  A better simulation would be to set the ADJ pins to slightly different values and make sure they balance correctly, by using different values for Rf representing a 1% spread.  Plot the output currents next to one another and see how high you can get before one of the regulators is overloaded.