Hero999

No I haven't. I reasoned that it probably wouln't help because the positive line is just as bad as the negitive.

I don't know how I can improve it anymore. I've tried adding 100nf to the output it it didn't get much better to I increased this to 1µf without seeing a significant improvement. I suppose I might be able make the internal leads connecting the PCB to the output shorter and thicker to reduce their inductance and skin effect but I don't know what else can be done.

__________________
What's so bad about Microsoft?

Get Opera it's simply a superb browser.

Roff

I haven't read "Spice uncovers regulator-stability problems" carefully, but I have seen ringing on LM317 sims. I have not made hardware comparisons to verify the sim results. I thought you might find this interesting.

__________________
Ron

eng1

Even if the LM317 is rated at 1.5 A output current, it can source up to 2.2 A. Beyond this limit the internal protection circuit limits the current. Also.. the chip may be turned off by the thermal protection circuitry, if its temperature rises above 170°C.

I remember that article... If the model of the LM317 doesn't cover stability correctly, you may see ringing on a scope, even if you didn't on the simulator.
But if you see ringing on the simulator...!?!

Roff

Well, it's possible that the spice model could be wrong about phase margin (stability) in either direction. You're probably right in thinking that it would more likely err on the side of being more stable vs less.

__________________
Ron

Hero999

The datasheet says the LM317 will current limit between 1.5A and 3.4A with 2.2A being typical so I suppose I have a pretty average LM317.

It seems silly that the spice models say it's more stable than it is. I would make it the other way round so when engineers actually use it they know it'll work when the simulation says it will.

I'll try using a 4.7nf capacitor on the adj pin and it the, then I'll see what I can do about the negitive supply.

__________________
What's so bad about Microsoft?

Get Opera it's simply a superb browser.

eng1

Anyway the LM317 wouldn't be a problem without the op-amp.
You have to see if an RC network can be found to shape the loop gain (you need a zero!).
If you take my advice, you may try to re-design the PSU with two op-amps and a voltage reference. I gave this direction to my ideas and I'm designing the circuit. But I need a new transformer and it will take some time.

______
See also AN-1148 (National): nice app note about linear regulators.

Roff

I didn't put a zero in the loop, and it worked fine. The pole just needs to be low enough frequency to ensure that the loop gain is below unity when the second pole in the loop (from the regulator) cuts in. The regulator/follower has very low output resistance, so it takes care of the transient response. The op amp just provides low frequency tracking. In fact, the feedback loop should be slow enough that it does not follow transients on the "master" supply.
The big advantage of Hero's scheme is the use of the integrated regulator as the series pass element in the feedback loop. It is virtually bullet-proof. Duplicating this robustness with discrete parts requires a LOT of parts and a lot of smarts.

__________________
Ron

eng1

Of course the regulators will be part of the PSU, but the Vadj can be generated by two op-amps used in a close loop configuration (offsets, mismatches,... must be properly evaluated / compensated).
This is a valid method and ensures good tracking.

eng1

In my opinion there are two l.f. poles: the first is set by the internal compensation (Cc); the second is from the op-amp. The loop gain should roll off with -40 dB/decade slope in the mid freq. range. Unfortunately I can't simulate the circuit.

Hero999

But I am having the problem with the LM217 even though it isn't connected in the feedback loop of an op-amp.

What you are forgetting is that if there are two low frequency poles there is double the phase shift, therefore 180 degrees will occur well before the op-amp reaches unity gain causing oscillation.

__________________
What's so bad about Microsoft?

Get Opera it's simply a superb browser.

eng1

Hi Hero999. If I was forgetting that, why did I write that you need a zero then?

What do you mean exactly?

Hero999

The transient response it just as poor on the positive supply as the negitive, even though it isn't using an op-amp.

__________________
What's so bad about Microsoft?

Get Opera it's simply a superb browser.

Roff

Your schematic doesn't have 100nF caps from the input pins to GND. These are recommended by some (all?) manufacturers for stability. Your big polarized caps don't satisfy this requirement.

__________________
Ron

FMadi

Hi,

As Ron H pointed out you really should add those input capacitors, last year I was working in the tech department of a medical device company here in Brazil, and we were having similar problems with a power supply that used 7805, and adding those small tantalum caps, (before and after) the regulators really helped with the ripple figures, here is a quote from the application notes on the datasheet from national:
“An input bypass capacitor is recommended. A 0.1µF disc or
1µF solid tantalum on the input is suitable input bypassing
for almost all applications. The device is more sensitive to
the absence of input bypassing when adjustment or output
capacitors are used but the above values will eliminate the
possibility of problems.”

What type of capacitor did you use for the adjustment pin bypass?

Can you measure the ripple frequency at the output?

This is my first post here, hope i could help, and sorry for any grammar mistakes, English is not my native language.

__________________
Felipe Madi.

Roff

Wow, Felipe! Your English is better than most of our English-as-first-language contributors.

__________________
Ron