VRDN: bipolar regulator PCB for line level ckts: ±11V to ±20V @ 1.5A with "De-Noiser"

[Mark Johnson]

6.8 nF ---> 22 nF is not a gigantic change; just 10dB . It may suggest that your VRDN was just barely over the line and into "you must oscillate" territory just a wee bit. Very pleased to hear that it's now doing what you want it to do. 33nF would probably be safer still but why fix it some more if it ain't broke no longer.

Maybe this will provoke a prolonged "okay buy WHY??!?" discussion since some but not all boards oscillated.

[Elvee]

The 317 and 337 have been designed to look like mirror images of one another, but that's not the case, and the 337 is in fact the evil twin of the 317: it is more demanding, more touchy, and its structure is different. It requires a higher no-load current, and it is known to oscillate in some legit situations, when it shouldn't.

[jean-paul]

A known fact since decades. The early 79xx also had this. It depends on brand too for a part. That is why the power LED usually was connected to the LM337. Omitting the input cap lead to oscillation in many cases just like an undefined state with strange voltages at startup (or even complete latchup). There are negative regulators that don’t exhibit this phenomenon I also recall shortlived adjustable versions of 78xx and 79xx both having these issues.

I was taught then to avoid LM337 (NOT LM317) and use simple discrete circuits or the forgotten XR4195 when a symmetric PSU was needed to supply opamps. Thankfully progress gave better negative regulators.

[Elvee]

Originally Posted by Mark Johnson

So the next thing I did was perform some simulations. I found that the circuit became EVEN MORE STABLE in simulation, if the series RC compensation network was changed. Instead of (6.8 nanofarads SERIES 15 ohms), I found that (33 nanofarads SERIES 1 ohm) was even more stable.....

Mark, since you are busy revising stability issues, I think you could have a look at the output series resistor (R22).
I have the feeling that, combined with the ESR of the cap, it could result in a largish value, which would be problematic if a new mask revision of the 337 has a larger loop gain than earlier versions.
A large value has the advantage that the regulator will tolerate otherwise insane combinations of external output caps (something some members love to indulge in), because it will damp almost anything, but when the regulator is on its own, it could cause instabilities.

Note that having symmetrical components for the - and + sections may look intellectually and aesthetically speaking satisfying, but since the 317 and 337 are in fact very different different regulators, there is no contraindication opting for different values.
Overcompensating the 317 will do no harm, but it will reduce the available correction gain at high frequencies, and reduce the effectiveness of the denoiser at those frequencies.
10nF is reasonable tradeoff, and for the 337, 22nF should be OK, maybe with a revision of the OP cap resistor

[thompsontechs]

Found this in the datasheet for the LM337.

john-paul and elvee, thanks for your input.

Mark, thanks for your continued work and not just blowing us off on this, no need to pay for parts, you do enough already.

[thompsontechs]

I was curious as to how close the 22n was to oscillation, so I soldered tacked in a few other values to see. 10,12,14,16,18, all still had issues though not as bad. It looks like 20n if finally becomes stable, but I did see a strange swing as I increased voltage as it it dropped once .5v and then went back up. I put the 22n in and it is stable, but I think that a bit to close, so I am going to go ahead go with a 33n and make the change to R14 as well. I'd rather have a larger safety margin than 2n.

[thompsontechs]

Update: Okay, unless the change in R14 to 1R was a typo, that brings back the oscillation. I get the proper output, but it a nasty looking waveform at Vdj. I went back to 15R and 33nf and all is well. I think I'll stop there....

[Mark Johnson]

Originally Posted by thompsontechs

A pic of the waveform and also the schematic with the yellow being where the AC is and the red OK designating that there is no AC. The IN of regulator is also clean.

They didn't have the 863-LM337TG regulator, so I used LM337KCSE3

I am starting to think: that might be significant. All of the VRDN boards I have built and tested, used LM317s pulled from the bag shown below. None of those boards oscillated, not even with the unmodified (6.8nF SERIES 15R) compensation network. I just tested the one I still have, and it doesn't oscillate with (6.8nF SERIES 15R). Not when load current = zero , also not when load current = 50 milliamps.

Although LM337KSCE3 is out of stock at Mouser, I found some at DigiKey and ordered them immediately. I also ordered some of the KSCE3 positive regulators, at the same time. When they arrive I'll build another VRDN and see whether I can get it to oscillate.

It might be enough to scare you away from simulation completely, since you have no idea whether the simulation macromodel you've got in hand, is for the manufacturer's chip that you bought, or for the one you didn't buy.

_

[thompsontechs]

Let's see what your results are, if you still can't verify, I can ship you mine, as I already have TGs on the way because I was convinced this was the issue. I know I didn't screw up or change anything, but that part, so I'm going with Occam.

The P never had an issue and functions flawless. FYI

Let me know what you find....

JT

[amandarae]

Hi,
Will the LM337BTG work as a suitable replacement for the 863-LM337TG?
https://www.mouser.com/ProductDetail...KxF%2F1w%3D%3D

Datasheet of the "BTG" https://www.mouser.com/datasheet/2/3..._D-1773565.pdf
seems to be identical to the "TG". The only difference I can see is the Operating Temp Range (-40 to 125 deg. C for the BTG, and 0 to +125 deg. C for the TG).
The BTG is available (1,271 pieces last time I checked) in Mouser.
Thank you!
Abe