TL431 + LM317 + PNP = confuzion

[Xymox]

OK I am now confuzed... I have run across a linear power supply with a circuit I am having trouble understanding.. With no load it has the TL431 sawtooth instability and about 100mV of it. With any load above about 75mA it changes to a zener like wideband random noise of about 3mV pk-pk and this noise does not vary with load. This odd behavior caused me to look at the circuit..

Why the high values for the TL431 adj side ? Does this give the TL431 gain ? Why ?

WHy is there a LIM317 hooked to the PNP ? WHy not just drive a darlington output and skip the LM 317. Is the 317 a 2nd stage of regulation and can then drive a std PNP ? Note the PNP is kinda a odd choice of part.

WHY do this design ? See schematic..

 

[betwixt]

Except for the obvious graphic error, I think it will.
I'm not sure there is any advantage to using the TL431 at all but it will still regulate. In fact if the TL431 is used, a fixed voltage regulator IC like a 7805 could be used instead of the LM317 with the 'ground' pin sitting on a variable pedestal.

Referring to post #18, all the current through the LM317 passes through the 10 Ohm resistor until the voltage dropped across it is high enough to bias the transistor into conduction, from then on it passes the remainder of the load current. That configuration has the advantage that if the LM317 shuts down, for example by overheating, it also shuts down the current path through the transistor.

Personally, I think adding the TL431 is overkill as it duplicates the function of the LM317 but if people are prepared to pay extra for it....

Brian.

 

[KlausST]

Hi,

Personally, I think adding the TL431 is overkill as it duplicates the function of the LM317 but if people are prepared to pay extra for it....

I agree.
It will improve "accuracy"... but...
it will not improve noise, since the LM317 noise is added anyways
I´m not sure about stability...

Klaus

 

[danadakk]

Post 18 sim -

But stability is in question -

I ran at lower current as I thought LM317 going into current limit protection, all that did was
drop oscillation rate due to higher Rload.

Transient response (Rload transient) also not looking good -

All in all worth more investigation.


Regards, Dana.

 

[d123]

Hi,

Everyone got there before me about the current bypassing transistor and stuff about TL431. Interesting how both TL431 and LM317 internal reference have approx 50mV droop with temperature [same thing, one for reference, other for power = TL431 is Vref x (1 + (R1/R2)) + (Iref x R1) and LM317 is Vref x (1 + (R2/R1)) + (Iref x R2)].

Had a play on an approximation to the circuit for the benefit of xymox and out of personal curiosity... Important aspect is that LM317 model used appears to have no temperature drift (!), other than that is seems to work normally/credibly. I was unable to find a model of the heavy-duty -250V, -15A, -1.2V NJW0302G PNP and got sick of spending time looking for one so used a tiny -3A device instead... I found a suitable inductor value by trial and error (it oscillates a lot with other values but looks acceptable with the 100 uH, so that'll do seeing as we're guessing. The source resistance of VG1 (impedance is not available, only source resistance can be defined in the sim tool) is 1 Ohm.

Interestingly, even for an approximation, temperature run at -50ºC, +50ºC, +150ºC shows 'considerable' Vout droop (14.1V down to 13.77V, from memory). I'd wondered if there was some theory about TL431 and LM317 and maybe even the BJT Veb leading to temperature-compensated Vout, seems not here, maybe the respective TL431 and LM317 internal reference droops do cancel out in reality to keep Vout constant across temperature range.

Dana's comment about TL431 stability vs capacitance is spot-on, the graphs in the datasheet clearly show 100nF is a risky choice...

Ahhh (...what good and often embarrassingly bad memories), this takes me back several years to the 7805 being the only voltage regulator I had used, and I even think Crutschow helped me understand how to implement the more sophisticated version:

As we all know, I'm not great at this stuff and have lagoons in my knowledge, but IMO, the circuit looks like 'a bit naïve, cheap and crummy, e.g. down to skipping 2 cents-worth of regulator protection diodes, etc., but who cares because the end user will never know', which is odd given the wasteful use of the TL431 which does seem to be intended to improve Vout regulation (the TL43 - which considered globally, is about as accurate as the LM317 internal reference anyway...). Frankly, I could have made this, which is patently no endorsement of the quality... LM317 datasheet has several schematics with a voltage reference on the adjust pin, by the way.

Xymox, I really do recommend suffering the learning curve of becoming well-versed in using LTSpice or any other simulation tool, you will actually save a lot of time in the long run, plus you can look at a wide range of signals especially under different conditions that it is impossible to reproduce at home, for example.

And here are the pertinent datasheets, for anyone interested:

 

[danadakk]

The Rload transient case w/o TL431, 1uF (dotted line) and 1 mF (solid line) Cload -

Regards, Dana.

 

[Xymox]

O M G......

You guys are AWESOME !!!!!!! EVERYBODY who worked on this THANK YOU !!!!!

VERY interesting.. SPICE is showing what i see in real life.. Its unstable with various kinds of loads..

Yes indeed, I need to learn SPICE.. The above is GREAT example as to why. But it really looks like its experence that counts. For example, I know AutoCAD. AutoCAD can be used by a novice, but, its so deep that it takes decades of daily experence to get good at and even then you never really "master" it and there is always something new to learn or new ways to use it. AutoCAD is best left to people who live and breathe it for thier whole lives. WHile I could learn SPICE, it might do me more harm then good as not being a expert with it I might spend lots of time with it and come to incorrect conclusions. PLUS modeling only gets you so far.. BUT.. I REALLY want to put SPICE on my tool belt.. I will look at it..

Man what you guys did was amazing.. I have some distracting meaningless work dribble I need to do today but I will swing back to this far more interesting project this evening..

I already made some mods to this thing playing around with it.. I pulled the inductor. It made things more unstable. I put a .68 uf cap on the TL431, which did help noise but its still in the instability zone. I ordered up some WIMAs in larger sizes and will fool with these.

I do want to test its regulation in more depth.

So.. Why is it using those large values on the TL431 for the adj ? Everything I have seen use far lower values ?

> stupid question alert < To get sensing, could I take both ends of the divider on the TL431 and extend those outward ?

Wow.. Thansk again for the SPICE modeling.. Wowee.. So cool. I need to look at it more later today.

--- Updated Jun 13, 2022 ---

So in short, there is NOTHING good about this circuit ? No trade off where some good parameter is maxed at the expense of others ?

--- Updated Jun 13, 2022 ---

Hi,

Everyone got there before me about the current bypassing transistor and stuff about TL431. Interesting how both TL431 and LM317 internal reference have approx 50mV droop with temperature [same thing, one for reference, other for power = TL431 is Vref x (1 + (R1/R2)) + (Iref x R1) and LM317 is Vref x (1 + (R2/R1)) + (Iref x R2)].

Had a play on an approximation to the circuit for the benefit of xymox and out of personal curiosity... Important aspect is that LM317 model used appears to have no temperature drift (!), other than that is seems to work normally/credibly. I was unable to find a model of the heavy-duty -250V, -15A, -1.2V NJW0302G PNP and got sick of spending time looking for one so used a tiny -3A device instead... I found a suitable inductor value by trial and error (it oscillates a lot with other values but looks acceptable with the 100 uH, so that'll do seeing as we're guessing. The source resistance of VG1 (impedance is not available, only source resistance can be defined in the sim tool) is 1 Ohm.

View attachment 176816

Interestingly, even for an approximation, temperature run at -50ºC, +50ºC, +150ºC shows 'considerable' Vout droop (14.1V down to 13.77V, from memory). I'd wondered if there was some theory about TL431 and LM317 and maybe even the BJT Veb leading to temperature-compensated Vout, seems not here, maybe the respective TL431 and LM317 internal reference droops do cancel out in reality to keep Vout constant across temperature range.

Dana's comment about TL431 stability vs capacitance is spot-on, the graphs in the datasheet clearly show 100nF is a risky choice...

Ahhh (...what good and often embarrassingly bad memories), this takes me back several years to the 7805 being the only voltage regulator I had used, and I even think Crutschow helped me understand how to implement the more sophisticated version:

View attachment 176819

As we all know, I'm not great at this stuff and have lagoons in my knowledge, but IMO, the circuit looks like 'a bit naïve, cheap and crummy, e.g. down to skipping 2 cents-worth of regulator protection diodes, etc., but who cares because the end user will never know', which is odd given the wasteful use of the TL431 which does seem to be intended to improve Vout regulation (the TL43 - which considered globally, is about as accurate as the LM317 internal reference anyway...). Frankly, I could have made this, which is patently no endorsement of the quality... LM317 datasheet has several schematics with a voltage reference on the adjust pin, by the way.

Xymox, I really do recommend suffering the learning curve of becoming well-versed in using LTSpice or any other simulation tool, you will actually save a lot of time in the long run, plus you can look at a wide range of signals especially under different conditions that it is impossible to reproduce at home, for example.

And here are the pertinent datasheets, for anyone interested:

Thank you D123, That is just awesome work...

 

[KlausST]

So in short, there is NOTHING good about this circuit ?

Sadly I´ve seen so many crap - especially for (or from so called) audio enthusiasts.
Often the "designer" have no idea in how to design and calculate really high quality electronics.
This circuit seems to me of no good qualtiy. I´d be interested in a "technical" discussion with the designer.

But for sure there are also good ones.

But it´s almost impossible for the customers to distinguish the good from the bad.

Klaus