LM317T - Stability under bypassed load

[TheFloyd]

 
Hi Everyone, first post here, so sorry if i have done it wrong, or put it in the wrong spot.   Sadly i have hit the frustrating point where that fun project has become a drag, and so much time has gone into it that it either needs to be right, or not at all.   (we've all been there )

so, of the couple of most common current bypass circuits for the 317T, that are out there, i decided not to use the BD139 on the Reg output, driving the stack of 3055's this time, and adopt the other most common approach of putting a PNP on the input of the 317, that then drives my TIP3055's to do the heavy lifting.   The image attached has values for R1,2,3,4,5 that I have been experimenting with, in an effort to solve the issue i am experiencing, but i am at a point now where i am wondering if i am trying to change unavoidable operation.
Many designs with different pnp/npn transistors have values for R1 as 22ohm or 15ohm, and R2 as anything from 150ohm to 5K.  most have R3 as 470 or 500 etc, so please just interpret what is in this snapshot of today's diagrams as the product of a lot of testing that yields the best solution so far. (if you have advice that shows my values to be silly, i would love to learn why, and how to correctly calculate them)

The Issue...  I have all these TIP3055's appropriately hooked up with their .1ohm 10W resistors to ensure they behave, because i want to build a number of these circuits as high-current voltage sources, with a set and forget voltage trim-pot.    The issue is cant seem to overcome is that under load, the supply doesnt maintain voltage.   I need this to easily deliver 10A at 12V, but with the values from the designs i found online, i found that if you dropped a 5A load across it, the Vout instantly dropped to a few volts. initially this was really easy to improve, as Q1 was totally bias wrong and an adjustment of R1 up from 22ohms,  R2 down to 120ohms and R3 down to 220ohms, meant that at 5A, the voltage drop was small.   But there still was a drop.  My issue is that i have a nice digital display for the 12V 10A supply, and it is disappointing to see it drop to 11.7v as soon as even 5A is pulled out of it.
I thought that the point of adding that feedback link from the power supply output, to the 317T output, would help the device regulate the voltage, but as soon as i put a 10A load across it, the Vout drops to about 11.2v (or less).     My next thought was that perhaps i need the feedback connection to come directly off the 3055's Emitter, rather than after the .1ohm resistor, but it has been too long between drinks on the maths, im afraid... (please help)

one thing to absolutely confirm is the input supply.   it has a honking-great transformer and about 60,000uF of low ESR filter caps feeding a Vcc, that never drops under about 17volts at the bottom of the remaining ripple.


so my questions is:   i really dont want to argue the fact that pretty much all other power supplies are better than these old beginner ones, there is something magical about these circuits, and i would dearly like to know if there are ways i can improve this (or the other design i mentioned at the start).    Is it unreasonable the think i should be able to get this to dish out 12v, until 10A when the fuse pops ?

nothing would make me happier than someone pasting in some formulas, calculations, explanation and suggestions on what i need to change to make this little project back into the joy it started off as!

thanks in advance!
Floyd.

[xavier60]

Because the value of R4 causes the LM317 to draw ~10mA which satisfies it's minimum current draw specification, the value of R1 is too high to prevent Q1 from being turned on before it's needed to be. I would try 10Ω for R1 and R2 as a starting point, then look for what other trouble there might be. 

[TheFloyd]

Thanks Xavier, I really appreciate how quick you jumped in and gave me a pointer on this.
I have already rebuilt it all out on a breadboard and made the following amendments:
R1=10Ω
R2=10Ω
R3=470Ω

Interesting points of note:
- R3 is pretty much always 470Ω in examples of this circuit (or similar).   so i had it in my diagram as 220Ω, simply because i am struggling to work out the appropriate calculation to tune this value correctly, given i am considering its function as part of the collector-base biasing.  as such i was tuning it with practical experimentation where i found its value in my circuit seemed to provide the most output stability at 220Ω.   for the sake of this discussion i have reset it back to the conventional wisdom value 470Ω.

- the first value i changed was R1=10Ω, and it immediate reproduced the initial and substantive issue, where Vreg would output 12v, and under 5A load, the output would drop to millivolts. (this is with R2=120Ω).     As you suggested, dropping that R2 value to 10Ω, immediately satisfied the 317's minimum current of 10mA and it made a more reasonable attempt at voltage regulation.

unfortunately, it has the same issue.   As this is an issue regarding regulation under load, i am no longer bothering to tune the pot as i change values, so i just record the values for Vreg output under no load than then under load.    with the changes above, the output variance was 12.39v@0A -> 11.27v@5A.

IMSAI Guy does a good video on youtube, and blows a lot of dust off my understanding of these circuits, as do a few others.    Even though there a countless examples of this circuit online to model (like this one, that i found to behave like mine : https://www.eleccircuit.com/best-dc-power-supply-3amp-to-adjust-1-2v-20v-3v-6v-9v-12v/ ), I am yet to find any clear and detailed explanation of the maths under this design.    I do understand the conceptual functions of R1 in this context, and am pretty solid on the use of a 317, but the functional and calculated relationship between R1, R2 and R3, is an area that i would love expert guidance!


Genuine question:   should i be able to correctly bias T1, such that my 317 can give me a rock solid Vout when under different loads?     I hand on heart ask this, because of the million times i have built 317 regulated circuits, they have either been fixed voltage @ under an amp, and as such, have not used any current bypass, or they have been variable supplies, typically of the design that has the Vreg output driving the base of NPN's.   in these instances, as i loaded up the output, i would adjust the voltage with R5, to keep it constant. When i came to build this project, i read a lot of forums where people commented that this design prevented the 317 from effectively regulating, so i thought the one i am experimenting with now should perform better.   so far, it yields the same behavior.

[TheFloyd]

A little more context, i have added a bigger view of the project.
this is to be the mother of all benchtop supplies, with everything from a handy variable voltage and current supply, and then two rock solid fixed supplies of 12v and 5v.   the front panel is kitted out with USB charging ports, Anderson plugs for integration into solar/battery boxes, and even modular power sockets that are lifted from an old corsair PC PSU, so that i can power PC hardware on my bench, like HDD's etc.

the best 317 circuit i have had success in terms of adding in the ability to variably limit output current  is circuit A.    This is actually extremely like the way that i have always made bypassed 317-based variable supplies, only it has that pot across the base of the BD139.  seems ok i guess, however i find that this circuit still needs some tuning, as it seems to limit the current when fully open, to less than i know it can deliver.    but i'll tune that.     This is the current bypass style that i have heard, impacts the 317's ability to regulate properly, but dont fully understand the details of why.

The issue at hand is my tussle with Circuit B.   I have only shown one instance of this, but there are two of them, with the other tuned for 5V@10A.

[Vovk_Z]

That circuit (B) is almost a datasheet example, so made of fine parts has to work fine. You may measure voltages at different points with and without load and it may give you a clue. (Possibly Q1 is not ok).

[xavier60]

The idea of the the Bypass design is that initially the LM317 only supplies output current. At about 60mA, the 0.6V drop across R1 begins to turn on Q1 which in turn drives the output transistors, suppling load current. The values of R1 and R2 are not critical, they just affect the current at which the bypass action begins.

As load current is increased, the LM317 should not need to supply much more than the 60mA. This can be checked by monitoring the voltage drop across R1.

The circuit could be oscillating, check with an oscilloscope if possible. I would remove the 10uF capacitors and shift the 100uF to the left side of R1.

Due to non-ideal layout, the LM317 might not be seeing the actual output voltage. All it cares about is the ~1.25V across R4.
The LM317 will do its best to regulate the voltage between its output pin and the bottom end of R5, as measured at R5.

[TheFloyd]

I agree Vovk_z, which is one of the main reasons i wanted to change my historical approach to doing this, over to the one that is in the ti.com datasheet.    Other than it using 2N2905 and TIP73, it is fundamentally the same, and is the reason that I started off with a 5KΩ as the base resistor.   

The one big difference is the that there is only a single NPN bypass transistor, and as such there is no need for the 0.1Ω current limiters on the emitter.     This has me wondering if it is because that there is no direct connection between the TIP3055 Emitters and the 317 Output pin that is causing this problem under load.    at around 5A, there is about 300-500mV dropped across those 0.1Ω emitter resistors (varies between them depending on the current each transistor is passing i guess), which makes me wonder if it is this voltage drop that is pulling down the regulator's output when under load.    if indeed this is the issue, are there any examples of how i can connect the emitters of the 3055's to the regulator's output differently?     

[TheFloyd]

Awesome.  thanks for this.   it is super helpful, and firms-up my understanding.

couple of direct responses:

The idea of the the Bypass design is that initially the LM317 only supplies output current. At about 60mA, the 0.6V drop across R1 begins to turn on Q1 which in turn drives the output transistors, suppling load current. The values of R1 and R2 are not critical, they just affect the current at which the bypass action begins.

R1 explanation is my understanding.  thanks for confirming this operation.   if i was to set R1 as 22Ω, is there a formula for calculating R2?

As load current is increased, the LM317 should not need to supply much more than the 60mA. This can be checked by monitoring the voltage drop across R1.

Thanks.   appreciate the tip.

The circuit could be oscillating, check with an oscilloscope if possible. I would remove the 10uF capacitors and shift the 100uF to the left side of R1.

I had wondered about this, as i have been stumped by that issue in years gone by.   I'll move that 100nF, good call, and pull the 10uF electrolytic.    I recall fixing an oscillation issue previously with a 330nF across the regulator output too.

Due to non-ideal layout, the LM317 might not be seeing the actual output voltage. All it cares about is the ~1.25V across R4.
The LM317 will do its best to regulate the voltage between its output pin and the bottom end of R5, as measured at R5.

the Vref  is pretty constant at around 1.25v (ish).


Also, fogot to confirm in my response to Vovk_z, I have a small pile of LM317T's and TIP32CG's, and can rule out them being defective.

[xavier60]

With  22Ω for R1, the current at which bypass starts would be ~0.6/22=27mA. R2 need not be any particular value and is kind of redundant. It's main function is supposed to be to protect Q1 from possible excessive Base current which is a lot for your transistor anyway.
Going too high with R2 adds to the regulator's over all drop-out voltage. It can stay at 10Ω.

Have you also measured the voltage between the LM317's output pin and the bottom of R5 while load current is being increased?

[PCB.Wiz]

As this is an issue regarding regulation under load, i am no longer bothering to tune the pot as i change values, so i just record the values for Vreg output under no load than then under load.    with the changes above, the output variance was 12.39v@0A -> 11.27v@5A.

That shounds like instability.
As above, you need to carefully check stability with any 'added external gain' circuit.
A good scope and a radio, and impulse / ramping loads helps test things.

I agree Vovk_z, which is one of the main reasons i wanted to change my historical approach to doing this, over to the one that is in the ti.com datasheet.    Other than it using 2N2905 and TIP73, it is fundamentally the same, and is the reason that I started off with a 5KΩ as the base resistor. 

TI use 22R and 5K as they want to scrub  off excess external gain.
A lower R1 drops the V/mA gain, and a high R2 (like they have here of 5k), drops V/mA further and acts as a low pass element.
Good Vin decoupling matters a lot here as the source impedance becomes part of the load stability.

  if i was to set R1 as 22Ω, is there a formula for calculating R2?

R2 is less critical but higher values mean lower added-gain. Too high and you need more voltage to drive the base, and step response is slower.

Besides DC load tests you should check the step response of the regulator.
Besides carefully adding C, you can experiment using series R-C elements like 4R7 + 100nF or 2R2 + 470nF etc to dampen things down.
The added series R lowers the gain, without excess phase at highest frequency.

3055's are ancient parts, and do not have great HFE vs Current performance, but they are available almost anywhere.

[xavier60]

I mistook the input cap for 100uF, 100nF shouldn't cause a problem in its original position.
Actually because the LM317 has some direct control over the output, it will be trying to keep the whole thing stable.
Stability should still be confirmed.

[TheFloyd]

Thanks again xav and PCB.     have to say, as a first time experience in this forum, everyone is amazing and i cant thank you all enough!

so...   oscillation.    This was absolutely a factor, and not one that i thought would be so profound in terms of the regulator's ability to do its job.
moving the 100nF to the left of R1 (as it is actually shown in the datasheet from ti), immediately made a positive improvement, but i went 1 step further and put the 1uF tant there instead.    interestingly, i found that the output voltage under load lifted considerably if i added the 100nF back in between R1 and R2.  Sounds a lot like the RC suggestion from PCB.Wiz

the circuit as it looks now is depicted below, and here are the measurements i get under no load and under load.  What i am finding though is how inconsistent this operates between tests.   

@0A
output=11.97v
I(R1)=9.8mA
317
-Vo=11.93
-Vi=18.85
-Va=10.92
-Vref=1.236

32CG
-Vc=53mV
-Vb=18.75V

@5A
output=11.7v
I(R1)=53.2mA
317
-Vo=11.7
-Vi=16.06
-Va=10.66
-Vref=1.22

32CG
-Vc=11.93V
-Vb=15.92V

[PCB.Wiz]

Thanks again xav and PCB.     have to say, as a first time experience in this forum, everyone is amazing and i cant thank you all enough!

so...   oscillation.    This was absolutely a factor, and not one that i thought would be so profound in terms of the regulator's ability to do its job.
moving the 100nF to the left of R1 (as it is actually shown in the datasheet from ti), immediately made a positive improvement, but i went 1 step further and put the 1uF tant there instead.    interestingly, i found that the output voltage under load lifted considerably if i added the 100nF back in between R1 and R2.  Sounds a lot like the RC suggestion from PCB.Wiz

Adding caps (or fingers, or scope probes....) that then change voltage is a classic sign of oscillations.
Try a small series R in that 100nF

@0A
output=11.97v
32CG
-Vc=53mV

that 53mV does not make sense ?
The R3 470R should make Vc = output until it starts driving current.

The web also shows this, with PNP parts only

[TheFloyd]

I agree about that 53mV seeming wrong!
I have also had at the options to build this entirely by using PNP's like the example below, but i have already fabricated the enclosure and mounted all of the TIP3055's etc, so  the only reason i have not bailed on the design, is because i dont want to have to go and buy all new power transistors (again).

It is interesting that in your example, they dont include an "R2" at all on the base of the PNP's.   is this a point of significance to consider?

I did also try a series resistor with the 100nF.   didnt make a material difference unfortunately.

thanks again for the help with this!

[iMo]

Sorry to be a bit off topic - I've been running one of my PSUs with a 3A version of the LM317 for 25y already and I am pretty "unhappy" with it - as it will be killing any wrongly wired experimental constructions. I smoked many precious chips while using it. This type of PSU is good for fixed voltages with well defined loads (like stand alone stuff).
I would highly recommend you a power source with a max current setting, at least 2 positions as the minimum - like 50mA (smoke test) and full.

PS: to your wiring - with large currents you need perfect wiring to avoid instabilities. I mean you have to carefully consider how you actually solder the wiring - perfect grounding in "star" configuration, avoiding ground loops, do use short thick wires, etc. The schematics on the paper is one thing, the actual implementation the other..

[PCB.Wiz]

It is interesting that in your example, they dont include an "R2" at all on the base of the PNP's.   is this a point of significance to consider?

They do have emitter resistors, and in transfer/gain terms, you can multiply that by HFE and move to the base, as a virtual resistor.

Sorry to be a bit off topic - I've been running one of my PSUs with a 3A version of the LM317 for 25y already and I am pretty "unhappy" with it - as it will be killing any wrongly wired experimental constructions. I smoked many precious chips while using it. This type of PSU is good for fixed voltages with well defined loads (like stand alone stuff).
I would highly recommend you a power source with a max current setting, at least 2 positions as the minimum - like 50mA (smoke test) and full.

Not off topic at all, as that is the next step in the design, plus field use reports are always on topic 

Current limit is essential, and certainly some choices are also useful on the bench.

I was thinking about this issue yesterday :
The OP may be better to switch to a LM317L, otherwise the regulator limit itself will push things well above 1A, even with no booster elements.

Or, a *2931A family regulator is another candidate, suitable for 12V out, with boost transistor examples and multiple packages.


The external PNP and NPN stages can be limited themselves using series diodes and emitter resistors.

More modern core regulator parts like NCV47721, have adjustable current monitor and current set features.

Google finds PNP+NPN examples using old stalwarts like the uA723, and it is quite well suited to a 12V bench supply.
https://powersupply33.com/1v-27v-3a-variable-dc-power-supply-circuit-diagram.html

[David Hess]

Below is the old recommendation from National Semiconductor for boosting the output of a regulator.  The power diode at the input compensates for the PNP transistor's Vbe to enforce better current sharing between the regulator and booster transistors so that the regulator's short circuit protection also protects the transistors.  A 1N5400 would be suitable, or any 3+ amp silicon rectifier.  Multiple 2N3055s can be used in parallel with emitter ballasting.

The input capacitor is the original recommended value of 0.1 microfarads for regulator stability.  The output capacitor should probably be like 20 to 100 microfarads per amp.

Note the single point ground at the output is used for best regulation.

[AnalogTodd]

Stability is a big topic when it comes to linear regulators; if you have something unstable then all DC voltage targets go out the window. A 5V regulator that is unstable can have the output oscillating at a DC average of 2.1V and a 1Vp-p change in the output. Always look at the circuit with an oscilloscope to verify things are stable before trusting any DC measurement.

With that said, looking back at your original circuit and comparing it to the datasheet (page 17 of this data sheet: https://www.ti.com/lit/ds/symlink/lm317.pdf) shows that you have a 1uF tantalum capacitor on the output where the data sheet shows 47uF. The increased output capacitor in the data sheet is not necessarily surprising as the addition of the external transistors are giving a significant amount of current gain that can throw off the loop response of the LM317T, hence the need for additional capacitance at the output to change where the output pole comes in to stabilize the loop.

The LM317 will normally be able to function without a lot of input capacitance, with 22 ohms from IN to the bulk input capacitors is usually low enough impedance for the regulator to look into. BE CERTAIN YOU HAVE ENOUGH BULK INPUT CAPACITANCE!!! You have two circuits that you show in a post, one shows 60,000uF of bulk capacitance and the other shows none. You need to remember that the wires from your supply to this circuit will act as an inductor so if you don't have enough bulk input capacitor the double pole formed by the LC (or the output pole of the supply feeding your circuit) can be a high enough frequency that the LM317T cannot reject it. I have seen more than one circuit where the impedance at the input of the regulator input was either too high or a double-pole was formed that was close to or above the unity gain frequency of the regulator and the entire circuit oscillated.

Last, as one person said, understand where you are making your measurements. The LM317T wants to regulate from the OUT pin as close to the regulator as possible to the ADJ pin as close as possible. Wires and traces have parasitic resistance and can be significant when you are talking about several amps of load. The circuit David Hess showed adds Kelvin sensing for the ground side of the regulator to lower the parasitic resistance drop there as much as possible.

If you want to go back and discuss your circuit 'A', I will point out that it will have poor regulation across load as the Vbe of the transistors you have on the output will change significantly as you go from no load to full load (the formula for this is dVbe = V_T * ln ( dI_C ) where V_T is 26mV at 25C and is proportional to temperature in degrees Kelvin). At room temperature, a 10X difference in load from 10mA to 100mA is ~60mV in Vbe change. That neglects base resistance and other effects!

[David Hess]

With that said, looking back at your original circuit and comparing it to the datasheet (page 17 of this data sheet: https://www.ti.com/lit/ds/symlink/lm317.pdf) shows that you have a 1uF tantalum capacitor on the output where the data sheet shows 47uF. The increased output capacitor in the data sheet is not necessarily surprising as the addition of the external transistors are giving a significant amount of current gain that can throw off the loop response of the LM317T, hence the need for additional capacitance at the output to change where the output pole comes in to stabilize the loop.

The input to a positive NPN regulator needs to be decoupled if it is located far from the bulk input capacitance.  Otherwise it is not required.

In this case it needs to be decoupled because of the input series resistance of the boost circuit.

The LM317 will normally be able to function without a lot of input capacitance, with 22 ohms from IN to the bulk input capacitors is usually low enough impedance for the regulator to look into. BE CERTAIN YOU HAVE ENOUGH BULK INPUT CAPACITANCE!!! You have two circuits that you show in a post, one shows 60,000uF of bulk capacitance and the other shows none. You need to remember that the wires from your supply to this circuit will act as an inductor so if you don't have enough bulk input capacitor the double pole formed by the LC (or the output pole of the supply feeding your circuit) can be a high enough frequency that the LM317T cannot reject it. I have seen more than one circuit where the impedance at the input of the regulator input was either too high or a double-pole was formed that was close to or above the unity gain frequency of the regulator and the entire circuit oscillated.

Solid tantalum capacitors have better high frequency characteristics, so when used for frequency compensation at the output of a regulator, can be a fraction of the capacitance of a standard aluminum electrolytic capacitor.

The boosted design should have more output capacitance however.

[AnalogTodd]

The input to a positive NPN regulator needs to be decoupled if it is located far from the bulk input capacitance.  Otherwise it is not required.

In this case it needs to be decoupled because of the input series resistance of the boost circuit.

Usually emitter-follower or source follower circuits don't need as much input capacitance because of the basic way they operate--changes to the output immediately command more or less drive to the pass element as the emitter or source is pulled on or off appropriately. You still need a low impedance from some level of capacitance at the input, whether from nearby bulk input capacitance or nearby decoupling. One still needs to be aware of potential sources of oscillation such as adding a boost circuit, long wires between regulator input and bulk input capacitance, or having the response of the supply feeding the circuit operating at close to the same bandwidth as the regulator (the two loops will beat against each other like crazy). The LM317T is a stable enough circuit that the 22 ohms alone is often not enough to require local decoupling if the bulk capacitance is close by (the LM317 won't draw huge currents and is very stable in that light load region).

Solid tantalum capacitors have better high frequency characteristics, so when used for frequency compensation at the output of a regulator, can be a fraction of the capacitance of a standard aluminum electrolytic capacitor.

The boosted design should have more output capacitance however.

Each type of capacitor has its benefits and drawbacks. Aluminum electrolytics and solid tantalums work well with a regulator like the LM317T because they have ESR. The LM317T was designed in an era when low ESR ceramics were not available in a size that would work as load capacitance for a regulator output such as the LM317T. If you are concerned about an LC circuit oscillating, the worst-case you can have is some nice long wires with local ceramic decoupling. The lack of ESR means it wants to ring at a higher amplitude for longer because it becomes a high-Q tank circuit with almost nothing to damp it out. Going with capacitors with higher ESR will damp these oscillations out.

I think what you are confusing between solid tantalum and aluminum in regulator loop compensation is not their high frequency characteristics but instead the effect of ESR. A pole is formed using the output capacitance and is directly related to the output capacitance amount. Without ESR, if you have a 10uF aluminum electrolytic it may set the unity gain bandwidth of the regulator at 30kHz and a 1uF solid tantalum would be 300kHz. What ESR gets you is a zero to come in and raise the phase so that you get a bump in phase margin where the gain cross unity. The downside of the added ESR is often a larger transient deviation for the same load step--less C means bigger voltage transient together with higher ESR means more voltage drop for the same current. Things can still be stable, and in some cases the additional bandwidth that comes from a smaller capacitor may give a better response, but it's not always the case.

[David Hess]

I think what you are confusing between solid tantalum and aluminum in regulator loop compensation is not their high frequency characteristics but instead the effect of ESR. A pole is formed using the output capacitance and is directly related to the output capacitance amount. Without ESR, if you have a 10uF aluminum electrolytic it may set the unity gain bandwidth of the regulator at 30kHz and a 1uF solid tantalum would be 300kHz. What ESR gets you is a zero to come in and raise the phase so that you get a bump in phase margin where the gain cross unity. The downside of the added ESR is often a larger transient deviation for the same load step--less C means bigger voltage transient together with higher ESR means more voltage drop for the same current. Things can still be stable, and in some cases the additional bandwidth that comes from a smaller capacitor may give a better response, but it's not always the case.

I was just going by old application notes which suggest using solid tantalum capacitors of 1/10th the size of an aluminum electrolytic in this application, with the high frequency impedance being what is important.  My own tests show the ratio to be more like 1/4 but NPN output positive regulators are very tolerant.

This comes up more for the negative regulators because they operate with a common emitter NPN output than with the positive regulators, but the positive regulators still benefit from better transient response when the output capacitor is included even though it is not strictly required.  The negative regulators *must* operate with an output bulk capacitor.

With the current boosted circuit, I would include the output capacitor and increase its size proportional to output current.  I have always gotten good results with this circuit but others seem to find it temperamental.

[xavier60]

For anyone who might not know, time shouldn't be wasted on Circuit A in post #3.
 It has previously appeared on the forum. It can't possibly work in any useful way.

[iMo]

Except the tantalum capacitors have more ESR than aluminum (and more than ceramic) ones they have much less parasitic inductance as well (even it may not play a big role here). The 50mA current limit (my number above) for a smoke test is a value which should not evaporate the chips (their metal layers), but still it is rather on the higher end. I would even go with 20mA.
The LM317L as the driver would not help you much as they usually deliver >100ma when shorted.
Also the output capacitance will contribute to the initial current surge (unless the current limiter could sink it), so the less capacitance the better.

[Uky]

Off-Topic...

One drawback regarding linear three-terminal regulators is that the internal reference is not bypassed (with a capacitor).
Thus, noise generated by that reference will find its way all the way to the output of the regulator circuit.
Attempting to supress this noise using (large) capacitors may lead to increased risk of self oscillation.

I know that my entry is a bit off-topic but I would recommend taking a look at the old 723 regulator. It may not be the best in all
cases but the basic functionality with a separate decoupled reference (6.2V - Zero drift vs temperature), its buffer and a separate
OP-Amp where the regulator loop can be designed and optimized and also with the posibility to optimize the time constants/frequency compensation
of the regulator. There are many circuit ideas on the internet that also shows how to implement foldback current limiting, a feature that
saves the series pass elements from getting overheated during prolonged short circuit conditions.

 

[AnalogTodd]

I think what you are confusing between solid tantalum and aluminum in regulator loop compensation is not their high frequency characteristics but instead the effect of ESR. A pole is formed using the output capacitance and is directly related to the output capacitance amount. Without ESR, if you have a 10uF aluminum electrolytic it may set the unity gain bandwidth of the regulator at 30kHz and a 1uF solid tantalum would be 300kHz. What ESR gets you is a zero to come in and raise the phase so that you get a bump in phase margin where the gain cross unity. The downside of the added ESR is often a larger transient deviation for the same load step--less C means bigger voltage transient together with higher ESR means more voltage drop for the same current. Things can still be stable, and in some cases the additional bandwidth that comes from a smaller capacitor may give a better response, but it's not always the case.

I was just going by old application notes which suggest using solid tantalum capacitors of 1/10th the size of an aluminum electrolytic in this application, with the high frequency impedance being what is important.  My own tests show the ratio to be more like 1/4 but NPN output positive regulators are very tolerant.

This comes up more for the negative regulators because they operate with a common emitter NPN output than with the positive regulators, but the positive regulators still benefit from better transient response when the output capacitor is included even though it is not strictly required.  The negative regulators *must* operate with an output bulk capacitor.

With the current boosted circuit, I would include the output capacitor and increase its size proportional to output current.  I have always gotten good results with this circuit but others seem to find it temperamental.

Old application notes really didn't talk about ESR of capacitors. It's one of those things that was never really considered until MLCC components came onto the market. I know this firsthand, I was designing linear regulators from this point in time forward. Yes, they found solid aluminum would allow you to use smaller capacitance values and they attributed this to 'high frequency characteristics' of the devices, but we now know that is the ESR of the capacitor that creates a zero in the loop and boosts the phase at that corner.

The thing is to not think about this in terms of whether it is positive or negative regulators, but instead to think of the basic regulator configuration. You have emitter-follower or source-follower configurations that are very forgiving in terms of output capacitor; these are devices like the LM317T and they are so forgiving because the main pole comes internally at low frequency and the second high-frequency pole comes from the power device and drive circuitry interacting with the output capacitance and doesn't move around much. From there, you have common-emitter or common-source regulators such as the LT1763; these are much more involved in terms of frequency compensation as one of the major poles is formed by C_OUT * R_LOAD which moves all over the place and one must pay attention to ESR when designing the part to make it as stable as possible with a wide range of output capacitor types. Last is a Sziklai pair configuration that can actually be inherently stable under a lot of output capacitor types and sizes, including no output capacitance. The LT3088 is an example of this.

If you want to learn about linear regulator theory of operation and compensation, this old National Semiconductor (now TI) app note is a good place to start: https://www.ti.com/lit/an/snva020b/snva020b.pdf. Dealing with low ESR output capacitors is another matter, and that is covered in this app note: https://www.ti.com/lit/an/snva167a/snva167a.pdf.

I've designed dozens of linear regulators in my career, both positive and negative and in all configurations I listed above.

Off-Topic...

One drawback regarding linear three-terminal regulators is that the internal reference is not bypassed (with a capacitor).
Thus, noise generated by that reference will find its way all the way to the output of the regulator circuit.
Attempting to supress this noise using (large) capacitors may lead to increased risk of self oscillation.

I know that my entry is a bit off-topic but I would recommend taking a look at the old 723 regulator. It may not be the best in all
cases but the basic functionality with a separate decoupled reference (6.2V - Zero drift vs temperature), its buffer and a separate
OP-Amp where the regulator loop can be designed and optimized and also with the posibility to optimize the time constants/frequency compensation
of the regulator. There are many circuit ideas on the internet that also shows how to implement foldback current limiting, a feature that
saves the series pass elements from getting overheated during prolonged short circuit conditions.

 

One thing that was often done was to add a capacitor such as the 10uF from the ADJ pin of the LM317T to ground as shown in the schematics. That provides a low impedance at frequency across the gain resistor and bypasses the high frequency gain of the resistor divider to improve noise and supply rejection (the circuit looks like a unity gain configuration at frequency as opposed to gain of X). This gives the lower output voltage range (can't do 3.3V or 5V with a 6.2V reference) that you can't get with the old LM723. Of course, there are now new regulators such as the LT3045 that can get even lower noise at output voltages down to zero.