Author Topic: TL431 linear power supply  (Read 9153 times)

0 Members and 1 Guest are viewing this topic.

Offline blackdog

  • Frequent Contributor
  • **
  • Posts: 580
  • Country: nl
  • Please stop pushing bullshit...
Re: TL431 linear power supply
« Reply #125 on: December 21, 2018, 08:57:55 pm »
Hi Kleinstein  :)

Thank you for finding the errors in de links en telling me!

Almost all links were broken, sometimes even a double address in one line and even one line whas started with http:www.eevblog.com in front of it...

And when I clicked on a link, the pdf was downloaded... spooky  :-DD

I hope it is now OK

Kind regards,
Bram

“Two things are infinite, the universe and human stupidity, and I am not yet completely sure about the universe.”
 

Offline David Hess

  • Super Contributor
  • ***
  • Posts: 9902
  • Country: us
  • DavidH
Re: TL431 linear power supply
« Reply #126 on: December 22, 2018, 03:09:18 am »
The LM358 is a slow opamp and with a well-designed circuit it should provide a reasonable power supply with the 2N3055 transistors.

I think there is some benefit up to a 3MHz operational amplifier (except for likely needing extra frequency compensation) but the old JFET input parts were 3MHz and had much higher slew rates which is even more important for designs where the voltage or current error control loop sits in saturation.  The old RC4136 was a weird 741 class operational amplifier which was 3MHz so this was not universal.

Old slow LM358/LM324s, 741s, and 310As could almost always be used without any extra frequency compensation and perform just fine.

Quote
The TIP41 and the 2N3055 are both "slow" transistors and they do not help to make the circuit stable.
Hint, make the base emittor resistor a little lower say 56 or 68 Ohm, which helps with the paracit capacitors of the 2N3055.

And also this, there are so may different 2N3055 transitoren, Ft 0f 0,8Mhz, Ft of 3Mhz etc,
So if someone present here a schematic with 2N3055 transistors, then you will actually have to take into account the worst version when it comes to loop compensation.

I do not think anybody has produced the slowest 2N3055s for a long time now.  If you want one of those, use the 2N3771G or 2N3772G.  See below.

The slow 2N3055 were the more robust old style transistors. One hardly finds them any more. However the once popular 2N3773 are usually still slow. For a high performance power supply it really helps if the transistors are fast, but for a simple low cost supply one can still use the normal 2N3055 or TIP35. It still helps if the second transistor in the Darlington configuration is fast.

Be real careful about the 2N3773; I thought they all used the 0.8MHz process but apparently all new production at least from On Semiconductor is 4MHz.  On the other hand, the 2N3773G has the highest secondary breakdown voltage of parts listed below at 100 volts.  The MJ15015 and MJ15016 are not far behind at 90 volts.

2N3771G NPN 40V 30A 150W 0.2MHz
2N3772G NPN 60V 20A 150W 0.2MHz

2N3055AG NPN 60V 15A 115W 0.8MHz
MJ15015 NPN 120V 15A 180W 0.8MHz

MJ15016 NPN 120V 15A 180W 2.2MHz

2N3773G NPN 140V 16A 150W 4MHz
MJ15022 NPN 200V 16A 250W 4MHz
MJ15023 PNP 200V 16A 250W 4MHz
MJ15024 NPN 250V 16A 250W 4MHz
MJ15025 PNP 250V 16A 250W 4MHz

Quote
A low power the BD139 is a good choice, at higher power something like D44H11.  A slower supply would need a larger output capacitor though.

The D44H11 is a RET (ring emitter transistor) style so it may have other advantages as well.  I commonly see them used in low noise and fast response regulators and of course audio power amplifiers.  Oddly enough, the final design of the problematical single ended high voltage inverter Tektronix used in their 465 series oscilloscopes changed from a selected 2N3055 to a much faster RET D44H/D45H series.
 

Offline spec

  • Frequent Contributor
  • **
  • Posts: 833
  • Country: england
  • MALE
Re: TL431 linear power supply
« Reply #127 on: December 22, 2018, 06:27:35 am »
Thank you all for the interesting information in the last posts #122 to #127 above. But, I am afraid that much of the good stuff cannot be embodied in this PSU because the components are a given by the OP, who probably has limited access to components in his local. As stated a few times now, given a free hand, I would not have have used the components or architecture in the PSU of post #89, especially the opamp, or the extra transistor stage, or at least not in the common emitter configuration.

I agree that the additional transistor stage is a PITA from a frequency compensation point of view, but it is the only way I could provide adequate output voltage swing and current sinking, while providing sufficient base drive for the output driver transistor. My scheme on the original design was, in the interests of frequency stability, to get the voltage gain of the transistor stage down to 1 at as low a frequency as practicable.

It may be possible to drive the output stage directly from the LM358 if another emitter follower were added to drive the output stage. There is another aspect though, that concerns the relationship between the constant voltage and constant current modes. This is always a difficult area and IMO the safest way to do it is by the 'decreasing gm' method which precludes the use of a voltage drive of the output transistors directly by the opamp.

The other aspect is that the OP wants to use an LM317 regulator and with its dropout voltage and 40V input limit that makes things difficult.

About the 24:1 voltage divider. While I agree that it may increase the noise and voltage drift, I am not sure that it is that significant. But the attenuation does reduce the feedback and should help with the frequency stability, at least that was my thinking. It also allows some flexibility in tailoring the open loop response, as has already been done.

If anyone has an improved complete, practical circuit, that meets all of the OPs requirements and is easy to read and that the OP can use to build a PSU, please post it (this is a genuine request).

Also if anyone has an approach, in addition to not1xor1's work, for compensating the reply #89 circuit please post a complete practical circuit (of the compensation that is). That would not only be interesting, to me anyway, but also very useful.

 

« Last Edit: December 22, 2018, 07:20:56 am by spec »
 

Offline Kleinstein

  • Super Contributor
  • ***
  • Posts: 6226
  • Country: de
Re: TL431 linear power supply
« Reply #128 on: December 22, 2018, 11:01:01 am »
The initial circuit right from the start is not that bad - it is mainly missing the frequency compensation. So with the first response one could call it problem solved. There are still some points to optimize and the selection of parts could be better, but nothing serious.

For just a simple power supply the "normal" modern (e.g. 3-4 MHt Ft) 2N3055, TIP35 or similar should be sufficient in speed. It helps if in the Darlington circuit the smaller transistor is a little faster. But this is easy, as the BD139 is nothing special. Even the D44H11 is not that exotic.
The voltage regulator with the Darlington emitter follower output and thus a low output impedance output stage normally does not need a fast OP for the regulation, unless one has high demands. So for normal use the LM358 is sufficient, at least if one does not divide down the feedback signal very much (e.g .down to 1.2 V), as this would need more gain.

If there is a second loop for current regulation (and not just the transistor for a crude current limit), than one might want a faster OP in the simple circuit, as the slew rate of the OP would limit how fast the current limit can engage.  The 0.3 V/µs of the LM358 would require some 100 µs to come down from 30 to near 0 in case of a short. Even with a faster OP, the current regulation is a little harder in this type of circuit.

With only up to 25 V the simple circuit type is perfectly fine. Going up to the usual 30 V starts to get tricky as the supply for the OPs is limited.

The LM317 was an idea for the OPs supply. Here the 40 V limit may indeed be a problem - though it only applies to the in to out difference, not the absolute voltage that is never seen by the LM317. Still I would prefer the simple Zener solution for the rather low current (e.g. 2 mA) needed for the OP.


 

Offline imo

  • Super Contributor
  • ***
  • Posts: 1763
  • Country: 00
Re: TL431 linear power supply
« Reply #129 on: December 22, 2018, 02:26:49 pm »
This works nice, imho..
It oscillates without the C4 capacitor.
Responses I_Load=10mA/5A for a list of C_Load capacitors.
« Last Edit: December 22, 2018, 02:28:53 pm by imo »
 

Offline Kleinstein

  • Super Contributor
  • ***
  • Posts: 6226
  • Country: de
Re: TL431 linear power supply
« Reply #130 on: December 22, 2018, 03:02:38 pm »
There still seems to be quite some ringing - so maybe C4 should be even larger. A suitable series resistor to C4 could also help.
 

Offline imo

  • Super Contributor
  • ***
  • Posts: 1763
  • Country: 00
Re: TL431 linear power supply
« Reply #131 on: December 22, 2018, 03:46:41 pm »
And with the current limiter, switching various R_Loads in/out.
 

Offline imo

  • Super Contributor
  • ***
  • Posts: 1763
  • Country: 00
Re: TL431 linear power supply
« Reply #132 on: December 22, 2018, 04:14:08 pm »
And with constant R_Load=10ohm (switched on/off) with various values of R2 (output voltage setting).
 

Offline Kleinstein

  • Super Contributor
  • ***
  • Posts: 6226
  • Country: de
Re: TL431 linear power supply
« Reply #133 on: December 22, 2018, 04:57:27 pm »
There are two small problems with the simple TL431 based regulator:
The first is relative obvious: one can not set a voltage below 2.5 V. This might be OK for a fixed supply to a circuit, but would be a slight limitation for a stand alone supply.

The second possible problem is that the loop gain and thus the suitable compensation tends to depend on the set voltage, if the voltage is set by adjusting the feedback divider. The way it is shown in the last simulation, by adjusting the lower resistor this is to a large part avoided, but at the cost of a rather nonlinear scale and even higher lower voltage limit.

So the circuit is nice and simple for a fixed supply, but not that suitable for a variable. For an adjustable supply it should be better to have a kind of OP to compare the feedback from a fixed divider with a variable set voltage.

For finding the right compensation from a simulation I found it rather useful to look at the output impedance: So similar to the load test with a constant current load a current sink (source) is used as a load, but this time in AC mode and first without much of the output capacitance. Looking than at the AC output voltage than directly gives the output impedance.  Ideally one want's a low output impedance of cause. The more important part is that the phase shift does not reach values above 90 degrees from a resistive load. Getting to close to an ideal inductance (90 degree phase) is also bad as this would cause much ringing if there is a corresponding capacitive load.

Some of the higher frequency problems can be fixed with those RC combinations at the output, but the lower frequency part has to come right from the regulator. Only at the very low end (e.g. below some 100 Hz) one can tolerate more phase shift, as it would take extreme, non realistic capacitors (e.g. C > 100 mF with ESR in the µOhms) to make the regulator unstable.

The nice point about looking at the output impedance instead of the more common loop gain is that a single simulation covers all possible load impedance cases. One still has to check different DC currents as the output stage can behave different at high and low currents. For the loop gain one would need to additionally test with different capacitors at the output to get the "whole" picture. Another point to check would be the hand over from current to voltage regulation and back.
 

Offline imo

  • Super Contributor
  • ***
  • Posts: 1763
  • Country: 00
Re: TL431 linear power supply
« Reply #134 on: December 22, 2018, 11:39:20 pm »
Bode
« Last Edit: December 23, 2018, 01:07:02 am by imo »
 

Offline not1xor1

  • Frequent Contributor
  • **
  • Posts: 376
  • Country: it
Re: TL431 linear power supply
« Reply #135 on: December 23, 2018, 09:09:31 am »
The initial circuit right from the start is not that bad - it is mainly missing the frequency compensation. So with the first response one could call it problem solved. There are still some points to optimize and the selection of parts could be better, but nothing serious.

The initial circuit is not that bad ... if one just looks for trouble>:D
So far have I been unable to make it work under all the possible conditions of load and feedback resistor.
TL431 is handy for a fixed voltage regulator, but unless you severely compensate it (achieving awful transient recovery times) it is unsuitable for a bench psu.

A PSU with fixed feedback opamp and variable voltage reference is not more difficult to build, but is much more reliable and easier to set up.
And with a range of 2.5-25V, as required by the original poster, there is no need for a further voltage boost stage.
I already showed two simple circuits which are simple yet offer much better performances than the other proposal.
Of course one is free to do what he/she likes... :horse:

 ;D

 

Offline not1xor1

  • Frequent Contributor
  • **
  • Posts: 376
  • Country: it
Re: TL431 linear power supply
« Reply #136 on: December 23, 2018, 09:11:12 am »
Bode

that's fine for a power oscillator  ;D
 

Offline imo

  • Super Contributor
  • ***
  • Posts: 1763
  • Country: 00
Re: TL431 linear power supply
« Reply #137 on: December 23, 2018, 10:33:55 am »
Stability test simulation made according to the AD app note

Vout 15V/1.5A with:

1. C4=100pF
2. C4=1nF
3. C4=1nF ser 56ohm
4. none comp C4 - here at 300kHz the phase margin is 0deg - it oscillates
« Last Edit: December 23, 2018, 01:18:54 pm by imo »
 

Offline not1xor1

  • Frequent Contributor
  • **
  • Posts: 376
  • Country: it
Re: TL431 linear power supply
« Reply #138 on: December 24, 2018, 08:17:43 am »
Stability test simulation made according to the AD app note

Vout 15V/1.5A with:

1. C4=100pF
2. C4=1nF
3. C4=1nF ser 56ohm
4. none comp C4 - here at 300kHz the phase margin is 0deg - it oscillates

you must not forget that TL431 is specified for a minimum current of 1mA (although in some cases might be fine with as little as 500µA).
R4 might work for the whole voltage range just for 1A of output current or so...
once you replace it with a constant current source the open loop gain soars and the nightmare begins  ;D
BTW it would be better to start from 1Hz or even below...
 

Online T3sl4co1l

  • Super Contributor
  • ***
  • Posts: 13747
  • Country: us
  • Expert, Analog Electronics, PCB Layout, EMC
    • Seven Transistor Labs
Re: TL431 linear power supply
« Reply #139 on: December 24, 2018, 10:41:37 am »
FWIW, TL431 works just fine at <1mA, it's just not going to be spot on 2.500V.  In particular, it drops to about 2.2V at light currents, around which it's pretty much off (~uA).

Knowing this behavior is helpful when working with a clamping or startup threshold sort of circuit: any time the threshold needs to be sharper than a comparable zener (a 2.5V zener is a damn good resistor in comparison), but also accurate at higher currents.  (For example, the fact that current rises from ~uA at ~2.2V, to 1mA+ at 2.50V, only means a clamping threshold is rounded off by a comparable amount.)

Tim
Seven Transistor Labs, LLC
Electronic design, from concept to prototype.
Bringing a project to life?  Send me a message!
 

Offline imo

  • Super Contributor
  • ***
  • Posts: 1763
  • Country: 00
Re: TL431 linear power supply
« Reply #140 on: December 24, 2018, 12:33:42 pm »
With above R4=4k7 the TL431's current will be from 10mA to 4mA for 3V/5A to 30V/5A output voltage/current..
Mind it is an simulation only, with values and parts taken as an example..
With 60V input and 3V/5A output the Q1 will dissipate 20W and Q2 260W of heat :)
« Last Edit: December 24, 2018, 12:48:35 pm by imo »
 

Offline not1xor1

  • Frequent Contributor
  • **
  • Posts: 376
  • Country: it
Re: TL431 linear power supply
« Reply #141 on: December 25, 2018, 08:07:23 am »
With above R4=4k7 the TL431's current will be from 10mA to 4mA for 3V/5A to 30V/5A output voltage/current..
Mind it is an simulation only, with values and parts taken as an example..
With 60V input and 3V/5A output the Q1 will dissipate 20W and Q2 260W of heat :)

We must take into account the case of output voltage set to 25V with maximum load.
In such case, when using a real transformer, the rectified voltage would drop several volts due to the transformer losses and the ripple, so the minimum rectified voltage has to be greater than:
> maximum Isense voltage drop (about 0.7V) +
   maximum ballast voltage drop (in case of paralleled power devices i.e. about 0.7V) +
   maximum compound Vbe drop (1.5-2V) or ce saturation voltage (whichever is greater) +
   peak-to-peak ripple (1-6V depending on the value of the capacitor(s) and the maximum output current)

And then the resistor feeding the base of the power devices should provide at least 1mA through the TL431 plus the base current for the power devices... let's make it a couple of mA in case of just 1A max output current, that makes : 2e-3 * 4.7e3 = 9.4V...
That's a lot... at least 9.4W wasted by the power devices for no real purpose.
Besides that the task of reducing the input ripple would be totally on the shoulders of that poor TL431...  :D

So now we have 3 options:
-1) use a low minimum voltage constant current source (making the TL431 more prone to oscillations)
-2) use a transformer with voltage much higher than required (wasting further 10W per A of output current)
-3) use a charge pump (2 diodes + 2 capacitors voltage doubler) to get a high voltage rail

The 3rd options might be the only viable in case of a TL431 PSU.
So in this case the minimum rectified voltage can be as low as 28V which would allow to get more than 50V (more likely 53-54V) on the high voltage (low current) rail.
If we just need 2mA (like in the above example) we must use a (50-28)/2e-3 = 11k base-feed resistor.
But we could better use 2 resistors, let's make it 3.3k and 8.2k (or 4.7k and 6.8k) and place a 100µF electrolitic cap in the resulting node to ensure both soft start and ripple reduction.

The maximum current through the resistor would be achieved with minimum output voltage and no load. In that case the high voltage rail might get as high as 70V-71V that would make less than 7mA...
still safe for the TL431 with just around 17mW of dissipation while it would be better to use 2 1W resistors (although they would dissipate about 1/4 W).

Of course, in case of higher maximum output current, the resistor values (and power) should be recalculated.
 

Offline not1xor1

  • Frequent Contributor
  • **
  • Posts: 376
  • Country: it
Re: TL431 linear power supply
« Reply #142 on: December 25, 2018, 08:13:20 am »
FWIW, TL431 works just fine at <1mA, it's just not going to be spot on 2.500V.  In particular, it drops to about 2.2V at light currents, around which it's pretty much off (~uA).

Knowing this behavior is helpful when working with a clamping or startup threshold sort of circuit: any time the threshold needs to be sharper than a comparable zener (a 2.5V zener is a damn good resistor in comparison), but also accurate at higher currents.  (For example, the fact that current rises from ~uA at ~2.2V, to 1mA+ at 2.50V, only means a clamping threshold is rounded off by a comparable amount.)

Tim

I know that TL431 would work with few hundreds of µA, but I can't imagine who would like a PSU whose output voltage may drop 10% or more depending on the load  ;D
 

Online T3sl4co1l

  • Super Contributor
  • ***
  • Posts: 13747
  • Country: us
  • Expert, Analog Electronics, PCB Layout, EMC
    • Seven Transistor Labs
Re: TL431 linear power supply
« Reply #143 on: December 25, 2018, 08:20:11 am »
I know that TL431 would work with few hundreds of µA, but I can't imagine who would like a PSU whose output voltage may drop 10% or more depending on the load  ;D

I know, it seems silly doesn't it, when we're designing a PSU whose output may drop 100% (hopefully not more!) depending on load! ;D ;D

...Hyperbole aside, actually, note that the TL431 is at maximum current when the PSU output is minimum load -- it's regulating best there!  So it really doesn't hurt anything, as long as, at the point where current limiting cuts in, on that very knee, there's still that extra 1mA available for the TL431.

And again, the consequence is merely a softening of the knee, which is hardly immaculate in this case because we've been talking about Vbe sensing action in this thread.

So, assuming the bias supply current is that precise, it really would work out just fine here. :-+ :-+

The bigger problem, then, would be ensuring current is actually that close -- in practice, it will be far less accurate (we're talking, say, a 1mA margin out of a 50mA or whatever total requirement!), and setting the guard band for that variance will take priority.

Good to know, if that guard band runs out, it'll behave itself though. :)

Tim
Seven Transistor Labs, LLC
Electronic design, from concept to prototype.
Bringing a project to life?  Send me a message!
 

Offline imo

  • Super Contributor
  • ***
  • Posts: 1763
  • Country: 00
Re: TL431 linear power supply
« Reply #144 on: December 25, 2018, 09:39:31 am »
With a 6.3mA current source and 2n2 ser 100ohm comp.
3V/5A
25V/5A
« Last Edit: December 25, 2018, 09:56:02 am by imo »
 

Offline not1xor1

  • Frequent Contributor
  • **
  • Posts: 376
  • Country: it
Re: TL431 linear power supply
« Reply #145 on: December 27, 2018, 08:14:20 am »
With a 6.3mA current source and 2n2 ser 100ohm comp.
3V/5A
25V/5A

I confirm the circuit works as showed in your screenshot.
Il looks like the problems arises from the placement of the AC source that yields awful results when placed like I did.

I've to check more carefully the subject to understand what is wrong.
Thanks

P.S.
since I suspected something is wrong in your measurement method I made a quick transient test, switching between 5mA and 1A on 25V output and the output goes awry... huge oscillations... see below

« Last Edit: December 27, 2018, 08:59:38 am by not1xor1 »
 

Offline imo

  • Super Contributor
  • ***
  • Posts: 1763
  • Country: 00
Re: TL431 linear power supply
« Reply #146 on: December 27, 2018, 02:25:04 pm »
Measurement method?? - That is the question! :)

I took the LT/ADI method - see the link above.
You previously did with a simple AC sweep.

Your oscillation - are you using zero ESR output capacitors?

Below with "real capacitors", no oscillation, a small ringing w/ 160mVpp amplitude..
« Last Edit: December 27, 2018, 03:28:16 pm by imo »
 

Offline not1xor1

  • Frequent Contributor
  • **
  • Posts: 376
  • Country: it
Re: TL431 linear power supply
« Reply #147 on: December 27, 2018, 07:40:45 pm »
Measurement method?? - That is the question! :)

I took the LT/ADI method - see the link above.
You previously did with a simple AC sweep.

Your oscillation - are you using zero ESR output capacitors?

Below with "real capacitors", no oscillation, a small ringing w/ 160mVpp amplitude..

That is a different method but should yield similar results.

I used models of real capacitors too...
I also tried to add a resistor of 0.5 ohm in series to each capacitor with no result.
BTW this is LTspice XVII x64 Jun 15 2018 running in linux.
I tried both alternate and normal solver, 4 threads and single thread.
No matter what I try I always get oscillations.
Might that depend on the TL431 model I'm using here ...?  :-//

in the comment I read:
Quote
** Released by: Analog eLab Design Center, Texas Instruments Inc.
* Part: TL431
* Date: 12/14/2009
* Model Type: Transient and AC 
* Simulator: PSPICE 
* Simulator Version: 16.0.0.p001
* Datasheet: SLVS543J - AUGUST 2004 - REVISED DECEMBER 2005
 

Offline imo

  • Super Contributor
  • ***
  • Posts: 1763
  • Country: 00
Re: TL431 linear power supply
« Reply #148 on: December 27, 2018, 08:39:54 pm »
Here is the latest LTspiceXVII, and with the latest Bordodynov library (LTspiceXVII_2018Dec21.zip).
Below the TL431s offerings..
PS: tried with several TL431 models and got similar results as above..

And the library is here, for the latest download go to "Additional links :"
« Last Edit: December 27, 2018, 09:20:48 pm by imo »
 

Offline xavier60

  • Super Contributor
  • ***
  • Posts: 1169
  • Country: au
Re: TL431 linear power supply
« Reply #149 on: December 27, 2018, 11:14:55 pm »


P.S.
since I suspected something is wrong in your measurement method I made a quick transient test, switching between 5mA and 1A on 25V output and the output goes awry... huge oscillations... see below

I think that C1 and R17 should be made larger, I'd start with 1uF and 1K
Hioki AS100D vom, HP 54645A dso, Fluke 87V dmm, AN8008 dmm, Agilent U8002A psu, New! FY6600 function gen.
 


Share me

Digg  Facebook  SlashDot  Delicious  Technorati  Twitter  Google  Yahoo
Smf