TL431 vs OPamp for smps feedback: transient response?

[grizedale]

I am doing isolated 350w half bridge with vin = 90-265vac
Vout is +-50V

I wish for excellent no-load to full-load transient response.
(its powering a guitar amp)

I believe the TL431, if used in feedback , will tend to pull-down its own supply voltage when on no-load.

…this will make it less responsive to sudden application of load(?)

..also, when on full-load, TL431 will tend to shut off its own supply current.
-this will make it less responsive to sudden load removal.

Therefore I wish to use opamp instead of TL431.

Do you agree with my reasoning?

 

[FvM]

Do you agree with my reasoning?

No. TL431 can be used with all kinds of frequency compensations you want to. A first prerequisite is, that you have an idea of the intended loop gain characteristic.

 

[Orson Cart]

A TL431 can be made to work well as a Vea controller, usually gain of 1 max at 1kHz, many 100's of thousands of SMPS have been made and sold using this device to control output volts. Regards, Orson Cart.

 

[grizedale]

If an SMPS goes into no load........

...then many smps will give a very slight overvoltage on the output....this is because many pwm controllers will still switch on for a tiny amount of time, before quickly switching off again, in each cycle of switchiong period.

...often this is due to the presence of Leading Edge Blanking.

-another reason an isolated smps may give slight overvoltage when on no load is due to peak-charging from the transformer leakage inductance.
...

Now, the TL431 has its Vcc supply pin, and its cathode as the same pin.

........that means , when there is the slight overvoltage, the internal opamp, will turn tht TL431's output transistor on more and more, and as such the supply voltage of the TL431 will drop too low.............the TL431 starts to turn off, and thus the output transistor turns off a bit, and so on and son on.

....in this way, the TL431 is not fully compus mentus when on no-load condicitons, and so its transient response from no-load is not that great.

-as you rightly point out, the TL431 is used in many supplies, but this is where sudden transients from no load to full load are not occurring regularly.

 

[FvM]

A said, the TL413 behaviour depends on the external compensation circuit. I don't agree, that the said effects are specific to TL431 as such. What's your exact circuit?

 

[grizedale]

my circuit is as in fig 3 of the following

**broken link removed**

Fig 1 of this document shows the TL431....yuo can see that TL431 is nothing more than an opamp feeding into a BJT.

...So if the "+" input of the TL431 opamp is slightly greater than the "-" input of the TL431's opamp, ..and this happens for long periods such as prolonged no-load, then it is basic electronics that that opamps output will go to its positive rail, and turn the BJT on into saturation, and thereby shorting out its own supply voltage.

This is because the TL431's supply pin and cathode pin are ona and the same pin.

The TL431 isn't a bad device, but it does have this shortcoming.
The TL431 is used in vast quantities because it is , pound-for-pound, the cheapest IC on the market, in the world.

So, in summary, a pin is saved by having the cathode pin and supply pin as the same pin, but then unfortunate effects occur, as we expect with all such compromises.

 

[Orson Cart]

Often the TL431 draws current thru an opto coupler (which is connected to Vcc) so at no load the TL431 is drawing the nominal max current thru the opto to keep the psu o/p to zero.
Since it is running linearly the transient response in this setup is quite good (as good as it can be for a TL431)
At full power out the current in the 431 is minimum (lowest drive to the opto) and this is one of the reasons the opto-led is often bypassed with a 470 ohm res to supply the bias current to the 431 (at least 400uA) so that it is still in a linear operating state at full power. In this type of use - full bandwidth and best possible transient response is obtained - the limitations being imposed by the power circuit and the PWM generation circuit as mentioned below, Regards, Orson Cart.

 

[FvM]

In a slightly simplified view, the shown "fig 3" circuit implements just a 1st order low-pass characteristic for the feedback path. It's very unlikely the compensation you would choose for optimal control loop operation. It's true, that circuit variants are restricted with TL431, but the device has more capabilities than utilized in this circuit.

 

[grizedale]

Another reason to NOT use the TL431,....exists when one is supplying audio class d power amplifiers......

...since audio usage very often involves sudden no-load to full-load transients, where the smps is actually severely overloaded for a few hundred milliseconds.....resulting in sagging of the output voltage rail.

.....you can clearly see that the circuit of fig 3 in the follwing would not properly be able to handle that situation......since the TL431 would simply have a momentary loss of supply current during the rail sagging.....

**broken link removed**

...even if the TL431's biasing resistor (Rbias in fig 3) was made very small, in cases of overload resulting in considerable output voltage sagging, the TL431 would staunch its own supply current, because the voltage on the TL431 cathode would be driving right up to the rail, and there simply would not be enough voltage across "Rbias" in fig 3 to give enough supply current to the TL431.

-The figure 1 clearly shows the TL431.......it is simply an op-amp into a NPN , with a reference included.

-imagine the output voltage having sagged down in a harsh full-load transient...........that transistor (output transistor of TL431) would be turned off.................its collector (which also happens to be the cathode of the TL431) would drift up to the rail voltage, and not enough supply current would get to the TL431.

-This would be the case even of the TL431 bias current was supplied by a bias resistor placed across the opto-coupler diode.

So would you agree that in power supplies suffering harsh no-load to full-load transients, (and vice-versa) with overloads of ~300ms which make the output rail voltage sag down, then the TL431 should be replaced with a discrete op-amp arrangement, since one can ensure that a discrete op-amp *always* has a full supply of operating current?

 

[FvM]

I don't agree, that the sayed problems are generic TL431 issues. Basically, two points have to be achieved:
- good loop compensation with sufficient phase margin
- large signal behaviour designed according to the dynamic load characteristic

It may be the case, that you'll find it more convenient to implement this features with an OP error amplifier. But I keep my opinion, that TL431 can do as well.

 

[Orson Cart]

since the TL431 would simply have a momentary loss of supply current during the rail sagging.....

Actually this works to address the sag, as if the rail to the 431 goes low there is LESS drive to the opto - and therefore the converter goes to full power - this trick is used a lot (feedforward control) to pre-empt and speed up the response of a feedback loop. Regards, Orson Cart.

 

[grizedale]

Hi
Right i accept that the full-load case is not bad with the TL431....

...but its the no-load case which catches out the TL431.

This is because the output rail is slightly above regulated value in no-load, and so the "+" input of the TL431 is slightly above the "-" (reference) input of the TL431, for long periods, so the TL431 opamp drives its output transistor into saturation , and shorts out its own supply by doing this, because in the TL431 , the supply pin, amd the collector of the output transistor are the same pin.

so in no-load, the TL431 is drifting in and out of "consciousness", and is not ready to serve a suddenly occuring heavy load transient.

---------- Post added at 11:00 ---------- Previous post was at 09:47 ----------

Hi,

To provide evidence that the TL431 does indeed short out its own supply voltage in times of no-load, please see the following image....

https://i53.tinypic.com/28vv9sk.jpg

....this shows a simulation of it....the trace is the TL431 cathode voltage in no-load....it goes down to 1v98...which means it has insufficient supply voltage because TL431 requires 2V5 at cathode to be fully "compus mentus".

Here is the simulation file, as well as the .sym and .sub files for the TL431.

SIMULATION FILE FOR LT Spice
2shared - download Forward converter _feedback loop TL431 __3 BEST 210 AND 375_with transient.asc


.asy file for TL431
2shared - download TL431A.asy

.sub file for TL431A
2shared - download TL431A.asy

The .sub file gets dropped into the "sub" folder......
the .asy file gets dropped into the "asy" folder.

-the .asc file is the simulation and saved as a .asc file so it can run




...Anyway, having now prooved conclusively that TL431 shorts out its own supply voltage in times of no-load, do you know of any method for mitigating this?

 

[FvM]

The TL413A.mod file is missing from the download. Why don't you simply append the simulation files as a *.zip archive to your post?

As a guess without having analyzed the example, isn't it a trivial case of control loop overshoot, that can basically happen with other feedback circuits as well? In so far, I doubt that anything has been proved yet.

 

[grizedale]

I will try and zip the files...but i dont see that zip archive facility in edaboard.

There is no .mod file for LT Spice.

All you need for the TL431 is the .asy (assembly) file which is the symbol for the schematic.

-also the .sub (subcircuit") file is needed.......this is probably called .mod in other simulators.

you have to put ".include TL431A.sub" as a spice directive in the schematic.
(EDIT > SPICE DIRECTIVE)

As you know, there is otherwise no TL431 in LT Spice, so you have to add it in yourself, as i have hopefully described.



Anyway, this all boils down to the fact that TL431 has the collector of its output transistor, and its supply pin, sharing the *same* pin, so this causes the problems of this post

 

[FvM]

You apperently intended to provide TL431A.sub through 2share, but repeated the TL431A.asy link instead.

I have a TL431 model with different pin mapping, incompatible to your circuit. Also I'm not sure if both models are equivalent. So I prefer to have your file for the simulation.

edaboard supports a limited number of extensions when appending files. The most convenient way to post simulation files is to zip them on your computer and append the zip archive to your post with the manage attachments dialog. Another option is to change the extension to something known, e.g. *.txt.

 

[grizedale]

-ok i have just attached the TL431A.sub, TL431A.asy and the .asc simulation file for LT spice

-you will see, when you run it on no-load that the TL431 has this problem in no-load...that is, the TL431 , shorts out its own supply voltage when the smps is on no-load.

This is the big problem with TL431.

Its output BJT and supply pin share the same pin.....this is an obvious recipe for problems

 

[Orson Cart]

you will see, when you run it on no-load that the TL431 has this problem in no-load...that is, the TL431 , shorts out its own supply voltage when the smps is on no-load." & "This is because the output rail is slightly above regulated value in no-load, and so the "+" input of the TL431 is slightly above the "-" (reference) input of the TL431

The TL431 draws max current through its cathode at no load - true - but for a properly set up system this current will only be 10-15mA say - if the power stage and its control can go smoothly to 0% duty cycle then the 431 will remain in a linear state and the ref pin will not go higher than normal. We have designed a 3kW telecom rectifer (180-300VAC to 55V, 50amps) that used a TL431 (standard) as the volt control and the dynamic performance was extremely good from 0-100% load stepping (and 100%-0) with less than 300mV deviation and a settling time of 1.5mS - so if the power stage and its control are well sorted the TL431 can be a good Vea. Regards, Orson Cart.

 

[grizedale]

-maybe your telecom system implemented the type of smps which when on no-load, genuinely does not pass energy to the output...so its output voltage does not rise up that tiny bit above regulation level.

-many cheap smps's do in fact report a slightly higher vout when on no load, and , for them , the TL431 will short out its own supply voltage, as described above.

I am sending the simulation to LT Spice forum, because if i am wrong about this, then LT Spice suimulator is faulty.........though i doubt very much that LT Spice is faulty.

LT Spice reports the TL431 taking its cathode to less than 2V when on no load........that means the TL431 is not fully on as it needs 2V5.

 

[Orson Cart]

From the below one would infer it is the power stage (with its control) and not the choice of Voltage error amplifier that determines the overall behaviour in your current project - perhaps better to sort out the power stage as all Vea's will be similar in operation unless you put clamps on the Vea output. Regards, Orson Cart.

 

[grizedale]

i egree, to a point, however, we are all generally bounded by cost.

Most of the cheap PWM controllers will give you an SMPS that gives a slightly high vout when on no-load, and fopr the sake of cost, we are stuck with this.

However, in our case, we need a rapid response from no-load to full-load, so it looks like we must pay for a discrete op-amp in place of the TL431.

Changing the power stage to cope with this would make it too expensive......an op-amp is going to be the cheapest way forward.