1

I am currently working with few 150 W SMPS designs for industial applications with heavy transient load condition. Current design works well by mean of line/load regulation and device reliability but facing some issues regarding transient loads, say 500 Hz to 1 kHz 50% to 100% load transients. It can't maintain output voltage and faces very large voltage dip.

I had tested some reputed brand SMPS, they have very good response in such load conditions without stability issues. Only change I notice was, they use op-amps in feedback loop, while I use TL431.

My questions are:

Is TL431 have any limitation over op-amps?

If I want to replace TL431 with op-amp, on which characteristics of op-amp should I focus while selecting?

winny
  • 13,064
  • 6
  • 46
  • 63
Dhaval Lalani
  • 45
  • 4
  • 3
    A TL431 is more like the combination of an op-amp and a voltage reference. TL431 has a GBW product of a couple MHz so I would suspect your loop compensation values and the optocoupler (if any) first. – Spehro Pefhany Oct 26 '20 at 05:30
  • 2
    A schematic is needed to make sense of your question. Nothing wrong with a TL431. – Andy aka Oct 26 '20 at 07:59
  • @SpehroPefhany I haven't deep knowledge for loop compensation and design, but follow PWM manufacturer's design tools with some trial-error method. So that may be issue. Any suggestion for book or app note for beginers to study loop compensation?? – Dhaval Lalani Oct 26 '20 at 09:46
  • I wrote this [book](https://www.amazon.com/Designing-Control-Linear-Switching-Supplies/dp/1608075575/ref=sr_1_1?ie=UTF8&qid=1350833636&sr=8-1) specifically for compensating switching power supplies. – Verbal Kint Oct 26 '20 at 10:28

1 Answers1

5

The TL431 is well suited for stabilizing switching converters. It is difficult to beat this 3-leg component in terms of simplicity and costs. However, you have to know some of its limitations and needs to obtain the best of its performance:

  1. bias current is of paramount importance and ensures performance. Keep in mind that there is no \$V_{cc}\$ pin on this part which is self-biased from it cathode voltage. The classical TL431 requires at least 1 mA of bias current. Too low a current and the open-loop gain already small (50-60 dB) further degrades leading to a poor closed-loop output impedance. See the below graph to see how it affects the response with or without sufficient bias:

enter image description here

Some more recent devices require less bias like the TLV431 (100 µA) but are limited in maximum voltage. The NCP431 might be a good option but watch the pinout in SOT-23. As you can see in the above picture, you can add bias by adding resistance in parallel with the LED which creates a free current source (\$\frac{V_f}{R_{bias}}\$)

  1. Operating current is also important depending on the output voltage. Indeed, if the LED series resistance is of too high a value, regulation can suffer especially in light load. Please note that this resistance plays a role in a the mid-band gain as well and should be carefully selected. The maximum value it can take on is described below:

enter image description here

  1. Make sure the loop is properly compensated and the TL431 is wired in a genuine type 2 configuration (one single capacitor) as described below. The pole and zero must be selected to cross over sufficiently high (but low enough considering the RHP zero) and, more importantly, you have to measure the open-loop gain once the prototype is assembled. You cannot skip this important point and transient response observation is not enough. When doing so, make sure the fast- and slow-lanes are tied together (see this AN for more details).

enter image description here

As a final note, make sure the optocoupler is well characterized, in particular its low-frequency pole \$C_{opto}\$ which will work against the theoretical phase margin you want. You have all the needed details in the APEC seminar I taught in 2018.

Verbal Kint
  • 20,420
  • 1
  • 16
  • 50
  • In ending of point-3, you suggested to tie slow and fast lane together.But some design notes suggest to tie fast lane before LC filter inductor. Is there any major effect of it?? – Dhaval Lalani Oct 26 '20 at 09:23
  • I observed effect of gain resistor value on transient response, but feared to make too low value (High gain) against stability. I found resistor ~ 47R to 68R for good response for 24V output. Any suggestions?? – Dhaval Lalani Oct 26 '20 at 09:27
  • 1
    Sorry for the confusion: you tie the slow- and fast-lanes for measuring the open-loop gain only. If you have correctly separated them because of the insertion of an \$LC\$ filter, this is fine. But as shown in the AN, either you independently measure each lane or tie them together for the measurement. I can't comment on the series LED resistance: the key is measuring the loop gain on the prototype or use an averaged model in a simulator. You can also use my SIMPLIS templates that I recently released. It could be a first approach to assess your crossover frequency. – Verbal Kint Oct 26 '20 at 09:34
  • I tried same by tie both lane just for measurment. I used Cleverscope CS320A-FRA for loop analysis of prototype. I brought it to study FRA, but it made me confused. The FRA have major change between constant amplitude sweep and varing amplitude sweep.Both have PM,GM and unity gain frequency totally different. Any specific reason ?? – Dhaval Lalani Oct 26 '20 at 10:09
  • Maybe [SPX2431](https://www.maxlinear.com/ds/spx2431.pdf) could help. Its pole is at a higher frequency than standard TL431 and the second pole is way higher too. – bobflux Oct 26 '20 at 10:12
  • Please update the post with a schematic diagram, we are only speculating without it. – Verbal Kint Oct 26 '20 at 10:27
  • I thought I'd see what price your book could be found for used. Sadly for me, it holds its price well :-). You may wish to (try to) do something about [**this**](https://www.yumpu.com/en/document/view/63752310/pdf-download-designing-control-loops-for-linear-and-switching-power-supplies-a-tutorial-guide-full-pdf-online/3) allegedly free version. [Some such are just ID traps - whether by signing up you really get access I know not - nor ever seek to find out on such sites]. – Russell McMahon Oct 27 '20 at 23:18
  • Alas, there are many download links like this one and there is not much publishing companies can do. I concur with your comment on the selling price and it seems former readers do not want to get rid of their book for a cheap price either: not good for the second-hand market : ) – Verbal Kint Oct 28 '20 at 07:00