TL431 PS Riple Rejection

[Phaedrus]

I'm using a TL431 for a voltage reference in a project of mine, and I was wondering how best to calculate the ripple rejection. You see, my input voltage source is fairly dirty and semi-unknown: +5VDC +/-5% <50mV_pk-pk. However that's simply the specification: it could end up being far worse, up to 100mV_pk-pk. I've also got some noise from other circuit elements.

Spice simulation shows only a few mV up to 200Hz, and hundreds of uV above 2kHz. But this seems better than I would expect from a TL431.

My noise requirements aren't actually *that* stringent; but I want to know how best to characterize PSRR of a shunt regulator with input and output filters before committing to layout.

Thanks!

[dannyf]

But this seems better than I would expect from a TL431.

The fact that the simulation runs counter to your expectation doesn't by itself mean the simulation is wrong.

[Phaedrus]

Not necessarily, no. But it's enough to do some double checking.

I've plugged in a few different simulations, including this one:
http://www.audio-perfection.com/voltage-regulators/linear-regulators/accurate-tl431-model-development-analysis-and-evaluation.html

More or less the same result, give or take some microvolts.

There's about 0.5µV_pk-pk @200MHz, probably the regulator action itself. Input noise seems to be attenuated by -17dB @20Hz, -35dB @200Hz, and -50-60dB @2kHz+. Including 555 and digital noise I'm around 50-100µV_pk-pk ripple altogether in LTSpice.

[T3sl4co1l]

Did your simulation specify ESR for any of the caps?

Transient simulations don't do noise, so the residual level is not representative of actual noise.  TL431s are usually in the 70uV (0.1-10Hz) RMS range, I think.

The two 470uF caps in parallel utterly dominate above maybe 200Hz, assuming their ESR is actually quite low (real electrolytics may be in the 1-10 ohm range, which is comparable, but not dominant to, the TL431's low frequency performance).

Note that it's not generally recommended to utterly swamp a regulator with huge assed capacitors.  Regulators work best when they "have room to think", so to speak.  Also, marginal values (like 0.01-1uF) cause oscillation.  So you're almost as well off with no capacitance at all, as with too much.  The main difference is at high frequency.

For low noise and high PSRR, I would suggest adding an RC filter or capacitance multiplier before the bias resistor, and one after the regulator if acceptable (any load will cause a DC voltage drop on the resistor, which can be compensated by tweaking the feedback loop... but that needs a few more components besides).

You can also replace the bias resistor with a current source, which will significantly increase PSRR.  High frequency performance will be limited by the transistor's Ccb figure, but by then, ferrite beads will be enough to play clean-up.

Tim

[Phaedrus]

I originally had two 10uF filter caps, one on the input and one on the output, with the 470uF and ferrite bead upstream filtering. Experimentation showed 470uF giving better results; though that was using TI's simulation.

[Conrad Hoffman]

Something most forget is that simple capacitor models that just include an esr value are only good at a single frequency. Thus, the sim predictions are only good for a single frequency. There are some more complex models available, but they're more difficult to incorporate, plus you need to know the specifics of the cap you're using.

[T3sl4co1l]

If they have ESR at all.  Most default caps don't have ESR (or ESL).  Which newbies forget.

LTSpice for example has default parasitics on most everything, which I like even less; newbies still forget to touch the numbers, and besides, the model hides within the schematic.  So I can't just read the model verbatim, I have to go out of my way to open the .ASC -- if one was even provided.  If not, I have to ask about the schematic, if it includes this or not.  Like I did above...

For simulation purposes, I much prefer having all the elements shown explicitly.  So if I were doing a sim of this, I would specifically have R, C and L in series for each cap that it matters on.  Labeled, so no one has to guess what their purposes or values are.

Tim

[Wh1sper]

My practical experience would say:
C3 and C4 are parallel. At least one of them is not needed and to big anyway
The Power for the 555 is giving you some switching noise backward. There is where you have to fight against it.
And do not connect the 555 power after the L but before. Or spend another L.
IMHO, not experience with any simulation, I like practical experience 

[Bored@Work]

What does a TL431 cost? Maybe 10 cent, 50 cent? Buy a handful of them, build that simple circuit on a copper clad PCB and measure. Do the measurement and the simulation agree? Congratulations. Do they disagree? Go with the measurements.

[dannyf]

Experimentation showed 470uF giving better results; though that was using TI's simulation.

That sentence is a little bit confusing. Did 470uf give better results in an actual measurement, or did 470uf give better results in a simulation?

[eneuro]

You see, my input voltage source is fairly dirty and semi-unknown: +5VDC +/-5% <50mV_pk-pk.

Why 0.5W resistor R3? Even 470 R3 0.25W resistor should do the job I think, while there is still > 1mA @ 5V input required curent for TL431I for proper operation.
I could consider also adding in series with R3 bigger resistor fast diode like 1N4148 to prevent discharging C3 & C7 capacitors by "strange loads" from 555 & digitals 

Is it even possible simulate TL431I while there are many versions of this IC?
I used to use TL431I (Industrial -40*C..85*C operarting range as remember)
At which temperarture ranges did you simulated this thing and which TL431 version?
It looks like mission imposible to simulate such things with high accuracy  while there are so many changing parameters in real PCB environment so prototype is a must I think while it is cheaper than... CPU power needed to simulate this with such high precision I guess 

[Phaedrus]

Aye carumba. I may just use an LM4040 with 2.5V fixed output. The BOM cost will be lower than messing with the TL431 in this case.

[David Hess]

What does a TL431 cost? Maybe 10 cent, 50 cent? Buy a handful of them, build that simple circuit on a copper clad PCB and measure. Do the measurement and the simulation agree? Congratulations. Do they disagree? Go with the measurements.

It is trivial to calculate the typical and minimum low frequency ripple rejection from the datasheet using the specified dynamic impedance.