Suggestions for stabilising a TL431 + MOSFET current sink

[Grant]

Hi there

I've built a one amp current sink and at present it is uncompensated
and it is performing quite nicely as a 200kHz power oscillator, with
about 2V p-p across the 2.579 Ohm compound current sense resistor.

Following text is based a screen scraping of a web page I just put
up, along with some pictures to describe this issue at:

http://grrr.id.au/current-sink/

I've not done much with MOSFETs in linear amplifier mode, so some
suggestions for stabilising this one Amp current sink circuit are
needed.

The circuit uses an N-channel MOSFET for the pass element, a 0.5%
accurate TL431B reference and a couple dozen 0.1% resistors provide
for accurate current sensing with a 2.579 Ohm shunt.

The circuit 'works' in the sense that the magic smoke stays where
it belongs and no components get overly warm. The circuit needs
stabilising as the thing is at present uncompensated and oscillating
around 200kHz.

The circuit diagram:

http://grrr.id.au/current-sink/CurrentSink-circuit-750x736.jpg

Circuit description: Fixed 1.000 Amp current sink. Of that one Amp,
983mA goes through the current sense resistor, the remaining 17mA
current goes via a trimpot scaling circuit to feed the TL431B's
Ref pin. A 13V zener limits gate drive voltage during circuit
saturation, and a 110 Ohm gate resistor is placed close to the
MOSFET.

A bank of 24 by 61R9, 250mW, 0.1% resistors to form R9, a 2.579
Ohm, six Watt current sense resistor. Calibrating the current
sink to 1.000A is done by adjusting the trimpot to so that 2.535V
is measured across the current sense resistor, with a 4 1/2 digit
Fluke multimeter.

Photo of the oscillation waveform:

http://grrr.id.au/current-sink/CurrentSink-waveform-R9.jpg

The oscillation waveform shows about 2V p-p across the 2.579 Ohm
current sense resistor. The MOSFET is happy enough on its large
heatsink, it gets very warm to the touch but it's not too hot
for continuous operation.

Photo of the built circuit:

http://grrr.id.au/current-sink/CurrentSink-photo.jpg

Photo of the thing shows placement of the MOSFET gate resistor
up close to the MOSFET gate where it belongs. The large heatsink
allows continuous operation, as the input voltage may go up
to 30V, from the nominal 24V.


What's needed are suggestions for what stabilising components
to use around the TL431B. Circuit frequency response is not
that critical as this current sink will be used to measure
the value of compound power resistors up to about ten Ohms,
while trimming their value.

The circuit is powered from a 12 cell 40AH or 100AH SLA battery.

Web page is here: http://grrr.id.au/current-sink/ where I'll add
further information as I play with the thing, and as ideas you
contribute are added.

Thanks,
Grant.

 

[Tim Williams]

Your MOSFET is shown as a P type?? 33N10 certainly isn't..

Compensation across the '431 is the way to go. Besides 10nF, you may need series resistance to its input, to set the RC time constant. I don't know what R5 and R6 are; R7 is small enough at 150 ohms to be tough to drive. It's not like the input takes much current, 10k ohms is plenty.

Tim

 

[Grant]

Your MOSFET is shown as a P type?? 33N10 certainly isn't..

Oops, I do mean N type, the part number is correct, last time I was
drawing MOSFETs, they were high side switching with P-channels, didn't
notice I drew the arrow wrong way round here.

Compensation across the '431 is the way to go. Besides 10nF, you may
need series resistance to its input, to set the RC time constant. I
don't know what R5 and R6 are;

Your eyes probably didn't believe the numbers 20, 12, 12 on the diagram
for trimpot, R5 and R6, the values are in Ohms. I did mention 17mA
going down that part. Yes, it low, I wanted fairly high current
through there to swamp the 20 Ohm trimpot's lower spec tolerance
so the adjustment range worked as expected (it did

R7 is small enough at 150 ohms to
be tough to drive.

Not from the ~2.5 Ohm source, it isn't

It's not like the input takes much current, 10k
ohms is plenty.

Well, I put 200 Ohms in there between R9 and TL431.R, to get
something to work against, then added 1.0nF from R to K, plus
15nF & 3k0 from R to K.

Just come back to the PC now to see if anyone replied to my post,
after adding some caps and stuff to the circuit. Funny thing is
that after writing up the problem, I had to problem guesstimating
a set of values that stabilised the thing first try, when I went
back to the soldering iron.

I think it was a confidence issue, because last time I tried to
stabilise a linear MOSFET power amp (8 n-channels in parallel) the
thing simply didn't work out for me. That was 18 months or two
years ago, and I think I was given some logic threshold devices
to work with.

This current sink is now over-compensated (I think), but I can't
tell until I make a little load switcher to give the thing a step
change so I can watch the response and fine tune it.

Drifts only about 1mV when it warms up, another mV variation from
short circuit to 20 Ohms load. So I think it will be okay for
what I want.

I'm expecting to trim compound power resistors to about one percent
accuracy with this current source, so it is good enough for that.

The compound resistors I'll be trimming only need to be accurate
enough for monotonic operation over a 6 bit range, so aiming for
1% should be okay. Unless, of course, I discover it's not okay.

Grant.

 

[Grant]

....
I recommend a cap from K to R on the TL431. 10 nF is probably more than
enough.

Thanks, see my reply to Tim, I put 1.0 nF in there, after adding 200 Ohm
to R9 for the cap to act against.

Grant.

 

[Tim Williams]

The compound resistors I'll be trimming only need to be accurate
enough for monotonic operation over a 6 bit range, so aiming for
1% should be okay. Unless, of course, I discover it's not okay.

And if it's not, you can always go cascode.

You could even make a "corrected cascode", so an op-amp drives the cascoding transistor so as to keep its source at a constant voltage. Something you're already doing with this transistor, so you just do it twice. That should essentially eliminate early effect from consideration. As is, TL431's gain should be more than high enough to not care about this yet.

I'm sure you've taken stock of such things as tempco? A fan might not be a bad idea; turbulence is supposed to make things much noisier, but that should still be less error than the error due to tempco, if the resistors are getting very warm. The resistors down inside that monster stack will run slightly warmer without forced air. Maybe not by much.

The TL431 itself is only about 1% (2.475-2.525V), so you'd need a much better reference to do anything more accurate anyway. At this point you start getting into precision *everything*, calibration procedures and, ehh maybe not as far as tracable references...

Tim

 

[Grant]

And if it's not, you can always go cascode.

You could even make a "corrected cascode", so an op-amp drives the
cascoding transistor so as to keep its source at a constant voltage.
Something you're already doing with this transistor, so you just do
it twice. That should essentially eliminate early effect from
consideration. As is, TL431's gain should be more than high enough
to not care about this yet.

MOSFET has Early effect? Or, do you mean the TL431?

I was concerned the TL431 wouldn't have enough gain, but it seems
good enough for this task.

I'm sure you've taken stock of such things as tempco?

No, such things I measure or get a feel for once built

A fan might not be a bad idea; turbulence is supposed to make
things much noisier, but that should still be less error than
the error due to tempco, if the resistors are getting very
warm. The resistors down inside that monster stack will run
slightly warmer without forced air. Maybe not by much.

The resistors sharing 2.5 Watts, 104mW each. Yes, they feel a
little hotter than the heatsink, maybe a fan _is_ good idea,
particularly if I'm using the thing in summer, it gets to 40'C in
here. At the moment, my heater is struggling to get the place
up to 20'C

The controller is bolted really close to the heatsink, there
is a noticeable drift on warmup, but only a mV or so. And, I
measure shunt voltage at the sensing end, not the where the
current is being carried -- four wire measurement.

The TL431 itself is only about 1% (2.475-2.525V), so you'd
need a much better reference to do anything more accurate
anyway. At this point you start getting into precision
*everything*, calibration procedures and, ehh maybe not as
far as tracable references...

Texas TL431B I'm using here is 2.495V +/- 12mV, 0.5% initial
accuracy, (TL431A is 1%, standard part is 2% -- TI datasheet)
plus I trim the current via voltage across the 0.1% sense
resistor, so circuit accuracy depends on the 0.1% shunt resistor
and tolerance of Fluke 4 1/2 digit multimeter (it _is_ quite
an old meter though, Fluke Model 8050A).

The circuit drifts about 1mV across the shunt for warmup, another
mV for load varying between near zero to 20 Ohms, and another mV
for line regulation (24.8 to 29V on charge, but it's not normal
to charge the batteries whilst using them for this sort of thing,
also not good for the batteries either as the charger not smart
enough to compensate for a load).

Since current is measured at about 2.5 mV per mA through the
shunt, the circuit is accurate to something better than 2mA
plus 0.1% shunt and <mumble> tolerance for the meter used to
adjust the shunt voltage to match the calculated value (2.535V).


I don't know how to tell LTSpice how to model the TL431 and
trust I'd get an accurate simulation. So it's the old soldering
iron method for me. I think it doesn't do trimpots either, but
it's been a while since I use LTSpice. Might have a go at it
later.

Grant.

 

[Grant]

I've built a one amp current sink and ...

....
It's now compensated and on overnight test New circuit is up too.

http://grrr.id.au/current-sink/CurrentSink-circuit-2-750x567.jpg

or, see http://grrr.id.au/current-sink/ for the revised web page.

Restoring C1 across the 13V zener had no effect, adding supply bypass
and the usual feedback compensation components worked first try. The
circuit may be under-damped, I need to make a load step switcher so I
can see how the circuit responds for dynamic response, another day.

Makes small sparks when shorting the output with a nickel plated
alligator clip to stainless steel resistor lug. Would you expect
that from a one Amp constant current sink running from 24V?

Power supply is fairly stiff, a pair of 12V, 100AH SLA batteries
through a 35A auto spade style fuse and a couple or three metres
of cable.

Grant.

 

[John KD5YI]

On 7/4/2010 6:22 AM, Grant wrote:

(snip)

I don't know how to tell LTSpice how to model the TL431 and
trust I'd get an accurate simulation. So it's the old soldering
iron method for me. I think it doesn't do trimpots either, but
it's been a while since I use LTSpice. Might have a go at it
later.

Grant.

Use the LT1431 which comes with LTSpice. Look in the Op-Amps folder.

Cheers,
John

 

[Tim Williams]

I hate that mosfet symbol. I draw mosfets as



|
|-
--------| |
|>
|


and


|
|<
--------| |
|-
|


which is much more obvious.

Well gee, that's just patently wrong. The arrow is the substrate, so you've drawn a P type substrate follower. Or something. Which has absolutely no physical meaning whatsoever, and is forward biased, which is completely useless.

On the other hand, this symbol is EXACTLY what it is.
http://webpages.charter.net/dawill/tmoranwms/Circuits_2010/Current_Sink.png
Gate: insulated. In order of contact with the gate oxide, the source, substrate and drain all appear on the other side. Source and substrate are tied by wiring. Substrate is a diode junction. The line and arrow mean "diode". That diode is reverse biased. FETs are not minority carrier devices, so that arrow MUST point backwards (except for cases like synchronous rectifiers, when you are using it backwards).

I don't suppose you use the symbols that show the extra zener as well? I've always been tempted to rebel against that symbol by drawing the substrate 'dash' with zener wings.

And you could draw a MESFET with a schottky "long S" gate, same as schottky BJTs. It would even be acceptable to use one line instead of a gate and three dashes, since they're pretty much all the same hunk of material, no insulators.

Tim

 

[Grant]

(1) Insert moderately high value resistor between pot ("VR1") tap and
"R" terminal of TL431

I did in new circuit, 200 Ohm, seems low but then there is 17mA going
past the 'R' terminal.

(2) Then a cap "K" to ""R"

The TL431 is, after all, an OpAmp of sorts

Yes, I tried the opamp classic style controller feedback: small cap
plus larger cap with series resistor, surprisingly it worked on
first try values, but might be underdamped. I'll get back to that
when I make a load delta switcher, just an oscillator driving a
big MOSFET switching in 5 or 10 Ohms extra load. Need one of them
anyway for testing responses, last one I made added voltage into
feedback loop, hardly reusable.

Caution! Still a potential for oscillation... the cap from "K" to "R"
may need a small series resistor to introduce a "zero"/phase-lead,
because of the MOSFET capacitances.

I don't see any oscillation, I've got a noisy mains ground here so
there's much noise on the CRO even when the thing is powered
off And the thing is powered from batteries isolated from earth.


I'm renting, so there's not much I can do about the bad earth --
did ask for an electrician to check it years ago, he pulled earth
wire out of the neutral bar, measured resistance between neutral
and earth and pronounced it okay! Ignoring the fact two other
units on same earth point were looping earth via their neutral
blocks.

Hardly a proper earthing test, and we have odd law here that says
bad earthing that met standards in place when the units were built
is okay -- even though running an earth connected to water pipe,
and going through taps before the pipe goes underground breaks
modern rules.

Grant.

 

[Grant]

On 7/4/2010 6:22 AM, Grant wrote:

(snip)



Use the LT1431 which comes with LTSpice. Look in the Op-Amps folder.
Thanks.

Cheers,
John

 

[Grant]

You could eliminate D3, R3, C2, C3, and R10. But it would be fairly
slow. A fast load box probably needs opamps.

Yeah, right -- and let the gate drive go up to 24 to 30V on no load?
No thanks.

Grant.

 

[Grant]

Would a small cap between GND and the junction of R4 and The mosfet gate
be of any use? SOmetimes a stopper resistor (R4) is not enough on its own.

Might try that once I can do some dynamic load testing, thanks.

Grant.

 

[Grant]

I use the old symbol. I don't like it, but there it is.

Yours looks a little like an IGBT. Might that cause confusion?

I use something like what I see in the data books, at the risk of
being confused about the arrow direction as I was this time

Tim's explanation that it shows the reverse biased junction diode
at least makes sense, rather than confusing with IG whatevers.

Grant.

 

[Grant]

You could eliminate D3, R3, C2, C3, and R10. But it would be fairly
slow. A fast load box probably needs opamps.

Twice you've suggested slugging a large cap directly across the TL431,
I'm wary of doing that because the MOSFET is inside the feedback loop.

What am I missing here?

Grant.