Two feedback paths (again)

Hi guys (which includes women in my vernacular)

I've got what looks like another two feedback loop thingie.
(It's simpler so maybe I can get a result.)

I use this as a low noise voltage source.

Vin-+------------------+
| |
+Vref-+ |
p |\ |
o<--|+\ |/
t | >~Rb-|
| +|-/ |>
GND ||/ |
+---+-R-+--+---+--\ ^---out/load.
| +-C-+ R | ---
R/4 idle +Rf-+
| | |
GND---------+ Cf
+-------+

Rb~500, R = 10k, C = 10-100nF, Ridle-10k,
Rf = 500, Cf = 100uF(Al)

It's a voltage reference/ opamp/ pass element/ filter.

There's some RC filtering between pot and opamp that is not shown.

So this has some problems.
It's got an output impedance of Rf/beta
several ohms, and the temperature variations in
the uncontrolled cap mult. output filter.
So I tried taking the RC feedback from the output.
Putting the cap mult in the loop.
Worked, but lots more noise at low frequency.
(broad peak at ~2kHz.)
The now uncontrolled node between pass transistor emitter and cap mult
collector/base now had about the same broad
2 kHz. noise peak.
So If I could split up the feed back task so everything slower than
(say) 100ms was from the output, with the faster stuff via the pass
element.... that might be nice.

So as I add HF feedback from the pass element node,
I've got to subtract it from the DC path.
Is that right?
(sorry, just thinking out loud at the SED.)

George H.

 

Sounds good. One more stage and you'd have a good old fashioned phase
shift oscillator. With just two, it looks something like a Sallen-Key
filter -- feedback delayed by another RC. The feedback phase shift isn't
90 degrees, so it's "guaranteed" not to oscillate...

In reality, a couple microseconds around the loop can fill in the last bit
of angle required to make an oscillator. Effectively, as the filter's Q
approaches infinity, that little extra bit will push it "just past
infinity", loop gain goes >1, and it oscillates.

Real numbers are modular at infinity (well, in electronics they are), so
"greater than positive infinity" is the same as "greater than negative
infinity". So if you drew an RLC equivalent circuit, the loss would be a
negative resistance. By extension, a negative resistance amplifier
necessarily has a negative Q.

Seconded.

Incidentally, this is also handy with op-amps (notoriously bad at driving
capacitance or capacitive cables), shunt regulators (like TL431) and LDOs
(i.e., an upside-down TL431) -- the difference with LDOs being, they don't
give you the internal FB node, so you're screwed and have to make do with
ESR.

Tim

 

Good idea.

 

Hi Jim, Well no real 'target' spec's. I just scribbled down the
circuit taking the feedback from the output... then tried it the next
day.

The circuit in use (with feed back after the pass transistor) drives
a 75 ohm heater from ~30V supply, so ~350 mA. The noise is in the
2-3nV/rtHz range. (I assume there is some 1/f corner at low frequency
too, but I don't know where it is.)

George H.

 

"No!!!!!", Flails arms over head and runs screaming from room.

George H.

 

OK I hadn't thought that far ahead.
re ~25mV/Ic ~ so most of the time less than an ohm.

George H.

 

to inverting input <----R=10k----+Hey I just stuck in a big ass cap in the feed back path.
(The 20uF as shown above.)
The output low pass has a tau of 0.5k ohm*100uF = 500ms.
and a high pass feed back of R/4 (2.5kohm) * 20uF = 500ms.

This seems to work.
Though I have to do some 'real' work today and can't test it totally.

George H.

 

No, it can be a good idea in voltage references!

You can often use the (boosted) output to source the drive to
the reference. So it is almost completely immune to supply variations.

 

Grin... yeah, Well that's a long story. But the heater is down the
bottom of a LN2 probe that is used to measure johnson noise of
resistors as a function of temperature. Depending on the resistors
used there would be some capacitive pickup from the lines feeding the
heater. (they shared a common cable.) So... cap multiplier to the
rescue!

Anyway I might have a need in the future for a nice low noise power
supply.
(And just thinking of ways to improve what I've already done.)

George H.

 

The opamp is an LT1013 (input voltage can go down near the negative
power rail)
and the transistors are TIP31C's...
Both just pieces I have from other projects.

George H.

 

Oh sorry I thought Jim was making a joke.
So feed the reference from the output voltage....
Thanks I'll remember that too.
In the current circuit I don't think it's noise on the supply rail
that is the limiting noise source.... not really sure what it is...
(the circuit works so I moved on. :^)

George H.

 

Oh sorry I was a bit loose with the numbers. (And a bit loose with
the schematic.)
The reference is an LM4040 - 5.0 volts
The supply is 30 volts, 0-25 volts across the 75 ohm heater


It's really a +/-15 volt supply, but the above circuit treats the -15V
rail as ground.... well that's not quite correct either. There's a
cap multiplier on the negative rail...and the output of that is what
I've called ground above.


I didn't want the heater to put any currents through the 'real'
ground.

George H.

 

Wed, 30 Jan 2013 07:47:52 -0800 (PST), George Herold

Voltage controlled current source?

 

Well now I feel clever; I "invented" that one myself couple of years ago
for same reason (noiseless).

 

Wow, cool.
I was wondering if I needed some resistance in the highpass feedback
path..?
(Would that add more gain?)

Say if you look at the emitter of Q1 do you see a noise peak at about
~2kHz?
I see just about the same level of noise with either feedback
topology.
I've no idea where it comes from... not a big deal 'cause I see no
sign of it on the output.

In real life or in spice?

George H.

 

Ahh, in real life? If so how?

George H.

 

Perhaps not altogether noiseless, but it *is* possible to make
active room-temperature circuits presenting a resistive impedance
with noise temperatures (well) below ambient.

Jeroen Belleman

 

Well, in real life in LTSpice.

You mean LTSpice isn't real life?

Another one is those handy 1E9H inductors as noiseless loads

 

Are you talking about active damping type circuits?
(The few I've seen always seem to have some other trade off...
like less noise but over a reduced bandwidth.)

Got any examples?

George H.

 

Indeed I do, and yes, there are trade offs. I've mentioned
this a few times on s.e.d., in threads on LNAs. I've made some
with input resistances that appear to have only one third of
the noise a real resistor of that value would have.

It's possible to do much better still if you're willing to
put up with a narrow bandwidth.

Jeroen Belleman