Is this circuit stable?

http://preview.tinyurl.com/36vncy

Please let me know if this schematic has any glaring errors or problems.

It's a DC gain stage, with adjustable offset and scale. Precise component
values might change, but the basic form will not, unless it is bad in some
way I can't see.

The first half of the op-amp makes a new ground, so that there is enough
negative voltage to allow through-zero offset tweak to set to zero. The
supply will be a small 1 watt 5V to 24V DC-DC power converter, so that the
op-amp can have >20V supply allowing a wide range output while the supply
is only +5V. The power converter will also isolate the supply input so that
the new ground can be connected to the signal input and output ground which
will be commoned to the chassis and earth. The op-amp and the 2K2/Zener
network are both fed directly from this isolated 24VDC supply.

The second half of the op-amp amplifies a DC signal on the non-inverting
input. It's set up so the input impedance is as high as possible, but a
resistor might be placed across the input in any finished circuit.

When I built this, it worked accurately and quietly on a pinboard, and on a
small PCB I made for it, when reading the output on a Fluke voltmeter, but
when wired to an LED panel meter in a small metal box the reading was
erratic. I traced this to the meter changing its current draw when
different segments lit, and the ground circuit current seemed to be
affecting the gain stage output, causing a feedback effect that I couldn't
eradicate, no matter how I arranged the ground wiring.

Is there something about this circuit that is inherently unstable or
inaccurate? If so, what should best be changed?

 

Hmm, meant to crosspost to s.e.d and s.e.c, and borked it, but nm....

 

This is the first appearance in this thread on S.E.D., I didn't get the
cross-post set correctly late last night. On S.E.C., please ignore the
earlier post, this one is updated with an extra paragraph near the end.

http://preview.tinyurl.com/36vncy ('DC Gain Stage' diagram, PNG format).

Please let me know if this schematic has any glaring errors or problems.

It's a DC gain stage, with adjustable offset and scale. Precise component
values might change, but the basic form will not unless it is bad in some
way I can't see.

The first half of the op-amp makes a new ground, so that there is enough
negative voltage to allow through-zero offset tweak to set to zero. The
supply will be a small 1 watt 5V to 24V DC-DC power converter, so that the
op-amp can have >20V supply allowing a wide range output while the supply
is only +5V. The power converter will also isolate the supply input so that
the new ground can be connected to the signal input and output ground which
will be commoned to the chassis and earth. The op-amp and the 2K2/Zener
network are both fed directly from this isolated 24VDC supply.

The second half of the op-amp amplifies a DC signal on the non-inverting
input. It's set up so the input impedance is as high as possible, but a
resistor might be placed across the input in any finished circuit.

When I built this, it worked accurately and quietly on a pinboard, and on a
small PCB I made for it, when reading the output on a Fluke voltmeter, but
when wired to an LED panel meter in a small metal box the reading was
erratic. I traced this to the meter changing its current draw when
different segments lit, and the ground circuit current seemed to be
affecting the gain stage output, causing a feedback effect that I couldn't
eradicate, no matter how I arranged the ground wiring.

The circuit works ok with supplies other than 24V, and the two 100K
resistors in the divider are meant to keep the zero point fixed even if the
zener voltage drifts with temperature. The zener value itself is chosen for
low thermal drift.

Is there something about this circuit that is inherently unstable or
inaccurate? If so, what should best be changed?

 

It looks like it should function.

That is a confusing dual use of the word "supply".

A complete schematic would help this description make more
sense.

Is the gain stage intended to produce only positive output
voltages?

By the way, there might be a better way to do the gain
adjustment. Your method puts DC current through the wiper
contact, and that might, eventually, make its contact noisy.

Since the opamp bias current is very low, you can avoid
almost all contact point current by connecting the wiper to
the input and adding a pair of resistors at each end to
complete the feedback divider. This method also gives you a
larger gain range for a given total divider resistance and
pot resistance.

A high resistance path to signal zero volts is handy if you
might ever disconnect the input signal. That will keep the
opamp from saturating under that condition.

The reference voltage section must also supply the full
"ground" return current the right opamp produces. If the
output current gets anywhere near the opamp current limit, I
would expect to see the reference voltage amplifier start to
malfunction.

An integrated voltage reference chip would probably do a
better job (less temperature sensitive, sharper voltage
regulation and lower noise) than any zener. Take a look at
this data sheet:
http://cache.national.com/ds/LM/LM431.pdf
With two resistors, you can program it to whatever voltage
you need.

Not so much inaccurate as just having a limited current
capability that you need to be aware of.

That is not to say this is the best circuit for the job,
because you have not very well specified exactly what the
job is.

What range of input voltage?
What minimum input impedance?
What gain range?
What load impedance?
What frequency response?
What temperature range?

 

John Popelish wrote:

Hold on. I just realized that the zener reference section
shows no connections at all to either the opamp supply pins
or the input/output common (the bottom right pin, I assume).

Either you missed some important connections in the
schematic, or this circuit cannot function.

 

Thankyou, lots of good stuff there, it helps me to think. I added some
context to that file.

http://preview.tinyurl.com/36vncy ('DC Gain Stage' diagram, PNG format).

It does, till I move it into its permanent home. The meter instability
is in 5mV range, at worst, 50mV, but did not exist at all until I moved the
circuit from pinboard to PCB is a metal case with chassis mounted BNC
sockets. Seems that context matters, so I agree, I should have posted more
info than I did.

Ok. First, the supply is +5V to +15V or so, sent to an LM2904T 5V LDO
regulator. This allows battery use as well as most cheap DC output PSU's.
The supply ground is commoned to the chassis at the regulator's tab, OR at
the sensor input ground if the LED panel meter's floating ground is shorted
to the supply ground, which I chose as best (and recommended) option,
isolating the regulator tab from the case to prevent a ground loop, which
lessened the reading instability problem but did not remove it.

The regulated 5V feeds the LED panel meter as directly as possible to keep
the meter's ground current from affecting the gain stage signal ground. The
gain stage gets 24V from a 1W power converter that takes from the 5V
regulator and isolates the 24V supply to the gain stage so that its new
ground can become the real system (signal) ground.

Nice, I'll give that some thought and testing time. If I make new boards
for this thing, I'll probably do that. It's not a high priority right now
though, I've seen similar devices do it the way I have without problems. I
can see it's not the best way, but I don't think that changing it will cure
the problem caused by the LED panel meter's changing current draw.

I'll load it lightly by meters with huge input resistances. This is what
bothers me; as there is not supposed to be a significant current, I don't
know how the LED panel meter's current draw can change the gain stage's
signal levels so much when it's directly supplied by the main 5V regulator.
I first thought of poor load regulation in the LM2904, but it's not that.

I've thought of using something like this, but would rather use a fixed
voltage 2-pin device that could occupy similar volume to the zener, if
possible. I abandoned the tightly packed box I was hoping to use, so I
don't mind if the thing stands up off the board a bit. It just has to
fit the space on its plane. I need to consider other sources of inaccuracy
first. If the op-amp offset voltage drifts much, that will be a weaker
point.

The divider made by two 100K resistors will halve the voltage at the zener,
whatever it is. I'm more concerned with asymmetry, if it can exist, between
the new ground set by that midpoint, and the offset adjust to zero made by
the offset tweak pot. Thermal drift in the op-amp might be one cause of
asymmetry, but that wouldn't cause the main problem of output changing when
the LED meter reading changes.

In my first use of this design, a laser power meter, an activated charcoal
coated TEC in a metal barrel to shield it from ambient heat and provide a
thermal mass. (I reconditioned a cheap Scientech head I got off eBay...)

+10 µV to +2V DC.

Unknown. I designed it for the highest input impedance a modest priced op-
amp can give me, so it reads the voltage of a sensor while drawing a
current so small I need not consider it. If I had to adapt it to something
requiring a load resistance, I'd add a resistor between input and ground.

About 11, for the TEC's voltage boosted to give a 1mV per mW output for
power to be read on a voltmeter. I'll adjust the values of the gain stage
network for wider adjustment than I currently have for scale, but this
isn't where the problem is...

High as possible. I noted your points about current, but I'd intended this
to read and write voltages. The Fluke meter I used, and the LED panel meter
that caused the bother, both have extremely high input resistances. If I
needed current drive, I'd add a voltage follower, or arrange an entirely
different circuit with that in mind.

Nothing strict. I chose an op-amp that would be stable without special HF
filter requirements for stability. I only need this to work fast enough to
show changes a human eye or ear can keep up with, so perhaps 200 Hz. At
fastest, I might supply an audio input with DC blocking disabled so I can
turn a wave recorder program into a 2KHz sample rate data logger, so
definitely never higher than 1 KHz.

Not thought of that much because the op-amp will drift a bit whatever else
I do... Any temperature a human might be comfortable in. Maybe extended to
0°C to 85°C would be good, more secure. I'll just calibrate the thing at a
settled room temperature, most times.

Mentioned in the original text, but I agree, it should be in the diagram.
Op-amp is supplied by the same 24V that feeds the zener and divider
network. I've added to that diagram:
http://preview.tinyurl.com/36vncy

 

To clarify: I'm not talking about repeated oscillation here. I know we can
hear to maybe 20 KHz, but a one-shot change shorter than about 5
milliseconds isn't easily resolvable. In short, I'm monitoring aperiodic
signal changes that might be perceptible to human senses.

I don't think any of this will bear on the meter reading instability
problem, but I might as well be complete in detail if I want nice full help
in answer.

 

< LOTS OF STUFF DELETED >

Ok, from what I have read here, it appears that you maybe getting
a common connection or, near it with the supply to your common ground.
The common in the meter is going to cause the virtual ground you
made to tie together.
Even if you have a meter with a isolated common on the input, some
where along the line it's going to join to the common of the chassis...

The supply your using should be totally isolated, no common of the
supply should be connected to the chassis. This means a supply with a
transformer in it or a DC-DC converter.

AT this point, you can use that virtual ground like you did but only a
higher current type so that you can supply the meter with its regulated
5 Volts from one side of the main source and the TAB (common) goes to
your virtual ground.
The virtual ground can also be connected to the chassis but not the
main supply..

So for example, lets say we use a 15 volt battery. The +&- of this
source should never come in contact with any commons/chassis grounds,
only does the virtual ground you made comes in contact with the common ..
You can then use the + side of the battery for example to operate
the regulator. The reg common goes to the virtual ground.. etc....

I think you get it.
P.S.
I think you need to use a higher amp handling OP-AMP of you want
to go this way.
It would be better to simply use a DC-DC with Dual output converter
for your voltage gain circuit and tie the CT (center Tap) of the
converter to the main common along with everything else.

If you want, you could always create a - volt generator that is
regulated via a -post regulator..
For example, assume you have a 15 volt supply, using a 555 timer in
astable mode into a CAP that is rectified for - volts, then pass that
into a -REG..
You can then use the + Reg witch will have plenty of current on
reserve to supply the meter and the -reg to supply the -side of your
voltage gain Op-amp.

The 555 is good for 200 ma's if memory serves?

This is a commonly done trick in cases where single supplies were used
and - voltage was needed.

 

(snip)

This helps a lot.

The LDO I am guessing is actually an LM2940T.
http://cache.national.com/ds/LM/LM2940.pdf
These things are notorious for oscillating without the
correct output capacitor effective series resistance (ESR).
The graph on page 10 shows this allowable range for the
resistance in series with the output capacitor. I usually
use a very low ESR capacitor with an external .47 ohm
resistor, to make them behave. If you have a scope you
might check yours. The circuit board may add less
resistance than the plug board did.

Again, the trouble sounds like it is related to the regulator.

If I am following your schematic, the regulator ground is
connected to the DPM, and then to the output of LF412A, yet
that output is trying to produce 2.5 volts with respect to
that ground, powered by the 5 volt supply. I don't think
you can ground the output of LF412A.


(snip)

I think the circuit has a major flaw relating to tying the
meter input ground to the power supply ground. I think the
meter input ground has to be only the 2.5 volt reference, so
cannot tie back to ground through the case.

(snip)

(snip)

The circuit components should work pretty well (with the
exception of reference drift from the zener).

 

The LED DPM must be muxing the display. This will generate noise on
the pseudo ground. Sounds like a bad idea to me.

TI used to make a pseudo ground generator. I don't have the number
handy, but I got a sample and used it in a project. It worked fine,
but this isn't rocket science. I also favor using a regulator over an
op amp for such applications.

 

Hold on again. I keep forgetting that the 24 volt supply is
isolated and powers the two opamps (I assume). So the
reference opamp does not so much produce a +2.5 volt level
as it forces the - rail of the 24 volt supply to -2.5 volts.

That voltage will be perturbed by the high frequency
capacitive currents that couple the switching in the 24 volt
supply to the output, relative to ground.

You cannot just bypass the -2.5 volt rail to ground with a
big capacitor, to absorb this noise current, because the
opamp doesn't like a big capacitor effectively on its output
(even if its effective output, in this case is its negative
supply pin). But a capacitor in series with a 10 to 100 ohm
resistor might keep it happy.

I still think the problem revolves around the LDO breaking
into oscillation and quenching out from the noise provided
by the isolation converter.

You also don't show any bypass capacitor across the power
supply rails of the opamps, which would help stabilize them
against breaking into oscillation, but perhaps the isolation
converter includes a good output capacitor that is very
close to them. Still, I would probably try adding a .1 uF
across each opamp. Another from the -2.5 volt rail to the
divider input to the reference opamp might help, too.

 

You don't have a lot of good choices for the capacitor on
the output of the LDO.

Here is an example of a low ESR capacitor that can use an
external resistor to make sure the total resistance is
inside the single decade that makes for stable operation:
http://www.panasonic.com/industrial/components/pdf/ee_sp_cap_cd_dne.pdf
I would probably use the 8 volt 22 uF cap because it has the
lowest ESR.

 

I thought of that, and scoped everything I could scope...
The capacirot is exactly as specified, 22µF with 0.5 ohm ESR. It's a
tantalum, SMT, soldered directly across GND and OUT right at the base of
the TO220 package. (And no, I didn't damage anything with heat.

If there is any better way to satisfy the exact requirements for this IC,
I'd like to know what it is. I also tried with a standard 78S05, just to be
sure. Same problem with the meter reading causing feedback to the gain
stage...

 

The power converter isolates between the 5V input and it's own 24V (+/-
12V) supply out, so you can. That op-amp's first half output is the gain
stage section's only contact with chassis or earth or any other kind of
ground that might exist.

 

wrote in

I had half an op-amp spare, and a tight space...
This thing works. It's only when it goes into the metal case and ground is
wired to it that things go agley. I considered basic proximity too, on the
pinboard, and lifted the gain stage (wired to pinboard, mounted on box
baseplate), inverted it, held it as close to the other stuff as I could,
and could not provoke the error that way. Only by deliberately complicating
the ground circuit and making the DPM draw current from points remote from
the regulator's own connection, could I reproduce the problem on the
pinboard.

The small power converter throws much more noise on the line than the DPM,
I scoped this, it was prominent, though low level. I couldn't see any noise
that I couldn't attribute to this, or an even lower level of general white
noise mush. Neither affected the working of the circuit. I tried using
capacitors to reduce the noise from the power converter, successfully, but
that didn't solve the problem.

 

Isolated... There is another reason why I have to consider that tie though,
the common mode to the DPM allows 1V excursions. Maybe I could risk a lower
zener voltage, below 2V, to allow floating meter input with excursion
within bounds, but as I tried the other recommendation on the meter's badly
uninformative data sheet, of placing a 10K resistor in place of that tie,
and so no change in results apart from a small (about 2mV) offset in the
readings I was getting with gain stage input shorted, I don't think the tie
was a problem. I got better results that way on the pinboard too. The metal
box made things worse to the point where the improvement was masked, but I
still think it's probably the right way. That's why I isolated the 5V
regulator tab from the box to break a ground loop. That helped, it lowered
the erratic reading variance. Before doing that, (and also making the main
supply to the box ground at the regulator gnd and not at the box at first
contact opportunity). I got the reading errors to range over about 0.8mV
instead of over 50mV as they had when I'd just grounded with no thought for
the loops that might be formed.

 

And if that drift is halved along with the actual reference itself? (Two
100K's...) Shouldn't it cancel, being symmetrical about the zero point?
That is the plan, anyway, it's why I didn't go for some more exotic and
precise reference. I did consider drift, even so, it's why I chose that
voltage, I read that Z5V1 (or sometimes Z5V6) has the smallest thermal
drift.

 

The LDO has it's best interest cared for. Tantalum chosen for the exact
best ESR as specified by data sheet, solder directly to its pins, can't get
closer or better contact than that.

I tried caps in various places. Not one cap placement affected the meter
reading causing feedback to the gain stage. Only messing with ground
circuitry did that.

I scoped the output, input, gnds, everything, at various frequencies and
gains. The only periodic waveform was a very low level spike at around 150
KHz (if I remember right) from the power converter. It got everywhere in
the system, but had no effect. I removed most of it with a cap, saw that
this did not make any change to the feedback problem, so omitted the cap.

The changes on the output were seen to be in sync with the LED changes.
It's not an HF noise thing, it;s a gross-current-change thing. BUT, it
happens even when the gain stage's gnd is deliberately not in a loop, or
even shares a length of wire, with the DPS's supply ground. This
arrangement solves the problem, on the pinboard, but not in the metal case.

The only thing I can think of is that the thing might still be HF or even
RF related, but for VERY short durations, during the change in current when
the reading updates. If the meter reads the instantaneous voltage it finds
at update time, it might be reading a tiny one-shot pulse caused by itself.
That sounds wrong, because you'd think that the measurement would be taken
and finished before the meter can change the display, but what if it's not
so? What if there is a small fraction of a second during which the reading
is changing at same time as input is being sensed? Any overlap could cause
a spike to be caused and read back, and the duration would be too short to
catch on my scope, which can't see one-shot events.

What I might do is just forget the DPM and provide the gain stage output to
a BNC socket as before, and also to a pair of 4mm sockets for a multimeter.
The DPM was just a nice thought, but the gain stage works perfectly if I
just use a multimeter. That won't have any shared power ground to think
about.

 

So true. I actually got lucky, I just happened to have the SMT tantalums
around, they had the exact spec wanted. Perfect fit on the legs of the IC
too, it really doesn't get better.

 

Thankyou, but look again. There is one. There is no fault of commoning
different grounds. If there was, this would never work, anywhere. It DOES
work, on the pinboard, it only gets problematic at 5mV levels or so when
boxed in a metal case. None of which has anything to do with a gross error
between a floating ground tied to a -2.5V line by mistake. The NMA0512DC
power converter provides the needed isolation, it's why I chose it.