On Jul 13, 9:55*am, Lostgallifreyan wrote:
George Herold wrote in news:c9ea2d68-021e-4387-8389-
:

Hi, *I usually reserve the name 'instrumentation amp' for those
differential input three opamp things that I use for bridge circuits.
That said, I do like the OPA277 for DC measurments. *Do you need the
10uV offset? *The OPA227 is a bit faster, but has a bit more DC offset
voltage.

It makes life easier if I have that low offset. I'd also strongly prefer
to use what I have, it means I can buy more cheaply and try to encourage
makers to persist in making and selling certain devices by staying with
those. I also like dual-amp IC's a lot, I find them very practical to lay out
compact boards for them.

I see what you mean about the three-amp devices, with no compromise on input
resistance between the two inputs of a single amp. I guess I use the term
'instrumentation' fairly loosely, based on intended purpose rather than the
device itself.

Have you looked at the OPA134? *There is also a newer version... I
think the number is OPA164??, (I couldn't get the TI website to work)
smaller voltage noise than the 134 but more cuurent noise and large
capacitance on the input. *What's the source impedance driving the
opamp?

Good question, and one I've yet to follow up, my understanding of these
things has only just reached that bit.. I learned that two input noise
figures can be divided one by the other to find out the ideal source
impedance to feed a given input with, but I only read that last night, these
things take time to explore... So far I've always used the basic logic that
is usually applied to avoid precise impedance matching: make sure the source
is very low, and the input very high. This is apparently fine for readio
reception and most audio couplings, so I assumed I could do it too. I took
the idea further, I assumed that if I keep the adapter as simple as possible
I can reduce noise more than it will rise due to thermal noise in large
resistor values, hence I used a passive adder with 1Meg resistors. I can
change this to 100K perhaps, at risk of drawing more power. This method
already works fine in my power meter design so I guess it's ok here too.

Oh I've been doing all sorts of noise stuff lately. I don't do much A-
D, so if I make a mistake there I hope someone will correct me.

So 20 bits is about 10^6, if you have 10 volts full scale that means
10 uV is your LSB. So if your noise is much greater than 10uV you are
losing resolution. Lets do the voltage noise first and use the
opa277. The voltage noise is 8nV/rtHz. Your band width is maybe
100kHz? (Do you have any filtering before the A-D) So the rms noise
from the opamp will be about 8nV *sqrt (100k Hz) ~ 2.5 uV. That looks
fine. What about the Johnson noise of your 100 kohm resistor?
(forget the 1Meg!) it's got 40 nV/rtHz. or about 14uV of noise...
that's starting to have an impact... And now the current noise. The
current noise ofthe opa277 is 0.2pA/rtHz. (Hey that's pretty good for
a BJT front end) times 100kohm is 20nV/rtHz of voltage noise or
about 7uV rms. (assuming I guessed you bandwidth correctly.) To get
the total noise you have to add in quadrature. (sqrt of the sum of
squares) Which means it's only the big one that matters. The 14
uVrms from the 100k resistor. Reduce the resistance a bit more and
this looks fine. (unless I've made some hugh blunder.)

George H.