three op-amp instrumentation amplifier offset in psoc1

[gary1wang]

Hi guys

how can i three op-amp instrumentation amplifier offset in psoc. three op-amp instrumentation amplifier has analog ground reference. and i am using bandgap-+bandgap.


Best Regards

[bobmarlowe]

As long as you stay with your inputs within the limit of Vss to Vdd there is no need to offset the amplifier. For compensating small offsets a resistive voltage-divider wold be sufficant.

Bob

[danadak]

Gary are you -

1) Trying to remove offset error and drift from the IA ?

2) or trying to offset signals into the IA so they meet common mode
input requirements ?

It would help if you describe characteristics of your input signal, what you
want to do to that signal, and why you are selecting IA (I assume because you
have a high CM environment or a bridge sensor application...) ? Also the precision,
resolution goals of design.

Regards, Dana.

[gary1wang]

I am trying to to get rid of offset and drift in the IA. Since i connect negative input on refmux as reference. and the positive input measures the stable voltage level . And this IA output to ADC.

In IA datasheet, the gain error is not specified, because i am using gain euqal to 1.

[danadak]

Gary, you do not list design goals, ie. I want 16 bit absolute accuracy ove T & V,
that guides you as to whether you can achieve it.

The errors you have to deal with are (Sensor, PSOC IA, PSOC A/D) :

1) Drift
2) Non Linearity
3) PSRR
4) Noise
5) Gain errors

Example, lets say you want 16 bit system resolution, and accuracy of 1 lsb. So
in PSOC with ref = Vbg +/- Vbg = 1.3 nom, then 1 lsb = 2.6 V/ 2**16 = ~ 40 uV.
The drift in just the ref, 40 mV over T & V, in lsbs = 40 mV / 40 uV = 1000 lsbs. This does
not include all other errors.

The IA gain error, at G = 1, is .2 % typical, that corresponds to a 9 bit system in absolute
accuracy. Or ~ 128 lsbs of error in a 16 bit system.

Typical Vdd noise in a PSOC designs is 200 - 400 mV, and IA CMRR typical is 60 db, so
that is 400 uV input referred, not counting frequency rolloff in IA CMRR, and that rep-
resents 10 lsbs.

So without doing a full analysis over T & V, and not having worst case numbers on many
specs, IA, A/D, already we are at 1138 lsbs of error !

So you start with your goal, that in turn establishes how many lsbs of error, and then you
look at each component in system to see how the components contribute.

The only way to achive very high absolute accuracy is to do a production cal on system,
with PSOC running a cal routine talking to high precision instruments that force hi precision
V and T to PSOC and establishing a table of errors, which then can be converted into a power
equation or least squares error fit equation to be used by PSOC for normal application correction
of sensor readings.

The only other partial fix is possibly use an SCBLOCK, under API control force inputs to a known
voltage or ref, and capture offset, and subtract out of measured. But this is only a partial work-
around, is does not compensate for Vref drift, or a lot of other issues like long term component drift.

If you do not need absolute accuracy, just high non linearity, than NL errors and noise are your
main concerns.

Or live with the errors and non worst case specs PSOC 1 is lacking. PSOC 3/5 improves on this,
but still a work in progress.

Regards, Dana.

[danadak]

One other possibility is to look at dual slope collection and error
subtraction.

http://en.wikipedia.org/wiki/Integrating_ADC

Regards, Dana.

[danadak]

Gary I found one other old doc that shows a S/H application, in an older version
of designer. May suffice to capture offset, on further investigation, you be the
judge if it can be used.

an2350.pdf

(57.29 KiB) Downloaded 63 times

Regards, Dana.

[dseguine]

Offset removal is covered in AN2226, works nicely with instrumentation opamp or pga.
this assumes that you are measuring the insamp output with an ADC. Offsets get down to the microvolt level.
---- Dennis Seguine

[danadak]

This is an interesting discussion of CDS and CHS techniques, comparisons,
strengths, weaknesses -

https://classes.yale.edu/04-05/enas627b/papers/enzCDS.pdf

Another approach, oversampling and decimation ( does not solve absolute
accuracy problem, helps resolution problem, possible eliminate PGA... ) -

http://www.silabs.com/Support%20Documents/TechnicalDocs/an118.pdf

Regards, Dana.