Hi, I am trying to develop a strain gauge amp for a custom load cell in an architecture project. I'm just reading the ADC, small signal amplification, offset and gain correction papers. As a test, I tried to measure the PGA output with G=1 from bench power supply input. Attached is a pencil sketch of the simple measuring circuit: The dev kit is the PSoC5 CY8CKIT-050. The PGA input pin is set to P0[2] and connected to the positive terminal of the power supply. The PGA output is set to P0[3] and connected to the HI terminal of the multimeter. The LO terminal of the multimeter and the minus terminal of the power supply are both clipped to the VSSA ground terminal of the dev kit board.

If some 50 mV are input, the output follows happily down to something like 24mV where it remains. I naively would have expected it to go down all the way to the negative rail. I didn't measure the offset but only read the power supply display, since it really doesn't matter if I can't go even near the full scale ouput of a load cell bridge (which at 2mV/Vex at 2.5Vex would ouput at most 5mV). The data sheet of the PGA module says about the output swing: Voh min = VDDA - 0.15V and Vol max = VSSA + 0.15V which is (assuming that Voh, ol stands for V output high and low) the opposite of what I would have expected. The technical reference manual says: RR to within 50 mV of Vss or Vdda, which is not RR at all.

So, what am I getting wrong here? I makes no sense to check offset and gain calibration if the most basic thing doesn't behave as expected. Any clue and nudge in the right direction would be highly appreciated. A

The TRM I think is incomplete.

The datasheet makes it clear it is R-R In, not R-R out.

These might help -

www.cypress.com/?docID=38337

http://www.cypress.com/?rID=49159

You could always write a test algorithim to capture the point at which output goes non-linear,

then subtract that out of the signal chain. Using DAC, S/H.......Although the problem is most

likely a strong f( Temp ). There are workarounds for that as well.

Regards, Dana. 

Did you set the power mode to High?

The PGA output swing is stipulated at power = high setting, 100K load,

and is 150 mV off the rails, Vdda - 150 mV and Vssa + 150 mV, guarenteed

min and max as stipulated.

It is not R-R out.

Regards, Dana.

Thank you very much for the reply, Dana. I couldn't find something about the output not going to the rail in the TRM or the data sheet of the module, obvioulsy overlooked it. I read the calibration papers you linked to and ran the associated project, but I am having a huge comprehension problem here:

The output does not go non-linear below 24mV but it is flat. At least as far as I could see on my Agilent 34410A 6.5 digit multimeter. Nothing moving but noise. So I can't calibrate that out because the PGA simply doesn't return any information below 24mV or so. The correlated double sampling notes are concerned with an offset (which is not linear and temp dependent, so you have to constantly resample it), but what I think I measured here is not an offset, but a cut-off. I did change the power setting to high but that made no difference, as far as I can see, the load is the GOhm input of the multimeter, so that should have no influence.

Dana, what you said about the output swing between Vssa +150mV to Vdda -150mV makes sense, I was just confused by the "min" and "max" subscripts which seem to be the wrong way round. But from that clarification comes an even more fundamental problem: How to measure outputs below 150mV?

If I take a commercial load cell with a full bridge I get 2mV/V excitation ouput. If I use Vex=2.5V the full scale output would be 5mV. If that is amplified with G=1000 to 5V than those 150mV where I don't get any information from the input are 150mV/5V=3%. For a 2kg load cell that would mean no measuring below 6 grams. That's obviously not what happens. So how to get round that problem? One could for instance offset the zero intentionally and then substract the multiple after amplification. But I don't see any such business going on in the example circuits.

I really breadboarded some circuits and read all available info. So I am not asking those questions out of laziness but because the stuff that is supposed to work simply doesn't seem to like it is advertised. But obviously it works very nicely in reality (I've got a precision scale dev board and precision load cell which goes down to fractions of grams from zero). So how is it done? And that doesn't even address my 24mV cut-off problem :) Thank you again for the input, highly appreciated. Regards, Arne

Why don't you post the link to this thread as a tech case -

www.cypress.com

“Support”

“Technical Support”

“Create a Case”

I concur, the PGA out, the rail compression, not normally what one sees.

And its not a common "threshold" one would expect. I do not have enough

chip level knowledge to help here. The need for such hi gain clearly

places some severe constraints on offsets as well as PSRR, CMRR.

A 2 kg load cell, with .1 g resolution, implies an A/D resolution of ~ 16

bits to 1/2 LSB. So A/D is not at issue. I will do another post with more

thoughts in a few minutes.

A couple more references, general info -

http://www.cypress.com/?docID=34330

This reference (same as above) has at bottom two authors email addresses, maybe contact them to see if they have done more work.

http://www.tmworld.com/design/characterization/4389520/Weighing-scale-design-Measure-signals-accurately-4389520

Regards, Dana.

In the ref the Cypress guys fix the excitation error problem by going

ratiometric. But overall that does not fix some of the more daunting

issues.

First and foremost a PGA is a single ended device, and you need lots

of CMRR, hence an IA solution is necessary for the G stage as the

A/D internal amp is limited to 16X. And the fact sensor is basically

a bridge detector with large CM.

This is one approach, but author does not give a worst case noise and

offset and temperature and.....analysis of its performance -

http://www.eetindia.co.in/STATIC/PDF/201006/EEIOL_2010JUN16_SIG_AN_02.pdf?SOURCES=DOWNLOAD

Just a little more ref material, might be useful -

http://www.ti.com/ww/en/industrial/sensors/weigh/design.html

If you start by doing a goal and backing into it with an error analysis that might

assist you. Normalize everything to lsbs, and you can do a spreadsheet where

all the critical problems arise, and what their performance has to be. Its tedious,

but it's the way your 6.5 digit DVM got built. I think it will point you to doing the IA

with a seperate high performance IA when all is done. That device, if any good,

takes care of most, not all, CMRR and PSRR and noise and nonlinearity.

Regards, Dana.

Lastly, for very high precision stuff, like your 6.5 DVM, the attached method

is used to use low cost components in very high precision systems.

The foil basically explains it.

Regards, Dana.