PGA204: Linearit of PGA204

Hi David, 

Thanks for posting on e2e! 

From a quick glance, I expect that the input bias current is contributing to the error introduced by the larger resistors.

If you keep the same ratio but lower the resistance, does the linearity improve? 

I am having a bit of a hard time understanding your table, could you explain what IN+, IN-, and the other side 300 ohm of your PGA204 output is?

All the best,
Carolina

lowering the resistance helps

300 Ohm is just a load on output, I cut out the circuit, in hope to simplify the case

IN+ and IN- connects to Fluke 5100 calibrator

Thank you!

Hi David, 

Could you please let me know what the value of VCC is and what gain the PGA204 is in?

All the best,
Carolina

+/-15V, which is provided by DCP010515, and the circuit is isolated/floating, the gain on PGA204 is at 1

Hi David, 

For my understanding, could you please tell me if the blue circuit or the red circuit is more representative of your setup: 

All the best,
Carolina

Blue

Since this circuit is floating, the ground is the not solid, let me do some tests about it first

The "ground" is not in the center of +/-15V, it can be several volt away from the center, would this cause the non-linearity issue?

Hey David, 

Interesting... are you saying the red ground is shifting or the yellow? 

All the best,
Carolina 

Yellow. As mentioned above, the circuit is isolated/floating

Hi David, 

I don't expect the ground shift (shown in yellow) to affect the measurement. Here is a quick excel I assembled that shows that when there is ground shift, the signal is simply counted as "common-mode" and therefore ignored by the instrumentation amplifier: ground_shift.xlsx

Could you probe the VIN+ & VIN- (Pins 4 and 5) of the PGA204 and tell me if they match the hypothetical I have calculated? I would like to isolate the error introduced by the device and input signal. 

All the best, 
Carolina

what is the max CM for +/-15V supply? in my case, I suspect CM can reach 8V, with normal mode signal around +/- 4V

Hey David, 

The common is relative to the power supply, for ±15V supplies, here is the specification from the datasheet: 

The ground shift would have to be either 7V or -11V to begin to be an issue:

All the best,
Carolina 

So CM doesn't seem to be the factor to cause poor linearity

Hi David, 

I agree, were you able to complete the following? 

Caro Walter said:

Could you probe the VIN+ & VIN- (Pins 4 and 5) of the PGA204 and tell me if they match the hypothetical I have calculated? I would like to isolate the error introduced by the device and input signal. 

All the best,
Carolina

Just came back from vacation. I double check the circuit, and noticed a section that I thought it was disabled by relay was actually involved. 



Since DN1 clamps the voltage at 13.6V referenced to floating "GND", if I use the GND as reference, here is what I read 

V1=786V
V2=-7.86V
V3=13.7V
V4=0.13V

Hi David,

Good to hear from you! 

If your 'equivalent' PGA204 output matches the highlighted row: 

Then I think it is fair to assume the largest remaining error contribution is the resistor values in the input bias current of the PGA204. 

All the best,
Carolina 

Gain error is expected, but I was hoping the error is linear so that I can use digital calibration to take care of it

But the result was not linear, like a dog leg (please see the first post of the thread)

Hi David, 

As you can see, this error is much larger than that which I would expect to see from the PGA204. 
Since the behavior extends to "a batch" of PGA204s instead of just 1, I suspect there is something else to contribute to the error. 

Could you repeat the experiment with inputs "V2" and "V4" (remove the resistors) and apply directly the smaller signal? This would help us isolate if the larger error contribution is the PGA204 or something else in the setup. 

What ppm are the resistors used? Is there an associated temperature range of the experiment? 

All the best,
Carolina 

When we remove the voltage dividers (5M/50K), and apply only small signals, it works fine. The resistor is rated 0.1% accuracy

Even in the midpoints? 

What is the ppm of the resistors? 

“When we remove the voltage dividers (5M/50K), and apply only small signals, it works fine. The resistor is rated 0.1% accuracy”, yes, very good result everywhere with signal between +/-10V

Correction: I tracked down a few from the first batch that we thought it was good and tested them again, and confirmed they also had this "dog leg" like problem, just not as as strong so our cal program didn't catch it. Again, when the divider is removed for small signals, PGA204 works fine in old batch

Based on the tests, PGA 204 all shows similar "error" when having that big voltage divider/float ground in front of it.

Hi David, 

Could you please share the resistor part numbers?

All the best,
Carolina

2.5M   https://www.digikey.com/en/products/detail/stackpole-electronics-inc/RMCF1206FT2M49/1759724

49.9K  https://www.digikey.com/en/products/detail/yageo/RT0805BRD0749K9L/1075926?s=N4IgTCBcDaIJ4EMDmBGAnCsAXATgWgDsATEAXQF8g

Thank you !

You have a nice weekend!

Hi David, 

Thanks for sharing this, I will run some calculations and get back to you - happy weekend!

All the best,
Carolina 

Thank you very much for going extra length for us!

After we double check your suggestion/result, it seems to suggest a wider error, but not an explanation for non-linearity (dog-leg like) for the whole batch. 

Carolina, any chance to reopen the ticket? I would like to keep further discussion in the same thread if possible

Hey David, 

Of course, we can always keep working on it :)

When you tried the same ratio of resistors but tried different ones, could you share those part numbers? 
Are you willing to try a different instrumentation amplifier?

If not, could you implement a JFE2140 to isolate the large resistors from the input of the PGA204? JFE2140 Ultra-Low-Noise Pre-Amplifier This solution introduces a slight DC error that should be calibrated out by your program. 

All the best,
Carolina

We need the PGA to provide the gain for the front end

The other resistors were from Vishay, which we ordered long time ago, and the part number is no longer active

If maybe a challenge to add it in since the pcb is very crowded

I requested to download pspice from TI (waiting for approval), to see if we can take a closer look at PGA204 

Hi David, 

Okay so it seems the only real knob we can turn are the resistor values, here are a few in stock options on Digikey:

To reduce the input bias influence on the error, the same ratio can be kept, and the new resistors are two 1MEG and one 20k. 

The current consumed by R1 & R2 is 156mW each and R3 is 3mW. 

Here are two options for the 1MEG: 

AGTF1206-TE-1004BT Venkel | Resistors | DigiKey Marketplace

RS73F1JTTD1004B KOA Speer Electronics, Inc. | Resistors | DigiKey

Here are two options for the 20k: 

TFCR0603-10W-E-2002BT Venkel | Resistors | DigiKey Marketplace

ERA-3AEB203V Panasonic Electronic Components | Resistors | DigiKey

These two can be simulated with 0.1% tolerance: 

Important note: the resistor divider is now a factor of 101 and no longer 100.7996 as before. 

When I ran the monte carlo sim + input bias parametric sweep, here are the results. As you can see, the min/max values are significantly closer. The digital calibration will only help. 

Here is the linear regression fit line for this:

Under consideration but had no in stock options,

(1) Reduce the tolerance of 2.49MEG resistors, 

(2) Replace the two 2.49MEG resistors with a one 4.99MEG 0.1% resistor, that can support more than 100mW. 

The only in stock option feels too close to what the design would need: 

Hopefully this helps, 
Caro

with tighter smaller resistors, 5M -> 1M, I am sure it will lower the weird dog-leg effect, but to truly address the problem, we would like to know why it had such a strange non-linearity. The monte carlo sim only shows the non-linearity could be within the min-max windows, but it didn't explain why it would happen

Hi David, 

Could you provide more information on your end point calibration routine? 

I hypothesize the following:

0V, the output is 0V. 
800V, the output is 803V. 
Therefore the correction algorithm would -3V to each output. 

Could you confirm my hypothesis? 

All the best,
Carolina