ADS1235: Inconsistent Measurement Drift and Step Changes in Strain Gauge System

Hi Edgar Ripoll Vercellone,

How are you exciting the bridge - voltage or current? What is the magnitude of the excitation voltage / current?

Can you include a schematic?

Are you using the AC excitation feature in the ADS1235, or just the global chop mode?

Does the performance change / improve if you change the programmable delay to something >0us?

Have you tried similar measurements with a known good source? For example, a AA battery (you would have to change the gain), or using a fixed resistor divider (3 resistors) that generates a low level voltage across the middle resistor that is within the PGA common-mode range?

Let me know

-Bryan

Hello Bryan,

Bryan Lizon86 said:

How are you exciting the bridge - voltage or current? What is the magnitude of the excitation voltage / current?

I'm exciting the bridge with a voltage power supply. The bridge and the ADCs are powered at 5V.

Here's a block diagram of the supply of the device:

I made sure to configure the Reference Configuration (REF) Register (address = 06h) to REFP0 for both ADCs.

Bryan Lizon86 said:

Can you include a schematic?

Here's the schematic:

Bryan Lizon86 said:

Are you using the AC excitation feature in the ADS1235, or just the global chop mode?

I'm not using the AC excitation feature, only the global chop mode.

Bryan Lizon86 said:

Does the performance change / improve if you change the programmable delay to something >0us?

I'll perform this test and I'll let you know the results.

Bryan Lizon86 said:

Have you tried similar measurements with a known good source? For example, a AA battery (you would have to change the gain), or using a fixed resistor divider (3 resistors) that generates a low level voltage across the middle resistor that is within the PGA common-mode range?

I didn’t measure an AA battery because changing the gain would alter my setup, and I wanted to keep the test conditions identical. That’s why I tested my PCB using a Wheatstone bridge built with standard resistors (see results in the original post). However, I can try the 3-resistor setup and share the results.

Thanks! Best regards,

Edgar

Hi Edgar Ripoll Vercellone,

Thanks for providing the requested information, let me know when you are able to run the additional tests

-Bryan

Hi Bryan

Here are the results:

Bryan Lizon86 said:

Does the performance change / improve if you change the programmable delay to something >0us?

I conducted another 4-hour test in the lab to observe how our device behaves in a controlled environment (no mechanical strains, temperature controlled with air conditioning, and a stable laboratory power supply). This time, I set the conversion start delay to 67 µs instead of 0 µs, as you suggested.

Both ADCs still exhibit drift—one around 5,000 ADC levels and the other around 1,000 ADC levels. I’m starting to suspect that this is due to our sensor’s high sensitivity to temperature drift, as the strain gauges are referenced against a PCB-mounted voltage divider (consisting of two 1 kΩ resistors).

Bryan Lizon86 said:

Have you tried similar measurements with a known good source? For example, a AA battery (you would have to change the gain), or using a fixed resistor divider (3 resistors) that generates a low level voltage across the middle resistor that is within the PGA common-mode range?

I performed another 4-hour test using a fixed resistor divider (see block diagram) and set the conversion start delay to 67 µs.

One of the ADCs showed a slight positive drift, while the other exhibited the step change in measurements that occasionally appears at random. I still haven’t been able to determine the cause of this behavior.

Thanks for the support! Best regards,

Edgar

Hi Edgar Ripoll Vercellone,

Thanks for providing the requested information

It does seem like sensor is adding an error to your system, the plot where you are just measuring a resistor divider looks much cleaner (can you use the same units for all plots however? Some have a y-axis in volts, some in codes, it will be easier to compare if they are all the same. Codes should be fine)

Can you explain your reference circuit to me? The schematic shows what looks like a true ratiometric reference on REFP0 and REFN0, where the bridge excitation voltage is directly used as the ADC reference voltage. I am not sure what the VREF- net is however, as this appears to be grounded.

However, you have said that you are using a resistor divider for your reference, which I assume is the one shown in the bottom of the image below. It is not clear to me why would you use a resistor divider instead of the ratiometric reference, or why you have both. And are those matched resistors, or just two discrete components?

-Bryan

Hi Bryan

Bryan Lizon86 said:

It does seem like sensor is adding an error to your system, the plot where you are just measuring a resistor divider looks much cleaner (can you use the same units for all plots however? Some have a y-axis in volts, some in codes, it will be easier to compare if they are all the same. Codes should be fine)

Both have the same units—Codes. The thing is that the measurements with the sensors range from approximately 100K to 300K Codes, while those with the resistors are in the order of millions of Codes. To improve the readability, the latter plot is scaled by applying a factor of 1e6 (as indicated in the upper left corner of the plot). I realize the font size is quite small—I’ll make it larger next time for better clarity.

Bryan Lizon86 said:

Can you explain your reference circuit to me? The schematic shows what looks like a true ratiometric reference on REFP0 and REFN0, where the bridge excitation voltage is directly used as the ADC reference voltage. I am not sure what the VREF- net is however, as this appears to be grounded.

I based my design on the typical application circuit from the ADS1235 datasheet. In my implementation, Exc+ and Sen+ are the same wire (labeled as V_Left+ for one ADC and V_Right+ for the other). Each sensor is powered through an eFuse, that's why there are 2 Exc+ lines (V_Left+ and V_Right+). Similarly, Exc- and Sen- are the same wire and are grounded at the PCB.

I made this design choice to simplify wiring and reduce costs by using smaller connectors and fewer wires, given that each PCB measures 2 sensors. I assumed this would come with the following trade-offs:

1. Increased noise in the measurements: However, based on field data, the measured deformations are significantly larger than the noise, making the added noise acceptable.
2. Voltage drop across the cable: I expect this drop to be small and stable. Any offset introduced by it can be calibrated at the start of the operation. Maybe I'm wrong on this one, but I don't see how this could be related to the step change or drift issues I'm observing, especially the step changes, which also appeared when measuring the fixed resistor divider with very short wiring.

Bryan Lizon86 said:

However, you have said that you are using a resistor divider for your reference, which I assume is the one shown in the bottom of the image below. It is not clear to me why would you use a resistor divider instead of the ratiometric reference, or why you have both. And are those matched resistors, or just two discrete components?

You are right, it’s the resistor divider shown in the image you attached. They are 2 discrete resistors with 0.1% tolerance, though you are right that using matched resistor could improve performance.

We included the resistor divider because the deformation we’re measuring is complex, and we need to individually assess how each strain gauge deforms. Isn't my measure still ratiometric? Both the resistor divider and the strain gauges form a Wheatstone bridge, only that each adjacent arm is located in a different place, but they are excited from the same source.

I appreciate your support! Best regards,

Edgar

Hi Bryan,

Update: I left the last test (fixed resistors and changed delay) running since yesterday and I'm just processing the new data.

Fixed Resistor Divider:

The step change doesn't happen exactly at the same time at both ADCs.

Here are the 27-hour test results for each ADC, which doesn't show drift on the measurements but many step changes:

Gauges with start conversion delay increased:

Here are the 27-hour test results for each ADC. I'm only displaying one input per ADC, as the remaining inputs exhibit similar results for the same ADC.

-Edgar

Hi Edgar Ripoll Vercellone,

How are you sampling your data? You have 4x gauge inputs, are you measuring them sequentially and continuously? So, measure gauge1, then gauge2, then gauge3, then gauge4, then restart the process? If so, are you manually restarting the process each time by sending the START command (or toggling the START pin)?

Do all of the measurements have the same settings (sinc4 filter, global chop, 1200 SPS, no delay)?

Are you measuring all of your gauge inputs with respect to AIN1 (the resistor divider)?

Do you use a gain of 128 for all measurements (either the actual gauges or the external resistor divider I asked you to test)?

-Bryan

Hello Bryan

Bryan Lizon86 said:

How are you sampling your data? You have 4x gauge inputs, are you measuring them sequentially and continuously? So, measure gauge1, then gauge2, then gauge3, then gauge4, then restart the process? If so, are you manually restarting the process each time by sending the START command (or toggling the START pin)?

Measurements are taken every 35 ms, triggered by a Timer. This interval was chosen to ensure the ADCs have enough time to complete all four measurements with my configuration (sinc, chop, SPS, and delay). I verified this timing using an oscilloscope.

The inputs are measured sequentially by toggling the START pin. Both ADCs receive the START signal simultaneously, meaning gauge 1 of ADC1 is measured at the same time as gauge 1 of ADC2. The end of each measurement is managed via the DRDY interrupt, and the next set of measurements only begins once both ADCs have completed their current conversion.

Bryan Lizon86 said:

Do all of the measurements have the same settings (sinc4 filter, global chop, 1200 SPS, no delay)?

Yes, I don't change the configurations at any time. They are only configured once at startup.

Bryan Lizon86 said:

Are you measuring all of your gauge inputs with respect to AIN1 (the resistor divider)?

Yes, this is configured just before toggling the Start pin.

Bryan Lizon86 said:

Do you use a gain of 128 for all measurements (either the actual gauges or the external resistor divider I asked you to test)?

Yes, I use the same configuration for all the tests.

I’d love to hear your thoughts on my comments about the reference circuit—do you see any issues with my approach?

Edgar Ripoll Vercellone said:

I based my design on the typical application circuit from the ADS1235 datasheet. In my implementation, Exc+ and Sen+ are the same wire (labeled as V_Left+ for one ADC and V_Right+ for the other). Each sensor is powered through an eFuse, that's why there are 2 Exc+ lines (V_Left+ and V_Right+). Similarly, Exc- and Sen- are the same wire and are grounded at the PCB.

I made this design choice to simplify wiring and reduce costs by using smaller connectors and fewer wires, given that each PCB measures 2 sensors. I assumed this would come with the following trade-offs:

1. Increased noise in the measurements: However, based on field data, the measured deformations are significantly larger than the noise, making the added noise acceptable.
2. Voltage drop across the cable: I expect this drop to be small and stable. Any offset introduced by it can be calibrated at the start of the operation. Maybe I'm wrong on this one, but I don't see how this could be related to the step change or drift issues I'm observing, especially the step changes, which also appeared when measuring the fixed resistor divider with very short wiring.

Also, regarding the ratiometric measurements, does my reasoning make sense?

Edgar Ripoll Vercellone said:

We included the resistor divider because the deformation we’re measuring is complex, and we need to individually assess how each strain gauge deforms. Isn't my measure still ratiometric? Both the resistor divider and the strain gauges form a Wheatstone bridge, only that each adjacent arm is located in a different place, but they are excited from the same source.

Thank you for the support, Bryan! Best regards,

Edgar

Hi Edgar Ripoll Vercellone,

Edgar Ripoll Vercellone said:

I’d love to hear your thoughts on my comments about the reference circuit—do you see any issues with my approach?

Edgar Ripoll Vercellone said:

Also, regarding the ratiometric measurements, does my reasoning make sense?

This should be fine, it was not immediately clear to me what you were doing. But now it appears you are basically measuring a variable half bridge sensor, then using a fixed half bridge on your board to generate a differential measurement

-Bryan

Hi Bryan

Yes, that's our system. Do you have any suggestions on how I can further investigate the cause of these issues or what my next steps should be?

Thank you for your support! Best regards,

Edgar

Hi Edgar Ripoll Vercellone,

I don't have anything meaningful to offer here. This does not appear to be an issue with the ADC, it looks like >99% of the readings are correct in the fixed resistor divider case. My guess is the event causing the step change is something external

I will say that the average shift you are getting for the "fixed resistor divider" case is on the order of 2000  and 10000 codes (ADC1 and ADC2 respectively). At VREF = 5V and gain = 128, that is an LSB size of 4.66nV. So that is a shift of 9.3uV or 46.6uV, respectively. As you can see, not very large signals here, but noticeable by the ADC

I also generally observed that the time these noticeable step changes occurs happens right around the end of the day (6PM) and the beginning of the day (9AM). Not sure if there is any correlation there, maybe doors opening/closing as people leave/enter, causing vibrations? At the these low levels, you can tap on the table/surface on which your board is sitting and see small changes in the output. I'd wonder if the time of this behavior is repeatable?

-Bryan