LMP90100: Need help removing offset

Robbie,


I'm not sure what's wrong, but I did have a few questions and a few things to try.

Are you using the device in a gain of 1? Or do you change gains between the first measurement and any writing of the offset? I know that seems a bit obvious, but I thought I'd check. Do you change the gain coefficient before taking data? If you can, could you provide your configuration and how you set up the offset and gain?

Based on Figure 54, the value stored in CHx_SCAL_OFFSET should be subtracted from the VIN coming from the ADC. However, it seems you're either getting the wrong value or the offset subtraction isn't working at all (the change between 1000h and F98h is 68h - about 100 codes).

Instead of entering in the value directly, have you used the system calibration offset coefficient determination mode outlined in 9.3.7.2.1? This should get the correct value as well. I would run through that set of instructions and then read back the result of CHx_SCAL_OFFSET just to see if it matches what you would have gotten.

Another test to run would be to put in a unusually large number into the CHx_SCAL_OFFSET register to see if it is being modified by gain, or used in an unusual way. If you are currently putting in F98h, put in something larger like 3E69h (4x) and see what it returns back.

Naturally, I'd also check to see that when taking offset or running a calibration, that the offset value is really 0V in, and that there isn't some sort of inputs settling going on.

Regardless, read through my comments, run a couple of tests and let me know how it goes.


Joseph Wu

Joeseph,

Thank you for the response!  I am using a gain of 1 (not changing from default) and I've tried writing the offset both as the first configuration message and the last configuration message.  Said differently, I've tried configuring the offset first before any other register and last after all other registers have been configured.  I have verified I can read the offset value I program into the channel 0 offset register, so if I write a FF I read an FF and if I write a 0x77 I can read a 0x77.  This seems to have no affect on the actual offset though.

To take a step back, should I be able to read every single register in the register space and see what I've written?  If I configure all the register should I be able to read all those values back and see the exact same values I programmed in? 

Here is what I'm programming into the registers for 2 channel (ch0, ch1) repeated sample mode.

LMP90100 configuration register writes
----------------------------------------------
0x01: 0x08
0x02: 0x00
0x03: 0x00
0x09: 0x1A
0x0A: 0x02
0x0E: 0x40
0x10: 0x00
0x11: 0x83
0x12: 0x00
0x13: 0x00
0x17: 0x00
0x1F: 0x88
0x20: 0x01
0x21: 0x01
0x22: 0x13
0x23: 0x01
0x30: 0x00
0x31: 0x00
0x32: 0xFF
0x33: 0x00
0x34: 0x00
0x35: 0x00

Here is what I'm reading out of those registers:

Here is a more complete register read:

Here is what my data looks like with the 0xFF offset loaded as shown in the register read above:

Here is my register reads with no offset:

Here is the data from the no offset configuration above:

All registers are configured in the order shown in the text list of registers and the values I'm writing to them.  I'm assuming the order in which you write to the registers doesn't matter?

Robbie,


The order of the register write shouldn't make a difference.

Right now you have the CH0_SCAL_GAIN set to 000000h. The default is 800000h. Can you keep that value set to 800000h and check the result?


Joseph Wu

Also, here is the hardware configuration

Here are the register reads when the ch0 offset is set to 0x800000 and the read results.

Config registers reads:

Here are the sample results with the 0x800000 offset configured:

Oh I apologize I did misunderstand.  I think this is what you were looking for.  

Here are the register values:

Here are the sample values:

Why is the default value for CH0_SCAL_GAIN 0x800000?

Joseph,

I think that was my mistake.  I believe what I'm getting now makes sense and the offset subtraction may be working.

Below is my register values with no offset and sample values with no offset (I am only displaying ch0 now to make it simpler)

Here is what I get when I set the offset value in the CH0_SCAL_OFFSET to 0x3A0.  It looks like it is subtracting.

As a general question, does this make sense and for some basic linear lab supplies does this offset make sense?  It is not a large distance from 0V I don't think?

1. As for acquiring data do people generally just average a lot of these values together? 

2. There is a lot of noise.  Should I right shift to get rid of some of the lower bits? 

3. Do you right shift until you are using the same ENOB that is described in the data sheet for my configuration?

Should I figure out the ENOB and drop the extra least significant bits?

Robbie,


I don't really know what you're measuring so it's a bit difficult to make a determination of what you want to do. You could determine the ENOB and drop the LSBs if you want. However, you'd still have some noise in the measurement because ENOB here is a measurement of the RMS value and you might be more interested in the peak to peak value. To get something noise free, you might have to increase the RMS value by a factor of 6 (if you equate the RMS value with standard deviation, this would be 3 standard deviations from mean).

You could also do some averaging to reduce the noise and see if you can drop the noise floor significantly. In general, with n samples, you'll reduce the noise by √n. However, you're likely to do better by reducing the data rate by the equivalent amount.

I'd start with verifying that the noise is where it should be for the given operation. Then clean up noise sources if there are any. After that, the data is there and you can do what you want to do with it.


Joseph Wu

Ok that makes sense.  I forgot to mention my setup.  I think the noise ENOB aspect makes sense now.

My current setup is a dual output power supply driving both inputs (Vin+ and Vin- of ch0) to 2.00V with the ADC_REF at 5V.  I am trying to walk the differential voltage up in value by taking to positive input from 2.00V to 2.10V in 10mV (0.01V) increments while leaving the negative input VIN- at 2.00V.  When I step up in 10mV increments I see the output of the ADC flip from being positive to negative as shown in the image below.

The blue is steady state with both outputs roughly at 2.000V.  The green is when I increase the vin+ by 10mV.  The red is when I increase the vin+ by another 10mV.

I initially thought I had a noise problem, but I have some very good power supplies and am verifying their voltage outputs with a fluke meter calibrated to 1mV.  In this particular example I'm using no calibration values for the offset.

For completeness these are the register values:

How could the output change that much when the input is verified to be stable and separated in a postive manner i.e. the vin+ is ~20-30mV higher than vin-.

I'm using the Rigol DP832 power supply to generate both inputs. It has two adjustable outputs so I connect each of those to the channel 0 inputs. I am measuring the input voltages from the lab supply separately with two multi-meters. I'm just verifying that the positive input is increasing basically. Thanks for plotting those values out. I will be working on trying to get this part up and running reliably again this week so if you think of anything let me know! Thanks again!

Robbie,


It looks like your on your way to making some good measurements. Using a lab supply as your measurements should work, but I'd be sure to use a precision multimeter to measure the input to calibrate your results. With a 10mV change, the LMP90100 should easily be able to seperate out the differences.

The Rigol should have a low enough ripple so that the basic measurement should work with 10mV steps. Just the same, I'd still check to see what happens with larger steps.

If you have any issues feel free to post back. Good luck!


Joseph Wu

Thanks Joseph,

I worked on this all day today.  I wanted to get your opinion on my current data.  I found an example/demo video from TI online and it looked like when connecting the two leads VIN- and VIN+  together (say for channel 0) the code read back in this video was very very low without calibration and even lower with.  I feel like the codes I'm seeing on my setup when my VIN+ and VIN- for ch0 are shorted together (the numbers seem to be the same whether I leave them floating or tie them to ground)  are high.  Below are the values I'm seeing.

Theoretically I have this working and this is my "factory" offset.  Does this seem reasonable?  It is the same basic setup with no gains, no buffer (internal buffer amp) and running at 1.6sps. 

Thanks

Robbie,


I'm not sure about the values that you are showing. If you assume that the average value is about 110h, that would be about 272d, which translates to an offset of 160uV, assuming a reference of 5V.

If this is the pre-calibration number then maybe this might be high, but it may not be out of the realm of possibility. I just don't have any guess as to what this value would normally be. If this is the post-calibration number, then I'd say something is wrong.

Generally, I would short the inputs together to get the offset value also, so I wouldn't expect anything to be wrong unless there were some sort of settling issue.

I would also note that the offset may change once the gains and buffer are enabled. If you intend on using any additional gain or buffer, I would run a calibration for that setting.


Joseph Wu

That is a pre-cal number currently. I have shorted the inputs together and am using a 5V reference.

From what I understood in the datasheet, you are supposed to try to zero out the offset as a first step when calibrating and then apply the gain. Is that correct? It sounds like you are recommending I set the gain and buffer, then attempt to zero out the offset.

Joseph,

I have this up and running now the offset zeroing functionality seems to be working pretty well. Now that this is working it sort of brings up a problem I didn't think about.

Since the number is signed and when my offset is zero'd out I still have some noise that causes the number to bounce between a small positive number and small negative number. Said differently, when my strain gauge offset is zeroed out, I see the numbers at steady state fluctuate like below:


AFE ch1 value: FFFFCE
AFE ch1 value: 3F
AFE ch1 value: FFFFFD
AFE ch1 value: FFFFCC
AFE ch1 value: FFFFB7
AFE ch1 value: FFFFD3
AFE ch1 value: FFFF88
AFE ch1 value: FFFFDF
AFE ch1 value: FFFF71
AFE ch1 value: B
AFE ch1 value: FFFFD0
AFE ch1 value: 3A
AFE ch1 value: FFFFB3
AFE ch1 value: FFFFFB
AFE ch1 value: FFFF76
AFE ch1 value: FFFFF6
AFE ch1 value: FFFFE1
AFE ch1 value: 35
AFE ch1 value: FFFFEF
AFE ch1 value: FFFFF2
AFE ch1 value: 3C
AFE ch1 value: FFFFBD
AFE ch1 value: 1C
AFE ch1 value: 2E
AFE ch1 value: FFFFE5
AFE ch1 value: 1
AFE ch1 value: 2B

Just as an example. So i'm bouncing between positive and negative. Now, when I apply a gain (say of 128) this blows up on my because I get that gain applied to the negative number (which makes a much bigger negative number) and a positive number (which makes a much bigger positive number).

How should this situation be handled? I am new to this sort of data acquisition and don't see an option with this device to deal with this. Even if my analog side was very quiet and there was very little noise, as long as the values/codes go between positive and negative it seems this issue would exist and getting exactly 0 for an offset isn't really possible.

Do you have any guidance/recommendations?

Thanks!
Rob

Rob,


I not sure how to answer this question. Regardless of the measurement, you'll have some amount of noise. Applying gain, you'll always gain up the noise. The trick really is to reduce the noise to as low as possible while maximizing the input signal range. This is a combination of using the data converter at a lower data rate, averaging, a clean layout, and low noise reference.

As far as putting the device in gain and having the output code go positive and negative, you could do is truncate the output code so that the noise is smaller than the least significant bit that is visible. With a 0 input, you'd never see the negative codes.


Joseph Wu