Noise analysis, ADS1247, 4-wire mode with RTD

Rich,

There are a couple of things that are not clear.  Do you have your RTDs connected to the EVM?  Are you using 2 wire RTDs?  Can you send us a schematic of how you are connecting things up, as I can't visualize how you are doing this?  Can you send us some data?  This can either be raw data or pictures from ADCPro if that is what you are using for analysis.

Noise is a problem with any system.  In particular, the noise performance tables you mentioned in the datasheet are best case scenarios with shorted inputs.  In practice it is really hard to match this in a system.  From your discussion it appears you have a good understanding of the noise problem.    Layout is a huge concern in overall performance.  Also, I would gain to the maximum possible gain staying within the common mode input range.  The actual noise becomes less dominant with gain.  This seems counter intuitive, but the signal to noise ratio does increase.

If you are using the EVM, there are a couple of things to consider.  The first is the Rbias resistor should be placed close to the reference inputs.  The EVM doesn't really allow you to do that easily.  As you mentioned, noise should cancel in a ratiometric system.  However, any filtering will affect the phase difference between any filtered input and filtered reference.  The EVM has a 10uF cap near the reference inputs, and analog inputs also have filtering.  The analog filtering also effect the current path.  Another problem that can add noise with the EVM is the protection diodes (D1 and D3) at the external reference inputs.

Another thing to consider is aliasing, and any high frequency signals (EMI/RFI) that can noise up your readings.  To eliminate this issue you need a low pass filter at the input.  If you are using a two wire RTD, any filtering done within the current path will affect the result.  With a three wire RTD you can have two current paths and cancel the series voltage drops.  If you are making a single measurement across all four RTDs at once, which is what I think you are doing, I would remove all components relative to AIN0 (R4, C7 and C24) and actually make the measurement across AIN2 and AIN3.

Best regards,

Bob B

Hi Bob:

Thank you for your detailed response. I will try to answer your questions.

I am measuring the temperature of a block of copper that is about 50mm square and about 6mm thick. I am using four 2-wire RTDs that are embedded in the copper. The leads of each RTD protrude from the block and are soldered to pads on a PCB that is mounted to the copper block. The traces on the circuit board then connect the four RTDs in series. At the 'top' and 'bottom' of the RTD 'pole' I have two traces meeting at the RTD lead. This constitutes my Kelvin connection ( I realize that there is still some lead resistace error in the traces that connect one RTD to the next). The four PCB traces (+/- Force and +/- Sense) route to a connector. Then through a connector the four wires are brought to the EVM. The + Force wire is connected to screw terminal AIN0, + Sense is connected to AIN1, -Sense is connected to AIN2. The -Force wire is connected to header J8 pin 20 (Ref+). My 2.5 kohm 5 ppm resistor is soldered between pins 20 and 18 of J8 (Ref+ to Ref -).  As you said this places the Rbias resistor more than an inch from the ADS1247 chip.

So my circuit looks near identical to the Figure 4 of apps note SBAA180 (Four wire RTD Example) except that I am using AIN1 as the (+) sense and AIN2 as the (-) sense. Is that a problem?

I have attached a screen shot from ADCPro. Note PGA gain =4, Data Rate 5sps. The time scale is such that the complete sweep is 410 sec.

For this measurement, my copper block had been in a highly insulated cylinder overnight. The input sensitivity of four 100 ohm RTDs in series (and IDAC = 1mA) is 1.54 mV/degC. The vertical scaling shown is 3 uV/box which translates to 1.95 mdegC/box. The range of the temperature window shown is 23.3 mdegC. With your eye you can remove the DC component and try to estimate RMS noise. When I do that I come up with a pkpk noise of about 5 boxes or 9.75 mdegC. Assuming gaussian noise that gives you an RMS noise of about 1.62 mdegC (divide pk-pk by 6).

Converting my measurements to ADC counts (or LSBs): Because the PGA gain is 4, the full scale input to the ADC is not +/-2.5V it is +/- 2.5V / 4 = +/- 0.625V (RTI). So the value of an LSB is 1.25V / 2^24 = 74.5 nV.. As mentioned above, my vertical scaling is 3 uV / box. This translates to 3 uV / 74.5 nV = 40.2 LSBs per box.

Also, from above, my noise is about 5 boxes pk-pk which is also then 5 x 40.2 = 201 LSB pk-pk or (again dividing by 6) = 33.5 LSBs rms.

When I short the two inputs together under this gain and data rate I measure 5.27 LSBs rms (or 0.393 uV rms which matches the data sheet value of 0.37 uVrms)

In summary, with shorted inputs I get noise of 5.27 LSB rms. When have the RTDs in the circuit I get noise of 33.5 LSB rms --- larger by a factor of 6.4.

Looking closely at the EVM schematic I see now the problems that could be generated by the input filters to the ADC and especially the 10uF cap (C20) that is in parallel with my 2.5k Rbias resistor. By the way that the EVM is implemented where C7 goes across AIN0 and AIN1 while C8 goes between AIN2 and AIN3 it seems to suggest that voltage measurement should be taken across either AIN0/AIN1 or AIN2/AIN3. I'm measureing across AIN1 to AIN2 which seems not quite right for the filtering setup.

Above you say to remove R4, C7 and C24 and measure across VIN2 and VIN3. Should I not also remove C8, C20 and D1 and D3 ?

Thanks, Rich

Bob:

Great ideas. Just the kind of insight I was hoping for. I now have a number of things to try.

Especially easy will be to try the internal ref ( I think that requires switching the physical switch S1). As I understand it, I should see an increase in noise with the internal reference because I've lost the ratiometric advantage. And you are saying that there is no way that using the internal ref should be less noisy than using external ref.

My Rbias resistor is a Vishay-type radial leaded part so that gives me a lot of options for placing on the board.

I'll keep you posted.

Rich

Hi Bob:

A follow up from our last correspondence. I was able to make significant strides in lowering system noise. Many of your suggestions helped but one of the bigger helps I found on my own.

For reference, when I first started this thread I was measuring an input referred noise of about 2.5 uVrms, now I am getting noise RTI of 1.05 uVrms. This is a big difference. (conditions: Vref = 2.000V, PGA = 4, sample rate 5sps) As you know from working with noise power, to decrease the measured rms noise by a factor of 2, you need to remove almost 87% of the original noise sources. [ sqrt(1^2 - 0.866^2)] = 1/2

These are the changes I initially made to the EVM circuit and configuration:

1. Changed the wiring of the sense leads so I use AIN2 and AIN3 as inputs
2. Removed D1 and D2 (showed measureable difference)
3. Moved my 2k Rbias resistor closer to the ADC as you outlined
4. Removed C20, the 10 uF cap around Rbias.
5. Removed and shorted R4, removed C24 and C7

With the changes above, I still saw what seemed to be bursts of noise lasting 1 to 4 seconds that would mess up my readings. This looks a lot like popcorn-type noise.  To investigate this, I used a very low noise high gain scope pre-amp (Tek ADA400A) and measured directly across my RTD to look at the voltage that the ADC is seeing. What I found on the scope was this popcorn noise effect correlated exactly with what I was seeing in the EVM (LabView) interface.

My reading of this is that the popcorn noise is coming from the IDAC source. I switched IDACs and saw the same noise. So what I did to significantly reduce this effect (although I can still see it) was to:

1. Put a capacitor back into the location of C24, but instead of 47 pF I used the 10uF cap that I had removed from C20.

This 10 uF capacitor, along with the series resistor values of R4 and my RTD (nominally 400 ohm) places a low pass filter on the IDAC at 35.6 Hz.

So with the above changes, and the big filter on IDAC, I was able to reduce my input referred noise by a factor of 2.5.

My temperature measuring system now has noise in the range of 700 uCrms (micro degrees C) and this is without averaging which could beat the noise down even further.

Pretty awesome device... If the IC designers were able to reduce the popcorn noise on the IDACs it would be even more awesome.

Rich