• State Resolved
  • Locked Locked
  • Replies 3 replies
  • Answers 1 answer
  • Subscribers 46 subscribers
  • Views 1628 views
  • Users 0 members are here
Support feedback
Options
Options
Related

This thread has been locked.

If you have a related question, please click the "Ask a related question" button in the top right corner. The newly created question will be automatically linked to this question.

ADS1262: 1 mV Absolute Accuracy over 5V FSR without calibration. Problems with absolute accuracy.

Mehmet Akif Erol
Prodigy 130 points
Part Number: ADS1262

Hi all,

I am trying to obtain 1 mV absolute accuracy over 5V full-scale range with ADS1262. This is to be achieved without any offset or gain calibration. ( Calibration is kind of hard for our production process. )

So, I post my schematics first :

Circuit Details :

1) AIN0-AIN1 have 100 ohm ADC input resistors. 

2) AIN2-AIN3 have 1kohm ADC input resistors. ( I am trying to achieve 1 mV accuracy with these two inputs )

3) AIN4-5-6-7-8 have 12k/2.4k resistive dividers to make these inputs 30V compatible.

4) AIN9 is used to monitor a current in the other part of the board.

5) All inputs except for AIN9 have Schottky protection diodes.

6) I use REF5050  5V reference voltage ( as it is better than internal reference). Reference voltage is connected to AVDD pin of ADS1262 in order not to lose any analog inputs.

7) I think layout is pretty good as well with ADC basics are all taken into consideration.

ADC Details :

1) No offset calibration or gain calibration is done.

2) Sample rate is 2400 SPS.

3) PGA is bypassed.

4) Sinc4 filter is used.

5) Chop mode off.

6) Internal Reference voltage is enabled. ( For temperature monitoring, external reference is used for ADC )

My initial reasoning:

+ Absolute input range is VAVSS-0.1 to VAVDD+0.1. So, 0V-5V full-scale range is OK.

+ Non-calibrated maximum offset voltage is 800 uV, which is 0.8mV. 

+ 150 nA input current creates 0.15mV over 1kohm input resistance.

+ CMRR, PSRR and INL values are all too good be considered for 1 mV accuracy.

+ Gain error is not important because PGA is bypassed. 

+  0.8mV+0.15mv = 0.95mV. So, i guess that 1 mV absolute accuracy could be a reasonable target. I know temperature drift will affect my absolute accuracy but i am OK with 1 mV absolute accuracy at room temperature.

Test results :

A) I conducted tests with AIN2 and AIN3 over 5V FSR. Measurements are compared with calibrated Fluke multimeter. ( whose absolute accuracy is 1 mV as well )

B) At lower input voltages, an accuracy of 1 mV or better is observed. However, absolute error increases linearly towards 5V. It reaches maximum of 5mV-6mV near 5V input.

Evaluation :

1) I wanted to see if i can get better results with reduced sampling rate. I reduced sampling rate to 25 SPS and only marginal improvement was achieved.

2) REF5050 reference voltage is not 5000mv, Fluke reads it as 4999mV ( I don't know how close it is actually to 5000mV). So, this may incur some error. I am actually going for a better reference voltage IC. But i don't have the new IC yet.

3)  ADS1262 Datasheet says :

"A third option for ADC reference is the internal analog power supply. However, an increase of linearity error
results with this connection, and therefore, use this option only for less-critical applications, such as ADC selfdiagnostics.

For critical applications, do not not use power-supply reference option. For applications that use the power supply voitage as the reference voltage, connect the power-supply voltage to the external reference inputs, and
select the appropriate external reference bits in the REFMUX register. "

This is exactly what i did. I connected external reference to AVDD in order not to lose any of my analog inputs.  Datasheet even suggests that even if i want to use AVDD as reference, i should do this by connecting AVDD to one of reference AIN pins. 

So, my first question is that could this be the reason for my very high absolute error ?

4) ADS1262 EVM has some nice C0G capacitors for input filtering. I don't have those but Schottky diodes have some capacitance and along with 1k input resistance, low-pass filter is formed with 9 MHz cut-off frequency. 

My second question is that could lack of input filtering results in such a huge error ?

5) I am surprised by the linear increase of my error.

Offset error shape would be :

Gain error shape would be :

So, my error looks more like a gain error. As i turned off PGA , i would expect my absolute error to be more like a flat error of offset error. This is obviously not the case. 

Again, i check the datasheet. Even though i turned off PGA, i would like to introduce gain error just like G = 1. Datasheet states that maximum uncalibrated gain error is 300 ppm.

This corresponds to G = 1,0003 for worst-case scenario.  For Vin = 4900 mV, i get VADC = 4901.47 mV.  1.47 mV maximum error.

I calculated 0.95 mV from offset voltage and input current. To this, i add 1.47 mV from gain error for a worst-case absolute error of  2.42 mV.

My third question is that is gain error still valid for bypassed PGA scenarios ?

------------------------------------------------------------------------------------------------------------------------------------------------------------------------

As you can see, i am getting much more than that towards 5V FSR.  I feel like i have one grave mistake that creates such a high absolute error. I would like take some advise on what i am doing wrong. 

Thanks in advance for you replies.

  • 0
    TI__Mastermind 44680 points

    Hi Mehmet,

    I think over temperature performance and not performing any calibration is going to make it really difficult to achieve 1mV accuracy (or 0.01% or about 100 ppm FSR)...

    Your calculation of the total ADC error is pretty spot on. The biggest sources of error for the ADC are going to the initial offset (~35ppm TYP) and the gain error drift (~50ppm TYP), which also applies to the PGA bypassed case (see Figure 9 in the datasheet). When adding errors you can either sum them to estimate a worst case error (as you showed in your example), or calculate the root sum of squares (RSS) which is lower and shows a more typical value for uncorrelated error sources. If you go by the ADS1262's typical errors and calculate using the RSS method, the ADS1262 accuracy will be somewhere around 79 ppm, without applying calibration. However, using maximum error values puts the "inaccuracy" as high as 507 ppm (unlikely since all performance specs would have to be near maximum values).

    However, I think the real problem lies to two additional error sources: The reference voltage (most likely cause), and the combination of input bias/leakage current flowing across the series resistor.

    • Reference voltage errors will take on a similar effect as a gain error (and result in the "bow-tie" shaped plot). The REF5050's initial accuracy of 0.05% is already 5x times higher than your targeted accuracy (assuming a 100% FS input signal). At 20% FS (i.e. 1V) the "gain error" of the reference causes an approximate 1mV error. Granted this is a maximum error specification, not a typical, but it still gives an idea of what can be expected without performing any calibration to compensate for this error. Second to the initial reference error, the reference source's temperature drift is also somewhat significant (3ppm/degC max, means a possible 30 ppm MAX error with only a 10 degC temperature swing).

    • The input bias current flowing through the input series resistance doesn't appear to be significant at first glance, but this might be made worse by the input protection diodes. Most diodes will have a significant leakage current associated with them, especially at hot temperatures. Without knowing the exactly leakage current I cannot say how large of an error this is creating, but I mention it to make you aware of this effect. As much as you have control over, you can try to select lower-leakage (TVS) diodes OR perhaps decrease the resistor values to minimize the offset voltage and offset drift that results from the diode leakage current.

    Regarding your concern over lacking C0G-type filtering capacitors, it is a good consideration but I would say that the ADS1262's digital filter is probably more than compensating for the lack of analog RC filters (At 2400 SPS the filter's noise bandwidth is only about 570 Hz). The only exception might be due to aliasing of high frequency signals back into the digital filter's passband at integer multiples of the modulator sampling frequency. While it wouldn't hurt to increase the input capacitance a bit to prevent this, I don't think it would have too much effect on the overall accuracy.

    In summary, I think you are seeing a large gain error in your system. Even without using the internal PGA, the delta-sigma modulator will still have some inherent gain error; AND, since the ADC is comparing the input signal to the reference voltage, any error in the reference voltage will translate into a gain error seen in the ADC's output. Since the initial reference accuracy is pretty key to your system, is there any way you might be able to take advantage of the REF5050's trim pin to perhaps tune the reference voltage? Otherwise, I think calibrating your system will probably be the easiest way to achieve your target accuracy.

    Best regards,
    Chris

  • 0 in reply to Christopher Hall
    Prodigy 130 points

    Hi Christopher,

    Thank you for great insight. 

    1) ADS1262 and REF5050 calibration is the best way to go but we don't want to depend on manufacturer level calibration for several reasons. So, i want to see what i can get without calibration.

    2) I suspected REF5050 as well and currently going for a better reference with 0.02 % initial accuracy and 0.75 ppm / C temperature drift. That would produce 1 mV absolute error for 5V input. 

    3) Schottky diodes are very good point and actually i suspected them too but i thought that the way they are connected wouldn't create "bow-tie" error shape :

    Both BAT54 schottky diodes are in reverse-mode.

    + For 0V input;  BAT54 on the left has 5V reverse voltage, BAT54 on the right has negligible reverse voltage.

    + For 5V input;  BAT54 on the left has negligible reverse voltage, BAT54 on the right has 5V reverse voltage.

    So, i expect that effect of diodes should be same for 100mV input and 4900mV. In both cases, roughly same error current will be sourced-sunk by BAT54 diodes. But let's assume as if this is not the case.

    From BAT54 datasheet :

    BTW, i am unable to make sense of BAT54 datasheet : IR = 0.5 uA typical ( 2 uA max.) is specified for 25 C in the table. But in the graph at 25 degrees, it is more like 0.02 uA. That's a huge difference. I suspected ONSemi and checked Diodes Incorporated counterpart and it is exactly same. 

    Anyway, taking the worst-case value of 2uA for VR=25V, I estimate that ( according to graph), for VR=5, the leakage current should be at least 10 times smaller. ( due to logarithmic graph). Let's take 0.2uA for VR=5V. That would create 0.2 mV error over 1k input resistor.  That value is not insignificant but still too low to explain my large error. ( and i used worst-case values)

    That's my calculation, maybe i am missing a point. What do you think ? 

    4) You didn't comment on tying external reference to AVDD pin of ADS1262. This is not recommended in the datasheet. Do you think this may incur any errors ?

    ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------

    I will now order second prototype of my board with better external reference and lower-leakage current Schottky diodes.  I will report back what i get with the new configuration. If you have any more suggestions, i would be more than happy to hear them.

    Regards,

    Mehmet

  • 0 in reply to Mehmet Akif Erol
    TI__Mastermind 44680 points
    Hi Mehmet,

    I'm sorry for the delayed response.

    I think the reference voltage error is more likely to be the dominant source of the "bow-tie" shaped error than the diode leakage current, as long as the diode leakage current is less than 1 uA. NOTE: Above 85C the leakage current increases exponentially with temperature and around 1uA leakage the induced offset voltage across the 1kOhm series resistor becomes an issue.

    From figure 3, it looks like leakage current just about doubles from VR = 0 to VR = 5V; however, you are right that the figure does not appear to agree with the Electrical Characteristics tables of the BAT53SL datasheet.

    Regarding whether or not to use the 5V analog supply as a reference source. I wouldn't be too concerned about the additional linearity error (NOTE: The datasheet is referring to using the internal connection to AVDD/AVSS), as this is probably not too significant; however, I still wouldn't depend on the 5V supply to be as accurate as a dedicated reference source. Also, power supplies tend to be a bit noisier too.

    The only other thought I have about improving accuracy is would your leakage current induced offset voltage be decreased by placing the diodes before the series resistors (on the sensor side)? If the output impedance of your signal source is low, it should be able to source this current without much loading effect. Then only the ADC's input bias current would flow through these resistors. Just something to consider.

    Best regards,
    Chris