ADXL206 Zero G Bias Repeatability

In the ADXL206 spec sheet on page 3, the "0 g bias repeatability" is listed as "+/- 10 mg."  Sorry if this question is obvious, but could someone clearly define what that means?  Is that the turn-on to turn-on repeatability of the sensor?  What I'm curious about is if I temperature calibrate / compensate an ADXL206 at one turn on for scale factor and bias errors, then cycle power on the ADXL206 off and attempt to use my previous temperature compensation, assuming that compensation was perfect will there be a possible additional error of +/- 10mg?  Thanks to anyone who can shed further light on this!

  • The answer to this question is not obvious and you have all the right to ask such a crucial question. As a matter of fact, I believe it must be the sensor producer’s responsibility the answer this question in their spec sheets. However as you have noticed, they don’t do so.

    Repeatability is a measure of how much error you should expect on calibration parameters (including bias, scale, temperature, g-dependence etc.) when you turn on the system. Essentially, the repeatability must have been defined for each calibration parameters separately. However, as in the case of ADXL206, MEMS producers usually define it for bias and simply ignore the rest. (It is an interesting coincidence that bias repeatability is also the easiest to compute.)

    Repeatability has 2 aspects: i)unit to unit repeatability, ii)in-time (turn-on/off) repetability. The producer must have clearly indicated which one of these they mention in the specs sheets. However, again as in the case of ADLX206, they never do that. (Why? I don’t know).

    Unit to unit repeatability determines the variance of calibration parameters between 2 different units before you calibrate them. Figure 4 of ADXL206 says something about it. Apparently, this value is much more than 30mg for ADXL206. Therefore, I believe they use the term “repeatability” for in-time (turn-on/off) repeatability.

    In-time (Turn-on/off ) repeatability characterizes how much error you should expect on calibration parameters after you calibrate them. Therefore, +/-10 mg means that if you calibrate the system now and decide to use it later (how much later?) your true bias calibration value must be within +/-30mg of the previously computed value with a probability of %99.9 regardless of any other environmental change.

    However, as you probably notice in the above definition, the repeatability value MUST BE ASSOCIATED with a time-of-validity parameter. In other words, the repeatability parameters can only be defined for certain period of time. Therefore, an acceptable repeatability statement must have been in the form of “+-10mg in X days/months/years”. Without such a time-of-validity component the repeatability definition is essentially meaningless.Unfortunately as in the case of ADXL206, the MEMS producers never specify that value.

    MEMS producers do not specify the time parameters because they also do not know what cause the calibration parameters to change in time. For instance, is it the temperature rather than "the time" that causes the bias/scale repeatability errors? If yes, can we calibrate with respect to temperature? Then, what about the temperature repeatability? Which kind of test can we perform to separate the effect of temperature, bias and scale factor repeatability on sensor outputs? (Have you seen any MEMS producer ever gives an answer to any of these questions? I have not.)

    As a result, a repeatability value without a time-of-validity tag is essentially useless. You should never rely on that value. As a matter of fact, I suspect that you will see a much bigger difference between 2 succesive calibration test results performed within a single hour. Therefore, you probably have to make up some kind of value by yourself depending of several calibration tests results.

    Or, if you do not have any specific reason to choose adxl206, you may also want to look at military sensor producers such as Honeywell whose units have been all around quite long time. For instance, Honeywell claims their bias repeatability is 4mg for one year. Apparently, they know how to use the terminology unlike the MEMS producers.

  • 0
    •  Analog Employees 
    on Aug 2, 2018 4:34 PM
    This question has been assumed as answered either offline via email or with a multi-part answer. This question has now been closed out. If you have an inquiry related to this topic please post a new question in the applicable product forum.

    Thank you,
    EZ Admin
  • HI There

    I have the same question as above and i don't think this ever got answered by AD

    The data sheet has a value of +/- 10g for 0g bias repeatability-  i'm only interested in the range upto +/- 1g. So my assumption is this sensor only gives 1% accuracy even after calibration (ie 10mg/1g x 100)??

    Kind Regards

    Martin

  • 0
    •  Analog Employees 
    on Feb 4, 2019 1:43 PM in reply to martin.stokes@sparkpi.co.uk

    I am sorry that this problem was missed in the past.  In the past, we have struggled to develop consistent coverage for this series of products. I believe that with our most recent organization adjustments, this has been appropriately recognized and will have more consistent coverage, in the future.  

    While I was not on this team, I can offer that this specifications typically represents what we would expect, for end of life drift, which would typically represent a lifetime of 10 years, at +25C (with de-rating for higher temperatures).  

    Bias drift is independent of the stimulus, so I would assume a linear model for the worst-case error: 

    Acceleration x (328/312) + 10mg. 

    One clarification: assuming you calibrate the device to address initial sensitivity error (1- 328 / 312) and temperature dependent errors (0.3%), we can offer some perspective, on end-of-life drift.  I am aware of a number of applications, which used similar sensors, in different packages, who have offered that the end of life drift (10 years, +25C) for sensitivity will be in the region of 1%.  Based on what I have observed, this is typically done with HTOL testing, then using the Arrhenius relationships to project those drifts towards an equivalent lifetime, at +25C.  I suspect that similar thinking could be applied towards the timeframe that this could be supported, at higher temperatures of operation, as well, but that is farther than I have been able to take such consideration.  

    Does this help?  

  • Many thanks for taking time to answer my question

    If I understand correctly, I can calibrate the unit to remove the initial sensitivity errors (and many others) but I still have a potential 10mg end of life drift error, and this is typically a 10y figure.

    I think you were implying this "bias repeatability" is linear over time? does that mean for short missions (say 1y) the post calibration bias repeatability error is up to 1mg?

    Kind Regards

    Martin