Post Go back to editing

ADXL001 noisy output

Dear community users,

at the moment I try to set up a vibration monitoring system using the MEMS accelerometer ADXL001. To "check" the ADXL001 I have just made a small pcb. On this pcb I soldered the ADXL001 and two capacitors. One capacitor at the power supply pin VDD to Ground (0.1uF) and the other one at the Output Pin Xout to Ground with 1000pF=1nF. The power for the ADXL001 is delivered by a linear regulator board with a 7805 fixed regulator. I  have connected 100nF capacitor from input to ground and output to ground of the 7805. My data acquisition board is a DAQ NI-9234 from National Instruments. The samplerate is 51.2k (maximum). My vibration source for the ADXL001 pcb board is a small electrodynamic shaker with sinus excitation.

In the time domain I see a lot of noise on the signal... I have just put the time domain and the FFT of the time signal in the attachment. The data acquisition and signal processing was performed with MATLAB.  Any ideas what to do against the noise ? I actually want to use the ADXL001 up to his 22kHz bandwiths but I get a lot of noise beginning at 1kHz... When I design a low pass / anti aliasing filter, then I have select the cut of frequency at 22kHz.. (lower than the half of the maximum samplerate).

The amplitude in the time domain plot is already calculated in acceleration g with the sensitivity setting of 24.2mV/g  at 5V power supply (refer to ADXL001 datasheet).

Thank you for your help and please apologize my bad english.
  • Have you taken measurements on this setup when the device was not shaking? It would be interesting to see if the RMS level of noise meets the expectations in the ADXL001datasheet when the device is rest, so we can start to determine how pure the stimulus motion is. 

  • Hello NevadaMark,

    I will post the non-shaking time signal as fast as possible. At the moment I at home and everything is in the lab... is it enough to make a 0g measurement ? When the measurement direction is 90 degree to the horizontal surface.

    Thank you very much for your quick reply !

    best regards,


  • Yes, a 0g measurement will work. Thank you!

    Best regards,


  • Hello NevadaMark,

    in the attachment I put the time signal calculated in acceleration with sensitivity 24.2mV/g and the time signal in voltage without any converting from voltage to acceleration (g). Also I put the sampled data file of the time signal in acceleration and in voltage. A picture of the ADXL001 board with placement is also attached.

    The rms value is calculated by the command "rms" in MATLAB.

    RMS Value for the Acceleration: 0.3613 g

    RMS Value for the Voltage: 2.5019V

    The data for acceleration and voltage was not made by one measurement.. i performed two measurements.

    Also in the attachment you can find the used MATLAB script.

    All files are compressed into one ZIP File.

    Thank you for your help!

    best regards,


  • UPDATE (2/14/13, NevadaMark): My assessment of the cut-off frequency (5kHz) in this post is not correct. The ADXL001 source impedance is not defined but it is much lower than 32kohms. When combined with the suggested external capacitor of 1000pF, it will not have an substantial impact on the frequency response expectations that are in the ADXL001 datasheet.

    Thank you for offering this information.  We may need or FormerMember to jump in on this, but this is what I would expect.  If we use a single-pole filter assumption, the noise bandwidth will be ~1.57 times greater than the cut-off frequency, which is ~5kHz when using a 1000pF capacitor. For the 70g version of the ADXL001, we can estimate the total noise of this configuration to be:

    Noise = 0.0033g/sqrt(Hz) x sqrt (1.57 x 5000) = 0.292g

    For the 250g version, this would be equal to ~0.321g and for the 500g version, this number would ~0.376g. Which version of the ADXL001 are you using? 

    Unfortunately, I was not able to grab a Matlab license so I was not able to compare the FFT from this most recent data. If this data indicates that the >1kHz noise increases when the device is in motion, the next step for consideration would be to determine if that response represents actual motion that the sensor is experiencing. The core sensor resonance is 22kHz. Based on your pictures, the PCB size could support resonance somewhere in the 1-5kHz region. It might help to determine the the quality of the sine-wave that your vibrating platform is capable of producing as well.

    Message was edited by: NevadaMark

  • Hello NevadaMark,

    thank you for your quick reply.

    Sorry, that I forgot the important information ! => I am using the 70g Version.

    The electrodynamic shaker I have got in the lab only supports max. 1000 Hz sine motion. The first data I have attached in my initial posting was made with 70 Hz excitation sine of the electrodynamic shaker. I will check the the quality of the sine wave with a "calibrated" accelerometer like from Brüel&Kjaer or Endevco tomorrow.

    Can you give me more details about your noise calculations  ?! Where is the factor 1.57 (half of pi ?!) coming from.

    When I use 1000pF the cut-off frequency is 5kHz why ? How can I calculate the cut off frequency as a function of the capacitior value ?! In other ADXL sensor datasheeds there are some tables and equations to calculate this but in the ADXL001 datasheet I can't find any output resistance of the Xout Pin to calculate any cut-off frequency. Is it a simple RC Filter ?! With or without DC coupling ?!

    Thank you very much !

    best regards,


  • We appreciate your feedback on the ADXL001 datasheet and are always looking for ways to improve their value. I didn't produce this datasheet but I suspect that this information was not provided, since the resonance at 22kHz introduces a more complex variable in the bandwidth estimation process.  This will be passed on to our team for future update consideration. Again, thanks!  Back to your questions:

    UPDATE (2/14/2013, NevadaMark): The accuracy of the following three lines (which have been changed to a lighter shade of gray)  has been questioned. I sincerely apologize for this, but it appears like the ADXL001's output impedance is not 32kohms.  Please see most recent post in this discussion for follow-up insights and suggestions. Thanks to for sharing this with me.


    The output impedance of the ADXL001 is 32kohms, which is consistent with may other accelerometer products, including the ADXL203.

    The 5 kHz came from calculating the cut-off frequency (-3dB) of the RC combination (32k/1000pF).

    fc = 1/(2*pi*R*C) = 1/(2*3.1416*32000*10^-9) = ~5000Hz


    If you are unfamiliar with this, the following reference may be helpful:

    For a basic description of the relationship between "noise bandwidth" and "cut-off frequency," check out the following link.

    For a more detailed, academic study of these principles, I learned them from the following textbook:

    Please keep in mind that the resonant behavior at 22kHz will extend the effective noise bandwidth and increase the total noise estimate.  Calculating a more precise expectation would require a more complex approach and may not yield any further value in this discussion.  My primary purpose was to provide a simple model for estimating the noise so we could ask, "are we seeing a reasonable amount of noise when the device is still?" I believe that the answer to that question is, "yes."  You are already working to answer the next question, which is, "what type of acceleration is my ADXL001-70 setup experiencing," through the use of your reference accelerometer.

    I hope that helps.


    Message was edited by: NevadaMark

  • While the ADXL001 does not provide 32kohm of source impedance, you can use an external resistor for the same purpose, by placing it in between the output of the ADXL001 and the 1000 pF capacitor, to introduce a single pole filter at 5kHz.  You stated that you need the entire 22kHz bandwidth, so you may not do this, but we wanted to clarify what you would do, if you wanted to lower the noise by lowering the bandwidth. While we work out the details on how to predict the total noise in a given configuration, you can still work on the original question of: "Is the noise in your original observation the result of actual inertial motion or from the natural noise behaviors of the ADXL001 itself."  In your validation test, you might want to couple your ADXL001 test board to the reference accelerometer and take simultaneous data. Keep in mind that the attachment method can also contribute to differences in the two sensor responses.

    Very best regards,


  • Hello opcode,

    Just curious if you have taken this any further. Hope you are doing well.


  • 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