Post Go back to editing

comparement ADXL380 ADXL355 spectral density noise

Thread Summary

The user is comparing the noise floor of the ADXL380 and ADXL355 accelerometers in their setup. The ADXL380 shows a higher noise level, which could be due to differences in sampling frequency and FFT bin width. The ADXL380 is configured in high-performance mode with the EQ filter active, and the measurement time was 16384ms for both devices. The user applied a Hanning window and compensated the FFT amplitude by a factor of 2.
AI Generated Content
Category: Datasheet/Specs
Product Number: ADXL380

Dear ADI-Engineer-Team,

 I've got some of the rare EVAL-ADXL380-EB-1Z and I start with measurements to compare    

  1. the new ADXL380 at EVAL-ADXL380-EB-1Z against 
  2. the ADXL355 in our application housing  - calls here “BM Fair30m” and
  3. the ADXL355 at a PMDZ eval board.  

Especially I look on the level of the noise floor in the frequeny spectrum.

 

short my Setup:

The ADXL355 in our application housing is good mounted on a linear shaker and the Eval-Boards ADXL355 und ADXL380 are glued with some tape on the housing in the same direction.

The ADXL355 in our application is driven by a FPGA via SPI and the MEMS gets a precise external clock from the FPGA as well.

The MEMS at the eval boards use their own clock and are driven by the SPI-interface of an Raspberry4  –(FIFO of the MEMS are used).

 

The MEMS datasheets says me, that the new ADXL380 should be has a lower noise level as the ADXL355 – (Noise Density are the same as Noise Spectral Density)

But my own frequency spectra charts of my measurements looks like that the ADXL380 has a 2 times higher noise level as the ADXL355 has?

At my opinion the 20bit amplitude resolution at +/-2g range of the ADXL355 and the 16bit amplitude resolution at +/-4g range of the ADXL380 should not matters with the absolute spectral noise density – or I’m wrong.  

 

Here are some  facts:

 

Datasheet of ADXL355:

   

Datasheet of ADXL380:

 

 

 

By default the EQ-Filter should be enabled at the ADXL380 after power- or a soft-reset.

…

Fourth-Order Equalizer Filter (EQ)

The ADXL380 includes a fourth-order digital equalizer filter designed to flatten the sensor frequency response to increase the sensing bandwidth up to 4 kHz (see Figure 62).

The EQ feature is only available when operating in HP mode.

The EQ is automatically bypassed in other operational modes.

The EQ, IIR1, and LPF cannot be used concurrently.

The FILTER register (Register 0x50) is used to disable the EQ.

In HP operational mode, the user must set the LPF_MODE bits in the FILTER register (Register 0x50, Bits[5:4]) to b’00 and

set the IIR1_EN bit in the TRIG_CFG register (Register 0x49, Bit 5) to b’0 prior to enabling the EQ (default configuration).

The EQ_BYPASS bit in Register FILTER (Register 0x50, Bit 6) is set to b’0 by default to enable EQ.

The user must set the EQ_BYPASS bit to b’1 to bypass the EQ.

The EQ is automatically bypassed in all other operational modes

…

  

I’m sure that ADXL380 is configured in high perfomance (HP) operational mode and that the EQ filter is active.

Here are some log lines of my program:

               …

Analog Devices ID: 0xAD

Analog Devices MEMS ID: 0x1D

Device ID: 0x17 (octal: 27)

REG_OP_MODE (0x26): 0b1100

REG_FIFO_CFG0 (0x30): 0b1100000

REG_FILTER (0x50): 0b0

REG_TRIG_CFG (0x49) : 0b0

REG_DIG_EN (0x27) : 0b11111000

                …

 

 

# frequency spectrum at ambient ‚no‘ - excitation

 

 

 

 

# small approx. 1,2mg Excitation with 333Hz sinus signal by shaker  (25Hz peak unknown infrasound, 50Hz and 100Hz peak comes probably from body sound transfer of the power supply coil in the analog amplifier of the shaker)

 

 

 

 

  • HI @askemos,

    To make a proper comparison between the two FFTs, we need to consider the acquisition and processing parameters, particularly the sampling frequency, number of samples, window and FFT normalisation. 

    These parameters can change the apparent noise floor even if the devices have the same noise density.

    As an example, assuming 8 kHz ODR for the ADXL380 in HP mode and 4 kHz for the ADXL355, if the same number of sample and the same window are used, the frequency bins of the ADXL380 FFT are twice as wide. Since the noise amplitude in each bin scales with the square root of the bin bandwidth, the ADXL380 could show approximately √2 higher noise amplitude per bin even for the same noise density.

    I hope this helps

  •  Thank you @EFernandez,

     I didn't really expect an answer after 1 3/4 year and am now pleasantly surprised.

    The ScreenShots of the power spectra of the ADXL355 (up to 2 kHz) and the ADXL380 (up to 4kHz) were based on the respective maximum possible sample rate of the chips of 4kHz and 8kHz respectively – i.e. not both with fixed ODR = 4000 as assumed.

    The measurement time, on the other hand, was identical and amounted to 16384ms and before the FFT, a Hanning window was applied across the board – especially for the ambient measurements – and after the FFT, the signal amplitude of the power spectrum was compensated by a factor of 2.

    The simultaneous comparative measurement between EV board ADXL355 (22.5 μg/√Hz) and EV board ADXL380 (20 μg/√Hz (XY) at that time showed practically twice as high noise carpet in the experiment and, like its "successor" ADXL318, will not be considered further for the time being.

    To make it easier to select the MEMS for your own application, I would like to specify the noise level in relation to specific frequency bands, as is usual with analog piezo sensors, if the value is not constant over the entire frequency band.  

  • Hi @askemos,

    Sorry for the delay in our initial reply, and thanks for the detailed feedback.

    With the same acquisition time and window, the FFT measurements should be comparable. Knowing now that you were running the ADXL355 at 4 kHz ODR, I went back to the characterisation data, since the datasheet noise specification is given at 500 Hz ODR.The characterisation data shows that at 4 kHz ODR, the typical noise density is actually lower than the datasheet value:

    • X: ~12µg/√Hz
    • Y: ~16.µg/√Hz
    • Z: ~15µg/√Hz

    I think these values are more consistent with what you observed experimentally and help explain why the ADXL355 noise floor appeared lower than expected compared with the ADXL380.

    Thanks also for your suggestion regarding noise versus frequency. The ADXL355 noise does show some frequency dependence across the measurement bandwidth. For the ADXL380, the noise density is quite flat across its 4 kHz bandwidth. For newer products, such as the ADXL383, we have included a noise spectral density versus frequency plot directly in the datasheet.

    Regards

Before You Switch


Switching languages will make ADI Explorer unavailable. Resume your session by switching back to English and reopening ADI Explorer.