Anti Alias Performance of the BMI160

[EDC]

I am interested to understand the anti alias performance of the BMI160 accelerometer for low frequency structural vibration measurements.  There may be some out of band vibration that I am concerned may be aliased.

The spec sheet refers to 'pre-filtered' data being feed to the ADC.  Does this imply an alagoue filter prior to digitizing?and if so what is the type, cutoff and order of that filter?  I assume it would be a fixed cutoff given the accel is sampled at 1600Hz.

Is this filter and sample rate sufficient to protect from aliasing in the ADC stage? 

My understanding is the digital filters wont protect me if aliasing is introduced in the ADC stage.

Thanks for any any assistance or guidance that can be offered.

[o_o]

Hi EDC,

In Normal mode, the bandwidth of the signal is always lower than 1/2 the sample rate, therefore aliasing is not an issue.
Of course, the MEMS mechanical design has a certain mechanical frequency response, but all the filtering is done digitally.


Note that, with the exception of BMA400 (that is a unique selling point), low-power modes of all our accelerometers do introduce aliasing, due to the duty-cycling of the accelerometer. With BMI160 for example, the accelerometer is always sampled at 3.2kHz, regardless of the working mode. In normal mode, this 3.2kHz data is filtered to the desired ODR and bandwidth.

o_o

[EDC]

Thanks for your reply.

I dont think I fully understand yet!  My background is with vibration measurement more than digital system design, so help on understanding how the electronics work is really appreciated. 

I dont believe your statement "the bandwidth of the signal is always lower than 1/2 the sample rate, therefore aliasing is not an issue"  really holds.  Aliasing is avoided by ensuring vibration frequencies outsde that bandwidth are excluded from the signal before it is digitized.  This doesnt just happen because you chose a certain sample rate(or bandwidth) they need to be physically removed.  

My understanding is that for a conventional ADC a sharp analogue low pass filter is required to ensure all out of band frequencies (i.e. frequencies above the nyquist frequency or half sample rate) are sufficiently attenuated before digitization. If this doesnt occur then they will be aliased into the data stream and can not be removed by any amount of digital filtering post digitization.

So my question was what the characterisitc of the analogue front end filter actually are?  Or is there some other ADC methodology being used?

The only other means of avoiding aliasing that I know of is via a delta-sigma ADC approach that significantly oversamples, low pass filters and then decimates.  But as the sample rate of the accelerometer is only 3200Hz (actually 1600Hz according to the spec sheet) and max ODR is 1600 Hz, I assume its not a delta-sigma ADC. 

As a conventional ADC, the analogue pre filter characteristics are significant. This would need to have a sharp cuttoff, e.g. 4th order or higher butterworth, no higher than 800Hz (or 1600Hz if sample rate is actually 3200 Hz).  

thanks in advance for any responses.

[o_o]

Hi EDC,

The BMI160 accelerometer contains physical moving masses that have a certain frequency response. The MEMS itself serves as an antialias filter. The signal is massively oversampled within the analog frontend, before being decimated and filtered down to the available output data rates.

There are no aliasing artifacts in normal mode. In case your application is sensitive to aliasing and requires ultra-low current consumption at the same time, you may consider BMA400, which uses a different method for achieving low-power, by accepting an increase in noise density, rather than an increase in signal bandwidth,

In addition, for BMI160, the 7 ADCs (3 for accel, 3 for gyro, 1 for temperature) are derived from a single clock source and sampled simultaneously to ensure synchronization. This clock source also triggers the update of the sensortime register, as well as the drive of the gyroscope oscillation.

o_o

[EDC]

Thanks for trying to assist.

Unfortunately I am wanting to know the specifics of the sampling/filtering setup to determine if a specific unit I am looking at that uses the BMI160 will be acceptable for my application, and what you are telling we is just very generic, rather vague and inconsistent with the data sheet.

e.g. What does 'massively' oversampled actually mean? Where does the data sheet describe this?

I was just hoping to get more specifics than the data sheet provides.

Thanks again for trying to assist.