[ADXL354-ADXL355] Frequency response for seismic purposes

1. We do not provide poles and zero as in most applications this is not critical because the flat passband range is used not approaching the filter response.  In applications where this is critical, there is usually a factory calibration performed on the final assembly which would help account for resonances of the module itself.

2. Again, we do not typically provide filter coefficient details in the design.  Much of the response is shown in the plots of the datasheet.

Seismic sensors are typically lower frequency and extremely low noise.  The ADXL355 can be used, but there are specific sensors designed for this application, albeit at higher costs.  If you need more control over the filtering characteristics, then the ADXL354 (analog output) plus external ADC may be a better option to investigate so you have full control over the digital filter.

Goodmorning Anthony, thanks for answer but keep in mind this for my purposes:
I'm aware seismic instruments (sensors and dataloggers) work at low frequencies with much less noise, but my target is design a cost-effective (compared to professional instruments such as Nanometrics, Kinemetrics, Quanterra, Sara and so on) device.

Obviously with "lower" performances

Of course I can buy professional accelerometer for seismic application with no noise, but it works at low frequencies, or  I can buy narrow band accelerometer, with higher cost and high noise.

My purpose is to compare professional seismic accelerometers (such Titan and Episensor) and MEMS based seismic accelerometers (GeoSig) with commercial MEMS (in sensor arrays such Suricat or stand-alone such Raspberry shields).

I can get PGA (peak gain acceleration) values during an earthquake from several 4$ MEMS (180ug/sqrt(Hz)), but it would be better if I find for same price (35$ in this aapplication for me stands for "more or less") a 50ug/sqrt(Hz) so I can detect smaller events.

At same time, I can use MEMS accelerometer for monitor earthquake effects on a building, installing one device for every floor. Which is the difference between a "cheap" LIS344ALH (just for example, to take a "concurrent" sensor) and an ADXL354 (with lower noise)?

How much will noise affects results on monitoring building?
How far are results obtained from a professional accelerometer (Titan SMA) and a MEMS based device?
Could I use a MEMS based device instead professional one? If so, keep in mind for the same price of Titan SMA I can fill a building with at least 10-15 devices or, for same budget, monitor 10-15 buildings instead only one.

So you can understand comparing "on the fly" professional instruments and cost-effective deevices is not so easy just reading sensor datasheet.
Then you can see comparing different sensors on shake-table will give you different results in terms of phase delay (and so in terms of waves analysis) even if transfert function (magnitude) is more or less the same (that's why researchers who will perform analysis asks me poles/zeros and/or filter coefficients).

Ok, let's come back to ADXL: ADXL354 is perfect for our characteristics in terms of selectable range, frequency range (0-1000Hz), cost and self noise.
I selected this MEMS also because same "core" is available in both analog and digital version.

As you correctly said in your answer I can add an external ADC for ADXL354, but this means I can increase space and components number: between LIS344ALH (I use this as example because I've already seen it in seismic devices) with external signal conditioning and ADC, and ADXL355, which do you think I'll prefer if I need to design a board for Intel Edison?

I understand you can't tell me factory "secrets" of ADXL sensors, but for my purpose just "take a look to datasheet plots" doesn't help me so much.

For ADXL001 (I've found a thread in this forum) it's reported on datasheet both resonant frequency and quality factor, so there was a Matlab script for modeling accelerometer core.

I used same script with no success.

s=tf('s');
wao=2*pi*2400;
qao= 10;
qao= 11*1.276;

NUMa=wao^2;
DENa=[1 wao/qao wao^2];
Ha=tf(NUMa,DENa);

freq(1:5000)=1:5000;
omega(1:5000)=2*pi*freq(1:5000);
[ma,pa]=bode(Ha,omega);
Ma(1:5000)=(abs(ma(1,1:5000)));

% reference from ADXL354 graph
x_ref=[1500,2132, 2400,2795, 3000, 4000, 5000];
y_ref=[1.09,2, 6.7, 2,1, 0.2, 0.1];

loglog(freq,Ma,'-b',x_ref,y_ref,'g*', 1500, 2*1.09, 'rx');

grid on;
hold on
ylim([0.01 10]);
xlim([10 5000]);
ylabel('Xout [g]');
xlabel('FREQUENCY [Hz]');
title('ADXL354 frequency response for X-Axis');
legend('hypothetical accelerometer core (Wr=2.4kHz, Q=11)', 'ADXL354 datasheet', 'core value (@1500 Hz)');

Here results:

ADXL354 datasheet reports gain 0,5 (@1500 Hz) due to low-pass internal filter.

Green stars are reference points from ADXL354 datasheet frequency response (fig. 7).

Due to low-pass filter, frequency response should be 0.5 instead 1.09 (first green star), so I suppose blue line would be "core" response, but in this case value at 1500 Hz should be in red cross.
Regardless Q value of second order system response (wao is given in datasheet) I can't reach red cross value...

If I apply a second order low-pass filter (yellow line) output response (violet line) fits better, but not exactly as datasheet plot so I suppose it isn't correct model.

So, is there any way to model core accelerometer and analog (is it really an analog sinc filter? It feels strange datasheet would be wrong but I've never noticed "analog" sinc filters...) filter?

About ADXL355, low-pass filter (at 4kHz ODR) is reported as 1000 Hz.... but in this case it's assumed corner value -6dB (0.5) instead -3dB (0.7), right?

Thanks, Michele.

We are also curious about the shape of the filter response for the decimation and interpolation filters.  As mentioned above, data about the shape of the curve will not be provided.  Can you comment whether the phase is flat at frequencies less than ODR/4 for the combined decimation/interpolation filters? (I'm guessing not, but just want to confirm)

Sean

Hi. Did you manage to find an answer to this question? I'm too trying to find the transfer function of ADXL355 but with no luck so far.