Smallest achievable spectral resolution with ADXL357

Dear Sirs,

We are currently evaluating design possibilities for low frequency - 0.7Hz-3Hz - accelerometer. We are considering either piezo (we have experience in that) or MEMS sensor. We are inclining to MEMS because of faster design time, lower possibility of error in mechanical design and low risk of EMI problems with small signals as we consider sensors with digital output (am I correct at assuming that MEMS accelerometer with digital output is virtually immune to EMI interference and strong magnetic fields - up to 0.3T). Currently we are evaluating ADXL355 as a candidate. According to the Datasheet Resolution is 3.9ug/LSB and Noise Density is 25ug/rtHz (for 2g range), is it possible to push spectral noise observed in the RMS amplitude spectrum of accelerometer signal down to 4ug given that we will have long enough (>=15sec) FFT frame length, averaging and any other DSP features if necessary? While trying to push down noise so low what else Datasheet parameters we have to take into consideration (at the first glance neither DC offset nor Velocity Random Walk should not influence our measurements)? Any recommendations on using built-in filters to improve resolution, reduce noise floor and lower computational load and memory requirement for external MCU?

Also how will sensor perform if we set LPF corner to ~5-10Hz, range to ±2g and apply 5g short impact or ±5g sinusoidal vibration?

Any additional tips and hints are welcome.

Thanks!