I am making measurements with the EVAL-ADMP521Z-FLEX using Clock: 1.78 MHz, Vdd=3.3v. I sample the PDM output bit train into a PC computer and make my analysis by S/W (I simulate a 4-pole LPF that produces signed 16-bit samples at 8.0 Ksamp/sec). I noticed that the reconstructed waveform has a negative offset of ~2.7% of the full dynamic range (that's to say the resulted waveform has a DC of -1750 when full dynamic range is -32768 to +32767). I also verified that the DC offset comes from the PDM input (and not from the data processing...): In a silent place, when sampling the PDM input I can see that the average number of '0' samples is ~5% higher than the average number of '1' samples.
Please let me know if this is a known issue or is this something that has to do with my setup.