I try to understand some measurement results of an accelerometer MEMS sensor. I used a mikrocontroller which is sampling an accelerometer's analog signal with the built in analog-digital-converter. Sampling rate (timer controlled task, starting the ADC) is 760 Hz, the signal is decimated to 100 Hz (cutoff @ 50 Hz) and then lowpassed to 3 Hz.
My inital goal was to have a look at the noise performance, i.e. how many ADC LSB noise is present. Therefore the µC logged the values for ~16 h and sent the data to a computer via a very slow serial interface. Transfer rate was only 1/5 Hz (otherwise a long time logging would have been complicated).
In the pictures you can see the results of three adc channels. To get an idea of the noise the first sample has been subtracted from all channels in the first plot. There is noise -2 to 1 lsb digit in all channels (x,y,z).
My software is showing a power spectrum automatically (using an hanning window), these are the results (first plot for original data, second plot for data with 1st sample subtracted)
My question is if I can interprete something from the periodic peaks in the original signal which have a distance of 0.0096 Hz which is about 104 seconds in time domain. And why there are no peaks in the "pure" noise signal (plot 2)? Are they simply an artifact?
Are there any deterministic components in this signal (which I might be able to correct) or is this noise stochastic?
Thank you very much!