I assume that SM randomizer primitive has linear spread of random values. So, when generating random freqiencies with octave-related equal spread (logarhitmic - same amount of hits per octave than per each band of same Hz; thus in lower range there will be more hits) - output requires some modification. I did this by converting Rand output into decibel mode, and then added two parameters. Highest generated frequency and how many octaves down the Rand should cover (thus it produces lowest generated frequency too).
Could someone check if my thinking is correct? Or is there simpler way?