The problem is the size of the range in position 1 with that time window.
For a 0.02s window with a 44100 sample rate audio the minimal frequency bucket is:
Num_samples = 44100 * 0.02 = 882 (now we search for a window size power 2 to fit that number samples (1024), this is the size of our FFT analysis)
So each FFT result position represent a band of 44100 / 1024 = 43.0660 hz
Now you have a band of 120 - 90 = 30 Hz, that value falls below the band size of the FFT
If you request a range under that minimal you obtain a 0.
With a timewindow of 0.03 you can reduce the minimal size to work with your bounds.
44100 * 0.03 =1323 -> next power 2 value 2048
44100 / 2048 = 21.53 hz
So you can increase the time window by 0.03 sec to reduce the minimal range and obtain a value for all your curent bands (or you can increase the size of that band over 43)