Issues with Spectrogram

Since our last post on the Spectrograms and their utility, we have spent some time thinking about the current challenges to the approach and how do we refine our spectrograms to meet the expectations.

Interestingly, all the answers are contained within the way ObsPy creates spectrogram, or any spectrogram is created for that matter. Here is a good link which explains some of the technical aspects of a spectrogram.

In our last post concerning spectrogram, we talked about poor resolution of the spectrograms that we created. It appears to us that whole issue was due to the three important variables in ObsPy spectrogram that we had failed to understand clearly:

1. Wlen (stands for Window Length): In concept, every spectrogram is created by dividing enitre data stream into small windows, and then FFT of each window is taken and stacked side by side to produce the plot. Therefore 'wlen' defines size of each such window. Now, a very large wlen implies less stacks (hence poorer resolution) while very low wlen implies too less data for single FFT(hence poorer resolution on color scale). We had to carefully tweak the settings for our purpose.
2. Over Lap: To produce better quality spectrograms, two consecutive windows in a spectrogram usually have some overlapping data points. But a very high overlap can disturb your X axis scaling.

We had test these settings one by one.By tweaking above two parameters, we were able to create somewhat better spectrograms, like the one displayed below.

Several spectrograms have been stacked one over another for LADWP building, with decreasing floor numbers(top to bottom)