Waterfall windowing problems

[elvischan917]

I am writing a program to make the waterfall and have serveral problems, hope you can help:

1. I am using square window, is it correct?

2. When I move the window, shall I keep the window size or shrink it in each step?

I cannot find any comprehensive guide or standard on the web, please help.

Thank you very much!

Alvis

[bwaslo]

I'm assuming you mean the "Cumulative Spectral Decay" plot? Some other things are also called 'waterfall' plots sometimes, but usually in Loudspeakers it's a CSD.

Square WIndow: "correct" depends on what you want to optimize. Most often a Hanning or Blackmann Harris half-window is used. Oversimplifying: you trade off the spectral splatter of a too-abrupt window with the loss of resolution of a too-smooth one.

In a CSD, the right (later) window edge is held fixed and the left (earlier) one is shifted to the right. As a side result, the frequency over which each successive FFT has meaning decreases as the left window moves (because there is less time left within the shrinking window). You have to keep the right window fixed to reject reflections (echoes) from the room and setup, it is usually set just ahead of the first reflection.

[elvischan917]

Thanks bwaslo for solving my first 2 problems.

Now I have another problem.

I simulated an impulse response for a butterworth 4th order bandpass filter with f1=500Hz and f2=2kHz. In my CSD, I cut through it with a square shrinking window.

a) The energy around 1kHz goes down to below -30dB after 6ms, is this duration normal? Or my calculation is wrong?

b) The energy around 100Hz~500Hz going up and down after 1ms, and it goes down to below -30dB after 4ms. Energy below 500Hz is suppressed before the first milliseond (this is what I expected) but it comes up afterward in the CSD. Is it wrong? Is it caused by sharp edge of the cutted impulse response on the left of the window?

[phase_accurate]

As an estimation this sounds plausible.

Do you have PSPICE ? Then you could use this to check further for plausibility.

Regards

Charles