Examination window size and FFT size in Goldestein filtering

Hi Everybody,
I extracted canopy height model using interferometric technic and TanDEM-X data, after making interferogram step, I did Goldstein filtering in snap software. image
I have tried several window ( 33, 55, 7*7) and FFT sizes and (32, 64, 128, 256 ). my final result was good for 32 FFT size. I don’t know what is mean of FFT size and what difference is with window size.

FFT stands for fast Fourier transform. Basically, small sample patches of the interferogram are transformed into a 2D representation and filtered in this domain, thus making the filter parameter flexible [0-1] for regions of different noise levels.
Richard Goldstein published a paper on this filter technique. It however doesn’t explicitly mention the role of FFT for the filtering:

Goldstein & Werner (1998): Radar interferogram filtering for geophysical applications

Better explanations are given in papers adapting his idea: