What is the reason for 128 scan minimum in the 'ddt' method of segmentation?

[Michael Head]

The math here is really complex and just out of my league so I don't have a good answer for you. You'll have to read the reference given in s_ddt.m. The 128 requirement is placed on the s_segmen routine which is simply reviewing the statistical distance determined by s_ddt. If you read the reference for the comparison routine, s_segmen, you'll find there isn't one...! s_segmen is making a arbitrary judgment of ddt statistical distance and the 128 is just part of the arbitrary part.

Customers often don't use the fancy ddt and instead simply divide the profile into fixed-length segments of a meter or 50 cm or some other size. This approach simply ignores trying to identify statistically similar regions in the profile (ddt's approach). You could try using the 'scans' method to obtain shorter than 128 scan profiles or you could simply change the 128 minimum on s_segmen.

You'll find that the measured dissipation depends strongly upon how the profile is segmented. No one to my knowledge has tried to compare the 'ddt' method with the 'scans' method so I can't even say which is better.

Also, there is a Fourier transform step that is subsequently applied to the segment however you define it. This step requires a minimum number of samples to make any estimate at all, with more samples leading to a better estimate. So you can't simply decrease the segment size towards zero without consequences even in 'scans' mode.

Sorry I can't be more help!