When windowing non-overlapped segments of data stream, part of the data is being "ignored" at the boundaries of the segments due to the tapering of window functions (non-rectangular). Some information is lost in further processing.

My question is: for a given window function, does overlapping the segments by $50 \%$ preserve $100\%$ of the data stream information ?


Depending what do you mean by preserve $100\%$ of data stream information. Usually $50\%$ of overlap is enough and you are not loosing too much information. Amount of data within your observation window is definitely sufficient for further analysis.

Although there are some windows, such as Hanning or modified Gaussian window which that are complementary. This allows you to perfectly reconstruct original signal from windowed segments.

| improve this answer | |
  • $\begingroup$ I am wondering if there are figures/numbers as to what percentage of the data is preserved, I heard some saying in the $70\%$ range, but I can't confirm this nor find any source. $\endgroup$ – Gilles Aug 17 '14 at 16:36

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.