I am using a window of 400 samples to analyze a signal with 1280 samples (sampling frequency used is 6.4 kHz). How do I find the overlap length in this case?
I believe that there is no an exact answer for that.
However, you may consider some implications of your heuristic:
- Obviously, you need some time to buffer the initial samples to start your process. Is this time relevant?
If yes, you should consider working with fewer samples in your window to avoid undesired delays.
- About the increment/slide ratio:
If you intend to use a small increment (e.g., 20 samples in your example), you will be able to have a more dynamic influence of your signal in your windows.
However, using longer increments (e.g., 200 samples in your example) will not only diminish the final number of your windows considerably but also imply in an emphatic low-pass filter effect in your signal. Note that when buffering those 200 samples you will not be able to perform any action (if your signal is used to control something). This means that longer increments tend to offer more stable signals but at the cost of greater delays.
What you need to balance is the dynamics x the delay of the system.
If you do not have time constraints, I would say that generally, to use 30% ~ 50% of overlap provides good signal stability and reasonable responsiveness.
Hope I could help you.
- Badges
- Science topic
- Similar topics
- Engineering
- Electrical Engineering
- Signal Processing