I am running a MatLab script from Chronux for multi-taper spectral analysis and am trying to accurately determine the number of tapers I should use. Anyone have any suggestions?
Hi Greta,
From what i know Matlab has a default number of 3 when it comes to the number of data windows (nw). What is important also is the number of tapers (k), which normally being estimated as 2*nw-2. This value is by experience a very good choice.
So, nw=3, k=4 should be able to take you further....
Thank you so much Konstantinos! I had been using the Chronux default of nw=3 and k=5 but will try changing k to 4.
In general from what I remember, increasing the number of tapers will decrease the variance of your spectrum.
If you are interested in a leakage free (well at least leakage reduced) spectrum use a single taper if you desire a very smooth spectrum use more. The number of tapers depends on what you desire to achieve.
Is there any paper taking about the choices of taper number? I found my result is sensitive to the number of taper. Thanks!
This paper cites a few that will go over the issue a bit. To start you can use 1,1 which is essentially no taper to see if you have a robust result.
http://dx.doi.org/10.1016/j.jneumeth.2016.04.015
- Badges
- Science topic