Multitapering

Having established that lower values of $M$ perform well when using Tukey tapering, we can use equations 16 and 13 with $N=196$, to give a multitaper with approximately the same spectral density estimate variance. For $M=10$ this gives $NW=13$. We compare this with $NW=4$, a commonly used value in the literature and corresponds to $M=32$. The results are shown in figure 14 for the TR=3 secs data. $NW=13$ performs much better and is similar to the Tukey estimator with $M=10$, which is not surprising since they correspond to the same variance and therefore to similar amounts of spectral density smoothing. Again, increased spectral density smoothing compared with that which is normally recommended is better.