To keep the blurring effect of the spectral window small, the bandwidth B,, of the 
Lee 
window must be small compared with the spectrum bandwidth B,. By, = mes is a popular 
guide. 
As explained in Sections 2.5.2—2.5.5 the following tendencies, as shown in Table 
2.3, are now Clear. 
Table 2.3. Statistical effects of the change of M. 
From sampling theorem it is known that, if [s(w)] is calculated from the frequency 
TON . : 4 Rn . : 
interval Aw = Ww 5(@) is completely determined, although since s(w) is a continuous 
function, it would be quite in order to evaluate it over a much finer set of points. It has 
1 
been shown that we can generally use the value Aw = 3 Bw for Daniel’s window, where 
27 
By, = Wr accordingly Aw = va When we use Aw = va the spectrum window effect 
will be replaced by a weighted mean of the ordinates, 
M 
S(w) = » aiPy(o +i). (2.118) 
i=-M 
Values of a; for several windows are shown in Table 2.4. 
This table also gives values for the coefficient proposed by H. Akaike,?> for 
minimizing the bias and variance of the estimate for the spectrum window. W> or its 
simplified form Q is generally used in most of our work, although W3 would give a 
better spectrum with steeper peaks and deeper valleys. As a general procedure for 
choosing the best of W; to W3, Akaike advises trying them all and, if no improvement 
in spectral form is recognized, adopting the window of the lower order. 
45 
