Geology Reference
In-Depth Information
on changing the integration variable to t =−
t . Thus, the Fourier transform of the
time reverse of the window w( t ) is the complex conjugate of W ( f ). Given that the
windoww( t ) is real and even, W ( f ) is real. The convolution satisfies
w n ( s )w n + m ( s
τ) ds
=
w n ( s )w n + m
s ) ds
−∞
−∞
w n + m ( u )w n ( u
=
w n + m ( u )w n
u ) du
=
τ) du ,
(2.377)
−∞
−∞
on changing the integration variable to u
s , and utilising the real and even
properties of the windows. Using the foregoing properties and the convolutional
relations (2.269) and (2.274), expression (2.374) becomes
E S gg, n ( f ) S gg, n + m ( f )
= τ
I 2
f 1 ) W 2 ( f 1 ) df 1 2
1
=
S gg ( f
−∞
I 2
b n ) W 2 ( f 1 ) df 1 2
1
f 1 )exp i f 1 ( b n + m
+
S gg ( f
.
(2.378)
−∞
From (2.360), we recognise that
I 2
f 1 ) W 2 ( f 1 ) df 1 2
E S gg, n ( f ) 2
1
S gg ( f
=
,
(2.379)
−∞
the square of the expected value of the sample spectral density estimate. Substitut-
ing in (2.365), we find for the covariance,
cov S gg, n ( f ),
S gg, n + m ( f )
I 2
b n ) W 2 ( f 1 ) df 1 2
1
f 1 )exp i f 1 ( b n + m
=
S gg ( f
−∞
b n ) W 2 ( f ) df 2
S 2
gg ( f )
I 2
exp i f ( b n + m
,
(2.380)
−∞
for spectral densities assumed to vary only slightly over the passband of W 2 ( f ).
The window w n + m ( t ), centred on b n + m , is displaced along the time axis by m
Δ
s
from the window w n ( t ), centred on b n , where
s is the separation of successive
windows. Now consider the convolution of the windoww n ( t ) with the time reverse
of the windoww n + m ( t ),
Δ
w n ( t )w n + m ( t
J m (τ)
=
τ) dt .
(2.381)
−∞
 
Search WWH ::




Custom Search