Digital Signal Processing Reference
In-Depth Information
for all
ω,
where
ω
s
=
2
π
T
See the review of sampling theory given in Appendix G.
10.2.2.A Formulating the optimization problem
The integral in Eq. (10.7) can be written as
A
(
ω
)
B
(
ω
)
dω/
2
π,
where
A
(
ω
)=
S
1
/
2
qq
(
jω
)
(
ω
)=
|
H
c
(
jω
)
|
,
(10
.
15)
|
H
(
jω
)
|
T
are non-negative quantities. Given
A
(
ω
)
≥
0, find a function
B
(
ω
)
≥
0such
that
ψ
=
∞
−∞
A
(
ω
)
B
(
ω
)
dω
2
π
(10
.
16)
is minimized under the constraint
∞
B
(
ω
+
kω
s
)=1
(10
.
17)
k
=
−∞
for all
ω.
Since the left-hand side is periodic, it is su
cient to ensure the condition
for 0
≤
ω<ω
s
.
For convenience we rewrite the objective function as
ψ
=
ω
s
0
∞
A
(
ω
+
kω
s
)
B
(
ω
+
kω
s
)
dω
2
π
(10
.
18)
k
=
−∞
Since
A
(
ω
) is non-negative, the non-negative
B
(
ω
) which minimizes Eq. (10.18)
should be such that, for each frequency
ω
0
in 0
≤
ω
0
<ω
s
,
the sum
∞
A
(
ω
0
+
kω
s
)
B
(
ω
0
+
kω
s
)
(10
.
19)
k
=
−∞
is minimized under the constraint (10.17). Let
k
0
be an integer such that
A
(
ω
0
+
k
0
ω
s
)
≤
A
(
ω
0
+
kω
s
)
for all integer
k.
♠
Lemma 10.1
.If
B
(
jω
) is chosen such that
B
(
ω
0
+
kω
s
)=
1for
k
=
k
0
0
(10
.
20)
otherwise,
then Eq. (10.19) is minimized, and the constraint (10.17) is satisfied as well.
♦
Proof.
Let
a
k
be a fixed set of non-negative numbers arranged such that
a
0
≤
a
1
≤
a
2
...
(10
.
21)
Search WWH ::
Custom Search