Digital Signal Processing Reference
In-Depth Information
thesamespace
V
h
,
the only difference being that the former is an orthonormal
basis. Thus, given any function
y
(
t
)
∈
V
h
, we can write
y
(
t
)=
n
e
(
n
)
h
new
(
t
−
nT
)
,
(5
.
32)
where
e
(
m
) are the coe
cients of expansion of
y
(
t
)usingthebasis
{
h
new
(
t
−
nT
)
.
Thus we have
∞
}
e
(
n
)
∞
−∞
=
n
y
(
t
)
h
new
(
t
nT
)
h
new
(
t
−
mT
)
dt
h
new
(
t
−
−
mT
)
dt
−∞
=
n
e
(
n
)
δ
(
m − n
)
(using Eq. (5.29))
so that
∞
y
(
t
)
h
new
(
t
−
mT
)
dt
=
e
(
m
)
.
(5
.
33)
−∞
Defining the function
g
(
t
)=
h
new
(
−
t
)
(5
.
34)
we see that
y
(
t
)
h
new
(
t
mT
)
dt
=
y
(
t
)
g
(
mT
−
−
t
)
dt,
so that the preceding equation can be rewritten as
∞
y
(
t
)
g
(
mT
−
t
)
dt
=
e
(
m
)
.
(5
.
35)
−∞
g
∗
(
nT
Note that
span the same space
V
h
.
The process of
extracting the coe
cients
e
(
m
)from
y
(
t
) can therefore be interpreted as filtering
followed by sampling, as shown in Fig. 5.11.
Summarizing, given the output
y
(
t
) of the channel
h
(
t
) as in Fig. 5.12(a), it
can be expressed in the form (5.32) using the orthonormal basis
h
new
(
t
{
h
(
t
−
nT
)
}
and
{
−
t
)
}
nT
).
Thecoe
cientsofexpansion
e
(
n
) can be obtained by filtering
y
(
t
)withthe
receiver filter
g
(
t
) having frequency response
H
new
(
jω
), that is,
−
H
∗
(
jω
)
S
h
(
e
jωT
)
G
(
jω
)=
,
(5
.
36)
as indicated in Fig. 5.12(a) on the right. This is simply the normalized matched
filter, whose sampled output has its noise component whitened. These samples
e
(
m
) contain the full information about
y
(
t
) in view of Eq. (5.32). Using a noble
identity we can redraw the system as in Fig. 5.12(b). Since
e
(
m
) contains all the
information about
y
(
t
), the sampled version
r
(
n
) of the output of the matched
filter
H
∗
(
jω
) also contains this information.
Search WWH ::
Custom Search