Digital Signal Processing Reference
In-Depth Information
y
bp
(
t
)=
x
bp
∗
h
bp
)(
t
)
=Re
(
x
lp
∗
h
lp
)(
t
)
e
jω
c
t
e
jω
c
t
s
c
(
k
)
p
(
t − kT
)+
jp
(
t − kT
)
=Re
∗ h
lp
(
t
)
.
k
This can be rewritten as
y
bp
(
t
)=Re
k
kT
)
e
jω
c
t
,
s
c
(
k
)
h
lp,e
(
t
−
(2
.
105)
where
h
lp,e
(
t
)=
p ∗ h
lp
+
jp ∗ h
lp
(
t
)=
h
r
(
t
)+
jh
i
(
t
)
.
(2
.
106)
Here
s
c
(
k
)
,p
(
t
)
,
and
p
(
t
) are real, but
h
lp
(
t
) is complex. So the real and imagi-
nary parts
p
(
t
kT
) get mixed up by convolution with
h
lp
(
t
). Nor-
mally the PAM/SSB demodulator just multiplies the received signal by cos
ω
c
t
and performs lowpass filtering. This removes
−
kT
)and
p
(
t
−
p
(
t
) and recovers
p
(
t
)
.
But, since
h
lp
(
t
) mixes up
p
(
t
)and
p
(
t
)
,
the receiver now has to be smarter, and must
perform a demodulation with cos
ω
c
t
and one with sin
ω
c
t
(like a QAM demod-
ulator). This recovers the real and imaginary parts of
s
c
(
k
)
h
r
(
t
kT
)
.
−
kT
)+
jh
i
(
t
−
k
So the received digital signal after sampling has the simple form
y
d
(
n
)=
k
s
c
(
k
)
h
d
(
n
−
k
)
,
(2
.
107)
where
h
d
(
n
)=
h
r
(
nT
)+
jh
i
(
nT
)
.
One may argue that we can simply keep the real part of
y
d
(
n
) and minimize
complication. But this is like wasting the power of the signal in the imaginary
part. For example, if
h
d
(
n
) happens to be purely imaginary, this is a disaster
because the real part of
y
d
(
n
) has no information about the signal. So we take
the entire complex signal
y
d
(
n
), perform “time-domain equalization” (TDE),
and take the real part to get
y
d
(
n
)=
k
s
c
(
k
)
h
d
(
n
−
k
)+
q
(
n
)
,
(2
.
108)
where
h
d
(
n
)isrealand
q
(
n
) is real-valued noise. Thus at this point the digital
signal processor simply sees a real PAM signal
s
c
(
k
) and a real channel
h
d
(
n
).
Search WWH ::
Custom Search