Академический Документы
Профессиональный Документы
Культура Документы
Graduate Program
Department of Electrical and Computer Engineering
Overview
Signal parameter estimation
Likelihood function
Carrier recovery & symbol synchronization in demodulation
Sem. I, 2012/13
Sem. I, 2012/13
s(t) = Re sl (t) e j 2 f ct
where
Sem. I, 2012/13
2.
Sem. I, 2012/13
p( r ) =
p( r ) p( )
p( r )
Overview
Signal parameter estimation
Likelihood function
Carrier recovery and symbol synchronization in signal
Demodulation
Sem. I, 2012/13
[rn sn ( )] 2
1
exp
2
2
2
n =1
p( r ) =
Where
rn =
r(t) f
T0
Sem. I, 2012/13
(t) dt
and
s n ( ) =
s(t; ) f
(t) dt
T0
[rn sn (t; )] 2
n =1
1
N0
2
[
]
r(t)
s(t;
)
dt
T0
( Show this?)
2
( ) = exp [r(t) s(t; )] dt
N 0 T0
Sem. I, 2012/13
10
Overview
Signal parameter estimation
Likelihood function
Carrier recovery & symbol synchronization in demodulation
Sem. I, 2012/13
11
Carrier phase
estimate for
reference signal
generation for
correlator
12
13
14
15
16
Overview
Signal parameter estimation
Carrier phase estimation
Sem. I, 2012/13
17
Sem. I, 2012/13
18
c(t) = cos(2 f c t + )
such that
1
1
c(t) s(t) = A(t) cos( ) + A(t) cos(4 f c t + + )
2
2
1
y(t) = A(t) cos( )
2
Sem. I, 2012/13
19
c c ( t ) = cos( 2f c t + )
c s ( t ) = sin( 2f c t + )
Sem. I, 2012/13
20
1
1
y I (t) = A(t) cos( ) + B(t) sin( )
2
2
And multiplying s(t) by cs(t) and low pass filtering yields the
quadrature component
1
1
y Q (t) = B(t) cos( ) + A(t) sin( )
2
2
Results:
Sem. I, 2012/13
21
Overview
Signal parameter estimation
Carrier phase estimation
Sem. I, 2012/13
22
1
2
(r(t) s(t, )) dt
( ) = exp
N 0 T0
1
2
1 2
2
r
(t)
dt
r(t)
s(t,
)
dt
s
(t,
)
= C exp
+
N 0 T
N
N
0
0
T
T
0
0
0
r(t)
s(t,
)
dt
Hence, ( ) = C exp
N 0 T
0
Sem. I, 2012/13
23
2
r(t) s(t, ) dt + lnC
N 0 T0
2
2
L ( ) =
r(t) s(t, ) dt + lnC
r(t) s(t, ) dt
N 0 T0
N 0 T0
Sem. I, 2012/13
24
N 0 T0
dL ( )
= r(t) sin(2 f c t + ML ) dt = 0; yields
d
T0
ML
r(t) sin 2 f c t dt
= tan 1 0
r(t) cos 2 f c t dt
Sem. I, 2012/13
25
26
Observe ML
T r(t) cos 2 f ct dt
0
27
Overview
Signal parameter estimation
Carrier phase estimation
Sem. I, 2012/13
28
Phase-locked Loop
Phase-locked loop (PLL) consists of a multiplier, a loop filter,
and a voltage-controlled oscillator (VCO)
Assume that the input to thePLL is a cos(2fct+) and the
output of the VCO sin(2 f c t + )
Then
Sem. I, 2012/13
1
1
= sin( ) + sin( 4 f c t + + )
2
2
29
Phase-locked Loop
The loop filter is a low-pass filter with transfer function
1 + 2s
G(s) =
1 + 1s
1 and 2 are design parameters (1 >> 2 ) that control the
bandwidth of the loop
Output of the loop filter gives control voltage (t) for VCO
The VCO is basically a sinusoidal signal generator with an
instantaneous phase given by
2 f c t + (t ) = 2 f c t + k v( ) d
30
Phase-locked Loop
Neglecting the double-frequency term, the PLL may be
implemented as shown below
31
Phase-locked Loop
Frequency response of the closed-loop transfer function
Sem. I, 2012/13
32
Overview
Signal parameter estimation
Carrier phase estimation
Sem. I, 2012/13
33
34
Sem. I, 2012/13
35
36
37
=
2
N o Beq
Ac2
N0 2
= 2 Beq
Ac 2
N 0 Beq
L
38
exp(L cos )
p( ) =
2 I 0 (L )
Where I0(.) is the modified Bessel function of order zero
Exact values of the variance can be obtained from the
above for the 1st order PLL
The figure in the next slide shows this for 1/
The linear approximation is exact for > 3
Sem. I, 2012/13
39
Comparison of VCO
phase variance for exact
and approximate (linear
model) first-order PLL.
Sem. I, 2012/13
40
Overview
Signal parameter estimation
Carrier phase estimation
Sem. I, 2012/13
41
Decision-directed Loops
How to maximize () or L() when the signal s(t; )
carries the information sequence {In}
Carrier recovery when the signal is modulated uses
decision-directed loops
In such cases, one can use one of two approaches
1. Assume {In} is known or
2. Treat {In} as a random sequence and average it over its statistics
Sem. I, 2012/13
42
Decision-directed Loops
In this case s(t; ) is completely known except for the
carrier phase
Consider decision-directed phase estimate for linear
modulation technique for which the received equivalent low
pass may be expressed as
rl (t) = e j I n g(t nT)+ z(t) = sl (t) e j + z(t)
Sem. I, 2012/13
43
Decision-directed Loops
The likelihood and log-likelihood functions for the equivalent
low pass signal are
1
j
( ) = Cexp Re
r
(t)
s
(t)
e
dt
l
l
N 0 T0
1
L ( ) = Re
r
(t)
s
l
l (t) dt
N 0 T0
j
e
j 1 K 1 (n +1)T
I n rl (t) g (t nT) dt
L ( ) = Re e
N 0 n =0
nT
j 1 K 1
= Re e
I n yn
N 0 n =0
Sem. I, 2012/13
44
Decision-directed Loops
Where
(n +1}T
yn =
nT
Then, L ( ) = Re 1
N
0
I yn cos Im
n =0
N0
K 1
I yn sin
n =0
K 1
ML
K 1
= tan Im I n yn
n =0
1
K 1
Re I n yn
n =0
45
Decision-directed Loops
Sem. I, 2012/13
46
Decision-directed Loops
For a DSB PAM signal of A(t)cos(2fct+), where A(t)=Amg(t)
and g(t) is assumed rectangular pulse of duration T
Carrier recovery with a decision-feedback PLL is shown
below
Quadrature
carriers
Sem. I, 2012/13
47
Decision-directed Loops
Output of the first multiplier and input to the integrator is
given by
1
[A(t) + nc (t)]cos
2
1
ns (t) sin + double frequency terms
2
e(t) =
Then:
Sem. I, 2012/13
48
Decision-directed Loops
The desired component
A2(t)sin , which
contains the phase
error, drives the loop
filter
Carrier recovery system
for M-ary PSK using
decision feedback PLL
is shown in the figure
Sem. I, 2012/13
49
Decision-directed Loops
For the case of M-ary PSK (DF PLL), the received signal is
demodulated to yield the phase estimate
2
m =
(m 1) m = 1,2,...... M
M
1.
1
r(t) cos(2 f c t + ) sinm = [Acosm + nc (t)] sinm cos( )
2
1
[Asinm + ns (t)] sinm sin( ) + double freq. terms
2
Sem. I, 2012/13
50
Decision-directed Loops
2.
1
r(t) sin(2 f c t + ) cosm = [Acosm + nc (t)]cosm sin( )
2
1
[Asinm + ns (t)]cosm cos( ) + double freq. terms
2
1
1
e(t) = Asin( ) + nc (t) sin( m )
2
2
1
+ ns (t) cos( m ) + Double frequency terms
2
The error signal drives the loop filter that provides the control
signal to the VCO
Note that the two quadrature noise components are additive
Sem. I, 2012/13
51
Decision-directed Loops
There are no product of the noise terms and thus no
additional power loss associated with decision-feedback PLL
The ML estimate of given by the earlier equation is also
appropriate QAM
(Readings on non-decision directed loops such as the
squaring loop, Costas loop etc pages 352-358 of text)
Sem. I, 2012/13
52
Overview
Signal parameter estimation
Carrier phase estimation
Symbol timing estimation
ML timing estimation
Sem. I, 2012/13
53
54
Sem. I, 2012/13
55
s(t; ) = I n g(t nT )
Assume the information symbols from the output of the demodulator are known transmitted sequences
Then the log-likelihood function has the form
L ( ) = C L
r(t) s(t; ) dt
T0
= C L I n r(t) g(t nT ) dt = C L I n yn ( )
n
Sem. I, 2012/13
T0
56
yn ( ) = r(t) g(t nT ) dt
T0
dL ( )
d
r(t) g(t nT ) dt
= In
d
d T0
n
d
= I n [ yn ( )] = 0
d
n
The above result suggests the implementation of the
tracking loop shown in next slide
Sem. I, 2012/13
57
58
Sem. I, 2012/13
59
Overview
Sem. I, 2012/13
60
*
L ( , ) = Re r(t) sl (t; , ) dt
N0 T
j
sl (t; , ) = e I n g(t nT ) + j J n w(t nT )
n
n
Sem. I, 2012/13
61
*
+
I
y
(
)
jJ
x
(
)
n n n
n n
Where
yn ( ) = r(t) g * (t nT ) dt
T
xn ( ) = r(t) w* (t nT ) dt
T
Sem. I, 2012/13
62
Define
A( ) + jB( ) =
Such that
1
N0
(I
yn ( ) + jJ n* xn ( )
L ( , ) = A( ) cos B( ) sin
The condition for the joint ML becomes
L ( , )
= A( ) sin B( ) cos = 0
L ( , ) A( )
B( )
cos
sin = 0
=
Sem. I, 2012/13
63
ML
And
B(
)
ML
= tan 1
A( ML )
A( )
B( )
+ B( )
=0
A( )
= ML
64
65
Overview
Sem. I, 2012/13
66
bias = E{ }
Where is the true value. If the bias is zero, then the
estimate is unbiased and the variance of the estimate is
67
E{[ ] }
E{[ ] 2 }
E [
ln p(x )] 2
E{[ ] 2 }
1
2
E [ lnp( x )]
2
2
E [ ln p(x )] = E [ ln ( )] = E 2 ln ( )
Sem. I, 2012/13
68
1
2
E
2 ln (
The lower bound is very useful since it provides a benchmark for comparing any practical estimates
Any unbiased estimate whose variance achieves the lower
bound is called an efficient estimate
Such efficient estimates are rare
Sem. I, 2012/13
69
r(t) sin(2 f
+ ML ) dt = 0
T0
Where
N0
N 0 2 T0
1
=
=
;
2
2
A T0 1 / 2 A
L
A2
where L =
is the loop SNR
2 N o Beq
Sem. I, 2012/13
70
Sem. I, 2012/13
71