CHAPTER 6

Problem 6.1 :

N
N
Using the relationship r(t) = n=1 rn fn (t) and s(t; ψ) = n=1 sn (ψ)fn (t) we have :
  
N 2
1
N0
[r(t) − s(t; ψ)]2 dt = 1
N0 n=1 (rn − sn (ψ))fn (t) dt

1 
N
N
= N0 n=1 m=1 (rn − sn (ψ))(rm − sm (ψ))fn (t)fm (t)dt

1
N
N
= N0 n=1 m=1 (rn − sn (ψ))(rm − sm (ψ))δmn

1
N
= N0 n=1 (rn − sn (ψ))(rn − sn (ψ))

N
= 1
2σ2 n=1 [rn − sn (ψ)]2

where we have exploited the orthonormality of the basis functions fn (t) : T0 fn (t)fm (t)dt = δmn
and σ 2 = N20 .

Problem 6.2 :

A block diagram of a binary PSK receiver that employs match filtering is given in the following
figure :

Received signal Matched filter Output data

✲ X❧ ✲ h(t) = g(T − t) ✲ Sampler and ✲
Detector
✻
✻

✲ Carrier phase cos(2πfc t + φ̂)

recovery

✲ Symbol
synchronization
æ
As we note, the received signal is, first, multiplied with cos(2πfc t + φ̂) and then fed the matched
filter. This allows us to have the filter matched to the baseband pulse g(t) and not to the
passband signal.

122
If we want to have the filter matched to the passband signal, then the carrier phase estimate is
fed into the matched filter, which should have an impulse response:

h(t) = s(T − t) = g(T − t)cos(2πfc (T − t) + φ̂)

= g(T − t)[cos(2πfc T )cos(−2πfc t + φ̂) + sin(2πfc T )sin(−2πfc t + φ̂)
= g(T − t)cos(−2πfc t + φ̂) = g(T − t)cos(2πfc t − φ̂)

where we have assumed that fc T is an integer so that : cos(2πfc T ) = 1, sin(2πfc T ) = 0. As we

note, in this case the impulse response of the filter should change according to the carrier phase
estimate, something that is difficult to implement in practise. Hence, the initial realization
(shown in the figure) is preferable.

Problem 6.3 :

(a) The closed loop transfer function is :

G(s)/s G(s) 1
H(s) = = = 2 √
1 + G(s)/s s + G(s) s + 2s + 1
The poles of the system are the roots of the denominator, that is
√ √
− 2± 2−4 1 1
ρ1,2 = = −√ ± j √
2 2 2
Since the real part of the roots is negative, the poles lie in the left half plane and therefore, the
system is stable.

(b) Writing the denominator in the form :

D = s2 + 2ζωn s + ωn2

we identify the natural frequency of the loop as ωn = 1 and the damping factor as ζ = √1
2

Problem 6.4 :

(a) The closed loop transfer function is :

K
G(s)/s G(s) K τ1
H(s) = = = 2
= 2 1 K
1 + G(s)/s s + G(s) τ1 s + s + K s + τ1
s + τ1

The gain of the system at f = 0 is :

|H(0)| = |H(s)|s=0 = 1

123
(b) The poles of the system are the roots of the denominator, that is
√
−1 ± 1 − 4Kτ1
ρ1,2 =
2τ1
In order for the system to be stable the real part of the poles must be negative. Since K
is greater than zero, the latter implies that τ1 is positive. If in addition we require that the
damping factor ζ = 2√1τ1 K is less than 1, then the gain K should satisfy the condition :
1
K>
4τ1

Problem 6.5 :

The transfer function of the RC circuit is :

1
R2 + Cs 1 + R2 Cs 1 + τ2 s
G(s) = 1 = =
R1 + R2 + Cs 1 + (R1 + R2 )Cs 1 + τ1 s
From the last equality we identify the time constants as :
τ2 = R2 C, τ1 = (R1 + R2 )C

Problem 6.6 :

Assuming that the input resistance of the operational amplifier is high so that no current flows
through it, then the voltage-current equations of the circuit are :
V2 = −AV1
 
1
V1 − V2 = R1 + i
Cs
V1 − V0 = iR
where, V1 , V2 is the input and output voltage of the amplifier respectively, and V0 is the signal
at the input of the filter. Eliminating i and V1 , we obtain :
1
R1 + Cs
V2 R
= 1
R1 + Cs
V1 1+ 1
−
A AR

If we let A → ∞ (ideal amplifier), then :

V2 1 + R1 Cs 1 + τ2 s
= =
V1 RCs τ1 s

124
Hence, the constants τ1 , τ2 of the active filter are given by :
τ1 = RC, τ2 = R1 C

Problem 6.7 :

In the non decision-directed timing recovery method we maximize the function :


2
ΛL (τ ) = ym (τ )
m

with respect to τ . Thus, we obtain the condition :

dΛL(τ )  dym (τ )
=2 ym (τ ) =0
dτ m dτ
Suppose now that we approximate the derivative of the log-likelihood ΛL (τ ) by the finite differ-
ence :
dΛL(τ ) ΛL (τ + δ) − ΛL (τ − δ)
≈
dτ 2δ
Then, if we substitute the expression of ΛL (τ ) in the previous approximation, we obtain :

2
2
dΛL (τ ) m ym (τ + δ) − m ym (τ − δ)
=
dτ  2δ
 2  2
1 
= r(t)g(t − mT − τ − δ)dt − r(t)g(t − mT − τ + δ)dt
2δ m
where g(−t) is the impulse response of the matched filter in the receiver. However, this is the
expression of the early-late gate synchronizer, where the lowpass filter has been substituted by
the summation operator. Thus, the early-late gate synchronizer is a close approximation to the
timing recovery system.

Problem 6.8 :

An on-off keying signal is represented as :

s1 (t) = A cos(2πfc t + φc ), 0 ≤ t ≤ T (binary 1)
s2 (t) = 0, 0 ≤ t ≤ T (binary 0)
Let r(t) be the received signal, that is r(t) = s(t; φc ) + n(t) where s(t; φc ) is either s1 (t) or
s2 (t) and n(t) is white Gaussian noise with variance N20 . The likelihood function, that is to be
maximized with respect to φc over the inteval [0, T ], is proportional to :

2 T
Λ(φc ) = exp − [r(t) − s(t; φc )]2 dt
N0 0

125
Maximization of Λ(φc ) is equivalent to the maximization of the log-likelihood function :
2 T
ΛL (φc ) = − [r(t) − s(t; φc )]2 dt
N0 0
2 T 4 T 2 T
= − r 2 (t)dt + r(t)s(t; φc )dt − s2 (t; φc )dt
N0 0 N0 0 N0 0

Since the first term does not involve the parameter of interest φc and the last term is simply a
constant equal to the signal energy of the signal over [0, T ] which is independent of the carrier
phase, we can carry the maximization over the function :
T
V (φc ) = r(t)s(t; φc )dt
0

Note that s(t; φc ) can take two different values, s1 (t) and s2 (t), depending on the transmission
of a binary 1 or 0. Thus, a more appropriate function to maximize is the average log-likelihood
1 T 1 T
V̄ (φc ) = r(t)s1 (t)dt + r(t)s2 (t)dt
2 0 2 0

Since s2 (t) = 0, the function V̄ (φc ) takes the form :

1 T
V̄ (φc ) = r(t)A cos(2πfc t + φc )dt
2 0

Setting the derivative of V̄ (φc ) with respect to φc equal to zero, we obtain :

ϑV̄ (φc ) 1 T
=0 = r(t)A sin(2πfc t + φc )dt
ϑφc 2 0
1 T 1 T
= cos φc r(t)A sin(2πfc t)dt + sin φc r(t)A cos(2πfc t)dt
2 0 2 0

Thus, the maximum likelihood estimate of the carrier phase is :

T
r(t) sin(2πfc t)dt
φ̂c,M L = − arctan  0T
0 r(t) cos(2πfc t)dt

Problem 6.9 :

(a) The wavelength λ is :

3 × 108 3
λ= 9
m = m
10 10
Hence, the Doppler frequency shift is :
u 100 Km/hr 100 × 103 × 10
fD = ± =± 3 = ± Hz = ±92.5926 Hz
λ 10
m 3 × 3600

126
The plus sign holds when the vehicle travels towards the transmitter whereas the minus sign
holds when the vehicle moves away from the transmitter.

(b) The maximum difference in the Doppler frequency shift, when the vehicle travels at speed
100 km/hr and f = 1 GHz, is :

∆fD max = 2fD = 185.1852 Hz

This should be the bandwith of the Doppler frequency tracking loop.

(c) The maximum Doppler frequency shift is obtained when f = 1 GHz + 1 MHz and the
vehicle moves towards the transmitter. In this case :
3 × 108
λmin = m = 0.2997 m
109 + 106
and therefore :
100 × 103
fD max = = 92.6853 Hz
0.2997 × 3600
Thus, the Doppler frequency spread is Bd = 2fD max = 185.3706 Hz.

Problem 6.10 :

The maximum likelihood phase estimate given by (6-2-38) is :



K−1 ∗
−1
Im n=0 In yn
φ̂M L = − tan 

K−1 ∗

Re n=0 In yn

 (n+1)T
where yn = nT r(t)g ∗(t − nT )dt. The Re(yn ), Im(yn ) are statistically independent compo-


nents of yn . Since r(t) = e−jφ n In g(t − nT ) + z(t) it follows that yn = In e−jφ + zn , where the
pulse energy is normalized to unity. Then :


K−1 
K−1 
In∗ yn = |In |2 e−jφ + In∗ zn
n=0 n=0

Hence : K−1
   2
E Im |In |2 e−jφ + In∗ zn = −K I¯n sin φ
n=0

and K−1
   2
E Re |In |2 e−jφ + In∗ zn = −K I¯n cos φ
n=0

127
  
Consequently : E φ̂M L = − tan −1 −cos
sin φ
φ
= φ, and hence, φ̂M L is an unbiased estimate of the
true phase φ.

Problem 6.11 :

The procedure that is used in Sec. 5-2-7 to derive the pdf p(Θr ) for the phase of a PSK signal
may be used to determine the pdf p(φ̂M L ). Specifically, we have :


K−1 ∗
−1
Im I y
n=0 n n
φ̂M L = − tan 

K−1 ∗

Re n=0 In yn

 (n+1)T

where yn = nT r(t)g ∗(t − nT )dt and r(t) = e−jφ n In g(t − nT ) + z(t). Substitution of r(t)
into yn yields : yn = In e−jφ + zn . Hence :


K−1 
K−1 
K−1
In∗ yn = e−jφ |In |2 + In∗ zn
n=0 n=0 n=0

U + jV = Ce−jφ + z = C cos φ + x + j(y − C sin φ)

2

K−1 ∗
n=0 |In | and z =
where C = K−1 n=0 In zn = x + jy. The random variables (x,y) are zero-mean,
Gaussian random variables with variances σ 2 .Hence :
1 −[(U −C cos φ)2 −(V −C sin φ)2 ]
p(U, V ) = e
2πσ 2
√
By defining R = U 2 + V 2 and φ̂M L = tan −1 VU and making the change in variables, we obtain

p(R, φ̂M L ) and finally, p(φ̂M L ) = 0∞ p(r, φ̂M L )dr. Upon performing the integration over R, as in
Sec. 5-2-7, we obtain :
1 −2γ sin 2 φ̂M L ∞ √ 2
p(φ̂M L ) = e re−(r− 4γ cos φ̂M L ) /2
dr
2π 0

where γ = C 2 /2σ 2 . The graph of p(φ̂M L ) is identical to that given on page 271, Fig. 5-2-9. We
observe that E(φ̂M L ) = φ, so that the estimate is unbiased.

Problem 6.12 :

We begin with the log-likelihood function given in (6-2-35), namely :

  
1
ΛL(φ) = Re r(t)s∗l (t)dt jφ
e
N0 T0

128



where sl (t) is given as : sl (t) = n In g(t − nT ) + j n Jn u(t − nT − T /2). Again we define
 (n+1)T  (n+3/2)T
yn = nT r(t)s∗l (t − nT )dt. Also, let : xn = (n+1/2)T r(t)s∗l (t − nT − T /2)dt. Then :
 jφ


K−1 ∗ 
K−1 ∗
ΛL (φ) = Re eN0 n=0 In yn − j n=0J xn n
= Re [A cos φ + jA sin φ]

K−1 ∗
K−1 ∗
where A = I y
n=0 n n −j n=0J x n n. Thus : ΛL (φ) = Re(A) cos φ − Im(A) sin φ and :
dΛL(φ)
= −Re(A) sin φ − Im(A) cos φ = 0 ⇒
dφ


K−1 ∗ 
K−1 ∗
−1
Im n=0 In yn −j n=0 Jn xn
φ̂M L = − tan 


Re K−1 ∗
n=0 In yn −j K−1 ∗
n=0 Jn xn

Problem 6.13 :

Assume that the signal um (t) is the input to the Costas loop. Then um (t) is multiplied by
cos(2πfc t + φ̂) and sin(2πfc t + φ̂), where cos(2πfc t + φ̂) is the output of the VCO. Hence :
umc (t)
= Am gT (t) cos(2πfc t) cos(2πfc t + φ̂) − Am ĝT (t) sin(2πfc t) cos(2πfc t + φ̂)
Am gT (t)   Am ĝT (t)  
= cos(2π2fc t + φ̂) + cos(φ̂) − sin(2π2fc t + φ̂) − sin(φ̂)
2 2
ums (t)
= Am gT (t) cos(2πfc t) sin(2πfc t + φ̂) − Am ĝT (t) sin(2πfc t) sin(2πfc t + φ̂)
Am gT (t)   Am ĝT (t)  
= sin(2π2fc t + φ̂) + sin(φ̂) − cos(φ̂) − cos(2π2fc t + φ̂)
2 2
The lowpass filters of the Costas loop will reject the double frequency components, so that :
Am gT (t) Am ĝT (t)
ymc (t) = cos(φ̂) + sin(φ̂)
2 2
Am gT (t) Am ĝT (t)
yms (t) = sin(φ̂) − cos(φ̂)
2 2
Note that when the carrier phase has been extracted correctly, φ̂ = 0 and therefore :
Am gT (t) Am ĝT (t)
ymc (t) = , yms (t) = −
2 2
If the second signal, yms (t) is passed through a Hilbert transformer, then :

Am ĝˆT (t) Am gT (t)

ŷms (t) = − =
2 2

129
and by adding this signal to ymc (t) we obtain the original unmodulated signal.

Problem 6.14 :

(a) The signal r(t) can be written as :

 
r(t) = ± 2Ps cos(2πfc t + φ) + 2Pc sin(2πfc t + φ)
  
 Ps
−1
= 2(Pc + Ps ) sin 2πfc t + φ + an tan
Pc
   
−1 Pc
= 2PT sin 2πfc t + φ + an cos
PT

where an = ±1 are the information symbols and PT is the total transmitted power. As it is
observed the signal has the form of a PM signal where :
 
−1 Pc
θn = an cos
PT

Any method used to extract the carrier phase from the received signal can be employed at the
receiver. The following figure shows the structure of a receiver that employs a decision-feedback
PLL. The operation of the PLL is described in the next part.

t = Tb
v(t) ✎  
✲× ✲ Tb (·)dt  ✲ Threshold ✲
✍✌ 0
✻
✲ DFPLL
cos(2πfc t + φ̂)

(b) At the receiver (DFPLL) the signal is demodulated by crosscorrelating the received signal :
   
−1 Pc
r(t) = 2PT sin 2πfc t + φ + an cos + n(t)
PT

with cos(2πfc t + φ̂) and sin(2πfc t + φ̂). The sampled values at the ouput of the correlators are :
 
1  1
r1 = 2PT − ns (t) sin(φ − φ̂ + θn ) + nc (t) cos(φ − φ̂ + θn )
2 2
 
1  1
r2 = 2PT − ns (t) cos(φ − φ̂ + θn ) + nc (t) sin(φ̂ − φ − θn )
2 2

130
where nc (t), ns (t) are the in-phase and quadrature components of the noise n(t). If the detector
has made the correct decision on the transmitted point, then by multiplying r1 by cos(θn ) and
r2 by sin(θn ) and subtracting the results, we obtain (after ignoring the noise) :
1  
r1 cos(θn ) = 2PT sin(φ − φ̂) cos2 (θn ) + cos(φ − φ̂) sin(θn ) cos(θn )
2
1  
r2 sin(θn ) = 2PT cos(φ − φ̂) cos(θn ) sin(θn ) − sin(φ − φ̂) sin2 (θn )
2
1
e(t) = r1 cos(θn ) − r2 sin(θn ) = 2PT sin(φ − φ̂)
2
The error e(t) is passed to the loop filter of the DFPLL that drives the VCO. As it is seen only
the phase θn is used to estimate the carrier phase.

(c) Having a correct carrier phase estimate, the output of the lowpass filter sampled at t = Tb
is :
 
1 Pc
r = ± 2PT sin cos−1 +n
2 PT

1 Pc
= ± 2PT 1 − +n
2 PT
  
1 Pc
= ± 2PT 1 − +n
2 PT
where n is a zero-mean Gaussian random variable with variance :

Tb Tb
σn2 = E n(t)n(τ ) cos(2πfc t + φ) cos(2πfc τ + φ)dtdτ
0 0
N0 Tb
= cos2 (2πfc t + φ)dt
2 0
N0
=
4
Note that Tb has been normalized to 1 since the problem has been stated in terms of the power
of the involved signals. The probability of error is given by :
  
2PT Pc
P (error) = Q 1−
N0 PT
The loss due to the allocation of power to the pilot signal is :
 
Pc
SNRloss = 10 log10 1 −
PT
When Pc /PT = 0.1, then SNRloss = 10 log10 (0.9) = −0.4576 dB. The negative sign indicates
that the SNR is decreased by 0.4576 dB.

131
Problem 6.15 :

The received signal-plus-noise vector at the output of the matched filter may be represented as
(see (5-2-63) for example) : 
rn = Es ej(θn −φ) + Nn
where θn = 0, π/2, π, 3π/2 for QPSK, and φ is the carrier phase. By raising rn to the fourth
power and neglecting all products of noise terms, we obtain :
√ 4 √ 3
rn4 ≈ Es ej4(θn −φ) + 4 Es Nn
√ 3 √ 
≈ Es Es e−j4φ + 4Nn

averaging the received vectors {rn4 } over K signal intervals, we have

If the estimate is formed by√


the resultant vector U = K Es e−jφ + 4 K n=1 Nn . Let φ4 ≡ 4φ. Then, the estimate of φ4 is :

Im(U)
φ̂4 = − tan −1
Re(U)
Nn is a complex-valued Gaussian noise component with zero mean and variance σ 2 = N0 /2.
Hence, the pdf of φ̂4 is given by (5-2-55) where :
 √ 2
K Es K 2 Es KEs
γs = 2
= =
16 (2Kσ ) 16KN0 16N0
To a first approximation, the variance of the estimate is :
1 16
σφ̂24 ≈ =
γs KEs /N0

Problem 6.16 :

The PDF of the carrier phase error φe , is given by :

φ2
1 − e2
p(φe ) = √
2σ
e φ
2πσφ
Thus the average probability of error is :
∞
P̄2 = P2 (φe )p(φe )dφe
−∞

∞ 2Eb
= Q cos2 φe p(φe )dφe
−∞ N0
  
∞ ∞ 2
1 
1 2 φ
= 2Eb
exp − e
x + 2 dxdφe
2πσφ −∞ N0
cos2 φe 2 σφ

132
Problem 6.17:

The log-likelihood function of the symbol timing may be expressed in terms of the equivalent
low-pass signals as
  
ΛL (τ ) =  N10 T0 r(t)sl ∗ (t; τ )dt
 

=  N10 T0 r(t) n In ∗ g ∗ (t − nT − τ )dt


1 ∗
=  N0 n In yn (τ )

where yn (τ ) = T0 r(t)g ∗(t − nT − τ )dt.
A necessary condition for τ̂ to be the ML estimate of τ is
dΛL (τ )
= 0 ⇒
d


τ
∗
[ n In y n (τ ) + n In yn ∗ (τ )] = 0 ⇒


dτ
∗ d
d ∗
n In dτ yn (τ ) + I y
n n dτ n (τ ) = 0

If we express yn (τ ) into its real and imaginary parts : yn (τ ) = an (τ ) + jbn (τ ), the above
expression simplifies to the following condition for the ML estimate of the timing τ̂
 d  d
[In ] an (τ ) + [In ] bn (τ ) = 0
n dτ n dτ

Problem 6.18:

We follow the exact same steps of the derivation found in Sec. 6.4. For a PAM signal In ∗ = In
and Jn = 0. Since the pulse g(t) is real, it follows that B(τ ) in expression (6.4-6) is zero,
therefore (6.4-7) can be rewritten as

ΛL (φ, τ ) = A(τ ) cos φ

where
1 
A(τ ) = In yn (τ )
N0
Then the necessary conditions for the estimates of φ and τ to be the ML estimates (6.4-8) and
(6.4-9) give

φ̂M L = 0
and
 d
In [yn (τ )]τ =τ̂M L = 0
n dτ

133
Problem 6.19:

The derivation for the ML estimates of φ and τ for an offset QPSK signal follow the derivation
found in Sec. 6.4, with the added simplification that, since w(t) = g(t − T /2), we have that
xn (τ ) = yn (τ + T /2).

134