Chapter 3

Representation of Bandpass Signals

Let x t
be a signal with Fourier Transform X f
. We say that x t
is narrowband about f c of bandwidth W if
fc  W and
X f
 0  f  fc 
W  f  fc  W 
 X f



 
 
 
 
 
 
 
 f
 fc  W  fc  fc  W fc  W fc fc  W

Let gLP t
be the impulse response of an ideal low pass filter


1  W  f   W
GLP f

0  elsewhere.

GLP f

 f

Let  W
∞
W
xc t

 ∞ x t  u
 2 cos 2π fc t  u

gLP u
du

and ∞
xs t
   x t  u
!  2 sin 2π fc t  u

#" gLP u
du
∞
Claim:
x t
 xc t
 2 cos 2π fct
$ xs t
 2 sin 2π fct

Proof: We will show that the following system is an ideal bandpass filter

3-1
3-2 CHAPTER 3. REPRESENTATION OF BANDPASS SIGNALS

 2 cos 2π fct
 2 cos 2π fct

xc t

 

 
   



 GLP 

x t
y t



 




 xs t

   



 GLP 

 

  2 sin 2π fct
  2 sin 2π fct

Figure 3.1: System for obtaining representations of bandpass signals
 3-3

Let the input to the system be an impulse at time τ. The output is then
∞
y t
 2 cos 2π fct
  δ u  τ
cos 2π fc u
gLP t  u
du
∞
∞
 2 sin 2π fct
  δ u  τ
sin 2π fc u
gLP t  u
du
∞
 2gLP t  τ
cos 2π fct
cos 2π fc τ
 2gLP t  τ
sin 2π fct
sin 2π fc τ

y t
 2gLP t  τ
cos 2π fc t  τ

Since the output at time t due to an impulse at time τ depends only on t  τ the above system is time-invariant
with impulse response
h t
 2gLP t
cos 2π fct

The Fourier transform of the impulse response is

H f
  ∞
  h t
e j2π f t
dt
∞

  2gLP t
cos 2π fct
e  j2π f t dt

∞

∞
∞ 
e j2π fct  e j2π fct  j2π f t
   2gLP t


∞ 2  e dt
  

 e j2π  f fc  t  e j2π  f  fc  t  dt
∞
   gLP t 
∞

  gLP t
e  j2π  f  fc  t dt    gLP t
e  j2π  f  fc  t dt

∞ ∞

∞ ∞
 GLP f  fc
 GLP f  fc

 W  f  fc  W
 1


or
 W  f  fc  W

 0 elsewhere
 1 fc  W  f  fc  W
 1  fc  W  f   fc  W
0 elsewhere 

So the above system is bandpass. Hence for signals which are narrowband we can write

x t
 xc t
 2 cos 2π fct
 xs t
 2 sin 2π fct


xL t
 xc t
 jxs t

x t
 Re  xL t
 2e j2π fct " 
The envelope of the bandpass waveform is

xe t
 x2c t
 x2s t


The phase of the bandpass waveform is
  xs t

θ t
 tan 1
xc t


x t
 xe t
 2 cos 2π fct  θ t


3-4 CHAPTER 3. REPRESENTATION OF BANDPASS SIGNALS

x(t)
0

-2
0 0.5 1 1.5 2 2.5 3 3.5 4
time
2
|X(f)|

0
-50 -40 -30 -20 -10 0 10 20 30 40 50
frequency

2
Phase

-2

-4
-50 -40 -30 -20 -10 0 10 20 30 40 50
frequency

Figure 3.2: Bandpass signal representation

Also
1
Xc f
  X f  fc
 X f  fc
" GLP f

 2
1
Xs f
  X f  fc
 X f  fc
" GLP f

j 2

Below we illustrate this representation for a simple example. Figure 1.2 shows a signal that is bandlimited
to the frequency range 30-34Hz. The top figure is the time domain representation while the bottom two are the
magnitude and phase in the frequency domain. Figure 1.3 shows the signals after mixing with the sinusoids. It is
clear that there is a baseband term and a double frequency term. Figure 1.4 shows the signal after low pass filtering.
We now have a complex signal. Finally Figure 1.5 shows the reconstructed signal along with the envelope.
Now consider the signal
x̂ t
 xc t
 2 sin 2π fct
 xs t
 2 cos 2π fct

The spectrum of this signal is given by
1 1
X̂ f
  Xc f  fc
 Xc f  fc
#"   Xs f
 fc
 Xs f  fc
"
j 2  2
1
  X f  2 fc
 X f
" GLP f  fc
  X f
 X f  2 fc
" GLP f  fc


2j
1
  X f
 X f  2 fc
" GLP f  fc
  X f  2 fc
 X f
" GLP f  fc


2j
1
 X f
GLP f  fc
 X f
GLP f  fc


j
 3-5

2 2

1 1

xc1(t)

xs1(t)
0 0

-1 -1

-2 -2
0 0.5 1 1.5 2 0 0.5 1 1.5 2
time time

2 2

1.5 1.5
|Xc1(f)|

|Xs1(f)|
1 1

0.5 0.5

0 0
-100 -50 0 50 100 -100 -50 0 50 100
frequency frequency

Figure 3.3: Bandpass signal representation

1 1

0.5 0.5
xc(t)

xs(t)

0 0

-0.5 -0.5

-1 -1
0 1 2 3 4 0 1 2 3 4
time time

2 2

1.5 1.5
|Xc(f)|

|Xs(f)|

1 1

0.5 0.5

0 0
-100 -50 0 50 100 -100 -50 0 50 100
frequency frequency

Figure 3.4: Bandpass signal representation

3-6 CHAPTER 3. REPRESENTATION OF BANDPASS SIGNALS

1.5

0.5

-0.5

-1

-1.5

-2
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

Figure 3.5: Bandpass signal representation

Essentially the signal x̂ has the negative frequencies multiplied by  j (shifted by π 2) and the positive frequencies
multiplied by  j (shifted by  π 2). This is called the Hilbert transform of the original signal x.

Now consider the signal x t
 jx̂ t
. This signal has spectrum

X f
 jX̂ f
 X f
GLP f  fc
 X f
GLP f  fc

 X f
GLP f  fc
 X f
GLP f  fc

 2X f
GLP f  fc

Thus the above signal has only positive frequencies. The signal x̂ t
is useful when considering single-sideband
systems.
 3-7


X f

 
 
 
 
 
 
 
  f
 fc  W  fc  fc  W Xc f
fc  W fc fc  W

 f

 W W
 Xs f

W 
 W f

Figure 3.6: Alternative representation of lowpass equivalent signals using Hilbert transforms.
3-8 CHAPTER 3. REPRESENTATION OF BANDPASS SIGNALS

cos 2π fct

 cos 2π fc t

 



xc t

 


   




 

 HT


sin 2π fct

y t



 


x t

 
cos 2π fct


 

 HT


xs t




   



  


 


sin 2π fct

sin 2π fct

 3-9

∞
xc t
   x t  u
 2 cos 2π fc t  u

gLP u
du
∞
∞
xs t
   x t  u
   2 sin 2π fc t  u

#" gLP u
du
∞
x t
 xc t
 2 cos 2π fc t
 xs t
 2 sin 2π fct

xL t
 xc t
 jxs t

x t
 Re  xL t
 2e j2π fct "
xe t
 x2c t

 x2s t

 xs t

θ t
 tan 1 
xc t

x t
 xe t
cos 2π fct  θ t


Xc f
  X f  fc
 X f  fc
#" GLP f

1
Xs f
 X f  fc
 X f  fc
GLP f
"
j
3-10 CHAPTER 3. REPRESENTATION OF BANDPASS SIGNALS

Representation of Narrowband Random Processes

The representation of narrowband signals can also be applied to narrowband stationary processes. A stationary
random process n t
is said to be narrowband if the power spectral density S n f
satisfies

Sn f
 0  f  fc 
W  f  fc 
W 
It is clear then that we can represent n t
as

n t
 nc t
cos 2π fct
 ns t
sin 2π fct

where ∞
nc t
   2 n u
cos 2π fc u
gLP t  u
du
∞
and ∞
  2 n u
sin 2π fc u
gLP t  u
du 
ns t

∞

Let n t
have zero mean and autocorrelation function (also covariance) Kn s  t
 Kn t  s
.
We define the complex random processes z t
and v t
as

z t
 nc t
 jns t

and

v t
 2 n t
cos 2π fc t
 j2 n t
sin 2π fct

 2e
 j2π f ct
n t

Clearly
∞
z t
   gLP t  u
v u
du
∞

Sz f
  GLP f
 2 Sv f

Claim:
Sz f
 4Sn f  fc
GLP f

Proof:

Kv s  t
 E  v s
v t
"
 4e
    E  n s
n t
"
j2π f c s t

  
4e j2π fc  s t  Kn t  s
 4e j2π fc τ Kn τ

where τ  t  s.
The power spectral density of v t
is
∞ 
Sv f
   Kv τ
e j2π f τ
dτ
∞
∞ 
 4 j2π f c τ
Kn τ
e j2π f τ dτ
 ∞e
∞  j2π  f  fc  τ
 4
 Kn τ
e∞
dτ  4Sn f  fc

Now it is clear that

Sz f
 4Sn f  fc
 GLP f
 2
 4Sn f  fc
GLP f

Note that Sz f
 0  f  W .
 3-11

Hence
∞
Kz τ
   4Sn f  fc
GLP f
e j2π f τ d f
∞
W
   4Sn f  fc
e j2π f τ d f
W

Claim:
E  z s
z t
"$ 0 s t
Proof:
∞ ∞
E  z s
z t
#"  E  gLP s  u1
v u1
du1 
 ∞  ∞ gLP t  u2
v u2
du2
∞ ∞
    gLP s  u1
gLP t  u2
E  v u1
v u2
" du1 du2
∞  ∞

    gLP s  u1
gLP t  u2
E  4e jωc  u1  u2  n u1
n u2
#" du1 du2
∞ ∞


∞ ∞

    gLP s  u1
gLP t  u2
4e  jωc  u1  u2  E  n u1
n u2
" du1 du2

∞ ∞

∞ ∞

4e jωc  u1  u2  ∞ 
  Sn ω
e jω  u2 u1  dωdu1 du2
∞ ∞
     gLP s  u1
gLP t  u2

∞ ∞ 2π ∞

  Sn ω
  gLP s  u1
e  ju1  ωc  ω du1   gLP t  u2
e  ju2  ωc  ω du2 dω

2 ∞ ∞ ∞

π ∞ ∞ ∞

Now

  gLP s  u1
e     du1   ∞ gLP u1
e  j  s  u1   ωc  ω du1

∞
ju1 ωc ω
∞  ∞

∞ 
  gLP u1
e ju1  ωc  ω du1 e js  ωc  ω
∞
 
GLP  ω  ωc
e js  ωc  ω  GLP ω  ωc
e js  ωc  ω

Similarly

  gLP t  u2
e     du2    gLP u2
e  j  t      du2

∞ ∞
ju2 ωc ω u2 ωc ω
∞ ∞

  gLP u2
e ju2  ωc  ω du2 e  j  uc  ω

∞

∞

GLP ωc  ω
e jt  ωc ω 


 
 GLP ω  ωc
e jt  ωc ω 

So
2 ∞
E  z t
z s
"   ∞ Sn ω
GLP ω  ωc
GLP ω  ωc

exp  js ωc  ω
 jt ωc  ω

dω
π
But GLP ω  ωc
GLP ω  ωc
 0 ω
Thus
E  z s
z t
"$ 0 s t
Using the above result we can obtain relations between the correlation functions for n c t
and ns t
as follows.

0  E  z s
z t
"  E  nc s
 jns s

nc t
 jns t

#"

 Knc s  t
 Kns s  t
 j Knc ns s  t
 Kns nc s  t


 
3-12 CHAPTER 3. REPRESENTATION OF BANDPASS SIGNALS

Since the real and imaginary parts each must be zero we have that

Knc s  t
 Kns s  t

Knc ns s  t

  Kns nc s  t



Kz s  t
 E  z s
z t
#"  E  nc s
 jns s

nc t
 jns t

"

 Knc s  t
 Kns s  t
 j Kns nc s  t
$ Knc ns s  t

 

 2Knc s  t
 j 2Kns nc s  t



From the above the below facts become obvious.

Facts:

1
Knc τ
 Re  Kz τ
#"
2
1
Kns τ
 Re  Kz τ
#"
2
1
Knc ns τ
 E  nc t
ns t  τ
#"  Im Kz τ



2
Kns nc τ

  Kns nc τ


1
S nc f
 F  Re  Kz f
" "
2
1
 F  Kz τ
 Kz τ
#"!"
4
1
  Sz f
 Sz  f
#"
4
  Sn f  f c
 Sn  f fc
#" GLP f

  Sn f  fc
 Sn f  fc
" GLP f

1
S ns nc f
 F  Im  Kz f
" "


2
  Sn f  f c
 Sn  f  fc
#" GLP f

S nc ns f


  S ns nc f


In the Karhuenen-Loève Expansion of a complex process X t

b
Zj   X t
ϕ j t
dt
a

We showed that
E  Z j"  0 j

E  Z j Zk "  δ jk λ j
Now if X t
is a complex process determined from a narrowband process then E  X s
X t
"  0 for all s  t so that
b b
E  Z j Zk "  E  X s
ϕ j s
ds  X t
ϕk t
dt
a a 
b b
   ϕ j s
ϕk t
E  X s
X t
" dsdt
a a
 0 j  k including j  k

1. 3-13

Let Zk  Xk  jYk j    1 now

E  Zk2 "  0 0  E  Xk  jYk
2
"
 E Xk2 "  E  "  2 jE  XkYk "
Yk2

Thus E  Xk2 "  E  Yk2 " and E  XkYk "  0

E   Zk  2 "  λk E   Zk  2 "  E  Xk2 "  E  Yk2 "  λk
E  Xk2 "  λk 2
E  Yk2 "  λk 2
The density of Zk is then

fzk zk

1  x2k
2 λ2k 1  y2k
2 λ2k
 e e
 2π λk
2  2π λk
2
1  zk  2

 e λk
πλk
To determine the joint density function we need to determine the correlations between the real and imaginary
parts of Zk and Zl . First observe the following identities.
If u  uR  juI and w  wR  jwI then
1
uR wR  Re  uw  uw #"
2
and
1
uR wI  Im  uw  uw "
2
Consider
1
E  Re  Zk " Re  Zl " "  ReE  Zk Zl  Zk Zl "
2
For k   l both of these terms are zero. Now consider
1
E  Re  Zk " Im  Zl "!"  ImE  Zk Zl  Zk Zl "
2
For k   l both terms on the right are zero as shown above while for k  l the first term is zero and the second term
is purely real so the right hand side is zero. Thus we have independence among the Z’s.
If the original (bandpass) process is (essentially) white with power spectrum N0 2 then the complex (lowpass)
representation will be (essentially) white with power spectrum N0 . Thus in the K-L representation Zl will be
(complex) Gaussian with mean zero and E   Zl  2 "  N0 . (λl  N0 ). If the process is (real) white noise then the K-L
representation will have E   Zl  2 "  N0 2 

1. Problems

X1
1. Let X = X2  be jointly Gaussian with zero mean and covariance matrix

σ2 σ2
K  σ2 σ2 

Y1
Express X as WY where W is a 2 by 2 orthogonal matrix and Y = Y2  where Y1 , and Y2 are independent
Gaussian random variables.
3-14 CHAPTER 3. REPRESENTATION OF BANDPASS SIGNALS

2. Let X  X1  jX2 , and Y  X3  jX4 where X1  X2  X3  X4
is a jointly Gaussian random vector with mean 0
and covariance
1 0 0  1

K

 0
0 
1
1
 1
1
0
0 
 1 0 0 1
Show that X and Y are uncorrelated but not independent.


3. Show that K s  t
= min s  t
s  t 0 is a nonnegative definite. (Hint: write min s  t

Ix y
 1 if y  x and is 0 otherwise. Then interchange integrals)


 ∞
0 Is u
It u
du where

4. Prove P

M A

i 1 i
 ∑Mi 1 P Ai
.
5. Let N1    Nn be independent identically distributed (i.i.d.) Gaussian random variables with mean zero and
variance N0 2. Let N̂i  Ni  n. Show that for any ∆ 0

  ∑  
n
lim P N̂i2  N0 2 ∆  0
n ∞


i 1

6. Let H0 and H1 be two events. Let pi x1  
xn be the conditional density of X1 
 Xn given Hi occurred.
(a) Assume that given Hi ,
X j
nj 1 is a sequence of i.i.d. Gaussian random variables with mean  1
i  E
and variance N0 2. Use the Chernoff bound to show that

P p1 X
 p0 X
 H0
 e nE N 0

(b) Now let X1     Xn be independent discrete random variables taking values +1 and  1 with
P Xi   1  H0
 p
P Xi   1  H0
 1  p
P Xi   1  H1
 p
P Xi   1  H1
 1  p

Thus if the number of components of x equal to +1 is d then p 0 x
 pd 1  p
n  d and p1 x
 pn  d 1  p
d .

Use the Chernoff bound to show that

P p1 X
 p0 X
 H0

 
 e n ln    
4p 1 p

(c) Again let X1   Xn be independent discrete random variables with Xi taking nonnegative integer values
only. Let

p0 x1    xn

i 1
λxi e λ0

∏ 0 xi !
n
 xi  0 1  i  n

and
n 
λxi e λ1
p 1 x1     xn
 ∏ 1
i 1 xi !
If λ1 λ0 find the best Chernoff bound on
Pe 0 
 P p1 X
 p0 X
 H0


and
Pe 1
 P p0 X p1 X
 H1

Let si be the optimal value of s for minimizing the bound to Pe i . Show s1 
 1  s0 .
1. 3-15

(d) Again let X1     X2n be independent discrete random variables with Xi taking nonnegative integer values
only. Let
 x 
p0 x1 
n
 x2n
 ∏ 1
i 1
λxi e λ1 2n λ0i e λ0
∏
xi ! i n  1 xi !
 xi
 0 1  i  2n

and
n λxi e  λ0 2n λxi e  λ1
p 1 x1     x2n
 ∏ 0
xi ! i ∏
1
i 1 n 1 xi !
If λ1 λ0 find the best Chernoff bound on
Pe 0 
 P p1 X
 p0 X
 H0


and
Pe 1  P p0 X p1 X
 H1


7. (a) Show that ∑∞ i 1 φi t
φi s
 δ t  s
for any complete orthonormal set of functions φi t
. ( Hint: Show
that the above function satisfies the definition of a delta function, that
∞
  f t
δ t  s
dt  f s

∞

for all functions f t
).

(b) Let K s  t
be a positive covariance function of the zero mean Gaussian random process X t
. Let K
 1
2
be the negative square root of this function as defined in class. Show that the process


Y s
  K 1 2
s  t
X t
dt

is a white Gaussian noise process. (You need to show that E  Y s
Y t
"  δ t  s
).

8. Let K s  t
be a real covariance matrix of a random process with eigenvalues λ i and (real) eigenfunctions φi .
Define (as in class) K 2 s  t
as
K 2 s  t
  K s  u
K u  t
du
and K n s  t
as
K n s  t
  Kn 1
s  u
K u  t
du 

Also define eK s  t
as
∞
Kn s  t

eK  s t   ∑



n 0 n!
Show that
∞
eK  s t 

 ∑ eλi φi s
φi t



i 1

9. A bandpass signal x1 t
 Re  xL 1 t
e j2π f 1t " has center frequency f1 and bandwidth W1 f1 . It is desired

f1 and generate the signal x2 t
 Re  xL 1 e j2π f2t " . The following



to mix the signal up to frequency f 2  

methods are available

(i) Multiply the signal x1 t
by cos 2π f2  f1
t
and filter out the signal at frequency f 2  2 f1 .
(ii) Generate the signal via x2 t
 x1 t
cos 2π f2  f1
t

 x̂1 t
sin 2π f2  f1
t
.
(iii) Generate the lowpass complex representation of x1 t
x1 c t
 x1 s t

. Then generate x2 t
$ x1 c t
cos 2π f2t

x1 s t
sin 2π f2t
.
  

Show that each method does what is intended.

3-16 CHAPTER 3. REPRESENTATION OF BANDPASS SIGNALS

10. Show that the complex lowpass equivalent representation of a bandpass signal can be generated by

xc t
 x t
cos 2π fct
 x̂ t
sin 2π fct

xs t
 x̂ t
cos 2π fct
 x t
sin 2π fct

11. Consider a bandpass signal of the form x t
 xc t
cos 2π fc t
 xs t
sin 2π fct
. It is desired to mix to
baseband to recover xc t
and xs t
. However, the local oscillators have a phase offset θ and so the local
signals are instead cos 2π f c t  θ
and sin 2π fc t  θ
. The signal is low pass filtered after mixing to baseband.
(a) Determine the signals out of the two low pass filters.
(b) If the phase is known to the receiver, determine how to recover the signals x c t
and xs t
from the filter
outputs.
12. (a) Simulate and compare to analysis the performance of QPSK modulation for E b N0  0  2  4  6  8 dB.
(b) Simulate the performance of QPSK modulation with a local oscillator with phase error θ  10 degrees
for Eb N0  0  2  4  6  8 dB. Can you analyze this system?
13. A bandpass signal is the input to a bandpass filter. The bandpass filter has impulse response as

h t
 hc t
cos 2π fct
 hs t
sin 2π fct

while the input signal has lowpass complex equivalent representation

x t
 xc t
cos 2π fct
 xs t
sin 2π fct


The output of the filter is

y t
 yc t
cos 2π fct
 ys t
sin 2π fct


Determine the relation between yc t
, ys t
and the input xc t
and xs t
and the filter hc t
and hs t
.