Lectures 6&7 Modulation
Eytan
Modiano
AA Dept.
Eytan Modiano
Slide 1
Modulation
?
Representing digital signals as analog waveforms
?
Baseband signals
–
Signals whose frequency components are concentrated around zero
?
Passband signals
–
Signals whose frequency components are centered at some frequency
fc away from zero
?
Baseband
signals can be converted to
passband
signals through
modulation
–
Multiplication by a sinusoid with frequency
f
c
Eytan Modiano
Slide 2
Baseband signals
?
The simplest signaling scheme is pulse amplitude modulation (PAM)
–
With binary PAM a pulse of amplitude A is used to represent a
“1” and a
pulse with amplitude -A to represent a
“
0
”
?
The simplest pulse is a rectangular pulse, but in practice other type of pulses are used
–
For our discussion we will generally assume a rectangular pulse
?
If we let g(t) be the basic pulse shape, than with PAM we transmit g(t) to represent a
“
1
”
and -g(t) to represent a
“
0
”
g(t)
1 => S(t) = g(t)
A
0 => S(t) = -g(t)
Tb
Eytan Modiano
Slide 3
M-
ary PAM
?
Use M signal levels, A
1
…A
M
–
Each level can be used to represent Log
2
(M) bits
?
E.g., M = 4 => A
1
= -3, A
2
= -1, A
3
= 1, A
4
= 3
–
S
i
(t) = A
i
g(t)
?
Mapping of bits to signals Si
b1b2
S
1
00
S
2
01
S
3
11
S
4
10
Eytan Modiano
Slide 4
gt
tT
Signal Energy
T
T
E
m
=
∫
0
((
t
)
)
2
d
t
=
(
A
m
)
2
∫
0
()
2
d
t
=
(
A
m
)
2
E
g
S
m
g
t
?
The signal energy depends on the amplitude
?
E
g
is the energy of the signal pulse g(t)
?
For rectangular pulse with energy E
g
=>
T
E
g
=
∫
0
A
2
d
t
=
TA
2
=
>
A
=
E
g
2
/
g(t)
?
E
g
/
2
0
≤≤
()
=
? ?
0
otherwise
E
g
/2
Eytan Modiano
T
Slide 5
Symmetric PAM
?
Signal amplitudes are equally distant and symmetric about zero
-7
-5
-3
-1
0
1
3
5
7
A
m
= (2m-1-M), m=1…M
E
M
E
ave
=
g
∑
(
2
m
?
1
?
M
)
2
=
E
g
(
M
2
?
1
)
/
3
M
m
=
1
Eytan Modiano
Slide 6
Gray Coding
?
Mechanism for assigning bits to symbols so that the number of bit errors is minimized
–
Most likely symbol errors are between adjacent levels
–
Want to MAP bits to symbols so that the number of bits that differ between adjacent levels is
mimimized
?
Gray coding achieves 1 bit difference between adjacent levels
?
Example M= 8 (can be generalized)
A
1
000
A
2
001
A
3
011
A
4
010
A
5
110
A
6
111
A
7
101
Eytan Modiano
A
8
100
Slide 7
Bandpass signals
?
To transmit a
baseband
signal S(t) through a
bandpass
channel
at some center frequency
f
c
, we multiply S(t) by a sinusoid with
that frequency
S
m
(t)
U
m
(t)= S
m
(t)Cos(2
π
f
c
t)
= A
m
g(t) Cos(2
π
f
c
t)
Cos(2
π
f
c
t)
F[Cos(2
π
f
c
t)] = (
δ
(f-
f
c
)+
δ
(f+
f
c
))/2
Eytan Modiano
-f
c
f
c
Slide 8
Passband signals,
cont.
F[A
m
g(t)] = depends on g()
A
m
-w
w
F[A
m
g(t) Cos(2
π
f
c
t)]
A
m
/2
A
m
/2
-f
c
f
c
Eytan Modiano
Slide 9
Recall: Multiplication in time = convolution in frequency
()
Energy content of modulated signals
∞
∞
E
m
=
∫
?∞
U
2
t
d
t
=
∫
?∞
A
m
g
2
()
m
()
2
t
Cos
2
(
2
π
f
c
t
)
dt
Cos
2
α
=
1
+
cos(
2
α
)
2
A
2
∞
A
2
∞
2
()
+
2
m
∫
?∞
gt
Cos
2
(
4
π
f
c
t
)
dt
E
m
=
2
m
∫
?∞
gt
2
()
A
2
m
E
=
E
+
≈
0
m
2
g
?
The cosine part is fast varying and integrates to 0
?
Modulated signal has 1/2 the
energy as the
baseband
signal
Eytan Modiano
Slide 1
0
Demodulation
?
How do we recover the
baseband
signal?
U
m
(t)= S
m
(t)Cos(2
π
f
c
t)
= A
m
g(t) Cos(2
π
f
c
t)
U(t)
LPF
S
r
(t)
2Cos(2
π
f
c
t)
U(t)2Cos(2
π
f
c
t)= 2S(t)Cos
2
(2
π
f
c
t) = S(t) + S(t)Cos(4
π
f
c
t)
The high frequency component is rejected by the LPF and we are left with S(t).
Eytan Modiano
Slide 1
1
jf
Bandwidth occupancy
G(f) = F[g(t)]
g(t)
A
∞
T
Gf
2
π
()
=
∫
?∞
g
(
t
)
e
?
jf
t
d
t
=
∫
0
A
e
?
j
2
π
f
t
d
t
Gf
()
=
(
AT
)
Sinc
(
π
fT
)
e
?
π
T
T
|G(f)|
AT
-2/T
-1/T
1/T
2
/
T
?
First
“null
” bandwidth = 2(1/T) = 2/T
Eytan Modiano
Slide 1
2
Bandwidth efficiency
?
Rs = symbol rate = 1/T
–
Log
2
(M) bits per symbol =>
R
b
= bit rate = log
2
(M)/T bits per second
?
BW = 2/T = 2Rs
–
Bandwidth efficiency =
R
b
/BW = log
2
(M)/T * (T/2) = log
2
(M)/2 BPS/Hz
?
Example:
–
M=2 => bandwidth efficiency = 1/2
–
M=4 => bandwidth efficiency = 1
–
M=8 => bandwidth efficiency = 3/2
?
Increased BW efficiency with increasing M
?
However, as M increase we are more prone to errors as symbols are closer together (for a given energy level)
–
Need to increase symbol energy level in order to overcome errors
–
Tradeoff between BW efficiency and energy efficiency
Eytan Modiano
Slide 1
3
Energy utilization
E
M
E
ave
=
g
∑
(
2
m
?
1
?
M
)
2
=
E
g
(
M
2
?
1
)
/
3
,
E
g
=
basic pulse energy
M
m
=
1
2
g
(
)
/
After modulation E
u
=
E
s
=
EM
2
?
16
E
b
=
average energy per bit
=
(
M
2
?
1
)
E
g
6
Log
2
()
M
?
Average energy per bit increases as M increases
BW efficiency
energy efficiency
Eytan Modiano
Slide 1
4
M
Two-dimensional signals
?
S
i
= (S
i1
, S
i2
)
?
Set of signal points is called a constellation
S
1
=(1,1)
S
2
=(-1,-1)
S
3
=(1,-1)
S
4
=(-1,1)
?
2-D constellations are commonly used
?
Large constellations can be used to transmit many bits per symbol
–
More bandwidth efficient
–
More error prone
?
The
“shape
” of the constellation can be used to minimize error
probability by keeping symbols as far apart as possible
?
Common constellations
–
QAM: Quadrature
Amplitude Modulation
PAM in two dimensions
Eytan Modiano
–
PSK: Phase Shift Keying
Slide 1
5
Special constellation where all symbols have equal power
Symmetric M-QAM
x
y
x
y
/
S
m
=
(
A
m
,
A
m
),
A
m
,
A
m
∈
{
+
/
?
1
,
+
/
?
3
,
...,
+?
(
M
?
1
)
}
M is the total number of signal points (symbols)
signal levels on each axis
M
16-QAM
Constellation is symmetric
?
M=
K
2
,
for some K
Signal levels on each axis are the same as for PAM
Eg
x
y
..
,
4
?
QAM
?
A
m
,
A
m
∈
{
+
/
?
1
}
x
y
16
?
QAM
?
A
m
,
A
m
∈
{
+
/
?
1
,
+
/
?
3
}
Eytan Modiano
Slide 1
6
-3
-1
1
3
-3 -1
1 3
Bandwidth occupancy of QAM
?
When using a rectangular pulse, the Fourier transform is a
Sinc
g(t)
|G(f)|
A
AT
T
-2/T
-1/T
1/T
2
/
T
?
First null BW is still 2/T
–
K = Log
2
(M) bits per symbol
–
Rb =
Log
2
(M)/T
–
Bandwidth Efficiency =
Rb/BW = Log
2
(M)/2
–
=> “Same as for PAM
”
Eytan Modiano
Slide 1
7
Energy efficiency
x
y
E
sm
=
[(
A
m
)
2
+
(
A
m
)
2
]
E
g
x
EA
m
)
2
]
=
EA
m
)
2
]
=
K
2
?
1
=
M
?
1
,
K
=
[(
[(
y
3
3
M
E
=
2
(
M
?
1
)
E
s
3
g
Transmitted
energy
=
E
s
=
(
M
?
1
)
E
g
2
3
E
b
(
QAM
)
=
Energy
/
bit
=
(
M
?
1
)
E
g
3
Log
2
()
M
?
Compare to PAM:
Eb increases with M, but not nearly as fast as PAM
E
b
(
PAM
)
=
(
M
2
?
1
)
E
g
M
Eytan Modiano
6
Log
2
()
Slide 1
8
Bandpass QAM
?
Modulate the two dimensional signal by multiplication by orthogonal carriers (sinusoids):
Sin &
C
o
s
–
This is accomplished by multiplying the A
x
component by
Cos and
the A
y
component by sin
–
Typically, people do not refer to these components as x,y but rather A
c
or A
s
for
cos and sin
or sometimes as A
Q
, and A
I
for
quadrature
or in-phase components
?
The transmitted signal, corresponding to the
m
th
symbol is:
x
Ut
()
y
()
m
()
=
A
m
g
t
Cos
(
2
π
f
c
t
)
+
A
m
g
t
Sin
(
2
π
f
c
t
)
,
m
=
1
.
.
.
M
Eytan Modiano
Slide 1
9
Eytan Modiano
Slide 2
0
Modulator
Binary data
Map k bits
Into one of 2
k
symbols
A
m
=(A
x
, A
y
)
g(t)
g(t)
Cos(2
π
f
c
t)
Sin(2
π
f
c
t)
U
m
(t)
Demodulation: Recovering the
baseband
signals
U(t)
LPF
S
x
r
(t)
2Cos(2
π
f
c
t)
U(t)
LPF
S
y
r
(t)
2Sin(2
π
f
c
t)
?
Over a symbol duration, Sin(2
π
f
c
t) and Cos(2
π
f
c
t) are orthogonal
–
As long as the symbol duration is an integer number of cycles of the carrier wave (fc = n/T) for some n
?
When multiplied by a sin, the
cos
component of U(t) disappears and
Eytan Modiano
similarly the sin component disappears when multiplied by
cos
Slide 2
1
()
t
()
Demodulation, cont.
()
2
Cos
(
2
π
f
c
t
)
=
2
A
x
g
t
Cos
2
(
2
π
f
c
t
)
+
2
A
y
g
t
Ut
()
()
cos(
2
π
f
c
t
)
sin
(
2
π
f
c
t
)
Cos
2
α
=
1
+
cos(
2
α
)
2
x
x
x
x
=>
Ut
t
()
cos(
4
π
f
c
t
)
≈
St
()
2
Cos
(
2
π
f
c
t
)
=
S
()
+
St
(
)
=
A
g
(
t
)
Similarly
,
U
(
t
)
2
Sin
(
2
π
f
c
t
)
=
2
A
x
g
(
t
)
Cos
(
2
π
f
c
t
)
Sin
(
2
π
f
c
t
)
+
2
A
y
g
(
t
)
sin
2
(
2
π
f
c
t
)
Sin
2
α
=
1
?
cos(
2
α
)
2
=>
Ut
t
y
()
cos(
4
π
f
c
t
)
≈
St
()
2
Sin
(
2
π
f
c
t
)
=
S
y
()
?
St
y
(
)
=
A
y
g
(
t
)
Eytan Modiano
Slide 2
2
Phase Shift Keying (PSK)
?
Two Dimensional signals where all symbols have equal energy levels
–
?
I.e., they lie on a circle or radius
Symbols can be equally spaced to minimize likelihood of errors
E
s
E
s
?
E.g., Binary PSK
s
1
s
2
E
s
E
s
?
4-PSK (above) same as 4-QAM
Eytan Modiano
Slide 2
3
()
)
M-PSK
A
i
x
=
Cos
(
2
π
i
/
M
),
A
i
y
=
Sin
(
2
π
i
/
M
),
m
=
0
,
...,
M
?
1
x
Ut
()
()
y
m
()
=
g
t
A
m
Cos
(
2
π
f
c
t
)
?
g
t
A
m
Sin
(
2
π
f
c
t
)
?
+
Notice
:
Cos
(
α
)
Cos
(
β
)
=
Cos
(
αβ
)
+
Cos
(
αβ
)
2
?
+
Sin
α
Sin
(
β
)
=
Cos
(
αβ
)
?
Cos
(
αβ
)
2
Hence
,
U
m
(
t
)
=
g
(
t
)
Cos
(
2
π
f
c
t
+
2
π
m
/
M
)
/
φ
m
=
2
π
mM
=
phases
shift
of
m
th
symbol
Ut
()
m
()
=
g
t
Cos
(
2
π
f
c
t
+
φ
m
)
,
m
=
0
.
.
.
M
?
1
Eytan Modiano
Slide 2
4
M-PSK Summary
?
Constellation of M Phase shifted symbols
–
All have equal energy levels
–
K = Log
2
(M) bits per symbol
?
Modulation:
Binary data
U
m
(t)
Map k bits
Into one of 2
k
symbols
A
m
=(A
x
, A
y
)
g(t) g(t)
Cos
(
2
π
f
c
t)
-Sin(2
π
f
c
t)
?
Notice that for PSK we subtract the
sin component from
the cos component
–
For convenience of notation only.
If we added, the phase shift would have been
negative but the end result is the same
?
Demodulation is the same as for QAM
Eytan Modiano
Slide 2
5