Digital Modulation

Saravanan Vijayakumaran
sarva@ee.iitb.ac.in

Department of Electrical Engineering

Indian Institute of Technology Bombay

August 8, 2012

1 / 45
 Digital Modulation
Definition
The process of mapping a bit sequence to signals for
transmission over a channel.

Example (Binary Baseband PAM)

1 → p(t) and 0 → −p(t)

p(t) −p(t)

t t

−A

2 / 45
 Classification of Modulation Schemes
• Memoryless
• Divide bit sequence into k -bit blocks
• Map each block to a signal sm (t), 1 ≤ m ≤ 2k
• Mapping depends only on current k -bit block
• Having Memory
• Mapping depends on current k -bit block and L − 1 previous
blocks
• L is called the constraint length
• Linear
• Modulated signal has the form
X
u(t) = bn g(t − nT )
n

where bn ’s are the transmitted symbols and g is a fixed

waveform
• Nonlinear

3 / 45
Signal Space Representation
Signal Space Representation of Waveforms
• Given M finite energy waveforms, construct an
orthonormal basis

s1 (t), . . . , sM (t) → φ1 (t), . . . , φN (t)

| {z }
Orthonormal basis

• Each si (t) is a linear combination of the basis vectors

N
X
si (t) = si,n φn (t), i = 1, . . . , M
n=1

 T
• si (t) is represented by the vector si = si,1 · · · si,N
• The set {si : 1 ≤ i ≤ M} is called the signal space
representation or constellation

5 / 45
Constellation Point to Waveform
φ1 (t)

si,1 ×

φ2 (t)

si,2 ×

.. ..
. .
+ si (t)
φN−1 (t)

si,N−1 ×

φN (t)

si,N ×

6 / 45
 Waveform to Constellation Point
φ∗1 (t)

si,1
R
×

φ∗2 (t)

si,2
R
×

.. .. ..
. . .
si (t)
φ∗N−1 (t)

si,N−1
R
×

φ∗N (t)

si,N
R
×

7 / 45
Gram-Schmidt Orthogonalization Procedure
• Algorithm for calculating orthonormal basis
• Given s1 (t), . . . , sM (t) the k th basis function is

γk (t)
φk (t) = √
Ek
where
Z ∞
Ek = |γk (t)|2 dt
−∞
−1
kX
γk (t) = sk (t) − ck ,i φi (t)
i=1
ck ,i = hsk (t), φi (t)i, i = 1, 2, . . . , k − 1

8 / 45
 Gram-Schmidt Procedure Example

s1 (t) s3 (t)

1 1

3 t
2 t

-1

s2 (t) s4 (t)

1 1

2 t 3 t

-1 -1

9 / 45
 Gram-Schmidt Procedure Example

φ1 (t) φ3 (t)

1
√1
2

2 3 t
2 t

-1

φ2 (t)
√ T
s1 = 2 0 0
 √ T
√1
2 s2 = 0 2 0
2 t √ T
s3 = 2 0 1
− √1
2
 √ T
s4 = − 2 0 1

10 / 45
 Properties of Signal Space Representation
• Energy

Z ∞ N
X
Em = |sm (t)|2 dt = |sm,n |2 = ksm k2
−∞ n=1

• Inner product
hsi (t), sj (t)i = hsi , sj i

11 / 45
Modulation Schemes
 Pulse Amplitude Modulation
• Signal Waveforms

sm (t) = Am p(t), 1≤m≤M

where p(t) is a pulse of duration T and Am ’s denote the M

possible amplitudes.
• Usually, M = 2k and amplitudes Am take the values

Am = 2m − 1 − M, 1≤m≤M

Example (M=4) A1 = − 3, A2 = − 1, A3 = + 1, A4 = + 3
• Baseband PAM: p(t) is a baseband signal
• Passband PAM: p(t) = g(t) cos 2πfc t where g(t) is
baseband

13 / 45
 Constellation for PAM
M=2

0 1

M=4

00 01 11 10

14 / 45
 Phase Modulation
• Complex Envelope of Signals
π(2m−1)
sm (t) = p(t)e j M , 1≤m≤M

where p(t) is a real baseband pulse of duration T

• Passband Signals
h√ i
p
sm (t) = Re 2sm (t)e j2πfc t
√ 
π(2m − 1)

= 2p(t) cos cos 2πfc t
M
√ 
π(2m − 1)

− 2p(t) sin sin 2πfc t
M

15 / 45
Constellation for PSK
QPSK, M = 4

01 11

00 10

Octal PSK, M = 8

011 001

010 000

110 100

111 101

16 / 45
 Quadrature Amplitude Modulation
• Complex Envelope of Signals

sm (t) = (Am,i + jAm,q )p(t), 1≤m≤M

where p(t) is a real baseband pulse of duration T

• Passband Signals
h√ i
p
sm (t) = Re 2sm (t)e j2πfc t
√ √
= 2Am,i p(t) cos 2πfc t − 2Am,q p(t) sin 2πfc t

17 / 45
Constellation for QAM

16-QAM

18 / 45
Power Spectral Density of Digitally Modulated
Signals
PSD Definition for Digitally Modulated Signals
• Consider a real binary PAM signal
∞
X
u(t) = bn g(t − nT )
n=−∞

where bn = ±1 with equal probability and g(t) is a

baseband pulse of duration T

• PSD = F [Ru (τ )] Not stationary or WSS

20 / 45
 Cyclostationary Random Process
Definition (Cyclostationary RP)
A random process X (t) is cyclostationary with respect to time
interval T if it is statistically indistinguishable from X (t − kT ) for
any integer k .

Definition (Wide Sense Cyclostationary RP)

A random process X (t) is wide sense cyclostationary with
respect to time interval T if the mean and autocorrelation
functions satisfy

mX (t) = mX (t − T ) for all t,

RX (t1 , t2 ) = RX (t1 − T , t2 − T ) for all t1 , t2 .

21 / 45
Stationarizing a Cyclostationary Random Process
Theorem
Let S(t) be a cyclostationary random process with respect to
the time interval T . Suppose D ∼ U[0, T ] and independent of
S(t). Then S(t − D) is a stationary random process.

Proof Sketch
Let V (t) = S(t − D). We prove that V (t1 ) ∼ V (t1 + τ ).
Z T
1
P [V (t1 + τ ) = v ] = P [S(t1 + τ − x) = v ] dx
T 0
Z T −τ
1
= P [S(t1 − y ) = v ] dy
T −τ
Z T
1
= P [S(t1 − y ) = v ] dy
T 0
= P [V (t1 ) = v ]

22 / 45
Stationarizing a Cyclostationary Random Process
Proof Sketch (Contd)
We prove that V (t1 ), V (t2 ) ∼ V (t1 + τ ), V (t2 + τ ).

P [V (t1 + τ ) = v1 , V (t2 + τ ) = v2 ]
1 T
Z
= P [S(t1 + τ − x) = v1 , S(t2 + τ − x) = v2 ] dx
T 0
1 T −τ
Z
= P [S(t1 − y ) = v1 , S(t2 − y ) = v2 ] dy
T −τ
1 T
Z
= P [S(t1 − y ) = v1 , S(t2 − y ) = v2 ] dy
T 0
= P [V (t1 ) = v1 , V (t2 ) = v2 ]

23 / 45
Stationarizing a Wide Sense Cyclostationary RP
Theorem
Let S(t) be a wide sense cyclostationary RP with respect to the
time interval T . Suppose D ∼ U[0, T ] and independent of S(t).
Then S(t − D) is a wide sense stationary RP.

Proof Sketch
Let V (t) = S(t − D). We prove that mV (t) is a constant function.

mV (t) = E [V (t)] = E [S(t − D)] = E [E [S(t − D)|D]]

E [S(t − D)|D = x] = E [S(t − x)] = mS (t − x)

Z T Z T
1 1
E [E [S(t − D)|D]] = mS (t − x) dx = mS (y ) dy
T 0 T 0

24 / 45
Stationarizing a Wide Sense Cyclostationary RP
Proof Sketch (Contd)
We prove that RV (t1 , t2 ) is a function of t1 − t2 = kT + 

RV (t1 , t2 ) = E [V (t1 )V ∗ (t2 )] = E [S(t1 − D)S ∗ (t2 − D)]

1 T
Z
= RS (t1 − x, t2 − x) dx
T 0
1 T
Z
= RS (t1 − kT − x, t2 − kT − x) dx
T 0
1 T −
Z
= RS (t1 − kT −  − y , t2 − kT −  − y ) dy
T −
1 T −
Z
= RS (t1 − t2 − y , −y ) dy
T −
1 T
Z
= RS (t1 − t2 − y , −y ) dy
T 0

25 / 45
 Power Spectral Density of a Realization
Time windowed realizations have finite energy

xTo (t) = x(t)I[− To , To ] (t)

2 2

STo (f ) = F(xTo (t))

|STo (f )|2
Ŝx (f ) = (PSD Estimate)
To

PSD of a realization

|STo (f )|2
S̄x (f ) = lim
To →∞ To
To
|STo (f )|2
Z
1 2
−
*
−
) xTo (u)xT∗o (u − τ ) du = R̂s (τ )
To To − T2o

26 / 45
Power Spectral Density of a Cyclostationary Process
S(t)S ∗ (t − τ ) ∼ S(t + T )S ∗ (t + T − τ ) for cyclostationary S(t)
Z To
1 2
R̂s (τ ) = s(t)s∗ (t − τ ) dt
To − T2o
Z KT
1 2
= s(t)s∗ (t − τ ) dt for To = KT
KT − KT
2
K
Z T 2
1 1 X
= s(t + kT )s∗ (t + kT − τ ) dt
T 0 K K
k =− 2
Z T
1
→ E [S(t)S ∗ (t − τ )] dt
T 0
Z T
1
= RS (t, t − τ ) dt = RV (τ )
T 0

PSD of a cyclostationary process = F[RV (τ )]

27 / 45
Power Spectral Density of a Cyclostationary Process
To obtain the PSD of a cyclostationary process
• Stationarize it
• Calculate autocorrelation function of stationarized process
• Calculate Fourier transform of autocorrelation

or

• Calculate autocorrelation of cyclostationary process

RS (t, t − τ )
• Average autocorrelation between 0 and T ,
1
RT
RS (τ ) = T 0 RS (t, t − τ ) dt
• Calculate Fourier transform of averaged autocorrelation
RS (τ )

28 / 45
Power Spectral Density of Linearly Modulated
Signals
 PSD of a Linearly Modulated Signal
• Consider
∞
X
u(t) = bn p(t − nT )
n=−∞

• u(t) is cyclostationary wrt to T if {bn } is stationary

• u(t) is wide sense cyclostationary wrt to T if {bn } is WSS
∗ ]
• Suppose Rb [k ] = E[bn bn−k
P∞ −k
• Let Sb (z) = k =−∞ Rb [k ]z
• The PSD of u(t) is given by

  |P(f )|2
Su (f ) = Sb e j2πfT
T

30 / 45
 PSD of a Linearly Modulated Signal

Ru (τ )
Z T
1
= Ru (t + τ, t) dt
T 0
Z T ∞ ∞
1 X X
∗
= E [bn bm p(t − nT + τ )p∗ (t − mT )] dt
T 0 n=−∞ m=−∞
∞ ∞ Z −(m−1)T
1 X X
∗
= E [bm+k bm p(u − kT + τ )p∗ (u)] du
T
k =−∞ m=−∞ −mT
∞ Z ∞
1 X
∗
= E [bm+k bm p(u − kT + τ )p∗ (u)] du
T
k =−∞ −∞
∞ Z ∞
1 X
= Rb [k ] p(u − kT + τ )p∗ (u) du
T −∞
k =−∞

31 / 45
 PSD of a Linearly Modulated Signal
∞ Z ∞
1 X
Ru (τ ) = Rb [k ] p(u − kT + τ )p∗ (u) du
T −∞
k =−∞

Z ∞
p(u + τ )p∗ (u) du *
−
− |P(f )|
)
2
−∞
Z ∞
p(u − kT + τ )p∗ (u) du *
− 2 −j2πfkT
− |P(f )| e
)
−∞

∞
|P(f )|2 X
Su (f ) = F [Ru (τ )] = Rb [k ]e −j2πfkT
T
k =−∞
  |P(f )|2
= Sb e j2πfT
T
P∞ −k .
where Sb (z) = k =−∞ Rb [k ]z
32 / 45
Power Spectral Density of Line Codes
 Line Codes
0 1 1 0 1 1 1 0 1 0 1

Unipolar NRZ

Polar NRZ

Bipolar NRZ

Manchester

Further reading: Digital Communications, Simon Haykin,

Chapter 6

34 / 45
 Unipolar NRZ
• Symbols independent and equally likely to be 0 or A

1
P (b[n] = 0) = P (b[n] = A) =
2

• Autocorrelation of b[n] sequence


A2

 2 k =0
Rb [k ] =
A2


4 k 6= 0

• p(t) = I[0,T ) (t) ⇒ P(f ) = T sinc(fT )e −jπfT

• Power Spectral Density
∞
|P(f )|2 X
Su (f ) = Rb [k ]e −j2πkfT
T
k =−∞

35 / 45
 Unipolar NRZ

∞
A2 T A2 T X
Su (f ) = sinc2 (fT ) + sinc2 (fT ) e −j2πkfT
4 4
k =−∞
∞
A2 T A2 X  n
= sinc2 (fT ) + sinc2 (fT ) δ f−
4 4 n=−∞
T
A2 T A2
= sinc2 (fT ) + δ(f )
4 4

36 / 45
 Normalized PSD plot

1
Unipolar NRZ
Su (f )

0.5
A2 T

0
0.5 1 1.5 2
fT

37 / 45
 Polar NRZ
• Symbols independent and equally likely to be −A or A

1
P (b[n] = −A) = P (b[n] = A) =
2

• Autocorrelation of b[n] sequence

 2
 A k =0
Rb [k ] =
0 k 6= 0


• P(f ) = T sinc(fT )e −jπfT

• Power Spectral Density

Su (f ) = A2 T sinc2 (fT )

38 / 45
 Normalized PSD plots

1
Unipolar NRZ
Polar NRZ
Su (f )

0.5
A2 T

0
0.5 1 1.5 2
fT

39 / 45
 Manchester
• Symbols independent and equally likely to be −A or A

1
P (b[n] = −A) = P (b[n] = A) =
2

• Autocorrelation of b[n] sequence

 2
 A k =0
Rb [k ] =
0 k 6= 0


   
• P(f ) = jT sinc fT
2 sin πfT
2
• Power Spectral Density
   
2 2 fT 2 πfT
Su (f ) = A T sinc sin
2 2

40 / 45
 Normalized PSD plots

1
Unipolar NRZ
Polar NRZ
Manchester
Su (f )

0.5
A2 T

0
0.5 1 1.5 2
fT

41 / 45
 Bipolar NRZ
• Successive 1’s have alternating polarity

0 → Zero amplitude
1 → +A or − A

• Probability mass function of b[n]

1
P (b[n] = 0) =
2
1
P (b[n] = −A) =
4
1
P (b[n] = A) =
4

• Symbols are identically distributed but they are not

independent
42 / 45
 Bipolar NRZ
• Autocorrelation of b[n] sequence

A2 /2

 k =0
Rb [k ] = − A2 /4 k = ±1
0 otherwise


• Power Spectral Density

A2 A2  j2πfT
 
2 −j2πfT
Su (f ) = T sinc (fT ) − e +e
2 4
A2 T
= sinc2 (fT ) [1 − cos(2πfT )]
2
= A2 T sinc2 (fT ) sin2 (πfT )

43 / 45
 Normalized PSD plots

1
Unipolar NRZ
Polar NRZ
Manchester
Bipolar NRZ
Su (f )

0.5
A2 T

0
0.5 1 1.5 2
fT

44 / 45
Thanks for your attention

45 / 45