Introduction To Digital Communications: Olivier - Michel@

Introduction to Digital communications
olivier.michel@grenoble-inp.fr
ENSE3 Grenoble-INP, France
2009
Typical communication media
A,D
A,D A,D
A,D
A,D
A,D
= D =
Analog Communications
CSQ :
Perfect recovery in the presence of noise is not possible
Digital communications
Shannon theory / coding, redundancy = perfect transmission is

possible, at finite SNR.
Basic communication components
Modulator : from a ’baseband’ or lowpass signal, to a

’passband’ signal
widely used freq. ranges

c
fc
λ/10
Mapper example
{−3, −1, 1, 3}
•
• a [n] nT g(
!
Introduction to Digital communications !
m (t)
Basic communication components = n a [n] g (t − nT )
T =
Linear modulator example
[1, 3, −1, 1, 3]
‘
a[n] = 1, 3, −1, 1, 3.... -> ISI, synchronisation, channel response......
Transmit
Intersymbol and receive
Interference filters forconstraints
: problem,
bandlimited AWGN channels
! Two facts:
! Perfection ( ) means infinite bandwidth
! Requirement for zero ISI :
! The aim:
! Perfect discrete channel based on perfect finite bandwidth channel
Nyquist criterion
INPG TST 3A J.M. Brossier - LIS

‘
Ideal sinc solution
Ideal sinc solution
Fréquency
Strictly rectangular spectrum hard to implement:

• slowly decaying, infinitely long, noncausal
waveform
• sensitivity to timing errors
• truncation errors (Gibbs phenomenon)
INPG TST 3A J.M. Brossier - LIS ‘

Accounting for the (linear) channel dispersion (next

sections)
Root-Nyquist filtering
where (h1 ? h2 )(t) = h(t) must satisfy Nyquist criterion !

Communications channels, briefly
Linear model of signal propagation
Dispersion (phase distorsion), selective attenuation, multipath ....

noise = electronic, multi-access interference, co-channel
interference....
w(n) is Additive, white, Gaussian (AWGN)
Selective filtering for SNR improvment

= = Es
En
Multipath filter model (no fading)
X
x(t) = αn s(t − τn ) + n(t)
npath
X
x̃(ν) = αn e−j2πντn s̃(ν) + ñ(ν)
npath
X
x (t) = c1 s (t⇒ 1 ) + c=
− τh(t) 2 s (t − α
τ2n) δ(t
+ · ·−
· +τcnN) s (t − τN ) ,
x (t) = s (t) ∗ h (t) npath
x (t) = c s (t − τ ) + c2 s (t − τ2 ) + · · · + cN s (t − τN ) ,
1
Communications channels, briefly 1
x (t) = s (t) ∗ h (t)
h (τ ) = c1 δ (τ − τ1 ) + c2 δ (τ − τ2 ) + · · · + cN δ (τ − τN ) .
Multipath filter model example
H (f ) |
Fading scales
Fading scales
! Distance
! outdoor, indoor
! Slow fading
! Log-normal
6-10dB, 5 (indoor)-20m (outdoor)
! Fast fading
! Multipath propagation

Rayleigh Channel
Channelmodeling
model modeling
Flat
FlatRayleigh fading:NLOS
Rayleigh fading: NLOS
! Path: cluster
! Path: a micropaths:
cluster a micropaths:
pdf pdf module
module
! NLOS (No Line of Sight, urban) : CLT:

! NLOS (No Line of Sight, urban) : CLT:
pdf phase
pdf phase

1 (t) + c2 w2 (t)} = c1 W1 (f ) + c2 W2 (f )
F{cIntroduction to Digital communications
1w Notations, reminder, math tools
Basics W (f )
w (t) ⇒ {
|W (f ) | f =0
Real Bandpass signals, bandwidth
Notations, reminder, math tools
Basics
physical constraints
I
1
∆t∆ν ≥
4π
=> Small T implies large freq!
BUT Digital Comm. : seeks for narrow pulses and small freq.
bandwidth !!!
I Physical ’Dirac’ pulse
1
TsNotations, reminder, math tools
Basics
Fourier spectrum of a periodic deterministic signal

Basics
Fourier spectrum of a random noise (estd)

Basics
h (t)
LTI systems
H (f )
Basics
LTI in frequency domain

H (f )
Basics
Examples
Ideal LPF :
Ideal delayed LPF :

Basics
Real causal linear phase LPF, with group delay t0
We can do better : see linear phase FIR filter design...(last year

lecture)
Basics
filtering noise : 1
Ts
Sy (ν) = |H(ν)|2 Sx (ν)

Complex baseband : motivations and tools
A briel review of AM modulation

AM with suppressed carrier :
m (t)
Rk : if m(t) is real, AM transmitted spectrum is redundant : motivation

|M! (f ) |
∈ R for QAM f = 0
! "
(2πfc t) = Demodulation
ej2πf t + e−j2πf t
1 c
(if trivial channel,
c m (t) ∈ R is(fknown,
and fc |M )| f = 0 synchro
and perfect
2
is
S (f ) =
assessed) #∞
m (t) cos (2πfc t) e−j2πf t
dt
−∞
#∞ #
1 ∞
= 1
2 −∞
m (t) e−j2π(f −fc )t dt + 2 −∞ m (t) e
−j2π(f +fc )t
dt
= 1
2M (f − fc ) + 1
2 M (f + fc ) .
Introduction to Digital communications = {m (t) · 2cos2 (2πfc t)}
! "# $
1+cos(2π·2fc t)
= {m (t) + m (t) cos (2π · 2fc t)}

continued... = m (t) ,
LPF has passband Bp ≥ freq
cutoff ≤ 2f
W Bp , stopband cutoff Bs s.t.BBsp ≤ −W
W cand
Bs < 2fc − W :
When the receiver local oscillator has freq. and phase offset {δf ; δφ},
then
v (t) = m(t) cos(2πδft + δφ)
(time varying attenuation : left as exercice)

s (t) = m (t) cos (2πfc t) + Acos (2πfc t)
! "# $
= [m (t) + A] cos (2πfc t)

AM whith carrier
S (f ) = 1 [M (f −f 2 c) +M (f +fc ) +Aδ (f −fc ) +Aδ (f +fc )]
s(t) = (m(t) + A) cos(2πfc t)
modern systems : A << max(m(t))

Large carrier AM : A > max(m(t)), allow enveloppe detection based
receivers.
Rk : Carrier - AM transmitted spectrum is redundant, consumes

energy (carrier)
Rice representation of deterministic signals
Hilbert transform
hilbert(f)
-j
Hilbert filter :
H(ν) = −j.sign(ν)
Analytic transform of x
zx (t) = x(t) + j.H[x](t)

Complex enveloppe
ax (t) = zx (t)e−j2πν0 t
Zx(f)
X(f) Ax(f) 2
1
-fo f
f
Rk: x(t) real signal, but zx (t), ax (t) complex-valued signals.

Consequences of previous definitions
H[x](t) ∈ IR, x(t) ∈ IR ⇒ x(t) = Re[zx (t)]

and
x(t) = Re[ax (t)ej2πν0 t ] H[x](t) = Im[ax (t)ej2πν0 t ]
As ax (t) ∈ C ⇒ ax (t) = px (t) + j.qx (t)
x(t) = px (t) cos(2πν0 t) − j.qx (t) sin(2πν0 t)

H[x](t) = px (t) sin(2πν0 t) + j.qx (t) cos(2πν0 t)
"ax (t) : Baseband equivalent signal , relative to ν0 "
alternate formulation

x(t) cos ω0 t − sin ω0 t px (t)
=
H[x](t) sin ω0 t cos ω0 t qx (t)
thus
px (t) cos ω0 t sin ω0 t x(t)
=
qx (t) − sin ω0 t cos ω0 t H[x](t)
Bedrossian’s theorem
Let f (t), g(t) such that :

I F (ν).G(ν) = 0 ∀ν

f (t) is LF(∆F )
I such that min[G(ν)] >> 2∆F
g(t)is HF
then
H[f .g](t) = f (t).H[g](t)
Example
Let x(t) = m(t) cos(2πν0 t + φ), with ∆M(ν) << ν0
The complex enveloppe (relatively to ν0 ) of x is
ax (t) = m(t)ejφ
Introduction to Digital communications -Lecture 2-

-Lecture 2-
2009
Narrow-band signals, Bedrossian’s th., consequences
Bedrossian’s theorem
Let f (t), g(t) such that :

I F (ν).G(ν) = 0 ∀ν

f (t) is LF(∆F )
I such that min[G(ν)] >> 2∆F
g(t)is HF
then
H[f .g](t) = f (t).H[g](t)
Example
Let x(t) = m(t) cos(2πν0 t + φ), with ∆M(ν) << ν0
The complex envelope (relatively to ν0 ) of x is
ax (t) = m(t)ejφ
Actually for e.g. AM, transmitted signal is

x(t) = m(t) cos(2πν0 t +φ) = m(t) cos(φ) cos(2πν0 t)−m(t) sin(φ) sin(2πν0 t).
As H[cos(2πν0 t + φ)] = sin(2πν0 t + φ), it comes zx (t) = e2πν0 t+φ , the

complex envelope is ax (t), and
(
m(t)ejφ = px (t) + jqx (t)
x(t) = px (t) cos(2πν0 t + φ) − qx (t) sin(2πν0 t + φ)
Application to real valued passband signals
Def : x(t) is a deterministic real passband signal if ∃B ∈ IR+ s.t.


 +
X (ν) = X (ν) if ν>B
X − (ν) = X (ν) if ν<B

 − ∗
X (ν) = X + (−ν)
(
Zx (ν) = 2X + (ν)
therefore
Ax (ν) = 2X + (ν + ν0 )
Equalities for real passband signals
Exercice : As x(t) = px (t) cos(2πν0 t) − qx (t) sin(2πν0 t), prove that
I
X + (ν) = 12 [Px (ν − ν0 ) + jQx (ν − ν0 )]
X − (ν) = 21 [Px (ν + ν0 ) − jQx (ν + ν0 )]
I Ax (ν) = 2X + (ν + ν0 ) = Px (ν) + jQx (ν)
Px (ν) = X + (ν + ν0 ) + X − (ν − ν0 )
I
Qx (ν) = 1j [X + (ν + ν0 ) − X − (ν − ν0 )]
Spectral interpretation, real passband signals -1-
X(f) Xh(f)
1 1
j
- fc fc - fc fc
j
where
Xh (ν) = TF [H[x(t)](ν)
px (t) = x(t) cos(2πν0 t) + xh (t) sin(2πν0 t)
TF[x(t)cos(w_0t)](f)
1/2
- 2 fc - fc fc 2 fc
1/2 TF[xh(t)sin(w_0t)](f)
- 2 fc - fc fc 2 fc
TF[px(t)p](f)
1/2
- 2 fc - fc fc 2 fc

qx (t) = −x(t) sin(2πν0 t) + xh (t) cos(2πν0 t)
1/2 TF[x(t)sin(w_0t)](f)
1/(2j)
1/(2j)
- 2 fc - fc fc 2 fc
1/(2j)
1/(2j)
TF[xh(t)cos(w_0t)](f)
1/2
j/2 j/2
- 2 fc - fc fc 2 fc
j/2 j/2
1/2 TF[qx(t)](f)
- 2 fc - fc fc 2 fc
Typical application involving passband signals -1-
Let x(t), y (t) real valued bandpass signals, expressed in terms of

their respective in-phase and quadrature components (rel. to ν0 )
(
x(t) = px (t) cos(2πν0 t) − qx (t) sin(2πν0 t)
y (t) = py (t) cos(2πν0 t) − qy (t) sin(2πν0 t)
then
s(t) = x(t).y (t)
= 12 [px (t).py (t) + qx (t).qy (t)] + HF terms around 2ν0
Application : Complex baseband representation of QAM
Typical application involving passband signals -2- :

{mI (t) , m (t)}
QAM
Quadrature Amplitude Modulation
(TRIVIAL CHANNEL)
(
vI (t) = mI (t)
where if perfect synchronization.
vQ (t) = mQ (t) Bp ≥ W
Exercice
Replace 2 cos(2πν0 t) (resp. sin() by cos(2πν0 t + φ) to account for

lack of phase synchronization. Prove that
(
vI (t) = mI (t) cos(φ) + mQ (t) sin(φ)
vQ (t) = −mI (t) sin(φ) + mQ (t) cos(φ)
Show that φ 6= 0 leads to some coupling between the in-phase and

quadrature components, and to attenuation of both.
Complex baseband representation of QAM

Writing the complex baseband form
m̃(t) = mI (t) + jmQ (t)
ṽ (t) = vI (t) + jvQ (t)
yields the much simpler representation of QAM
where if r (t) = s(t), ṽ (t) = m̃(t)

Rice representation of random processes, baseband filtering
Rice representation of random processes

Let x(t) be a real-valued a second order stationary, zero-mean
random process.

zx (t) = x(t) + j.H[x](t)


 a (t) = z e−2jπν0 t
x X
ax (t) = px (t) + j.qx (t) = ρx (t)ejφx (t)



x(t) = px (t) cos(2πν0 t) − qx (t) sin(2πν0 t)
Expressing Rxx (τ ) = E[x(t)x ∗ (t − τ )] as a function of px , qx , yields


 Rpp (τ ) = Rqq (τ )

R (τ ) = −R (τ ) = −R (−τ )
pq qp pq
x(t) wide sense stationnary ⇒

 E[|ax (t)|2
] = 2R pp (0)


E[x(t)] = cst ⇒ Rpq (0) = 0
Narrow band random processes
Def : x(t) is wide sense stationary narrowband random process if its

PSD γx (ν) is narrowband.
Let ax (t) = px (t) + j.qx (t) be the complex envelope of x, relative to
ν0 , then
a(t) is a complex random process, verifying
γa (ν) = 4γ + (ν + ν0 )
γp (ν) = γq (ν) = 41 [γa (ν) + γa (−ν)]
γpq (ν) = 4j1 [γa (−ν) − γa (ν)]
Complex envelope transformation through filtering

(narrowband)
x(t) y(t)
h(t)
Y (ν) = H(ν).X (ν) ⇒ Zy (ν) = H + (ν)Zx (ν) = 2H + (ν)X + (ν)

Ax (ν) = 2X (ν + ν0 ) ⇒ 2Zy (ν + ν0 ) = 2H + (ν + ν0 )Zx (ν + ν0 )
+
yielding
Ay (ν) = H + (ν + ν0 )Ax (ν) = Heq (ν)Ax (ν)
Heq (ν) 6= Ah (ν) BUT Heq (ν) = 12 Ah (ν)

Heq (ν) is LF shifted version of H(ν), without correcting factor 2!
Time domain filter input-output relations for complex

baseband signals
ay (t) = [Heq ⊗ ax ](t)

= 21 [(ph + j.qh ) ⊗ (px + j.qx )](t)
= ....
py (t) = 12 [ph ⊗ px ](t) − 12 [qh ⊗ qx ](t)
qy (t) = 12 [qh ⊗ px ](t) + 12 [ph ⊗ qx ](t)
Baseband formulation of interference formula
h(t)
γb (ν) γs (ν) = |H(ν)|2 γb (ν)
No/2
where γb (ν) : -f_0 f_0
Applying previous formula leads to
γa b = 4γb+ (ν + ν0) = 2N0

γa s = |Heq (ν)|2 2N0
continued
furthermore
γps (ν) = γqs (ν) = 14 [γas + γas (−ν)]

= N20 [|Heq (ν)|2 + |Heq (−ν)|2 ]
and
N0
γpqs (ν) = [|Heq (ν)|2 − |Heq (−ν)|2 ]
2j
I γpqs (ν) = 0 if Heq (ν) is symetric (i.e. if H + (ν) is symetric around

ν0 )
I In-phase and quadrature component have identical variances
Z ∞
σ 2 = N0 |Heq (−ν)|2 dν
−∞
Complex baseband equivalent channel (linear modulations)
!
[s (t) ∗ h (t)] 2e −j2πf
Complex baseband equivalent channel (linear c t
=
modulations)
s (τ ) h (t − τ ) dτ · 2e−j2πfc t
!
= s (τ ) 2e−j2πfc τ h (t−τ ) e−j2πfc (t−τ ) dτ
contd... " # " #
= s (t) 2e−j2πfc t ∗ h (t) e−j2πfc t ,
as
s(t) ⊗ hbp (t)]2e−j2πfc t = [s(t)2e−j2πfc t ] ⊗ [hbp (t)e−j2πfc t ] :
h (t) e−j2πfc t
then reversing the order of the LTI systems :
h (t) e−j2πfc t
Consequences :
I mod/demod are transparent (with synch oscillators) :
m̃ (t) H (f + fc ) =
{h (t) e−j2πfc t }
m̃ (t) H (f +
F{h (t) e−j2πfc t }
I TF [hbp (t)e−j2πfc t ] :
m̃ (t) H (f + fc ) =
e−j2πfc t }
h (t) e−j2πfc t
h̃ (t)
Heq (ν) = H̃(ν)!
I finally, for the noiseless complex baseband channel :
h (t) e−j2πfc t h̃ (t)

Noisy channel (additive)
ṽn (t)
f ∈ [fc − Bs , fc + Bs ]
Noisy channel (contd)

I from previous studies on complex baseband random process,
N0
I if noise PSD is CONSTANT ( 2 ) over the freq-range of interest,
I as all contribution outside [−W , W ] will be filtered out,

f ∈ [−W, W ]
considering that H̃(ν) = 1∀ν ∈ [−W , W ] the noise ṽn is
(t) modeled by
a white (cst PSDw̃ 2N
(t) 0 ) noiseSw̃ (f ) = N0
Summary

-Lecture 3-
2009
Digital modulations
Definitions
Digital modulation definition
Discrete symbols an ∈ Ω → Continuous time series m̃(t)


an ' a(nT )
|Ω| = M

1
T = Rs "symbol rate"
where
Channel coding : uk ↔ sequence of bits

therefore
1
Rb = T log2 M bits/sec
m̃ (t) = a [n] g (t − nT )
n
s (t) = Re{m̃ (t) e
Digital modulations } j2πfc t
System
Digital communication system

ṽ (t) = {2r (t) e
−j2πf t c
}
y (t) = ṽ (t) ∗ q (t)
y [m] = y (mT ) P
-pulse shaping : m̃(t) = n a[n]g(t − nT )
Transmitter :
-modulation : s(t) = Re{m̃(t)ej2πfc t }
-demodulation : ṽ = LPF{2r (t)ej2πfc t }

Receiver : -filtering : y (t) = ṽ (t) ⊗ q(t)
-sampling : y [m] = y (mT )
!∞ " # F !∞
Introduction to Digital1communications -Lecture 3-
T k=−∞ δ f− k
T ↔ m=−∞ δ (t − mT ) ,
Digital modulations
System
%
Digital communication&
k system, contd
∞ ∞
1 $ F
$
P (f ) ∗
C-baseband : δ f− ↔ p (t) · δ (t − mT ).
T T m=−∞
k=−∞
' () * ' () *
P
∞ P
∞
1
T P (f − T
k
) p(mT )δ(t−mT )
k=−∞ m=−∞
p (mT ) = δ [m]
!∞where for" the noiseless
# F channel
1
T k=−∞ P f − T
k
↔ δ (t) ,
g(t) ⊗ h̃(t) ⊗ q(t) = p(t)
verifies NYQUIST ISI supression criterion
{a [n]}
m̃ (t)
!
m̃ (t) = n a [n] g (t − nT )
T =
m̃ (t)
[a [0] , a [1] , a [2] , a [3] , a [4]] [1, 3, −1, 1, 3]
g (t) m̃ (t)
Digital modulations
!
m̃ (t) = a [n] g (t − nT )
Linear modulations
n
T = modulations
Linear digital
baseband message :
X
[a [0] , a [1] , a [2]
m̃(t) = , a [3] , a [4]]− nT )
a[n]g(t [1, 3, −1, 1, 3]
m̃ (t) n
example for a[n] = [1, 3, −1, 1, 3], with non-realistic g(t) :

Digital modulations
Power spectral density of linear digital comm. signal
PSD of linear digital comm. signal
P
m̃(t) = n a[n]g(t − nT )
∗
P P (t − τ )]
Γm (t, τ ) = E[m(t)m
= k k 0 E[a[k ]a∗ [k 0 ]]g(t − kT )g ∗ (t − k 0 T )
Assuming that (a[k ]) is a wide sense stationary process :
E[a[k ]a∗ [k 0 ]] = Γa (k − k 0 )
P P
⇒ Γm (t, τ ) = l Γa (l) k g(t − kT )g ∗ (t − τ − (k − l)T )
| {z }
Depends upon both t and τ !
BUT
Γm (t + T , τ ) = Γm (t, τ ) ⇒ CYCLOSTATIONARITY
Digital modulations
PSD of linear digital comm. signal, contd
Z T
1
Γm (τ ) = Γm (t, τ )dt
T 0
if g(t) ∈ IR,
|G(f )|2 P
γ(f ) = T l Γa (l)e−j2πflT
(
ma = E[a]; σa2 = E[a2 − E[a]]
Letting
ã = a−E[a]
σa
|G(f )|2 P
γ(f ) = T l σa2 Γã (l) + |ma2 | e−j2πflT
Digital modulations
PSD of linear digital comm. signal, contd
As Γã (l) = Γ∗ã (−l),

2 P∞
γ(f ) = 2σa2 |G(fT )| l=1 Re Γã (l)e−j2πflT (1)
2
+σa2 |G(fT )| (2)
|m2 | P k 2
+ T 2a G( ) δ(f − k ) (3)
k T T
I (1)(2) : continuous part of the PSD; (3) : discrete part

I (3) = 0 if |ma | = 0
I G(0) = 0 ⇒ γ e (0) = 0
l→∞
I (1) is an ordinary function of f if Γa (l) −→ 0 quickly enough
Digital modulations
Pulse shaping
Most popular pulse shape

cos( παt
T ) t sin(πx)
g(t) = 1−( 2αt 2 sinc( T ), sinc(x) = πx
 T )

T |f | ≤ (1−α)
2T
πT 1−α
1−α 1+α
G(f ) = T cos2 |f | − ≤ |f | ≤


2α 2T 2T
1+α
2T
0 2T ≤ |f |
α = roll-off parameter
larger α ⇒ less time-spread,
more freq-spread
Digital modulations
Pulse shaping
General representations of Linear digital modulations
General expression of baseband message :

X
m̃(t) = ap [n]gp (t − nT ) + j.aq [n]gq (t − nT )
n
and
X
m(t) = ap [n]gp (t − nT ) cos(2πfc t) − aq [n]gq (t − nT ) sin(2πfc t)
n
Example : Pulse Amplitude Modulation (PAM)

P
m̃(t) = n ap [n]gp (t − nT ), a[n] = (2k − 1 − M), k ∈ {1, 2, . . . , M}
i.e. gp (t) = g(t); gq (t) = 0

Digital modulations
Pulse shaping
Signal space dimension
P
Definition : Let m(t) = n s(t − nT , an ). The dimension N of the
signal space is the dimension of the real-valued functional space
spanned by the signals s(t, a).
P
- PAM : m̃(t) = n a[n]g(t − nT ),
where a[n] ∈ IR and g(t) ∈ IR ⇒ m̃(t) ∈ IR, thus NPAM = 1
P
- QAM : m̃(t) = n g(t − nT )[ap [n] + j.aq [n]) where (ap , aq ) ∈ IR2 ,
and g(t) ∈ IR ⇒ m̃(t) ∈ C, thus NQAM = 2
Rk : For OFDM or FSK, N > 2

Digital modulations
Pulse shaping
Energy
P
Definition : Let m(t) = n s(t − nT , an ). The energy requested for
transmitting a single symbol is E(a) = ||s(a)||2 (1 )
The average energy spend per symbol is, for |Ω| = M and uniform
probability of all symbols :
1 X
Es = E(a)
M
a∈Ω
The average energy per bit is then
1
Eb = Es
log2 (M)
1 ||s(a )||2
RT
n = (n−1)T s2 (t, an )dt
Digital modulations
Pulse shaping
Signal space examples

TED © CC
COPYRIGHT PROTEC
PAM : P. COMON
000 001 011 010 110 111 101 100
TED © CC
-7A COPYRIGHT PROTEC
5A 7A
COMMUNICATIONS NUMÉRIQUES 29
PAM8
M 2 −1 2 2
Es,PAM = 3 A ( )
QAM : Figure 24: Constellation PAM8
ons PAM
(t) s(u). Alors la dimension de la constellation est N = 1. Pour une modulation PAM-M, nous
A A 3A A 3A 5A
omplexe: !
x̃(t) = A g(t − kT ) uk
QAM4
QAM16
Es,PAM = 2(M−1) A2 , (if ∃k /M

k QAM32
= k 2)
M , pour 1 ≤ m ≤ M . En d’autres termes, les symboles sont dans u ∈ {−M + 1, . . .3, M − 1}. Il
E =2A2
s
2
E = 10A
s
E = 20A
s
2
ire dmin = 2A et l’énergie moyenne par symbole

Figure 25: Constellations QAM
M2 − 1 2
Es = A (13)
3 n(2n−1)(2n+1)
2 12 + 22 + . . . + (2n − 1)2 = 3
2 2
E = 10A E = 20A
s s
Digital modulations
Figure
Pulse 25:
shaping Constellations QAM
PSK :
( P
m̃(t) = A n g(t − nT )ejφ(a[n])
P
m(t) = A n g(t − nT ) cos(2πfc t + φ(a[n]))
011
010 001
110 000
100
111
101
PSK8
Es,PSK = A2
Figure 26: Constellation PSK
Other modulations
Differential modulations
Perfect phase locking of the receiver : impossible

I phase rotation in PSK or QAMq → errors in symbol detection
⇒ Encode phase jumps, resulting in rotation invariant modulations
I phase rotation invariance, ’infinite modulation memory’ to encode
initial phase
I demod : yk = ej(φk −θ) , < yk , yk −1 >= ej(φk −φk −1 ) = eja[k ]
TED © CC
COPYRIGHT PROTEC
30
Example : trellis representation of a Differential Binary PSK (M = 2) P. COMON
0 a 0 0 0
a a
b b b
! b ! b ! b !
a a a
state changes are associated to phase jumps of π.

Figure 27: Treillis de la modulation DBPSK
Par ailleurs, il est évident que l’énergie moyenne par symbole vaut Es = A2 , puisqu’elle est la même pour tous les
symboles.
Dans certains cas, il peut être souhaitable d’éviter des sauts de phase brutaux, et on peut introduire une
fonction de mise en forme dans la phase, comme on l’a fait pour l’amplitude complexe avec la fonction g(t). Cela
nous donnerait une enveloppe complexe:
!
x̃ = A g(t − kT ) eφ(uk )gϕ (t−kT ) .
Other modulations
Offset modulation
( : phase jumps ⇒
Pb
freq spread
high amplitude fluctuations, out of linear range of HF amplifiers
Solution : π phase jumps are not allowed !
Example : Offset QPSK :
gp (t) = g(t − T2 ) and gq (t) = g(t)

P
m(t) = n ap (t)g(t − nT − T2 ) − aq (t)q(t − nT )
⇒ Re(m̃(t) and Im(m̃(t)CO
doPYnot PROTEC
change
RIGHT TED © CC
simultaneously :31
COMMUNICATIONS NUMÉRIQUES
Offset−QAM4 Offset−QPSK
Figure 28: Transitions autorisées dans les modulations décallées

Other modulations
FSK
Constant envelope modulation
Motivation : bounding the amplitude fluctuation (to ensure linear

range operation of the HF electronic devices)
→ Frequency modulation (FSK) with continuous phase (CPM).
Rt P
FSK : x(t) = cos 2πfc t + 2πfd −∞ m(τ )dτ , m(τ ) = n an gd (τ − nT )
Instantaneous frequency : fc + fd m(t)

Other modulations
FSK
FSK example
h
Let gd (t) = ΠT (t) and fd m(t) = a[n] 2T
Rt
I ⇒ phase of C-envelope (x̃) is f m(τ )dτ is piecewise linear.
−∞ d
h
I ⇒ if fd is PAM (as often), then frequencies are separated by T :
h : modulation index of FSK
I phase jump between 2 consecutives ’symbols’ = 0 → (f1 − f2 )T is
1 1
2 -integer (then h is 2 integer).
where
( Rt
x(t) = cos(2π f0 t + t0 m(τ )dτ
1
P
m(τ ) : 4T n a[n]gd (t − kT ), a[n] ∈ {−1; +1}
I this is Min Shift Keying (MSK) for a BPSK.

Other modulations
FSK
Modulation standard examples
Standard Modulation type

DECT GFSK3
GSM GMSK
UMTS QPSK
Modem v34kbit/s QAM-1664
3 GMSK = MSK where the binary data flow is pre-filtered (before F-modulation), to
reduce sideband power.
Transmission over a noisy (white gaussian) ideal channel
Performances
Performances
p (t)
First consider
g (t) the
q (t) transmission of a unique symbol :
baseband representations of the communication system :
⇔
where E[w̃(t)w̃ ∗ (t)] = N0 δ(t) (complex white noise)
 P
E{w̃ (t) w̃∗ (t − τ )} 
= N0 δ (τ )
ys [m] = R n a[n]p((m
= = − n)T, ) = a[m]p(0)
Es
En
E{|ys [m]|2 }
E{|yn [m]|2 }
g (t) , q (t)} y [m] ∞
no ISI ⇒ p(0) w̃ (t)= a−∞
Ideal channel,y [m] [n]
q(τ )g(−τ )dτ
y [m] y [m]

 p (t)
s n
Es = E[|ys![m]|2 ] = σa2 p(0)2

ys [m] = n a [n] p ((m − n) T ) = a [m] p (0)
"∞
p (0) = −∞
q (τ ) g (0 − τ ) dτ,
# " #2
Performances
Performances, contd
R∞
as yn [m] = yn (mT ) = −∞
q(τ )w̃(mT − τ )dτ ,
Z ∞
En = E[|yn [m]|2 ] = N0 |q(τ )|2 dτ
−∞
Then , the pdf of the observation is
1 (y [m]−a[m]p(0))2
f (y [m]|a[m]) = √ e− 2En
2πEn
Performances
Performances, contd
Considering e.g. PAM2 : a[m] ∈ {−1, +1} , and a simple threshold

detector (threshold η)
( R∞
PFA = Pr(y [m] > η|a[m] = −1) = η f (y |a = −1)dy
Rη
PM = Pr(y [m] < η|a[m] = +1) = −∞ f (y |a = +1)dy
Rx 2
Letting erf(x) = √1
x 0
e−t dt,

PFA (η) = 1
− 12 erf η+p(0)
√
2 2En
 PM (η) = 1
+ 1 η−p(0)
2 erf
√
2 2En
and Perr (η) = Pr(a[m] = −1)PFA (η) + Pr(a[m] = +1)PM (η)

Performances
Performances, contd
Perr (η) = Pr(a[m] = −1)PFA + Pr(a[m] = +1)PM ⇒ choosing η to

minize Perr ?

∂Perr En p0
∂η = 0 ⇒ ηopt = 2p(0) log p1
where p0 = Pr(a[m] = −1) and p1 = Pr(a[m] = +1)

Performances
Optimizing the receiver (ideal channel)

p(0) 1 p0
Let ρ = √
2En
' SNR, and let k = 4 log p1
then
p0 k p1 k
Perr (ηopt ) = erfc(ρ + ) + erfc(ρ − )
2 ρ 2 ρ
As minimizing Perr ⇔ maximizing ρ, and reexprssing ρ :
R∞
p(0) −∞
q(τ )g(−τ )dτ )
ρ= √ ∝ h i 21
2En √ R∞
N0 −∞ |q(τ )|2 dτ
By Cauchy-Schwartz inequality, ρ is maximum if ∃λ such that
q(t) = λg ∗ (−t)
Performances
Optimizing the receiver (contd)

q(t) = λg ∗ (−t) : MATCHED FILTER EQUATION
Then by properly choosing λ,
s
Eg
ρopt =
N0
and if p0 = p1 (then k = 0),

s !
1 Eg
Perr = erfc
2 N0
finally for Best performance of the ISI free channel :

-G(f )Q(f ) = P(f ) must satisfy the Nyquist criterion
-Q(f ) = G ∗ (f )ej2πft0
Performances
Optimizing the receiver (contd)

From preceding equation :
|G(f )|2 must satisfy the Nyquist criterion. One option is
p
G(f ) = Prc (f )
as Prc (f ), the raised cosine filter, is Nyquist. The ’square-root raised
cosine pulse’ is
(1 − α)sinc( Tt (1 − α)) 4α cos(π Tt (1 + α))

gsrrc (t) = +
1 − (4α Tt )2 π(1 − (4α Tt )2 )
Performances

-Lecture 4-
2009
Baseband communications : real channel
Real (linear) channel
Real channel, facts :
I Channel impulse response 6= δ(t − τ ) (Except on restricted band

: case of multiuser freq. multiplexing)
I Baseband signal PSD has infinite freq. support ⇒ multi-user
interferences : coder output MUST be filtered.
I Physical channel introduces
√ attenuation, dispersin; e.g. coax.
cable or paired wires ( f attenuation).
I Channel selectivity, dur to multiple paths ... (for modulations with
carriers)
Real (linear) channel
Root-Nyquist
No ISI condition : filtering
Accounting for the (linear) channel dispersion :
where
(h1 ? h2 )(t) = h(t) must satisfy Nyquist criterion :
P P
h(t) k δ(t − kT ) = δ(t) ⇔ H(ν) ⊗ T1 k δ(ν − Tk ) = 1
P
⇔ T1 k H(ν − Tk ) = 1
I This warrant the existence of a unique t0 over each time interval

T
I In general : P
y (t0 + nT ) = an r (t0 ) + k 0 6=0 an−k 0 r (t0 + k 0 T ) + w(t0 + nT )
where (n − k ) = k 0 , r (t0 ) = g ⊗ h1 ⊗ h2 )(t)
P 0
k 0 6=0 an−k 0 r (t0 + k T ) = ISI TERM
I
EYE Diagram
EYE diagram
Definition : X
z(t) = r (t − mT )
m
Definition : Eye diagram : set of all possible trajectories of z(t) over a

time interval T
Csq : if r (t) 6= O over [t0 − L1 T , t0 + L2 T ], then ∃(L1 + L2 + 1) different

sample segments of z(t)
(No ISI here

T
at t0 = 2)
EYE Diagram
EYE Diagram, contd
I if r (t) satisfies Nyquist criterion, not r (t) + w(t).

I Underlying hypothesis of perfectly synchronized system doest
not generally hold
k I∈ In general,
{nP − P2 ,Nyquist + P −1
· · · , nPcriterion
2 }
is not strictly met P ≥8
Consequences :
EYE Diagram
EYE Diagram, contd
The eye diagram accounts for ALL possible segments of z(t) ⇒ it is

T-periodic Important remarks :
I The eye diagram accounts for ALL possible segments of z(t) ⇒
it is T-periodic
I Satisfying Nyquist criterion IMPOSES
1
Freq support (R(ν)) ≥ T
1
"One cannot send a sequence of symbol at a rate of T over a frequency
bandwith smaller than T1 "
EYE Diagram
EYE Diagram, contd

Example for the square root raised cosine (r (t)) :
Constellations
Constellation diagrams
This is the plot Im[y (n)] vs Re[y (n)] for many integers n.
(Reminder : y (n) is the complex baseband representation of the
received signal, a(n) ∈ C)
if everything works well, eq for QAM16 or PAM4 :
∈ C a [n] ∈ R
Constellations
Constellation diagrams,
Im{ycontd
↑ [n]} Re{y↑ [n]}
Complex trajectory of the received signal : This is the plot Im[y (t)] vs
Re[y (t)] for all possible sequences (here QAM16, SRRC)
Making decision at the receiver
Decision regions
Remind the most popular modulations , and associated variance
(uniform proba over the alphabet)
∆
2
n] | }
∆2
! " ∆2
! " ∆2
σ 6 M2 − 1 12 M2 − 1 4sin2 (π/M )
Decision regions
Decision rule (DR) :
Nearest Neighbor mapping

y (n) −→ a(n) ∈ Ω
I Consequence :
Decision regions = Voronoi diagram of the constellation.
I Definition : Symbol Error Rate (SER) :
Proba[DR[Y (n)] 6= a|a(n) = a]
I for M-PAM, and gaussian noise (zero-mean, σn2 )
s !
M −1 3σa2
SERMP AM = erfc
M 2(M 2 − 1)σn2
M
M (M 2 − 1) σe2
Decision regions, contd

y [n] a [n] e [n] Re{e [n]} Im{e [n]}
I SER for M -QAM, circular white gaussian noise (zero-mean, σn2 )
2
σ2
additive σe2
e [n]⇒noise variance on Im[r ] and
[n]} on Re[r ]
n
σ I
2 Re{e [n]}
2 Im{e
I 4 (M − 2)
integration on C-plane M 2 − 4M + 4
I P [n] = a}4(M1−P
corner|apoints,
4 r{ |a [n] =M
− 2)r{edge points, a}2 − 4M + 4 interior points.
calculate Proba[Error |a[n] = a] = 1 − Proba[correct|a[n] = a],

(simpler).
" s !#2
(M − 1) 3σa2
SERM 2 −QAM =1− 1− erfc
M 2(M 2 − 1)σn2
log 2 M
SER and Bit Error Rate : Gray coding
y [n]
≈ if |Ω| = M 1 symbol error causes potentially M bit errors! Gray coding
allow to impose BER ' SER
Diversity coding : spread spectrum methods
Motivations, goals
Originally :
- Provide robustness wrt jamers (military or secured communications)
- Lower probability of interception by lowering PSD of emitted signals
Modern applications :
- Robustness wrt echoes (multipaths), multi-users interferences
- CDMA, FDMA
bandwidth in excess of the minimum necessary to send the information; the band
spread is accomplished by means of a code which is independent of the data, and a
2. synchronizedofreception
Overview with
Spread the code at techniques
Spectrum the receiver is used for despreading and
Spread spectrum technique overview
subsequent data recovery”. [1]
A 1980’s
Thus,formal definition
three elements of SS technology
adequately that holds unaltered
describe SS communications [2]: until today is:
Spread spectrum
! The
“Spread spread is
spectrum signal occupiesof a transmission
a means bandwidth much in greater
which than the minimum
the signal occupies a
bandwidth bandwidth required
in excess to send
of the the information.
minimum necessary to send the information; the band
spread! isThe signal is spread by means of a code which is independent of the data. The
accomplished by means of a code which is independent of the data, and a
same code is used at the receiving end to demodulate the data. This differs from
synchronized reception
standard with
modulation thedata
where code at the receiver
modulation spreads theisspectrum
used for despreading
in some way. and
subsequent
! The data recovery”.
receiver [1] to the code to despread the signal via correlation which
synchronises
is then passed through a narrow bandpass filter to recover the original data.
Thus,
R.L. three elements
Pickholtz, adequately
D.L. Schilling a.describe SS communications
L.B. Milstein, ”Theory of [2]:Spread-Spectrum
Figure 1 illustrates the basic principle behind SS communications where f s denotes
Communications-A Tutorial”, IEEE Transactions on Communications, vol.
! The spread
the data signal occupies
rate (expressed as Rd later)aand
bandwidth
f c the code much greaterasthan
rate (expressed the minimum
Rc later).
Com30, no. 5, May 1982, pp. 855-884
bandwidth required to send the information.
! The signal is spread by means of a code which is independent of the data. The
same code is used at the receiving end to demodulate the data. This differs from
standard modulation where data modulation spreads the spectrum in some way.
! The receiver synchronises to the code to despread the signal via correlation which
is then passed through a narrow bandpass filter to recover the original data.
Figure 1 illustrates the basic principle behind SS communications where f s denotes

the data rate (expressed as Rd later) and f c the code rate (expressed as Rc later).
Figure 1: Frequency spreading illustrating principle of SS technology1

Direct Sequence Spread spectrum
3.6. PRINCIPES DU SPECTRE ÉTALÉ À SÉQUENCE DIRECTE (DSSS) 33
DSSS principle
3.6 Principes du spectre étalé à séquence directe (DSSS)
30 CHAPITRE 3. LES COUCHES PHYSIQUES UTILIS ÉES EN SANS-FILS
Le cas le plus simple d’une chaı̂ne de transmission à spectre étalé est celui de la mo-
Increasing artificially
dulation BPSK,the
donnéedata rate
dans la figure 3.16.⇔ spreading the spectrum
data in Ts data out
0 Ts
cos ωt cos ωt
PN gen. PN gen.
sin ωt
Fig. 3.16 – Chaı̂ne de transmission DSSS-BPSK
I
Entrées
S :
data Les/ entrées du système peuvent être vues comme :
– Les Q
données (dt ), cadencées à un rythme Rs = 1/Ts .
P
– Le code pseudo-aléatoire (pn t ), cadencé à un rythme Rc = 1/Tc = Nc Rs
Etalement :
L’étalement consiste cos ωt
PN gen.en la multiplication pure et simple du signal de données BPSK
par le code pseudo-aléatoire (voir figure 3.15). L’effet de la multiplication du signal d t par
le code pnt est d’étaler, en bande de base, le signal de largeur de bande ! R s sur une
bande Rc par un facteur Nc , où Nc est le nombre de chips par symbole (pour des codes
court, c’est la longueur de symbole). On dit également que N c est le ”processing gain”,
i.e. le facteur de proportionalité entre la largeur de bande du signal de données et celle du
signal étalé.
Il est également très important de signaler que, parallèlement à l’élargissement du
spectre, la densité spectrale est diminuée par le facteur N c , en effet, la puissance totale
est conservée, et comme celle-ci est égale à l’ı́ntegrale de la densité spectrale de puissance,
cette densité spectrale de puissance doit diminuer 3 .
Désétalement :
Au récepteur, le signal étalé doit être multiplié par la même séquence PN qu’à l’émetteur
pour être détecté :
– Si pnr = pnt (et les deux séquences sont synchronisées), alors le signal binaire peut
−Rsêtre
0 Rrécupéré.
s L’effet de cette multiplicationfcest −R fc le signal,
dec désétaler fc +c’est-à-dire
Rc de
ramener la largeur de bande du signal à R s .
s’écrire sous différentes formes :
BWSS Rc Tb
Diversity coding : spread spectrum methods SF = Gp = = = = Nc
BWinf o Rs Tc
data in Ts data out
DSSS principle PN gen.

cos ωt cos ωt
PN gen.
0 Ts
TEMPS FREQUENCE
+1
dt
−Rs Rs
-1
pnt
−Rc Rc
txb
−Rc Rc
Fig. 3.17 – Principe de la modulation DS-BPSK-SS

signal aiming to overcome narrowband interference. The receiver in turn knows how
to generate the same code in order to extract the original data.
Appendix
Direct Sequence 1 provides
Spread spectrum a thorough coverage of the approach used to generate the data
and code sequences presented in this Assignment. In brief, these sequences are
generated via convolution of a series of pulses with a rectangle in the time domain
DSSS principle (TD), equivalent to multiplying their spectrums in the frequency domain (FD).
In Figure 4, we consider 4 cycles of a data signal with a period Td ! 1024 samples,

along: with
Example chipa code used 16*(bit
rate= to spread the signal having a period Tc ! 64
rate) samples.
Data
Code
1
1
Amplitude
0
0
-1
-1 0 200 400 600 800 1000
0 1000 2000 3000 4000 ZOOM
Data and code sequences in the time domain
5
x 10
10 4000
Amplitude
X: 0 X: 0
Y: 1.049e+006 Y: 4096
5 2000
0 0
-50 0 50 -50 0 50
Power spectrums of data and code rectangles ZOOM
6 6
x 10 x 10
2
6
Amplitude
4
1
2
0 0
-50 0 50 -50 0 50
Power spectrums of data and code sequences ZOOM
Figure 4: Data and code sequences in the TD and FD for a DSSS system
For DS systems, the G p is expressed as the ratio of the chip rate (or information bit
Diversity coding : spread spectrum3.5. PRINCIPES DE BASES : DSSS ET FHSS
methods 31

un code pseudo-aléatoire (qui peut être ou ne pas être différent sur les deux branches).
Le débit Rc du code pseudo-aléatoire (appelé chip rate) est nettement plus élevé que le
débit Rs des symboles de données et Rc = N Rs . La bande passante du signal résultant (à
FHSS principle spectre étalé) est à peu près égale à R c .
Lorsque le code pseudo-aléatoire est de longueur égale à la durée d’un symbole de
donnée, on parle de code court, lorsque le code est plus long que la durée d’un symbole,
on parle de code long. En général, le code pseudo-aléatoire a une durée qui est un multiple
PN generator drives instantaneous frequency :
entier de la durée d’un symbole.
sin ωpn t
data Modulateur
M-FSK
cos ωpn t
PN gen.
n
ωpn = ωc + P N ∗ ∆ω
hop
1 2 3 4 ... N
−Rs 0 Rs
Wss = NWbb = 2 ∗ N ∗ Rs
Fig. 3.14 – Principe du spectre étalé à saut de fréquence

On notera
Introduction to Digital communications que4-le fait de sauter d’une fréquence à l’autre se fait le plus souvent avec
-Lecture
Diversity coding : spreaddiscontinuité de phase, le récepteur utilisera, pour une fréquence de saut pas trop lente
spectrum methods
(i.e. si on saute persque aussi souvent d’une fréquence á l’autre que le débit symbole), une
Direct Sequence Spread spectrum non-cohérente.
démodulation
On notera également qu’on définit un systéme á saut rapide comme étant un systéme
Main types of DSSS and FHSS ou l’on change de fréquence plus vite que le débit symbole, et á saut lent sinon.
DSSS FHSS
0 1 1 0 0 1 1 0
dt dt
−fo
fh4
PN
fh3
fh2
P N ∗ dt
fh1
h
Nc Tc hop
Code Court saut Saut rapide
Tc
0 1 1 0
dt dt
fh4 +fo
PN
fh3
fh2
P N ∗ dt
−fo
fh1
Nc Tc
Code Long Saut Lent
Tc
chip Th Tc
saut Chip
Definition :
Fig. 3.15 – Principaux types de DSSS et FHSS
Wss Rc
Spreading Factor = Wd = Rd = SF
Where W = bandwidth, Rc =Chip rate, Rd =symbol rate
DSSS : despreading 3.6. PRINCIPES DU SPECTRE ÉTALÉ À SÉQUENCE DIRECTE (DSSS) 35

additif Gaussien comporte un intégrateur bloqueur après l’opération de désétalement. In-

tuitivement, cet intégrateur permet de ”moyenner” le bruit et de s’affranchir au maximum
de l’effet d’un leger asynchronisme entre les codes.
dr
data in rxt Ts data out dr
data in rxt Ts data out
0 Ts
0 Ts
Correct decoder PN sequence

PN gen. PN gen.
Different decoder PN sequence

PN gen. PN gen.
TEMPS FREQUENCE
pnr = pnt TEMPS FREQUENCE
pnr != pnt
Tc
Tc
rxt
−Rc Rc f rxt
−Rc Rc f
Tc
Tc
pnr
−Rc Rc f pnr
−Rc Rc f
Ts
+1
+1
dr
f dr
−Rs Rs
−Rc Rc f
-1
-1
Fig. 3.18 – Principe de la démodulation DS-BPSK-SS

Fig. 3.19 – Principe de la démodulation DS-BPSK-SS
La figure 3.19 montre clairement que si les deux codes pseudo-aléatoires sont différents,
le signal en sortie varie toujours à un rythme R c , ce qui ne permet pas de détecter le signal
reçu et n’effectue pas d’opération de désétalement.
La figure 3.20 illustre l’effet d’un léger asynchronisme entre la séquence pseudo-aléatoire
à l’émetteur et la séquence reçue. Le léger asynchronisme a pour effet de créer des glitches
dans le signal résultant de la multiplication du signal reçu et de la séquence PN décalée.
Après intégration et échantillonnage au temps T s , on observe une valeur positive ou
négative qui sera fournie à un organe de décision. On voit clairement que l’intégrateur/échantillonneur
Effect of a synchronization lag in the PN Code at the organe de décision
receiver data in rxt dr
0
Ts
Ts
data out
PN gen. PN gen.
TEMPS FREQUENCE
pnr (t) = pnt (t − τ ) : léger asynchronisme
Tc
rxt
−Rc Rc f
Tc
pnr
décalé
−Rc Rc f
Ts
+1
dr
−Rs Rs f
-1
Après intégration idéal : Nc

on décide dˆ = 1
idéal : -Nc
Fig. 3.20 – Principe de la démodulation DS-BPSK-SS : effet d’une désynchronisation de

3.7. DSSS : EFFET DU BRUIT ADDITIF, DE L’INTERF ÉRENCE ET D’UN CANAL MULTICHEMINS39
Effect of a DSSS and white gaussian additive noise

Ts
data in
PN gen.
−Rs Rs f
Tc
−Rc Rc f
N(t)
Tc
−Rc Rc f
rxt
Bruit Blanc de densité spectrale de puissance No
dr PN gen.
Ts
0
Ts
Ts
−Rs Rs f
data out Bruit Blanc de densité spectrale de puissance No
organe de décision
SS 0 jam jam
proportional to G p . This interference suppression can be summarised as follows: [4]

Interference rejection, multipath channels
1. Multiplication by the spreading code once spreads the signal W.

2. Multiplicationrejection
Interference by the spreading code twice recovers the original signal, while the
interference signal gets multiplied once, therefore it is spread.
This concept is demonstrated in Figure 8.

Assume the interferer has constant PSD Io over its bandwidth Winterf :
5
SNRssFigure
= Io W8:
S DSSS interference rejection capability
interf
SNRdespread = Io W S
d
!"
SNR gain
R C Grammenos: Study and simulation of SS technology = SF Laboratory Assignment 2008
7
gives an example of the various sources of interference that can d

signal but which can be reduced using DSSS. The system in F

Various Distorsions
DS-CDMA (examined in athrough SS transmitter
later Section) systems
to recover the original data.
e 9: Overview of DS-CDMA spreading process and interference rejec

SS summary
Figure 2: Principle of a) DSSS2 (left) and b) FHSS3 (right)
The block diagram of a typical SS system is shown in Figure 3.

Figure 2: Principle of a) DSSS2 (left) and b) FHSS3 (right)
The block diagram of a typical SS system is shown in Figure 3.
Figure 3: Simplified version of a practical SS system4

Summary of SS methods benefits
I Interference rejection (immunity to multipath fading, jamming

resistance).
I Energy density reduction (low probability of intercept)
I Fine time resolution (ranging, position determination, accurate
universal timing).
I Multiple access (resource sharing, selective addressing).
One question : DSSS or FHSS ?
Near-Far Effect : Emitter B much closer to Receiver than Emitter A =>
Received power from B (even with orthogobal DSSS PN sequence)
may mask signal from A
⇒ FHSS prefered (e.g. GSM)
Pseudo noise sequences basic properties
Pseudo random sequence
Important to notice : Pseudo random sequence behaves like noise,

although it is fully deterministic.
Main properties
I Balanced code : number of +1 ' number of −1 ⇒ code mean
'0
R N T /2
I autocorrelation : RPN (τ ) = −Nc c cTc /2 PN(t)PN(t − τ )dt should be
as close as possible to δ(t)
I crosscorrelation
R N T: /2
RPNi PNj (τ ) = −Nc c cTc /2 PNi (t)PNj (t − τ )dt ' 0∀τ →
I ’orthogonality’ between PN sequences if RPNi PNj (0) = 0
I More interesting : RPNi PNj (τ ) ' 0 ∀τ
Examples
I Balanced code :
PN = +1 +1 −1 +1 −1 −1 −1 → Σ = 0
I Auto-correlation
PN(0) = +1 +1 −1 +1 −1 −1 −1
PN(0) = +1 +1 −1 +1 −1 −1 −1
→
− − − − − − −− −−−−−−−−−−−−−−−−−
+1 +1 +1 +1 +1 +1 +1
RPN (0) = 7
I Cyclic auto-corr
PN(1) = +1 −1 +1 −1 −1 −1 +1
PN(0) = +1 +1 −1 +1 −1 −1 −1
→
− − − − − − −− −−−−−−−−−−−−−−−−−
+1 +1 −1 −1 −1 +1 −1
RPN (1) = −1
pn(0) = -Lecture
Introduction to Digital communications +1 +1 4-+1 -1 +1 -1 -1
pn(0) = +1 +1 +1 -1 +1 -1 -1
-----------------------------
+1 +1 +1 +1 +1 +1 +1 --> Ra(0)=7
pn(0) = +1 +1 +1 -1 +1 -1 -1
Examples, contd pn(1) = +1 +1 -1 +1 -1 -1 +1
-----------------------------
+1 +1 -1 -1 -1 +1 -1 --> Ra(1)=-1
Rapn (τ )
7
-1
RPN (τ ) :
I Cross-correlation
La raison pour laquelle on désire une autocorrélation la meilleure possible est que,
outre
+1 que+1 −1 −1
cette propriété est celle+1 +1blanc,−1
d’un bruit −1 également d’aider à la
elle permet
synchronisation et à l’estimation du canal multichemin dans le cadre du récepteur RAKE.
+1 +1 +1 +1 −1 −1 −1 −1
→
−− −− −−
Cross-corrélation −− −−
La cross-corrélation décrit −− −−entre−−
l’interférence deux codes différents :
! N .T /2
+1 +1 −1 −1 −1 −1 +1 +1
c c
Rcpn (τ ) =
ij pni (t).pnj (t + τ )dt (3.57)
RPNi PNj (0) = 0 −N .T /2
c c
On parle de codes orthogonaux si la fonction de cross-corrélation entre codes est

égale à 0 pour τ = 0.
Exemple 3.9 codes orthogonaux

Applications of orthogonality in PN sequences
I Orthogonal codes do not ’interfer’ in despreading process =>

mutli-user capabilities
I orthogonal codes often do not enjoy good auto /
cross-correlation properties for τ 6= 0
⇒
I design short orthogonal code sequences (allow to separate users)
I design long code sequences (with good cross and auto corr
properties (good transmission properties)
I Multiply the sequences to built a code with both properties

-Lecture 5-
2009
Multi-carrier modulations : OFDM
Multicarrier modulations principles
Motivations
Case of a unique carrier transmission system :

1
I Symbol rate Rs = Ts
I Echoes, multipaths, diffusion, difraction ⇒ time leakage of a
given symbol over ' MTs , M large
I Channel equalization complex
I the EYE diagram is almost closed ⇒ detection problems,
decision errors...
Pb related to unique carrier transmission : example
I Assume
I an optical path delay ∆l = 100m, radio waves,
I Rs = 100Msymbol/s
I then delay τ = 300ns, equivalently τ = 30 symbols
I Equalization FIR filter of order N ' 2τ = 60
I Compuational load = 60/Ts (C-values) = 240
multiplications/additions per symbol = 24 Gops/s !!!
Multicarrier solution
I Transmit Nc in parallel (using Nc sub-channels), each with

duration Tc = Nc Ts
I Each channel has carrier frequency fi = f0 + ∆f , of width
Wc = WNc where Ws = spectral width in the mono-carrier case.
s
I Equalization of each sub-channel is much simpler as

I Wc << Ws ⇒ less fluctuations over Wc
I delay is constant in time, but much mower as expressed in symbols
⇒ lower order FIR equalizer
I If Nc >> 1, each equalizer involves only one multiplication !
I Requires Tc >> τ , i.e. large Nc
I Channel Coherence width Wb ' τ1 , then
Wc = W
Nc << Wb ⇒ Nc >> 1
s
Multicarrier signal expression
!
X c −1
NX
s(t) = dm,k ψm (t − kTc )
k m=0
Major requirement : avoid inter sub-channel interferences

I separate the sub-channel spectral bandwidth → low global
spectral efficiency
I involves complex / expensive mixing an modulator devices
OFDM
OFDM solution
Allow overlapping frequency bands, but no interences, then carriers
signals must verify
Z Tc
?
ψm (t − kTc )ψm 0 (t − kTc )dt = δm,m0
0
Classical simple solution :

ψm (t) = √1T exp(j2πfm t) si t ∈ [0, Tc [
c
0 sinon
where fm = f0 + mδf = f0 + m W
Nc , f0 being the first sub-channel central
s
freq.
OFDM
OFDM solution, contd

as
Z (k +1)Tc
?
ψm (t − kTc )ψm 0 (t − kTc )dt
kTc
Z Tc
1
= exp(j2π(fm − fm0 )(t))dt
0 Tc
sin(π(fm − fm0 )Tc )
=
π(fm − fm0 )Tc
orthogonality is met if (fm − fm0 )Tc = l, l ∈ Z, or equivalently
Ws Ws
(fm − fm0 )Tc0 = (m − m0 )Tc = (m − m0 )Nc Ts
Nc Nc
this implies
Ts Ws = 1
OFDM
Pulse shapes
ψm puse shapes and assocoated spectral representations
OFDM implementation
OFDM implementation
Using complex orthogonal exponentials leads to

!
X c −1
NX
1
s(nTc ) = dm,k √ exp(j2πfm (n − k )Tc )
k m=0
Tc
!
X exp(j2πf0 (n − k )Tc ) NX c −1
mn
= √ dm,k exp(j2π )
Tc Nc
k m=0
| {z }| {z }
Delay of OFDM symbols IFFT of dk ,m sequences, of length Tc

of duration Tc Requires Nc log2 Nc ops (Cooley Tuckey)
OFDM implementation
FT based OFDM modulation system

OFDM implementation
OFDM performances
I For WGN additive channel, same perf. as single carrier

modulation
I Perf. degrades for freq. selective channels : attenuated
sub-channels will have high SER/BER (as high as 0.5!)
I Makes error correcting codes compulsory to reach singe carrier
equiv. perf., with lower implementation cost
examples : WiFi 802.11*, WLAN, ADSL

Introduction To Digital Communications: Olivier - Michel@

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Introduction To Digital Communications: Olivier - Michel@

Uploaded by

Copyright:

Available Formats

Introduction to Digital communications

Introduction to Digital communications

ENSE3 Grenoble-INP, France

Typical communication media

Shannon theory / coding, redundancy = perfect transmission is

Modulator : from a ’baseband’ or lowpass signal, to a

widely used freq. ranges

INPG TST 3A J.M. Brossier - LIS

Ideal sinc solution

Ideal sinc solution

Strictly rectangular spectrum hard to implement:

INPG TST 3A J.M. Brossier - LIS ‘

Accounting for the (linear) channel dispersion (next

where (h1 ? h2 )(t) = h(t) must satisfy Nyquist criterion !

Linear model of signal propagation

Dispersion (phase distorsion), selective attenuation, multipath ....

Selective filtering for SNR improvment

Multipath filter model (no fading)

x (t) = s (t) ∗ h (t)

INPG TST 3A J.M. Brossier - LIS

! NLOS (No Line of Sight, urban) : CLT:

INPG TST 3A J.M. Brossier - LIS

INPG TST 3A J.M. Brossier - LIS

Fourier spectrum of a periodic deterministic signal

Fourier spectrum of a random noise (estd)

LTI in frequency domain

Ideal delayed LPF :

Real causal linear phase LPF, with group delay t0

We can do better : see linear phase FIR filter design...(last year

Sy (ν) = |H(ν)|2 Sx (ν)

A briel review of AM modulation

Rk : if m(t) is real, AM transmitted spectrum is redundant : motivation

= {m (t) + m (t) cos (2π · 2fc t)}

v (t) = m(t) cos(2πδft + δφ)

(time varying attenuation : left as exercice)

= [m (t) + A] cos (2πfc t)

s(t) = (m(t) + A) cos(2πfc t)

modern systems : A << max(m(t))

Rk : Carrier - AM transmitted spectrum is redundant, consumes

zx (t) = x(t) + j.H[x](t)

Rk: x(t) real signal, but zx (t), ax (t) complex-valued signals.

Consequences of previous definitions

H[x](t) ∈ IR, x(t) ∈ IR ⇒ x(t) = Re[zx (t)]

x(t) = Re[ax (t)ej2πν0 t ] H[x](t) = Im[ax (t)ej2πν0 t ]

As ax (t) ∈ C ⇒ ax (t) = px (t) + j.qx (t)

x(t) = px (t) cos(2πν0 t) − j.qx (t) sin(2πν0 t)

Let f (t), g(t) such that :

Let x(t) = m(t) cos(2πν0 t + φ), with ∆M(ν) << ν0

The complex enveloppe (relatively to ν0 ) of x is

Introduction to Digital communications

ENSE3 Grenoble-INP, France

Let f (t), g(t) such that :

The complex envelope (relatively to ν0 ) of x is

Actually for e.g. AM, transmitted signal is

As H[cos(2πν0 t + φ)] = sin(2πν0 t + φ), it comes zx (t) = e2πν0 t+φ , the

Application to real valued passband signals

Def : x(t) is a deterministic real passband signal if ∃B ∈ IR+ s.t.

Equalities for real passband signals

Exercice : As x(t) = px (t) cos(2πν0 t) − qx (t) sin(2πν0 t), prove that

I Ax (ν) = 2X + (ν + ν0 ) = Px (ν) + jQx (ν)

Spectral interpretation, real passband signals -1-

Spectral interpretation, real passband signals -2-

px (t) = x(t) cos(2πν0 t) + xh (t) sin(2πν0 t)