Mobile Communication Slide

Abstract of the course
Content of the course

Course requirements and references
Mobile Communications
Instructor: Nguyen Le Hung
Email: nlhung@dut.udn.vn; nglhung@gmail.com
Department of Electronics & Telecommunications Engineering
Danang University of Technology, University of Danang
Mobile Communications Course Information 1
This undergraduate course helps students to understand
mathematical fundamentals and practical transmission
techniques in 4G mobile communications (i.e., WiMAX,
LTE).
The course lecture notes also provide some possible
research directions (in 4G mobile broadband
communications) that can be considered for nal-year
projects of undergraduate students.
Chapter 1: Introduction
History & development of 1G/2G/3G/4G networks.
Promises and future trends
Cellular mobile communications
Chapter 2: Mobile wireless channel models.
Path loss
Shadowing
Multipath fading channels
Chapter 3: Physical-layer transmission techniques.
Digital modulations
Performance of digital modulations over fading channels
Orthogonal frequency division multiplexing (OFDM)
fundamentals
Pre-requisite: Basis knowledge of statistics, stochastic
processes and digital communications systems.
Class lecture notes: based on the following references:
Gordan L. St uber, Principles of Mobile Communication,
Second Edition, 2002
A. Goldsmith, Wireless Communications, Cambridge
2005.
Recent IEEE journal and conference papers.
Course assessment:
Exercises and/or projects: 20%
Midterm exam: 30%
Final exam: 50%
Outline
Introduction
Chapter 1: Introduction to Mobile
Communications
Nguyen Le Hung
Mobile Communications Chapter 1: Introduction to Mobile Communications 1
Outline
Introduction
Outline of Chapter 1
1
Introduction
Development of mobile communication systems
Mobile broadband technology evolution
2
System model
Frequency reuse
Cellular concept
Outline
Introduction
Development of mobile communications systems
time
code
frequency
code
space
FDMA (1G)
e.g., AMPS ~ 1980s
TDMA (2G)
e.g., GSM ~ 1990s
OFDM, SDMA (4G)
e.g., WiMAX, LTE
2010s
CDMA (3G)
e.g., W-CDMA ~ 2000s
frequency
time
time
~ 1 Gbps (stationary),
~ 100 Mbps (mobile)
frequency
frequency
~ 14 Mbps (downlink),
~ 5.8 Mbps (uplink)
~ 50 Kbps
A new signal dimension will be exploited in 5G ?
Outline
Introduction
OFDM versus FDMA
Frequency
Outline
Introduction
LTE
Advanced
802.16m
R 2.0
WCDMA
3GPP
HSDPA HSPA
Mobile
WiMAX
802.16e
R 1.0
EVDO
Rev A
EVDO
Rev B
CDMA 2000
3GPP2
LTE
802.16e
R 1.5
2005 2009 2010 2006 2007 2008 2011 2012
CDMA based OFDMA based
R99
4G
IMT-Advanced
Outline
Introduction
multimedia services: Voice, Video distribution, Real-
time videoconferencing, Data, for both business
and residential customers:
Explosive traffic growth
Internet growth, VoIP, VideoIP, IPTV
Cell phone popularity worldwide
Ubiquitous communication for people and devices
Emerging systems opening new applications
Unified network: Single distributed network,
multiple services, packet architecture
Extracted from Digital Communication lecture notes, McGill Uni.
Outline
Introduction
System model
Frequency reuse
Cellular concept
System model of cellular mobile communications
BTS
LTE/LTE Advanced
Single Cell
Multicell
approach using
game theory
Uplink (SCFDMA),
limited feedback design
Downlink (OFDMA)
SingleUser
Multiuser
Precoding
(SDMA)
Multihop
Relay
BTS
BTS
BTS
BTS
Intercell
interference
Intercell
interference
Singleuser/Multihop:
Channel Estimation,
Synchronization (CFO),
Channel Coding, ...
Network Controller
STBC with highspeed users
(large Doppler spread)
Cognitive radio
Space Time Block Code: STBC; PeaktoAverage Power Ratio: PAPR; VOFDM
Usercooperation
(cooperative/multihop
communications)
Outline
Introduction
System model
Frequency reuse
Cellular concept
Frequency reuse
The available spectrum is partitioned among the base stations
(BSs).
A given frequency band is reused at the closest possible
distance under a certain requirement of co-channel
interference.
Smaller cells have a shorter distance between reused
frequencies = an increased spectral eciency.
Microcells are of great importance in improving spectral
eciency.
Under frequency-reuse, users in geographically separated cells
simultaneously employ the same carrier frequency.
Outline
Introduction
System model
Frequency reuse
Cellular concept
Cellular concept
The cellular layout of a conventional cellular system is quite
often described by a uniform grid of hexagonal cells or radio
coverage zones.
In practice the cells are not regular hexagons, but instead are
distorted and overlapping areas.
The hexagon is an ideal choice for representing macrocellular
coverage areas, because it closely approximates a circle and
oers a wide range of tessellating reused cluster sizes.
A tessellating reuse cluster of size N can be constructed if
=
2
+ +
2
, (1)
where and are non-negative integers and . It follows
that the allowable cluster sizes are = 1, 3, 4, 7, 9, 12, . . ..
Outline
Introduction
System model
Frequency reuse
Cellular concept
Cellular concept: Multicell layout with frequency-reuse
3-cell
4-cell
7-cell
Macrocellular deployment
with 7-cell clusters
Macrocellular deployment
with 3-cell clusters
Introduction
Wireless channel modeling
Chapter 2: Wireless Channel models
Mobile Communications Chapter 2: Wireless Channel models 1
Introduction
Multipath wireless propagation
Path loss, shadowing and fading
reflection and diffraction
Extracted from Digital Communication lecture notes, McGill Uni.
Introduction
The characteristic of (mobile) wireless channel is the variations of
the channel strength over time and frequency.
The variations can be divided into two types:
Large-scale fading is yielded by:
path loss of signal as a function of distance and
shadowing by large objects such as buildings and hills.
Small-scale fading is yielded by the constructive and destructive
interference of the multiple signal paths between transmitter and
receiver.
Introduction
An example of path loss, shadowing and fading
0 50 100 150 200 250 300 350
-150
-140
-130
-110
-100
-90
-80
-70
-60
-50
R
e
c
e
i
v
e
d
P
o
w
e
r
[
d
B
m
]
Traveled distance [m]
Pathloss
Fading +
Shadowing +
Pathloss
Shadowing +
Pathloss
Introduction
An example of path loss, shadowing and fading (cont.)
0
K (dB)
P
r
P
(dB)
t
log (d)
Path Loss Alone
Shadowing and Path Loss
Multipath, Shadowing, and Path Loss
Introduction
Path loss models
Shadowing
Fading channel model
Path loss models
It is well known that the received signal power decays with the
square of the path length in free space.
More specically, the received envelope power is
1
:
= 1
|
G
|
G
:
(
X
4tJ
)
2
, (1)
where:
1
is the transmitted power,

G
and G
are the transmitter and receiver antenna gains,

respectively
d is the radio path length.
Introduction
Path loss models
Shadowing
Path loss models (cont.)
The signals in land mobile radio applications, however, do not
experience free space propagation. A more appropriate theoretical
model assumes propagation over a at reecting surface (the earth).
1
:
= 41
|
(
`
t
4d
)
2
G
|
G
:
sin
2
(
2
b
n
`
t
d
)
, (2)
where
b
and
n
are the heights of the BS and MS antennas,
respectively.
Under the condition that d
b
n
, (2) reduces to
1
:
= 1
|
G
|
G
:
(
`
t
4d
)
2
, (3)
where we have used the approximation sin r r for small r.
Introduction
Path loss models
Shadowing
Path loss models (cont.)
The path loss is dened by
1
(J1)
= 10 log
10
(
1
|
G
|
G
:
1
:
)
= 10 log
10
{
4
(
`
t
4d
)
2
sin
2
(
2
b
n
`
t
d
)
}
(4)
Several useful empirical models for macrocellular systems have been
obtained by curve tting experimental data.
Two of the useful models for 900 MHz cellular systems are:
Hatas model based on Okumuras prediction method and
Lees model.
Hatas empirical model is probably the simplest to use. The
empirical data for this model was collected by Okumura in the city
of Tokyo.
Introduction
Path loss models
Shadowing
Okumura-Hata models
With Okumura-Hatas model, the path loss between two isotropic
BS and MS antennas is
1
(J1)
=
+1log
10
(d) for urban area
+1log
10
(d) C for suburban area
+1log
10
(d) 1 for open area
(5)
where
= 69.55 + 26.16 log
10
()
t
) 13.82 log
10
(
b
) o(
n
)
1 = 49.9 6.55 log
10
(
b
)
C = 5.4 + 2 (log
10
()
t
,28))
2
1 = 40.94 + 4.78 (log
10
()
t
))
2
18.33 log
10
()
t
)
Introduction
Path loss models
Shadowing
Okumura-Hata models (cont.)
and
o
[1.1 log
10
()
) 0.7]
1.56 log
10
()
) + 0.8 for medium or small city

{
8.28 [log
10
(1.54
)]
2
1.1 for )
200MHz
3.2 [log
10
(11.75
)]
2
4.97 for )
400MHz
for large city
(6)
Okumura-Hatas model is expressed in terms of:
the carrier frequency: 150 )
1000(MHz),
BS antenna height: 30
200(m),
the mobile station (MS) height: 1
10(m),
the distance: 1 d 20(km).
Introduction
Path loss models
Shadowing
Numerical results of Okumura-Hata models
1 5 10 15 20
100
150
200
250
300
350
400
450
P
a
t
h

l
o
s
s

(
d
B
)

Distance d (km) under scale of log10
urban area
suburban area
open area
Figure 1: Path loss for
= 1.5m,
= 50m, )
= 900MHz.
Introduction
Path loss models
Shadowing
Shadowing
A signal transmitted through a wireless channel will typically
experience random variation due to blockage from objects in the
signal path, giving rise to random variations of the received power at
a given distance.
Such variations are also caused by changes in reecting surfaces and
scattering objects.
Thus, a model for the random attenuation due to these eects is
also needed. Since the location, size, and dielectric properties of the
blocking objects as well as the changes in reecting surfaces and
scattering objects that cause the random attenuation are generally
unknown, statistical models must be used to characterize this
attenuation.
The most common model for this additional attenuation is
log-normal shadowing.
Introduction
Path loss models
Shadowing
Shadowing (cont.)
Empirical studies have shown that A
n
has the following log-normal
distribution:
j
(r) =
2
ro
2
exp
{
(
10 log
10
r
2
j
(dBm)
)
2o
2
}
j
(r) =
2
ro
2
exp
{
(
10 log
10
r j
(dBm)
)
2o
2
}
where:
A
and A
denote the mean envelop and mean squared levels of

received signal (where the expectation is taken over the pdf of the
received envelope).
o
stands for standard deviation.

j
(dBm)
= 30 + 10[log
10
A
2
]
j
(dBm)
= 30 + 10[log
10
A
]
Introduction
Path loss models
Shadowing
Shadowing (cont.)
Sometimes A
n
is called the local mean because it represents the
mean envelope level where the averaging is performed over a
distance of a few wavelengths that represents a locality.
This model has been conrmed empirically to accurately model the
variation in received power in both outdoor and indoor radio
propagation environments
Introduction
Path loss models
Shadowing
Two Main
Multipaths
Local
Scattering
The complex transmitted signal can be expressed by
:(t) = Re
[
r(t)c
2t}
|
]
. (7)
Over a multipath (1 physical paths) propagation channel, the
received signal can be obtained by
j
1J
(t) =
J1
l=0
c
l
(t):(t t
l
(t)) +n(t). (8)
Introduction
Path loss models
Shadowing
Fading channel model (cont.)
Substituting (7) into (8) yields the following
j
1J
(t) = Re
[
J1
l=0
c
l
(t)r(t t
l
(t)) c
2t}
(|r
(|))
]
+n(t)
= Re
[(
J1
l=0
c
l
(t)r(t t
l
(t))
)
c
2t}
|
]
+n(t)
= Re
[
j(t)c
2t}
|
]
+n(t)
As a result, the received baseband signal can be determined by
j(t) =
.
c
.
(t)r(t t
.
(t)) +n
b
(t). (9)
where n
b
(t) is the receiver (thermal) noise signal.
Introduction
Path loss models
Shadowing
Wireless channel modeling (cont.)
The next step in creating a useful channel model is to convert the
continuous-time channel to a discrete-time channel.
We take the usual approach of sampling theorem.
Assuming that the input waveform is band-limited to \, the
baseband equivalent can be represented by
r(t) =
n
r
n
sinc(\t n), (10)
where r
n
= r(n,\) and sinc(t)
sin(t|)
t|
.
This representation follows from the sampling theorem, which says
that any waveform band-limited to \,2 can be expanded in terms
of the orthogonal basis functions sinc(\t n) with coecients by
samples (taken uniformly at integer multiples of 1,\)
Introduction
Path loss models
Shadowing
As a result, the baseband received signal can be determined by
j(t) =
.
c
.
(t)
n
r
n
sinc (\(t t
.
(t)) n) +n
b
(t)
=
n
r
n
.
c
.
(t)sinc (\(t t
.
(t)) n) +n
b
(t).
The sampled outputs at multiples of 1,\ is j
n
j(:,\) then
j
n
=
n
r
n
.
c
.
(:,\)sinc (:n t
.
(:,\)\) +n
b
(:,\).
(11)
Introduction
Path loss models
Shadowing
Let | :n then one can have
j
n
=
l
r
nl
.
c
.
(:,\)sinc (| t
.
(:,\)\) +n
b
(:,\)
Then, the discrete-time channel model can be given by
j
n
=
l
r
nl
l,n
+n
b
(:,\) (12)
where
l,n
=
.
c
.
(:,\)sinc (| t
.
(:,\)\)
This simple discrete-time signal model is widely used in
physical-layer transmission techniques in OFDM systems (e.g., WiFi,
WiMAX, LTE)
Introduction
Path loss models
Shadowing
Examples of transmitted baseband signal
:
01
00
10
11
I
+1 1
1
+1
Q
b
0
b
1
0
1
I
+1 1
1
+1
Q
b
0
11 10
11 11 10 11
10 10
I
+1 1
1
+1
Q
b
0
b
1
b
2
b
3
+3
11 01
11 00 10 00
10 01
+3
00 10
00 11 01 11
01 10
00 01
00 00 01 00
01 01
3
3
BPSK
QPSK
16-QAM
Over multipath channels, the received signal at MS is:
j
n
=
l
r
nl
l,n
+n
b
(:,\) (13)
It is noted that multipath fading gains
l,n
(channel impulse
response) is time-variant (depend on time index :).
Introduction
Path loss models
Shadowing
Channel estimation in mobile communications
Source
encoder
Channel
encoder
Digital
modulation
Channel
Source
decoder
Channel
decoder
Digital
demodulation
S
h
r = Sh + n
Pilot
S
Data
S
Data
S
Pilot
S
Data
S
Data
S
Pilot
S
h h h h h h
Introduction
Path loss models
Shadowing
Literature Review of Channel Estimation in Wireless
Communications
Detection/decoding
in communications
3 dB f
Rx signal
vector
Tx signal
matrix
CIR
vector
Rx noise
vector
Noncoherent Coherent without using CSI
3-dB performance
loss
use CSI
require Channel Estimation (CE)
vector matrix vector
(CSI)
vector
r = Sh + n
require Channel Estimation (CE)
with channel parameters as:
Deterministic unknowns Random variables
Fisher approaches:
LS ML
Bayesian approaches:
MMSE MAP LS, ML, MMSE, MAP,
Multipath fading channel (freq. selective) in multi-carrier transmissions (e.g.,OFDM)
Time-invariant (quasi-static) Time-variant (Time-selective)
Perfect
Synch.
Imperfect
Synch.
Channel Estimation (CE)
Blind Pilot Semi-blind
Joint CE and Synch.
Semi-blind
Perfect
Synch.
Imperfect
Synch.
Channel Estimation (CE)
Pilot
Joint CE and Synch.
Semi-blind Pilot
Pilot design to minimize:
MSE CRLB
MSE CRLB
MSE BCRLB
Turbo-based
Decision-direct.
MSE CRLB MSE CRLB MSE BCRLB
Introduction
Path loss models
Shadowing
Time-variant path gain
|,:
under mobile speed of 5 km/h
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14
1
1.1
1.2
1.3
1.4
1.5
1.6
1.7
1.8
1.9
2
Time (in OFDM symbol duration)
A
b
s
o
l
u
t
e

v
a
l
u
e

o
f

a
m
p
l
i
t
u
d
e

o
f

o
n
e

p
a
t
h

g
a
i
n

h
l
Mobile user speed = 5 km/h,
f
c
= 2 GHz,
128FFT, CP length = 10,
f
s
= 1.92 MHz,
2 time slots in LTE are considered,
Jakes model is considered.
pilot OFDM symbol
for channel estimation
Introduction
Path loss models
Shadowing
|,:
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14
0.95
1
1.05
1.1
1.15
A
b
s
o
l
u
t
e

v
a
l
u
e

o
f

a
m
p
l
i
t
u
d
e

o
f

o
n
e

p
a
t
h

g
a
i
n

h
l
f
c
= 2 GHz,
128FFT, CP length = 10,
f
s
= 1.92 MHz,
Jakes model is considered
Data OFDM symbol
Introduction
Path loss models
Shadowing
|,:
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14
0.8
0.9
1
1.1
1.2
1.3
A
b
s
o
l
u
t
e

v
a
l
u
e

o
f

a
m
p
l
i
t
u
d
e

o
f

o
n
e

f
a
d
i
n
g

g
a
i
n

h
l
f
c
= 2 Ghz, 128FFT, CP length = 10, f
s
= 1.92 Mhz,
Jakes model is considered.
Data OFDM symbol
Digital modulation techniques
Signal Space Analysis
Receiver Structure and Sucient Statistics
Error Probability Analysis and the Union Bound
Passband modulation
Chapter 3: Physical-layer transmission techniques
Section 3.1: Digital modulations
Mobile communications-Chapter 3: Physical-layer transmissions Section 3.1: Digital modulations 1
Passband modulation
Outline of the lecture notes
1
Advantages over analog modulation
Main considerations in digital modulation techniques
Typical types of digital modulation techniques
2
Rational
Signal and system model
Geometric representation of signals
Practical examples
Signal space representation
3
General results
Proofs of sucient statistics for optimal detection
Decision regions and criterion
4
Error probability
The union bound on error probability
5
Passband modulation
General principles
Amplitude and phase modulation
Pulse amplitude modulation (MPAM)
Phase shift keying (MPSK)
Quadrature amplitude modulation (MQAM)
Passband modulation
The advances over the last several decades in hardware and
digital signal processing have made digital transceivers much
cheaper, faster, and more power-ecient than analog
transceivers.
More importantly, digital modulation oers a number of other
advantages over analog modulation, including:
higher data rates,
powerful error correction techniques,
resistance to channel impairments,
more ecient multiple access strategies, and
better security and privacy.
Passband modulation
Advantages over analog modulation (cont.)
Digital transmissions consist of transferring information in the
form of bits over a communications channel.
The bits are binary digits taking on the values of either 1 or 0.
These information bits are derived from the information
source, which may be a digital source or an analog source that
has been passed through an A/D converter.
Both digital and A/D converted analog sources may be
compressed to obtain the information bit sequence.
Passband modulation
Digital modulation consists of mapping the information bits
into an analog signal for transmission over the channel.
Detection consists of determining the original bit sequence
based on the signal received over the channel.
The main considerations in choosing a particular digital
modulation technique are:
high data rate
high spectral eciency (minimum bandwidth occupancy)
high power eciency (minimum required transmit power)
robustness to channel impairments (minimum probability of bit
error)
low power/cost implementation
Passband modulation
Often the previous ones are conicting requirements, and the
choice of modulation is based on nding the technique that
achieves the best tradeo between these requirements.
There are two main categories of digital modulation:
amplitude/phase modulation
frequency modulation
Frequency modulation typically has a constant signal envelope
and is generated using nonlinear techniques, this modulation
is also called constant envelope modulation or nonlinear
modulation
Passband modulation
Typical types of digital modulation techniques (cont.)
Amplitude/phase modulation is also called linear modulation.
Linear modulation generally has better spectral properties
than nonlinear modulation, since nonlinear processing leads to
spectral broadening.
However, amplitude and phase modulation embeds the
information bits into the amplitude or phase of the
transmitted signal, which is more susceptible to variations
from fading and interference.
Passband modulation
Typical types of digital modulation techniques (cont.)
In addition, amplitude and phase modulation techniques
typically require linear ampliers, which are more expensive
and less power ecient than the nonlinear ampliers that can
be used with nonlinear modulation.
Thus, the general tradeo of linear versus nonlinear
modulation is one of better spectral eciency for the former
technique and better power eciency and resistance to
channel impairments for the latter technique.
Passband modulation
Rational
Practical examples
Rational
Digital modulation encodes a bit stream of nite length into
one of several possible transmitted signals.
Intuitively, the receiver minimizes the probability of detection
error by decoding the received signal as the signal in the set of
possible transmitted signals that is closest to the one received.
Determining the distance between the transmitted and
received signals requires a metric for the distance between
signals.
Passband modulation
Rational
Practical examples
Rational (cont.)
By representing signals as projections onto a set of basis
functions, we obtain a one-to-one correspondence between the
set of transmitted signals and their vector representations.
Thus, we can analyze signals in nite-dimensional vector
space instead of innite-dimensional function space, using
classical notions of distance for vector spaces.
In this section we show:
how digitally modulated signals can be represented as vectors
in an appropriately-dened vector space, and
how optimal demodulation methods can be obtained from this
vector space representation.
This general analysis will then be applied to specic
modulation techniques in later sections.
Passband modulation
Rational
Practical examples
Transmitted signal
Transmitter Receiver
+
n(t)
AWGN Channel
s(t)
i 1
K
m ={b ,...,b }
^
1 K
m ={b ,...,b }
^ ^
r(t)
y
s
s
1
t)
M
t)) [0, T)
1
t) t)} s
Figure 1: Communication system model over AWGN channel (i.e., a
special case of wireless channel).
Consider a communication system model as shown in the
above gure.
Every T seconds, the sytem sends 1 = log
2
` bits of
information through the channel for a data rate of 1 = 1,T
bits per second (bps).
Passband modulation
Rational
Practical examples
Transmitted signal (cont.)
There are ` = 2
1
possible sequences of 1 bits and each bit
sequence of length 1 comprises a message
:
i
= {/
1
, ..., /
1
} , where = {:
1
, ..., :
A
} is the set of
all such messages.
The message :
i
has probability j
i
of being selected for
transmission, where
A
i=1
j
i
= 1.
Suppose that message :
i
is to be transmitted over the
AWGN channel during the time interval [0, T). Since the
channel is analog, the message must be embedded into an
analog signal for channel transmission.
Passband modulation
Rational
Practical examples
Transmitted signal (cont.)
Therefore, each message :
i
is mapped to a unique
analog signal :
i
(t) = {:
1
(t), ..., :
A
(t)} where :
i
(t) is
dened on the time interval [0, T) and has energy
1
-
T
0
:
2
i
(t)dt, i = 1, ..., `. (1)
When messages :
i
are sent sequentially, the transmitted
signal becomes a sequence of the corresponding analog signals
as follows
:(t) =
I
:
i
(t /T). (2)
Passband modulation
Rational
Practical examples
Transmitted and received signals
In the aforementioned model, the transmitted signal is sent
through an AWGN channel where a white Gaussian noise
process n(t) of power spectral density
c
,2 is added to form
the received signal
:(t) = :(t) +n(t). (3)
T 0 2T 3T 4T
s (t)
1
1
1
2
s (tT)
s (t2T)
s (t3T)
s(t)
...
m
1
m
1
m
1
m
2
s
1
t)
M
t)) [0, T)
1
t) t)} s
Passband modulation
Rational
Practical examples
Received signal
Given :(t), the receiver must determine the best estimate of
which :
i
(t) was transmitted during each transmission
interval [0, T).
This best estimate of :
i
(t) is mapped to a best estimate of
the message :
i
(t) and the receiver produces this best
estimate : =
{
/
1
, ...,
/
1
}
of the transmitted bit sequence.
The goal of the receiver design in estimating the transmitted
message is to minimize the probability of message error
1
c
=
A
i=1
j ( : = :
i
:
i
sent) j (:
i
sent) (4)
over each time interval [0, T).
Passband modulation
Rational
Practical examples
Introduction
By representing the signals {:
i
(t), i = 1, ..., `} geometrically,
one can solve for the optimal receiver design in AWGN
channels based on a minimum distance criterion.
Note that, wireless channels typically have a time-varying
impulse response in addition to AWGN. We will consider the
eect of an arbitrary channel impulse response on digital
modulation performance in the next sections.
The basic premise behind a geometrical representation of
signals is the notion of a basis set.
Passband modulation
Rational
Practical examples
Basis function representation of signals
Specically, using a Gram-Schmidt orthogonalization
procedure, it can be shown that any set of ` real energy
signals o = {:
1
(t), ..., :
A
(t)} dened on [0, T) can be
represented as a linear combination of ` real
orthogonal basis functions {c
1
(t), ..., c
A
(t)}.
We say that these basis functions span the set .
Each signal {:
i
(t) } can be represented by
:
i
(t) =
.
)=1
:
i,)
c
)
(t), 0 t < T, (5)
where
:
i,)
=
T
0
:
i
(t)c
)
(t)dt (6)
is a real coecient representing the projection.
Passband modulation
Rational
Practical examples
Basis function representation of signals (cont.)
These basis functions have the following property
T
0
c
i
(t)c
)
(t)dt =
{
1 i = ,,
0 i = ,.
(7)
The basis set consists of the sine and cosine functions
c
1
(t) =
2
T
cos (2)
c
t) (8)
and
c
2
(t) =
2
T
sin(2)
c
t) . (9)
where
2
T
is used to obtain
T
0
c
2
i
(t)dt = 1, i = 1, 2.
Passband modulation
Rational
Practical examples
Basis functions in linear passband modulation techniques
With these basis functions, one only obtain an approximation
to (7), since
T
0
c
2
1
(t)dt =
2
T
T
0
0.5 [1 + cos (4)
c
t)] dt = 1+
sin (4)
c
t)
4)
c
t
(10)
The numerator in the second term of (10) is bounded by 1,
and for )
c
T 1 the denominator of this term is very large.
As a result, this second term can be neglected.
Passband modulation
Rational
Practical examples
Basis functions in linear passband modulation (cont.)
With these basis functions, one can have
T
0
c
1
(t)c
2
(t)dt =
2
T
T
0
0.5 sin (4)
c
t) dt =
cos (4)
c
t)
4)
c
t
0
(11)
where the approximation is taken as an equality as )
c
T 1.
With the basis set c
1
(t) =
2,T cos (2)

c
t) and
c
2
(t) =
2,T sin (2)

c
t), the basis function representation
(5) corresponds to the complex representation of :
i
(t) in
terms of its in-phase and quadrature components with an
extra factor of
2,T as follows
:
i
(t) = :
i,1
2
T
cos (2)
c
t) +:
i,2
2
T
sin(2)
c
t) . (12)
Passband modulation
Rational
Practical examples
Basis functions in linear passband modulation (cont.)
In practice, the basis set may include a baseband pulse-shaping
lter p(t) to improve the spectral characteristics of the transmitted
signal:
:
i
(t) = :
i,1
p(t) cos (2)
c
t) +:
i,2
p(t) sin (2)
c
t) (13)
where the simplest pulse shape that satisfy (7) is the rectangular
pulse shape p(t) =
2,T, 0 t < T.
Passband modulation
Rational
Practical examples
Denitions used in signal space representation
We denote the coecients {:
i,)
} as a vector
s
i
= [:
i,1
, ..., :
i,.
]
.
which is called the signal
constellation point corresponding to the signal :
i
(t).
The signal constellation consists of all constellation points
{s
1
, ..., s
A
}.
Given the basis functions {c
1
(t), ..., c
.
(t)} there is a
one-to-one correspondence between the transmitted signal
:
i
(t) and its constellation point s
i
.
The representation of :
i
(t) in terms of its constellation point
s
i

.
is called:
its signal space representation and
the vector space containing the constellation is called the
signal space.
Passband modulation
Rational
Practical examples
Denitions used in signal space representation (cont.)
A two-dimensional signal space is illustrated in the below
gure, where we show s
i

2
with the ith axis of
2
corresponding to the basis function c
i
(t), i = 1, 2.
M=4, K=2
00 11
01
10
M=8, K=3
000
001
011
110
100
010
110
101
s
i1
s
i2
s
i1
s
i2
Passband modulation
Rational
Practical examples
With this signal space representation we can analyze the
innite-dimensional functions :
i
(t) as vectors s
i
in
nite-dimensional vector space
2
.
This greatly simplies the analysis of the system performance
as well as the derivation of the optimal receiver design.
Signal space representations for common modulation
techniques like MPSK and MQAM are two-dimensional
(corresponding to the in-phase and quadrature basis
functions).
In order to analyze signals via a signal space representation,
we need to use some denitions for the vector characterization
in the vector space
.
.
Passband modulation
Rational
Practical examples
In particular, the length of a vector in
.
is dened as
s
i
=
)=1
:
2
i,)
. (14)
The distance between two signal constellation points s
i
and s
I
is thus
s
i
s
I
=
)=1
(:
i,)
:
I,)
)
2
=
T
0
(:
i
(t) :
I
(t))
2
dt.
(15)
Passband modulation
Rational
Practical examples
Finally, the inner product :
i
(t), :
I
(t) between two real
signals :
i
(t) and :
I
(t) on the interval [0, T) is dened as
:
i
(t), :
I
(t) =
T
0
:
i
(t):
I
(t)dt. (16)
Similarly, the inner product s
i
, s
I
between two real vectors is
s
i
, s
I
= s
i
s
T
I
=
T
0
:
i
(t):
I
(t)dt = :
i
(t), :
I
(t). (17)
It is noted that two signals are orthogonal if their inner
product is zero.
Passband modulation
General results
Receiver structure and sucient statistics
Given the channel output :(t) = :
i
(t) +n(t), 0 t < T, we
now investigate the receiver structure to determine which
constellation point s
i
or, equivalently, which message :
i
, was
sent over the time interval [0, T).
A similar procedure is done for each time interval
[/T, (/ + 1)T).
We would like to convert the received signal :(t) over each
time interval into a vector, since it allows us to work in
nite-dimensional vector space to estimate the transmitted
signal.
However, this conversion should not compromise the
estimation accuracy. For this conversion, consider the receiver
structure shown in the next gure.
Passband modulation
General results
Receiver structure and sucient statistics (cont.)
) ( ) ( ) ( t n t s t r
i

T
dt
0
()
T
dt
0
()
1 1 1 ,
r n s
i

) (
1
t I
) (t
N
I
N N N i
r n s
,
Find i
i
m m
As shown in the above gure, the components of signal and
noise vectors are determined by
:
i,)
=
T
0
:
i
(t)c
)
(t)dt, (18)
and
n
)
=
T
0
n(t)c
)
(t)dt. (19)
Passband modulation
General results
Receiver structure and sucient statistics (cont.)
We can rewrite :(t) as
:(t) =
.
)=1
(:
i,)
+n
)
) c
)
(t) +n
(t) =
.
)=1
:
)
c
)
(t) +n
(t),
(20)
where :
)
= :
i,)
+n
)
and n
(t) = n(t)
.
)=1
n
)
c
)
(t)
denotes the remainder noise.
Passband modulation
General results
If we can show that the optimal detection of the transmitted
signal constellation point s
i
given received signal :(t) does not
make use of the remainder noise n
(t), then the receiver can

make its estimate : of the transmitted message :
i
as a
function of r = (:
1
, ..., :
.
) alone.
In other words, r = (:
1
, ..., :
.
) is a sucient statistic for :(t)
in the optimal detection of the transmitted messages.
Let exam the distribution of r. Since n(t) is a Gaussian
random process, if we condition on the transmitted signal
:
i
(t) then the channel output :(t) = :
i
(t) +n(t) is also a
Gaussian random process and r = [:
1
, ..., :
.
] is a Gaussian
random vector.
Passband modulation
General results
Proofs of sucient statistics for optimal detection (cont.)
Recall that :
i
= :
i,)
+n
)
. Thus, conditioned on the
transmitted constellation s
i
, we have
j
= 1[:
)
s
i
] = 1 [:
i,)
+n
)
:
i,)
] = :
i,)
(21)
since n(t) has zero mean, and
o
= 1
[
:
)
j
]
2
= 1[:
i,)
+n
)
:
i,)
:
i,)
]
2
= 1
[
n
2
)
]
.
(22)
With Cov [:
)
:
I
s
i
] = 1
[(
:
)
j
)
(:
I
j
) s
i
]
= 1 [n
)
n
I
]
and some manipulations, one can have
1 [n
)
n
I
] =

0
2
T
0
c
)
(t)c
I
(t)dt =
{
0
,2 , = /
0 , = /.
. (23)
Passband modulation
General results
Thus, conditioned on the transmitted constellation s
i
, the :
)
s
are uncorrelated and, since they are Gaussian and also
independent. Moreover, 1
[
n
2
)
]
=
0
,2.
We have shown that, conditioned on the transmitted
constellation s
i
, :
)
is a Gauss-distributed random variable that
is independent of :
I
, / = ,and has mean :
i,)
and variance
0
,2.
Thus, the conditional distribution of r is given by
j (rs
.
sent) =
=1
j (:
:
.
) =
1
(
0
)
2
exp
=1
(:
:
.,
)
2
.
(24)
Passband modulation
General results
It is also straightforward to show that 1[:
)
n
(t)s
i
] = 0 for
any t, 0 t < T. Thus, since :
)
conditioned on s
i
and n
(t)
are Gaussian and uncorrelated, they are independent.
Also, since the transmitted signal is independent of the noise,
:
i,)
is independent of the process n
(t).
We now discuss the receiver design criterion and show it is not
aected by discarding n
(t).
The goal of the receiver design is to minimize the probability
of error in detecting the transmitted message :
i
given
received signal :(t).
To minimize 1
c
= j ( : = :
i
:(t)) = 1 j ( : = :
i
:(t)), we
maximize j ( : = :
i
:(t)).
Passband modulation
General results
Therefore, the receiver output : given received signal :(t)
should correspond to the message :
i
that maximizes
j (:
i
sent:(t)).
Since there is a one-to-one mapping between messages and
signal constellation points, this is equivalent to maximizing
j (:
i
sent:(t)).
Passband modulation
General results
Recalling that :(t) is completely described by : = (:
1
, ..., :
.
)
and n
(t), we have
j (s
i
sent:(t)) = j ((:
i,1
, ..., :
i,.
) sent(:
1
, ..., :
.
, n
(t)))
=
j ((:
i,1
, ..., :
i,.
) sent, (:
1
, ..., :
.
), n
(t))
j ((:
1
, ..., :
.
), n
(t))
=
j ((:
i,1
, ..., :
i,.
) sent, (:
1
, ..., :
.
)) j (n
(t))
j ((:
1
, ..., :
.
)) j (n
(t))
= j ((:
i,1
, ..., :
i,.
) sent(:
1
, ..., :
.
)) . (25)
where the third equality follows from the fact that the n
(t) is
independent of both (:
1
, ..., :
.
) and of (:
i,1
, ..., :
i,.
).
Passband modulation
General results
This analysis shows that (:
1
, ..., :
.
) is a sucient statistic for
:(t) in detecting :
i
, in the sense that the probability of error
is minimized by using only this sucient statistic to estimate
the transmitted signal and discarding the remainder noise.
Since r is a sucient statistic for the received signal :(t), we
call r the received vector associated with :(t).
Passband modulation
General results
Decision regions
As aforementioned, the optimal receiver minimizes error
probability by selecting the detector output : that maximizes
the probability of correct detection
1 1
c
= j ( : sentr received).
In other words, given a received vector r, the optimal receiver
selects : = :
i
corresponding to the constellation s
i
that
satises
j (s
i
r) j (s
)
r) , , = i (26)
where j (s
i
r) j (s
i
sentr received) for the sake of
notational simplicity.
Passband modulation
General results
Decision regions(cont.)
Thus, the decision regions (7
1
, ..., 7
A
) corresponding to
(s
1
, ..., s
A
) are the subsets of the signal space
.
and
dened by
7
i
= (r : j (s
i
r) j (s
)
r) , , = i) . (27)
Once the signal space has been partitioned by decision
regions, for a received vector r 7
i
, the optimal receiver
outputs the message estimate : = :
i
The receiver processing consists of o) computing the received
vector r from :(t), /) nding which decision region 7
i
contains r, and c) outputting the corresponding message :
i
.
Passband modulation
General results
An example on decision regions
This process is illustrated in the below gure, that shows a
two-dimensional signal space with four decision regions
7
1
, ..., 7
4
corresponding to four constellations s
1
, ..., s
4
.
The received vector r lies in region 7
1
, so the receiver will
output the message :
1
as the best message estimate given
received vector r.
(t)
1
(t)
2
s
s
s
s
1
2
3
4
1
Z Z
3
Z
2
Z
4
x
Passband modulation
General results
Decision criterion
Using Bayes rule, one can have
j (s
i
r) =
j (rs
i
) j (s
i
)
j (r)
. (28)
To minimize error probability, the receiver output : = :
i
corresponds to the constellation point s
i
that maximizes
j (s
i
r), i.e., the detected transmitted constellation point s
can be determined by
s = arg max
s
j (rs
i
) j (s
i
)
j (r)
= arg max
s
j (rs
i
) j (s
i
) , i = 1, ..., `
(29)
where the second equality follows from the fact that j(r) is
not a function of s
i
Passband modulation
General results
Decision criterion (cont.)
Assuming that transmitted messages :
i
are equally likely
(i.e., j(s
i
) = 1,`), (29) becomes
s = arg max
s
j (rs
i
) , i = 1, ..., `. (30)
Let dene the likelihood function associated with the receiver
as
)(s
i
) = j (rs
i
) . (31)
Given a received vector r, a maximum likelihood (ML)
receiver outputs : = :
i
corresponding to the constellation
point s
i
that maximizes )(s
i
).
Passband modulation
General results
Since the log function is increasing in its argument,
maximizing s
i
is equivalent to maximizing the log likelihood
function, dened as )
A1
(s
i
) = log )(s
i
). Using (24) for
)
A1
(s
i
) = log )(s
i
) yields
)
1J
(s
.
) =
1
=1
(:
:
.,
)
2
+constant =
1
0
r s
.
2
+constant.
(32)
Based on (30), the detected transmitted constellation point s can
be determined by the ML criterion as
s = arg min
s
r s
.
2
, i = 1, ..., `. (33)
Passband modulation
General results
Under the aforementioned assumption of equiprobable
(transmitted) messages :
i
, the ML structure minimizes the
probability of detection error at the receiver.
Under the aforementioned assumption of equiprobable
(transmitted) messages :
i
, the ML structure minimizes the
probability of detection error at the receiver.
Passband modulation
Error probability
Error probability of ML detection
With j (:
i
sent) = 1,`, the error probability of the ML receiver:
s
1
s
s
s
s
s
s
s
2
3
4
5
6
7
8
0
1
Z
1
d
min
r=s +n
1
c
=
1
.=1
j (r , 7
.
:
.
sent) j (:
.
sent)
=
1
`
1
.=1
j (r , 7
.
:
.
sent)
= 1
1
`
1
.=1
j (r 7
.
:
.
sent)
= 1
1
`
1
.=1
j (r:
.
) dr
= 1
1
`
1
.=1
j (r = s
.
+ ns
.
) dr
1
1

Passband modulation
Error probability
As observed, (34) gives an exact solution to the error
probability but it is impossible to solve this error probability in
closed-form. Therefore, the union bound on error probability
is investigated.
Let
i,I
denote the event that r s
I
< r s
i
given that
the constellation point s
i
was sent.
If the event
i,I
occurs, then the constellation point will be
decoded in error since the transmitted constellation point s
i
is
not the closest constellation point to the received vector r.
However, event
i,I
does not necessarily imply that s
I
will be
decoded instead of s
i
, since there may be another
constellation point s
|
with r s
|
< r s
I
< r s
i
.
Passband modulation
Error probability
The union bound on error probability (cont.)
The constellation point is decoded correctly if
r s
i
< r s
I
/ = i. Therefore,
1
c
(:
i
sent) = j
I=1
I=i
i,I
I=1
I=i
j (
i,I
) . (35)
where the inequality follows from the union bound on
probability.
More specically, j (
i,I
) can be determined by
j (
.,|
) = j (s
|
r < s
.
r :
.
sent)
= j (s
|
(s
.
+ n) < s
.
(s
.
+ n))
= j (n + s
.
s
|
< n)
= j
(
2 n s
.
s
|
cos c < s
.
s
|
2
)
Passband modulation
Error probability
Since n, s
i
s
I
= n s
i
s
I
cos c, one will have
n = n cos c =
n,s
is a Gaussian random variable with

zero-mean and variance
0
,2
As a result, j (
i,I
) can be simplied to
j (
i,I
) = j
(
n <
s
i
s
I
2
)
= j
(
n
d
i,I
2
)
=

o
,
2
1
0
exp
(
n
2
0
)
dn = Q
(
d
i,I
2
0
)
.
Passband modulation
Error probability
Substituting (36) into (35), one can have
1
c
(:
i
sent)
A
I=1
I=i
Q
(
d
i,I
2
0
)
. (36)
where the Q function, Q(.), is dened as the probability that
a Gaussian random variable r with zero-mean and variance of
1 is bigger than ., i.e.,
Q(.) = j (r .) =

:
1
2
exp
(
r
2
2
)
dr. (37)
Summing (36) over all possible messages yields the union bound
1
c
=
1
.=1
j (:
.
) 1
c
(:
.
sent)
1
`
1
.=1
1
|=1
|=.
Q
(
d
.,|
2
0
)
. (38)
Passband modulation
Error probability
Note that the Q function cannot be solved in closed-form. It
can be obtained from the complementary error function as
Q(.) =
1
2
erfc
(
:
2
)
.
One can upper bound Q(.) with the closed-form expression
Q(.)
1
.
2
exp
(
.
2
,2
)
. (39)
and this bound is quite tight for . 0.
Let dene the minimum distance of the constellation as
d
nia
= min
i,I
d
i,I
, one can simplify (41) with looser bound
1
c
(` 1)Q
(
d
nia
2
0
)
. (40)
Passband modulation
Error probability
Using (39) for the Q function yields a closed-form bound
1
c

` 1
exp
(
d
2
nia
4
0
)
. (41)
Note that for binary modulation (` = 2), there is only one
way to make an error and d
nia
is the distance between the
two signal constellation points, so the bound is exact
1
o
= Q
(
d
nia
2
0
)
. (42)
Passband modulation
Error probability
Bit error rate
Recall that 1
c
is the probability of a symbol (message) error:
1
c
= j ( : = :
i
:
i
sent) where :
i
corresponds to a message
with |op
2
` bits.
However, system designers are typically more interested in the
bit error probability, also called the bit error rate (BER), than
in the symbol error probability, since bit errors drive the
performance of higher layer networking protocols and
end-to-end performance.
Thus, we would like to design the mapping of ` possible bit
sequences to a message :
i
, i = 1, ..., ` so that a symbol
error associated with an adjacent decision region, which is the
most likely way to make an error, corresponds to only one bit
error.
Passband modulation
Error probability
Bit error rate (cont.)
With such a mapping, assuming that mistaking a signal
constellation for a constellation other than its nearest
neighbors has a very low probability, we can make the
approximation
1
o

1
c
|op
2
`
. (43)
The most common form of mapping with the property is
called Gray coding, which will be discussed in more detail.
Signal space concepts are applicable to any modulation where
bits are encoded as one of several possible analog signals,
including the amplitude, phase, and frequency modulations as
discussed later.
Passband modulation
General principles
General principles
The basic principle of passband digital modulation is to
encode an information bit stream into a carrier signal which is
then transmitted over a communications channel.
Demodulation is the process of extracting this information bit
stream from the received signal. Corruption of the transmitted
signal by the channel can lead to bit errors in the
demodulation process.
The goal of modulation is to send bits at a high data rate
while minimizing the probability of data corruption.
In general, modulated carrier signals encode information in the
amplitude c(t), frequency )(t), or phase 0(t) of a carrier
signal.
Passband modulation
General principles
General principles (cont.)
Thus, the modulated signal can be represented as
:(t) = c(t) cos [2 ()
t
+)(t)) +0(t) +c
0
] = c(t) cos [2)
t
t +c(t) +c
0
]
where c(t) = 2)(t)t +0(t) and c
0
is the phase oset of the
carrier. This representation combines frequency and phase
modulation into angle modulation.
One can rewrite the right-hand side of (44) in terms of its in-phase
and quadrature components as:
:(t) = c(t) cos c(t) cos [2)
t
t] c(t) sin c(t) sin [2)
t
t]
= :
J
(t) cos [2)
t
t] :
Q
(t) sin [2)
t
t] (44)
where :
J
(t) = c(t) cos c(t) is called the in-phase component of :(t)
and :
Q
(t) = c(t) sin c(t) is called its quadrature component.
Passband modulation
General principles
General principles (cont.)
We can write :(t) in its complex baseband representation as
:(t) = Re
[
n(t)c
)2)
|
]
(45)
where n(t) = :
1
(t) +,:
Q
(t).
This representation is useful since receivers typically process
the in-phase and quadrature signal components separately.
Passband modulation
General principles
In amplitude and phase modulation the information bit stream
is encoded in the amplitude and/or phase of the transmitted
signal.
Specically, over a time interval of T
-
, 1 = log
2
` bits are
encoded into the amplitude and/or phase of the transmitted
signal :(t), 0 t < T
-
.
The transmitted signal over this period
:(t) = :
1
(t) cos [2)
c
t] :
Q
(t) sin [2)
c
t] can be written in
terms of its signal space representation as
:(t) = :
i,1
c
1
(t) +:
i,2
c
2
(t) (46)
where basis functions c
1
(t) = p(t) cos (2)
c
t +c
0
) and
c
2
(t) = p(t) sin (2)
c
t +c
0
), where p(t) is a shaping pulse.
Passband modulation
General principles
Amplitude and phase modulation (cont.)
To send the ith message over the time interval [/T, (/ + 1)T),
we set :
1
(t) = :
i,1
p(t) and :
Q
(t) = :
i,2
p(t). These in-phase
and quadrature signal components are baseband signals with
spectral characteristics determined by the pulse shape p(t).
In particular, their bandwidth 1 equals the bandwidth of g(t),
and the transmitted signal :(t) is a passband signal with
center frequency fc and passband bandwidth 21.
In practice we take 1 = 1
j
,T
-
where 1
j
depends on the
pulse shape: for rectangular pulses 1
j
= .5 and for raised
cosine pulses .5 1
j
1.
Thus, for rectangular pulses the bandwidth of p(t) is .5,T
-
and the bandwidth of :(t) is 1,T
-
.
Passband modulation
General principles
Since the pulse shape p(t) is xed, the signal constellation for
amplitude and phase modulation is dened based on the
constellation point: (:
i,1
, :
i,2
)
2
, i = 1, ..., `.
The complex baseband representation of :(t) is
:(t) = Re
[
r(t)c
)c
0
c
)2)
|
]
(47)
where:
r(t) = :
J
(t) +,:
Q
(t) = (:
.,1
+,:
.,2
) p(t).
The constellation point s
.
= (:
.,1
, :
.,2
) is called the symbol
associated with the log
2
` bits and
T
s
is called the symbol time and the bit rate for this
modulation is 1 bits per symbol or 1 = log
2
`,T
s
bits per
second.
Passband modulation
General principles
There are three main types of amplitude/phase modulation:
Pulse Amplitude Modulation (MPAM): information encoded in
amplitude only.
Phase Shift Keying (MPSK): information encoded in phase
only.
Quadrature Amplitude Modulation (MQAM): information
encoded in both amplitude and phase.
The number of bits per symbol 1 = log
2
`, signal
constellation (:
i,1
, :
i,2
)
2
, i = 1, ..., `, and the choice of
shaping pulse p(t) determines the digital modulation design.
The pulse shape p(t) is designed to improve spectral eciency
and combat inter-symbol-interference (ISI).
Passband modulation
General principles
Amplitude and phase modulation over a given symbol period
can be generated using the modulator structure shown in the
next Figure.
Note that the basis functions in this gure have an arbitrary
phase c
0
associated with the transmit oscillator.
Demodulation over each symbol period is performed using the
demodulation structure of Figure xx +1, which is equivalent
to the structure of for c
1
(t) = p(t) cos (2)
c
t +c) and
c
2
(t) = p(t) sin (2)
c
t +c).
Typically the receiver includes some additional circuitry for
carrier phase recovery that matches the carrier phase c at the
receiver to the carrier phase c
0
at the transmitter, which is
called coherent detection.
Passband modulation
General principles
If c c
0
= c = 0 then the in-phase branch will have an
unwanted term associated with the quadrature branch and
vice versa, i.e., :
1
= :
i,1
cos (c) +:
i,2
sin (c) +n
1
and
:
2
= :
i,1
sin(c) +:
i,2
cos (c) +n
2
can result in signicant
performance degradation.
The receiver structure also assumes that the sampling function
every T
-
seconds is synchronized to the start of the symbol
period, which is called synchronization or timing recovery.
Receiver synchronization and carrier phase recovery are
complex receiver operations that can be highly challenging in
wireless environments.
We will assume perfect carrier recovery in our discussion of
MPAM, MPSK and MQAM, and therefore set c = c
0
= 0 for
their analysis.
Passband modulation
General principles
We will start by looking at the simplest form of linear
modulation, one-dimensional MPAM, which has no
quadrature component (:
i,2
= 0).
For MPAM all of the information is encoded into the signal
amplitude
i
. The transmitted signal over one symbol time is
given by
:
i
(t) = Re
[
i
p(t)c
)2)
|
]
=
i
p(t) cos(2)
c
t), 0 t T
-
1,)
c
,
(48)
where
i
= (2i 1 `)d, i = 1, 2, ..., ` denes the signal
constellation, parameterized by the distance d which is
typically a function of the signal energy, and p(t) is the
shaping pulse.
Passband modulation
General principles
Pulse amplitude modulation (cont.)
2
Shaping
Filter
g(t)
Shaping
Filter
g(t)
s(t)
InPhase branch
Quadrature Branch
i1
i2
s
s
i1
s g(t)
s g(t)
i2
c
sin(2 f t+ )
cos(2 f t+ )
c

0
cos(2 f t+ )
c

0

0
Figure 2: Amplitude and phase modulator.
Passband modulation
General principles
The minimum distance between constellation points is
d
nia
= :in
i,)

i
)
= 2d. The amplitude of the
transmitted signal takes on M dierent values, which implies
that each pulse conveys |op
2
` = 1 bits per symbol time T
-
.
Over each symbol period the MPAM signal associated with
the ith constellation has energy
1
-
0
:
2
i
(t)dt =
2
i
p
2
(t) cos
2
(2)
c
t)dt =
2
i
. (49)
It is noted that the energy is not the same for each signal
:
i
(t), i = 1, ..., `.
Passband modulation
General principles
Assuming equally likely symbols, the average energy is
1
-
=
1
`
A
i=1
2
i
. (50)
i
1
^
m=m
Find i: x Z
i
i
T
T
InPhase branch
/2
g(Tt)
g(Tt)
cos (2 f t+ )
i1 1
2 i2 2
Quadrature branch
s
s
c
sin (2 f t+ )
c

r(t)=s (t)+n(t)
r =s +n
r =s +n
Figure 3: Amplitude and phase demodulator.
Passband modulation
General principles
The constellation mapping is usually done by Gray encoding,
where the messages associated with signal amplitudes that are
adjacent to each other dier by one bit value, as illustrated in
the below gure.
With this encoding method, if noise causes the demodulation
process to mistake one symbol for an adjacent one (the most
likely type of error), this results in only a single bit error in the
sequence of K bits. Gray codes can be designed for MPSK and
square MQAM constellations, but not rectangular MQAM.
M=4, K=2
00 01 11 10
M=8, K=3
000 001 011 010 110 111 101 100
2d
2d
Passband modulation
General principles
An example: For p(t) =
2,T
-
, 0 t < T
-
a rectangular
pulse shape, nd the average energy of 4PAM modulation.
Solution: For 4PAM, the
i
values are
i
= {3d, d, d, 3d}.
Hence, the average is
1
-
=
d
2
4
(9 + 1 + 1 + 9) = 5d
2
. (51)
The decision regions 7
i
, i = 1, ..., ` associated with the
pulse amplitude
i
= (2i 1 `)d for ` = 4 and ` = 8
as shown in the next gure. Mathematically, for any `, these
decision regions are dened by
7
i
=
(,
i
+d) i = 1,
[
i
d,
i
+d) 2 i ` 1
[
i
d, ) i = `
(52)
Passband modulation
General principles
One can see that MPAM has only a single basis function
c
1
(t) = p(t) cos(2)
c
t).
Thus, the coherent demodulator for MPAM reduces to the
demodulator shown in the next gure, where the
multi-threshold device maps r to a decision region 7
i
and
outputs the corresponding bit sequence
: = :
i
= {/
1
, ..., /
1
}.
Passband modulation
General principles
2d
2d
A
1
A A A
2 3 4
Z
1
Z Z
2
Z
3 4
A A A A
Z Z Z Z Z
1
Z Z
A
1
A
2 3
Z
2 3 4 5 6 7 8
A A
8 7 6 5 4
Figure 5: Decision Regions for MPAM.
i
s (t)+n(t)
s
x
g (T t) 0
X
cos(2 f t)
c
^
s
T
Multithreshold Device
2d
4d
2d
4d
(M2)d
(M2)d
m=m =b b ...b
i 1 2 K
}
Z
i
Figure 6: Coherent Demodulator for MPAM.
Passband modulation
General principles
For MPSK modulation, all of the information is encoded in
the phase of the transmitted signal.
Thus, the transmitted signal over one symbol time is given by
:
i
(t) = Re
[
p(t)c
)2(i1)A
c
)2)
|
]
, 0 t T
-
(53)
= p(t) cos
[
2)
c
t +
2(i 1)
`
]
= p(t) cos
[
2(i 1)
`
]
cos(2)
c
t)
p(t) sin
[
2(i 1)
`
]
sin(2)
c
t).
Passband modulation
General principles
Phase shift keying (cont.)
The constellation points or symbols (:
i,1
, :
i,2
) are given by
:
i,1
= cos
[
2(i1)
A
]
and :
i,2
= sin
[
2(i1)
A
]
for
i = 1, ..., `. 0
i
=
2(i1)
A
, i = 1, 2, ..., ` = 2
1
are the
dierent phases in the signal constellation points that convey
the information bits.
The minimum distance between constellation points is
d
nia
= 2sin(,`), where is typically a function of the
signal energy.
2PSK is often referred to as binary PSK or BPSK, while 4PSK
is often called quadrature phase shift keying (QPSK), and is
the same as MQAM with ` = 4 which is dened below.
Passband modulation
General principles
All possible transmitted signals :
i
(t) have equal energy:
1
-
0
:
2
i
(t)dt =
2
. (54)
Note that for p(t) =
2,T
-
, 0 t T
-
, i.e., a rectangular
pulse, this signal has constant envelope, unlike the other
amplitude modulation techniques MPAM and MQAM.
However, rectangular pulses are spectrally-inecient, and
more ecient pulse shapes make MPSK nonconstant
envelope.
Passband modulation
General principles
Analogous to MPAM, MPSK constellation mapping is usually
done by Gray encoding, where the messages associated with
signal phases that are adjacent to each other dier by one bit
value, as illustrated in the below Figure.
With this encoding method, mistaking a symbol for an
adjacent one causes only a single bit error.
M=4, K=2
00 11
01
10
M=8, K=3
000
001
011
110
100
010
110
101
s
i1
s
i2
s
i1
s
i2
Figure 7: Gray Encoding for MPSK.
Passband modulation
General principles
Phase shift keying: Decision region
The decision regions 7
i
, i = 1, ..., `, associated with MPSK
for ` = 8 are shown in the next gure.
If we represent r = :c
)0

2
in polar coordinates then these
decision regions for any ` are dened by
7
i
=
{
:c
)0
: 2(i .5),` 0 2(i +.5),`
}
. (55)
For the special case of BPSK, the decision regions simplify to
7
1
= (r : r 0) and 7
2
= (r : r 0).
Moreover BPSK has only a single basis function
c
1
(t) = p(t) cos(2)
c
t) and, since there is only a single bit
transmitted per symbol time T
-
, the bit duration T
o
= T
-
.
Passband modulation
General principles
Phase shift keying: Decision region (cont.)
Thus, the coherent demodulator for BPSK reduces to the
demodulator shown in the next gure, where the threshold
device maps r to the positive or negative half of the real line,
and outputs the corresponding bit value.
We have assumed in this gure that the message
corresponding to a bit value of 1, :
1
= 1, is mapped to
constellation point :
1
= and the message corresponding to
a bit value of 0, :
2
= 0, is mapped to the constellation point
:
2
= .
Passband modulation
General principles
Phase shift keying: Decision region (cont.)
Z
1
Z
2
Z
3
Z
4
Z
1
Z
2
Z
Z
Z
Z
Z
Z
3
4
5
6
7
8
Figure 8: Decision Regions for MPSK.
i
s (t)+n(t)
g (T t)
b 0 X
cos(2 f t)
c
^
T
b
Threshold Device
2
m=1 or 0
m=1
^
^
m=0
}
}
r
1
Z :r>0
Z :r<0
Figure 9: Coherent Demodulator for BPSK.
Passband modulation
General principles
For MQAM, the information bits are encoded in both the
amplitude and phase of the transmitted signal.
Thus, whereas both MPAM and MPSK have one degree of
freedom in which to encode the information bits (amplitude or
phase), MQAM has two degrees of freedom.
As a result, MQAM is more spectrally-ecient than MPAM
and MPSK, in that it can encode the most number of bits per
symbol for a given average energy.
Passband modulation
General principles
Quadrature amplitude modulation (cont.)
The transmitted signal is given by
:
i
(t) = Re
[
i
c
)0
p(t)c
)2)
|
]
(56)
=
i
cos(0
i
)p(t) cos(2)
c
t)
i
sin(0
i
)p(t) sin(2)
c
t).
where 0 t T
-
.
The energy in :
i
(t) is
1
-
0
:
2
i
(t)dt =
2
i
. (57)
that is the same as for MPAM.
The distance between any pair of symbols in the signal
constellation is
d
i,I
= s
i
s
I
=
(:
i,1
:
I,1
)
2
+ (:
i,2
:
I,2
)
2
. (58)
Passband modulation
General principles
Quadrature amplitude modulation (cont.)
For square signal constellations, where :
i,1
and :
i,2
take
values on (2i 1 1)d, i = 1, 2, ..., 1 = 2
|
, the minimum
distance between signal points reduces to d
nia
= 2d, the
same as for MPAM.
In fact, MQAM with square constellations of size 1
2
is
equivalent to MPAM modulation with constellations of size 1
on each of the in-phase and quadrature signal components.
Common square constellations are 4QAM and 16QAM, which
are shown in the below gure.
Passband modulation
General principles
Quadrature amplitude modulation: constellation and
decision regions
4QAM
16QAM
Figure 10: 4QAM and 16QAM
Constellations.
Z
1
Z
2
Z
3
Z
4
Z
5
Z
6
Z
7
Z
8
Z
9
Z
10
Z
11
Z
12
Z
13
Z
14
Z
15
Z
16
Figure 11: Decision Regions for
MQAM withM = 16.
Passband modulation
General principles
Quadrature amplitude modulation: constellation and
decision regions
These square constellations have ` = 2
2|
= 1
2
constellation
points, which are used to send 2| bits/symbol, or | bits per
dimension.
It can be shown that the average power of a square signal
constellation with | bits per dimension, o
|
, is proportional to
4
|
,3, and it follows that the average power for one more bit
per dimension o
|+1
4o
|
.
Thus, for square constellations it takes approximately 6 dB
more power to send an additional 1 bit/dimension or 2
bits/symbol while maintaining the same minimum distance
between constellation points.
Introduction
AWGN channels
Fading Channels
Section 3.2: Performance analysis over fading channels
Mobile communications - Chapter 3: Physical-layer transmission techniques Section 3.2: Performance analysis over fading channels 1
Introduction
AWGN channels
Fading Channels
Outline of the lecture notes
1
Introduction
2
AWGN channels
Signal-to-Noise power ratio and bit/symbol energy
Error probability for BPSK and QPSK
Approximate symbol and bit error probabilities for typical modulations
3
Fading Channels
Introduction
Outage probability
Average probability of error
Introduction
AWGN channels
Fading Channels
Introduction
We now consider the performance of the digital modulation
techniques discussed in the previous chapter when used over AWGN
channels and channels with at-fading.
There are two performance criteria of interest: the probability of
error, dened relative to either symbol or bit errors, and the outage
probability, dened as the probability that the instantaneous
signal-to-noise ratio falls below a given threshold.
Wireless channels may also exhibit frequency selective fading and
Doppler shift. Frequency-selective fading gives rise to intersymbol
interference (ISI), which causes an irreducible error oor in the
received signal.
Introduction
AWGN channels
Fading Channels
In this section we dene the signal-to-noise power ratio (SNR) and
its relation to energy-per-bit (1
b
) and energy-per-symbol (1
s
).
We then examine the error probability on AWGN channels for
dierent modulation techniques as parameterized by these energy
metrics. Our analysis uses the signal space concepts of previous
section.
In an AWGN channel, the modulated signal :(t) = Re
[
n(t)c
2t}
|
]
has receiver noise n(t) added to it prior to reception. The noise n(t)
is a white Gaussian random process with zero-mean and power
spectral density
0
,2.
The received signal is thus :(t) = :(t) +n(t).
Introduction
AWGN channels
Fading Channels
Signal-to-Noise power ratio and bit/symbol energy (cont.)
We dene the received signal-to-noise power ratio (SNR) as the
ratio of the received signal power 1
:
to the power of the noise
within the bandwidth of the transmitted signal :(t).
The received power 1
:
is determined by the transmitted power and
the path loss and multipath fading.
The noise power is determined by the bandwidth of the transmitted
signal and the spectral properties of n(t). Specically, if the
bandwidth of the complex envelope n(t) of :(t) is 1 then the
bandwidth of the transmitted signal :(t) is 21.
Introduction
AWGN channels
Fading Channels
Signal-to-Noise ratio and bit/symbol energy (cont.)
Since the noise n(t) has uniform power spectral density
0
,2, the
total noise power within the bandwidth 21 is
1
n
=
0
,2 21 =
0
1. So, the received SNR is given by
o1 =
1
:
0
1
. (1)
In systems with interference, we often use the received
signal-to-interference-plus-noise power ratio (SINR) in place of SNR
for calculating error probability. If the interference statistics
approximate those of Gaussian noise then this is a reasonable
approximation.
The received SINR is given by
o1 =
1
:
0
1 +1
J
. (2)
where 1
J
is the average power of the interference.
Introduction
AWGN channels
Fading Channels
The SNR is often expressed in terms of the signal energy per bit 1
b
or per symbol 1
s
as
o1 =
1
:
0
1
=
1
s
0
1T
s
=
1
b
0
1T
b
. (3)
where T
s
and T
b
are the symbol and bit durations, respectively. For
binary modulation (e.g., BPSK), T
s
= T
b
and 1
s
= 1
b
.
For data shaping pulses with T
s
= 1,1 (e.g., raised cosine pulses
with a = 1), one will have SNR = 1
s
,
0
for multilevel signaling
and SNR = 1
b
,
0
for binary signaling. For general pulses,
T
s
= /,1 for some constant /, we have / SNR = 1
s
,
0
.
The quantities
s
= 1
s
,
0
and
b
= 1
b
,
0
are sometimes called
the SNR per symbol and the SNR per bit, respectively.
Introduction
AWGN channels
Fading Channels
For performance specication, we are interested in the bit error
probability 1
b
as a function of
b
.
However, for M-array signaling (e.g., MPAM and MPSK), the bit
error probability depends on both the symbol error probability and
the mapping of bits to symbols. Thus, we typically compute the
symbol error probability 1
s
as a function of
s
based on the signal
space concepts of previous section and then obtain 1
b
as a function
of
b
using an exact or approximate conversion.
The approximate conversion typically assumes that the symbol
energy is divided equally among all bits, and that Gray encoding is
used so that at reasonable SNRs, one symbol error corresponds to
exactly one bit error.
These assumptions for M-array signaling lead to the approximations:
b

s
log
2
`
and 1
b

1
s
log
2
`
. (4)
Introduction
AWGN channels
Fading Channels
Consider BPSK modulation with coherent detection and perfect
recovery of the carrier frequency and phase. With binary modulation
each symbol corresponds to one bit, so the symbol and bit error rates
are the same. The transmitted signal is :
1
(t) = p(t)co:(2)
t
t) to
send a 0 bit and :
2
(t) = p(t)co:(2)
t
t) to send a 1 bit. Note
that for binary modulation where ` = 2, there is only one way to
make an error and d
n.n
is the distance between the two signal
constellation points, so the probability of error is also the bound:
1
b
= Q
(
d
n.n
2
0
)
. (5)
In previous chapter, we have d
n.n
= s
1
s
2
= () = 2.
The energy-per-bit can be determined by
1
b
=
0
:
2
1
(t)dt =
0
:
2
2
(t)dt =
2
p
2
(t) cos
2
(2)
t
t)dt =
2
.
(6)
Introduction
AWGN channels
Fading Channels
Error probability for BPSK and QPSK (cont.)
Thus, the signal constellation for BPSK in terms of energy-per-bit is
given by s
0
=
1
b
and s
1
=
1
b
. This yields the minimum
distance d
n.n
= 2 = 2
1
b
. Substituting this into (5) yields
1
b
= Q
(
2
1
b
2
0
)
= Q
(
21
b
0
)
= Q
(
2
b
)
. (7)
QPSK modulation consists of BPSK modulation on both the
in-phase and quadrature components of the signal. With perfect
phase and carrier recovery, the received signal components
corresponding to each of these branches are orthogonal. Therefore,
the bit error probability on each branch is the same as for
BPSK:1
b
= Q
(
2
b
)
.
Introduction
AWGN channels
Fading Channels
Error probability for BPSK and QPSK (cont.)
The symbol error probability equals the probability that either
branch has a bit error:
1
s
= 1
[
1 Q
(
2
b
)]
2
. (8)
Example: Find the bit error probability 1
b
and symbol error
probability 1
s
of QPSK assuming
b
= 7 dB. Solution: We have
b
= 10
710
= 5.012, then 1
b
= Q
(
2
b
)
= 7.726 10
4
and
1
s
= 1
[
1 Q
(
2
b
)]
2
= 1.545 10
3
.
Introduction
AWGN channels
Fading Channels
Approximate symbol and bit error probabilities for typical
modulations
Many of the approximations or exact values for 1
s
derived above for
coherent modulation are in the following form:
1
s
(
s
) c
1
Q
(
a
1
s
)
(9)
where c
1
and a
1
depend on the type of approximation and the
modulation type. In the below table, we summarize the specic values of
c
1
and a
1
for common 1
s
expressions for PSK, QAM, and FSK
modulations based on the derivations in the prior sections.
Modulation P
s
(
s
) P
b
(
b
)
BFSK: P
b
= Q
BPSK: P
b
= Q
2
b
QPSK,4QAM: P
s
2 Q
P
b
Q
2
b
MPAM: P
s

2(M1)
M
Q
6
s
M
2
1
P
b

2(M1)
M log
2
M
Q
6
b
log
2
M
(M
2
1)
MPSK: P
s
2Q
2
s
sin(/M)
P
b

2
log
2
M
Q
2
b
log
2
M sin(/M)
Rectangular MQAM: P
s

4(
M1)
M
Q
3
s
M1
P
b

4(
M1)
M log
2
M
Q
3
b
log
2
M
(M1)
Nonrectangular MQAM: P
s
4Q
3
s
M1
P
b

4
log
2
M
Q
3
b
log
2
M
(M1)
Introduction
AWGN channels
Fading Channels
Introduction
Outage probability
Introduction
In AWGN the probability of symbol error depends on the received
SNR
s
. In a fading channel, the received signal power varies
randomly over distance or time due to shadowing and/or multipath
fading. Thus, in fading
s
is a random variables with distribution
j
;
(), and therefore 1

s
(
s
) is also random.
The performance metric when
s
is random depends on the rate of
change of the fading. There are three dierent performance criteria
that can be used to characterize the random variable 1
s
:
The outage probability,
, dened as the probability that
falls
below a given value corresponding to the maximum allowable
.
The average error probability,
, averaged over the distribution of
.
Introduction
AWGN channels
Fading Channels
Introduction
Outage probability
Introduction (cont.)
If the power of the received signal (with fading) is changing slowly
(slow-fading), then a deep fade will aect many simultaneous
symbols. Thus, fading may lead to large error bursts, which cannot
be corrected for with coding of reasonable complexity. Therefore,
these error bursts can seriously degrade end-to-end performance.
In this case acceptable performance cannot be guaranteed over all
time or, equivalently, throughout a cell, without drastically
increasing transmit power. Under these circumstances, an outage
probability is specied so that the channel is deemed unusable for
some fraction of time or space.
Introduction
AWGN channels
Fading Channels
Introduction
Outage probability
Outage probability
The outage probability relative to 1
ou|
is dened as
1
ou|
= j (
s
<
0
) =
;
0
0
j
;
()d. (10)
where
0
typically species the minimum SNR required for
acceptable performance. For example, if we consider digitized voice,
1
b
= 10
3
is an acceptable error rate since it generally cannot be
detected by the human ear. Thus, for a BPSK signal in Rayleigh
fading,
b
< 7 dB would be declared an outage, so we set
0
= 7 dB.
In Rayleigh fading with j
;
() =
1
;
c
;
, one will have

1
ou|
=
;
0
0
1
s
c
;
d = 1 c
;
0
;
. (11)
Introduction
AWGN channels
Fading Channels
Introduction
Outage probability
Outage probability (cont.)
Example: Determine the required
b
for BPSK modulation in slow
Rayleigh fading such as 50% of the time (or in space),
1
b
(
b
) < 10
4
.
Solution: For BPSK modulation in AWGN, the target BER is
obtained at 8.5 dB (i.e., for 1
b
(
b
) = Q(
2
b
), one have
1
b
(
10
0.85
)
= 10
4
). Thus,
0
= 8.5 dB, since we want
1
ou|
= j(
b
<
0
), we have
b
=

0
ln(1 1
ou|
)
=
10
.85
ln(1 .05)
= 21.4 dB. (12)
Introduction
AWGN channels
Fading Channels
Introduction
Outage probability
The average probability of error is used as a performance metric
when
s
is roughly constant over a symbol time. Then the averaged
probability of error is computed by integrating the error probability
in AWGN over the fading distribution:
1
s
=

0
1
s
()j
;
()d. (13)
where 1
s
() is the probability of symbol error in AWGN channels
with SNR , which can be approximated by the expressions in the
aforementioned table.
For a given distribution of the fading amplitude : (i.e., Rayleigh,
Rician, log-normal, etc.), we compute j
;
() by making the change

of variable
j
;
()d = j(:)d:. (14)

Introduction
AWGN channels
Fading Channels
Introduction
Outage probability
Average probability of error (cont.)
For instance, in Rayleigh fading (refer to Zhengs paper on Modied
Jake channel model), the received signal amplitude r has the
Rayleigh distribution
j(:) =
:
o
2
c
:
2
(2c
2
)
, : 0. (15)
The SNR per symbol for a given amplitude : is
=
:
2
T
s
2o
2
n
. (16)
where o
2
n
=
0
,2 is the PSD of the noise in the in-phase and
quadrature branches.
Dierentiating both sides of this expression yields
d =
:T
s
o
2
n
d:. (17)
Introduction
AWGN channels
Fading Channels
Introduction
Outage probability
Substituting (16) and (17) into (15) and then (14) yields
j
;
() =
o
2
n
o
2
T
s
c
;c
2
(c
2
T
)
. (18)
Since the average SNR per symbol
s
is just o
2
T
s
,o
2
n
, one can
rewrite (18) as
j
;
() =
1
s
c
;;
, (19)
which is exponential. For binary signaling, this reduces to
j
;
() =
1
b
c
;;
. (20)
Introduction
AWGN channels
Fading Channels
Introduction
Outage probability
Integrating the error probability of BPSK in AWGN over the
distribution (20) yields the following average probability of error for
BPSK in Rayleigh fading:
1
b
=
1
2
(
1

b
1 +
b
)
1
4
b
. (21)
where the approximation holds for large
b
.
If we use the general approximation 1
s
c
1
Q
(
a
1
s
)
then the
average probability of symbol error in Rayleigh fading can be
approximated as
1
s

0
c
1
Q
(
a
1
)
1
s
c
;;
d
s
=
c
n
2
(
1
.5a
1
s
1 + .5a
1
s
)
c
1
2a
1
s
.
Introduction
AWGN channels
Fading Channels
Introduction
Outage probability
Average probability of error: Numerical results of BPSK
0 5 10 15 20
10
6
10
5
10
4
10
3
10
2
10
1
10
0
b
(dB)
P
b
AWGN
Rayleigh fading
Figure 1: Average probability of bit
error
for BPSK in Rayleigh Fading

and AWGN.
0 10 20 30 40
10
9
10
8
10
7
10
6
10
5
10
4
10
3
10
2
10
1
10
0
Average SNR
b
(dB)
A
v
e
r
a
g
e

B
i
t

E
r
r
o
r

P
r
o
b
a
b
i
l
i
t
y
m = 0.5
m = 1
(Rayleigh)
m = 1.5
m = 2
m = 2.5
m = 3
m = 4
m = 5
m =
(No fading)
Figure 2: Average
for BPSK in
Nakagami Fading.
Introduction
AWGN channels
Fading Channels
Introduction
Outage probability
Average probability of error: Numerical results of MQAM
0 5 10 15 20 25 30 35 40
10
6
10
5
10
4
10
3
10
2
10
1
10
0
b
(dB)
P
b
M = 4
M = 16
M = 64
Rayleigh fading
AWGN
Figure 3: Average
for MQAM in Rayleigh Fading and AWGN.

Introduction
Linear Block Codes
Section 3.4: Channel Coding
Mobile communications-Chapter 3: Physical-layer transmissions Section 3.4: Channel Coding 1
Introduction
Linear Block Codes
Purpose of using channel coding
Coding allows bit errors (introduced by transmission of a modulated
signal through a wireless channel) to be either (i) detected or (ii)
corrected by a decoder in the receiver..
Coding can be considered as the embedding of signal constellation
points in a higher dimensional signaling space than is needed for
communications.
By going to a higher dimensional space, the distance between points
can be increased, which provides for better error correction and
detection.
Introduction
Linear Block Codes
Overview of code design
The main reason to apply error correction coding in a wireless
system is to reduce the probability of bit or block error.
The bit error probability 1
b
for a coded system is the probability
that a bit is decoded in error.
The block error probability 1
bl
, also called the packet error rate, is
the probability that one or bits in a block of coded bits is decoded in
error.
Block error probability is useful for packet data systems where bits
are encoded and transmitted in blocks.
The amount of error reduction provided by a given code is typically
characterized by its coding gain in AWGN and its diversity gain in
fading.
Introduction
Linear Block Codes
BER performance of channel coding
Uncoded Coded
10
6
10
10
4
2
C
g1
C
g2
P
b
SNR (dB)
4 5 6 7 8 9 10 11
10
7
10
6
10
5
10
4
10
3
10
2
Eb/No (dB)
D
e
c
o
d
e
d

B
E
R
Uncoded
Hamming (7,4) t=1
Hamming (15,11) t=1
Hamming (31,26) t=1
Extended Golay (24,12) t=3
BCH (127,36) t=15
BCH (127,64) t=10
Introduction
Linear Block Codes
BER performance of turbo coding
4 5 6 7 8 9 10 11
10
7
10
6
10
5
10
4
10
3
10
2
Eb/No (dB)
D
e
c
o
d
e
d

B
E
R
Uncoded
Hamming (7,4) t=1
Hamming (15,11) t=1
Hamming (31,26) t=1
Extended Golay (24,12) t=3
BCH (127,36) t=15
BCH (127,64) t=10
0.5 1 1.5 2
10
-7
10
-6
10
-5
10
-4
10
-3
10
-2
10
-1
10
0
E
b
/N
o
in dB
B
E
R
1 iteration
2 iterations
3 iterations
6 iterations
10 iterations
18 iterations
Introduction
Linear Block Codes
Introduction
Binary linear block codes
Generator matrix
Parity Check Matrix and Syndrome Testing
Hard Decision Decoding (HDD)
Introduction
Linear block codes are simple codes that are basically an extension
of single-bit parity check codes for error detection.
A single-bit parity check code is one of the most common forms of
detecting transmission errors.
This code uses one extra bit in a block of n data bits to indicate
whether the number of 1s in a block is odd or even.
Linear block codes extend this notion by using a larger number of
parity bits to either detect more than one error or correct for one or
more errors.
Introduction
Linear Block Codes
Introduction
Generator matrix
A binary block code generates a block of n coded bits from /
information bits, called an (n, /) binary block code.
The coded bits are also called codeword symbols. The n codeword
symbols can take on 2
n
possible values corresponding to all possible
combinations of the n binary bits.
We select 2
|
codewords from these 2
n
possibilities to form the code,
such that each / bit information block is uniquely mapped to one of
these 2
|
codewords.
Introduction
Linear Block Codes
Introduction
Generator matrix
Binary linear block codes (cont.)
The rate of the code is 1
t
= /,n information bits per codeword
symbol.
If we assume that codeword symbols are transmitted across the
channel at a rate of 1
s
symbols/second, then the information rate
associated with an (n, /) block code is 1
b
= 1
t
1
s
=
|
n
1
s
bits/second. Thus, we see that block coding reduces the data rate
compared to what we obtain with uncoded modulation by the code
rate 1
t
.
A block code is called a linear code when the mapping of the /
information bits to the n codeword symbols is a linear mapping.
Introduction
Linear Block Codes
Introduction
Generator matrix
The vector space of binary -tuples
The set of all binary n-tuples 1
n
is a vector space over the binary
eld, which consists of the two elements 0 and 1.
All elds have two operations, addition and multiplication: for the
binary eld these operations correspond to binary addition (modulo
2 addition) and standard multiplication.
A subset o of 1
n
is called a subspace if it satises the following
conditions:
The all-zero vector is in
The set is closed under addition, such that if
and
,
then
.
An (n, /) block code is linear if the 2
|
length-n codewords of the
code form a subspace of 1
n
. Thus, if C
.
and C
are two codewords

in an (n, /) linear block code, then C
.
+C
must form another

codeword of the code.
Introduction
Linear Block Codes
Introduction
Generator matrix
The vector space of binary -tuples: an example
Example: The vector space 1
3
consists of all binary tuples of length
3: 1
3
= {[000], [001], [010], [011], [100], [101], [110], [111]}.
Note that 1
3
is a subspace of itself, since it contains the all zero
vector and is closed under addition. Determine which of the
following subsets of B3 form a subspace:
1
= {[000], [001], [100], [101]}
2
= {[000], [100], [110], [111]}
3
= {[001], [100], [101]}
Solution: It is easily veried that
1
is a subspace, since it contains
the all-zero vector and the sum of any two tuples in
1
is also in
1
.
2
is not a subspace since it is not closed under addition, as
110 + 111 = 001
2
.
3
is not a subspace since it is not closed under addition
(001 + 001 = 000
3
) and it does not contain the all zero vector.
Introduction
Linear Block Codes
Introduction
Generator matrix
Hamming distance
Intuitively, the greater the distance between codewords in a given
code, the less chance that errors introduced by the channel will cause
a transmitted codeword to be decoded as a dierent codeword.
We dene the Hamming distance between two codewords C
.
and
C
, denoted as d(C
.
, C
) or d
.
, as the number of elements in
which they dier:
d
.
=
n
l=1
C
.
(|) +C
(|), (1)
where C
.
(|) denotes the |th bit in C
.
.
For example, if C
.
= [00101] and C
= [10011] then d
.
= 3.
Introduction
Linear Block Codes
Introduction
Generator matrix
The weight of a given code
We dene the weight of a given codeword C
.
as the number of 1s in
the codeword, so C
.
= [00101] has weight 2.
The weight of a given codeword C
.
is just its Hamming distance d
0.
with the all zero codeword C
0
= [00 . . . 0] or, equivalently, the sum
of its elements:
n(C
.
) =
n
l=1
C
.
(|). (2)
Since 0 + 0 = 1 + 1 = 0, the Hamming distance between C
.
and C
is equal to the weight of C

.
+C
.
Introduction
Linear Block Codes
Introduction
Generator matrix
The weight of a given code (cont.)
Since the Hamming distance between any two codewords equals the
weight of their sum, we can determine the minimum distance
between all codewords in a code by just looking at the minimum
distance between all codewords and the all zero codeword.
Thus, we dene the minimum distance of a code as
d
min
= min
.,.=0
d
0.
, (3)
which implicitly denes C
0
as the all-zero codeword. It is noted that
the minimum distance of a linear block code is a critical parameter
in determining its probability of error.
Introduction
Linear Block Codes
Introduction
Generator matrix
Generator matrix
The generator matrix is a compact description of how codewords are
generated from information bits in a linear block code.
The design goal in linear block codes is to nd generator matrices
such that their corresponding codes are easy to encode and decode
yet have powerful error correction/detection capabilities.
Consider an (n, /) code with / information bits denoted as
U
.
= [n
.1
, . . . , n
.|
] (4)
that are encoded into the codeword
C
.
= [c
.1
, . . . , c
.n
] (5)
Introduction
Linear Block Codes
Introduction
Generator matrix
Generator matrix (cont.)
We represent the encoding operation as a set of n equations:
c
.
= n
.1
p
1
+ n
.2
p
2
+ . . . + n
.|
p
|
, , = 1, . . . , n, (6)
where p
.
is binary (0 or 1) and binary (standard) multiplication is
used. One can write these n equations in matrix form as
C
.
= U
.
G, (7)
where the / n generator matrix G for the code is dened as
G =
p
11
p
12
. . . p
1n
p
21
p
22
. . . p
2n
.
.
.
.
.
.
.
.
.
.
.
.
p
|1
p
|2
. . . p
|n
. (8)
Introduction
Linear Block Codes
Introduction
Generator matrix
If we denote the |th row of G as g
l
= [p
l1
, . . . , p
ln
] then we can
write any codeword C
.
as linear combinations of these row vectors
as follows:
C
.
= n
.1
g
1
+ n
.2
g
2
+ . . . + n
.|
g
|
. (9)
Since a linear (n, /) block code is a subspace of dimension / in the
larger n-dimensional space, the / row vectors {g
l
}
|
l=1
of G must be
linearly independent, so that they span the /-dimensional subspace
associated with the 2
|
codewords.
Thus, G has rank /.
Since the set of basis vectors for this subspace is not unique, the
generator matrix is also not unique.
Introduction
Linear Block Codes
Introduction
Generator matrix
A systematic linear block code is described by a generator matrix of
the form
G = [I
|
P] =
1 0 . . . 0 j
11
j
12
. . . j
1(n|)
0 1 . . . 0 j
21
j
22
. . . j
2(n|)
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
0 0 . . . 1 j
|1
j
|2
. . . j
|(n|)
,
where I
|
is a / / identity matrix and P is a / (n /) matrix
that determines the redundant, or parity, bits to be used for error
correction or detection.
The codeword output from a systematic encoder is of the form
C
.
= U
.
G = U
.
[I
|
P] = [n
.1
, . . . , n
.|
, j
1
, . . . , j
n|
] , (10)
where
j
= n
.1
j
1
+ . . . + n
.|
j
|
, , = 1, . . . , n /. (11)
Introduction
Linear Block Codes
Introduction
Generator matrix
An example: Problem statement
Systematic linear block codes are typically implemented with n /
modulo-2 adders tied to the appropriate stages of a shift register.
The resultant parity bits are appended to the end of the information
bits to form the codeword.
Find the corresponding implementation for generating a (7, 4) binary
code with the generator matrix
G =
1 0 0 0 1 1 0
0 1 0 0 1 0 1
0 0 1 0 0 0 1
0 0 0 1 0 1 0
,
Introduction
Linear Block Codes
Introduction
Generator matrix
An example: Solutions
The matrix G is already in systematic form with
P =
1 1 0
1 0 1
0 0 1
0 1 0
,
From (11), the rst parity bit in the codeword is
j
1
= n
.1
j
11
+ n
.2
j
21
+ n
.3
j
31
+ n
.4
j
41
= n
.1
+ n
.2
+ 0 + 0.
Similarly, j
2
= n
.1
+ n
.4
, j
3
= n
.2
+ n
.3
.
The codeword output is [n
.1
, n
.2
, n
.3
, n
.4
, j
1
, j
2
, j
3
]
+ + +
u
i1
u
u u
i2
i3 i4
p
1 2
p
3
p
Introduction
Linear Block Codes
Introduction
Generator matrix
The parity check matrix is used to decode linear block codes with
generator matrix G. The parity check matrix H corresponding to a
generator matrix G = [I
|
P] is dened as
H =
[
P
T
I
n|
]
. (12)
It is easily veried that GH
T
= 0
|,n|
, where 0
|,n|
denotes an
all-zero / (n /) matrix. Thus,
C
.
H
T
= U
.
GH
T
= 0
n|
, (13)
where 0
n|
denotes the all-zero row vector of length n /.
Multiplication of any valid codeword with the parity check matrix
results in an all-zero vector. This property is used to determine
whether the received vector is a valid codeword or has been
corrupted, based on the notion of syndrome testing, as dened now.
Introduction
Linear Block Codes
Introduction
Generator matrix
Parity Check Matrix and Syndrome Testing (cont.)
Let R be the received codeword resulting from transmission of
codeword C.
In the absence of channel errors, R = C.
If the transmission is corrupted, one or more of the codeword
symbols in R will dier from those in C. Therefore, the received
codeword can be written as
R = C+e, (14)
where e = [c
1
, c
2
, . . . , c
n
] is the error vector indicating which
codeword symbols were corrupted by the channel.
We dene the syndrome of R as
S = RH
T
. (15)
If R is a valid codeword (i.e.,R = C
.
for some i), then
S = C
.
H
T
= 0
n|
.
Introduction
Linear Block Codes
Introduction
Generator matrix
Thus, the syndrome equals the all-zero vector if the transmitted
codeword is not corrupted, or is corrupted in a manner such that
the received codeword is a valid codeword in the code that is
dierent from the transmitted codeword.
If the received codeword R contains detectable errors, then
S
|
= 0
n|
.
If the received codeword contains correctable errors, then the
syndrome identies the error pattern corrupting the transmitted
codeword, and these errors can then be corrected.
Note that the syndrome is a function only of the error pattern e and
not the transmitted codeword C, since
S = RH
T
= (C+e)H
T
= CH
T
+eH
T
= 0
n|
+eH
T
. (16)
Introduction
Linear Block Codes
Introduction
Generator matrix
Since S = eH
T
corresponds to n/ equations in n unknowns, there
are 2
|
possible error patterns that can produce a given syndrome S.
However, since the probability of bit error is typically small and
independent for each bit, the most likely error pattern is the one
with minimal weight, corresponding to the least number of errors
introduced in the channel.
Thus, if an error pattern e is the most likely error associated with a
given syndrome S, the transmitted codeword is typically decoded as
C = R+e = C+e +e. (17)

When the most likely error pattern does occur, i.e., e = e, then
C = C, i.e., the corrupted codeword is correctly decoded. The

decoding process and associated error probability will be covered in
later Sections.
Introduction
Linear Block Codes
Introduction
Generator matrix
Let C
u
denote a codeword in a given (n, /) code with minimum
weight (excluding the all-zero codeword).
Then C
u
H
T
= 0
n|
is just the sum of d
min
columns of H
T
, since
d
min
equals the number of 1s (the weight) in the minimum weight
codeword of the code.
Since the rank of H
T
is at most n /, this implies that the
minimum distance of an (n, /) block code is upper bounded by
d
min
n / + 1. (18)
Introduction
Linear Block Codes
Introduction
Generator matrix
The probability of error for linear block codes depends on whether
the decoder uses soft decisions or hard decisions.
In hard decision decoding (HDD) each coded bit is demodulated as
a 0 or 1, i.e. the demodulator detects each coded bit (symbol)
individually.
For example, in BPSK, the received symbol is decoded as a 1 if it is
closer to
1
b
and as 0 if it closer to
1
b
.
Hard decision decoding uses minimum-distance decoding based on
Hamming distance.
In minimum-distance decoding the n bits corresponding to a
codeword are rst demodulated, and the demodulator output is
passed to the decoder.
Introduction
Linear Block Codes
Introduction
Generator matrix
Hard Decision Decoding (cont.)
The decoder compares this received codeword to the 2/ possible
codewords comprising the code, and decides in favor of the
codeword that is closest in Hamming distance (diers in the least
number of bits) to the received codeword.
Mathematically, for a received codeword R the decoder uses the
formula:
choose C
subject to d (C
, R) d (C
.
, R) , i = ,. (19)
If there is more than one codeword with the same minimum distance
to R, one of these is chosen at random by the decoder.
Maximum-likelihood decoding picks the transmitted codeword that
has the highest probability of having produced the received
codeword, i.e. given the received codeword R, the
maximum-likelihood decoder chooses the codeword C
as
C
= arg max
.=1,...,2
j(RC
.
). (20)
Introduction
Linear Block Codes
Introduction
Generator matrix
Hard Decision Decoding (cont.)
Since the most probable error event in an AWGN channel is the
event with the minimum number of errors needed to produce the
received codeword, the minimum-distance criterion (19) and the
maximum-likelihood criterion (8.33) are equivalent.
Once the maximum-likelihood codeword C
.
is determined, it is
decoded to the / bits that produce codeword C
.
.
C
1
C
C
C
2
3
4
min
d
t
Outline
Introduction
Fundamentals of OFDM transmissions
Possible research problems in OFDM transmissions
Section 3.4: Orthogonal Frequency Division Multiplexing
(OFDM) Fundamentals
Email: nlhung@dut.udn.vn; nnguyenlehung@yahoo.com
Mobile Communications - Chapter 3: Physical-layer transmissions Section 3.4: OFDM fundamentals 1
Outline
Introduction
1
Introduction
Development of mobile communications systems - Revisit
Mobile broadband technology evolution - Revisit
2
Functional blocks in OFDM transmitter and receiver
Transmitted OFDM signals
Time-variant multipath channels
Received OFDM signals
MQAM symbol detection in the frequency domain
3
Carrier frequency oset (CFO) estimation
Doubly selective channel estimation
Outline
Introduction
time
code
frequency
code
space
FDMA (1G)
e.g., AMPS ~ 1980s
TDMA (2G)
e.g., GSM ~ 1990s
OFDM, SDMA (4G)
e.g., WiMAX, LTE
2010s
CDMA (3G)
frequency
time
time
~ 100 Mbps (mobile)
frequency
frequency
~ 5.8 Mbps (uplink)
~ 50 Kbps
Outline
Introduction
time
code
frequency
code
space
FDMA (1G)
e.g., AMPS ~ 1980s
TDMA (2G)
e.g., GSM ~ 1990s
OFDM, SDMA (4G)
e.g., WiMAX, LTE
2010s
CDMA (3G)
frequency
time
time
~ 100 Mbps (mobile)
frequency
frequency
~ 5.8 Mbps (uplink)
~ 50 Kbps
Outline
Introduction
LTE
Advanced
802.16m
R 2.0
WCDMA
3GPP
HSDPA HSPA
Mobile
WiMAX
802.16e
R 1.0
EVDO
Rev A
EVDO
Rev B
CDMA 2000
3GPP2
LTE
802.16e
R 1.5
2005 2009 2010 2006 2007 2008 2011 2012
CDMA based OFDMA based
R99
4G
IMT-Advanced
The last decade has witnessed numerous intensive studies in employing
orthogonal frequency division multiplexing (OFDM) for the emerging
broadband communications systems (e.g., WiFi, WiMAX, LTE) to exploit
its high spectral eciency and robustness against multipath (frequency
selective) fading channels.
Outline
Introduction
Burst structure
S/P
IFFT
Insert CP
DAC RF
Clk
Osc
RF
LO
Pilotinsertion
Conv.
Encoder
P/S
c
i
MQAM
mapping
Pilot OFDM
symbols
Data OFDM
symbols
Information
bits,u
i
RF
LO
Clk
Osc
MQAM
Demapper
RF ADC
S/P decoder
-1
FFT
CP
Information
bits,u
i
Outline
Introduction
Consider a coded OFDM system using `-ary quadrature amplitude
modulation (M-QAM).
The encoded bit stream is bit-interleaved. Then, the resulting
sequence of interleaved bits is organized as a sequence of Q-bit
tuples {d
|,n
} where Q = log
2
` and
d
|,n
= [d
|,n,0
, ..., d
|,n,Q1
]
T
.
The sequence is further mapped to a complex-valued symbol
A
|,n
where is the `-ary modulation signaling alphabet, and
: and / denote the indices of OFDM symbol and subcarrier,
respectively.
m
X
, 0
m N
X
, 1
1 , 0 m
X
1 , 1 m N
X
m k
X
,
1 , m k
X
T
Q m k m k m k
d d ] ,..., [
1 , , 0 , , ,
d
The m-th OFDM symbol
The k-th subcarrier
Outline
Introduction
Consider a coded OFDM system using `-ary quadrature amplitude
modulation (M-QAM).
The encoded bit stream is bit-interleaved. Then, the resulting
sequence of interleaved bits is organized as a sequence of Q-bit
tuples {d
|,n
} where Q = log
2
` and
d
|,n
= [d
|,n,0
, ..., d
|,n,Q1
]
T
.
The sequence is further mapped to a complex-valued symbol
A
|,n
where is the `-ary modulation signaling alphabet, and
: and / denote the indices of OFDM symbol and subcarrier,
respectively.
m
X
, 0
m N
X
, 1
1 , 0 m
X
1 , 1 m N
X
m k
X
,
1 , m k
X
T
Q m k m k m k
d d ] ,..., [
1 , , 0 , , ,
d
The m-th OFDM symbol
The k-th subcarrier
Outline
Introduction
Transmitted OFDM signals (cont.)
Each OFDM symbol consists of information bearing subcarriers
A
|,n
, / = 0, ..., 1, where is the size of the fast Fourier transform
(FFT) and inverse-FFT (IFFT) used in the multicarrier transmission.
m n
x
,
QAM symbols
m k
X
,
CP samples

IFFT

P/S
DAC RF
Burst structure
S/P
IFFT Insert CP
DAC RF
Clk
Osc
RF
LO
Pilotinsertion
Conv.
Encoder
P/S
c
i
MQAM
mapping
Information
bits,u
i
Outline
Introduction
Copy and paste the last N
g
samples

CP portion CP portion
m N N
g
x
,
.
m N
x
, 1 m
x
, 0 m
x
, 1 m
x
, 2
.
m N N
g
x
, 1 m N N
g
x
,
.
m N
x
, 1
one OFDM symbol of samples
After inverse FFT (IFFT) and cyclic prex (CP) insertion, the
transmitted baseband signal of the :th OFDM symbol can be written as
r
n,n
=
1
|=0
A
|,n
exp
(
,2/n
)
, (1)
where n {
, ..., 0, ..., 1},
denotes the CP length.

Outline
Introduction
g
samples

m N N
g
x
,
.
m N
x
, 1 m
x
, 0 m
x
, 1 m
x
, 2
.
m N N
g
x
, 1 m N N
g
x
,
.
m N
x
, 1
After inverse FFT (IFFT) and cyclic prex (CP) insertion, the
transmitted baseband signal of the :th OFDM symbol can be written as
r
n,n
=
1
|=0
A
|,n
exp
(
,2/n
)
, (1)
where n {
, ..., 0, ..., 1},
denotes the CP length.

Outline
Introduction
Two Main
Multipaths
Local
Scattering
j
n
=
l
r
nl
l,n
+n
b
(:,\) (2)
The |th (time-variant) channel tap gain that includes the eect of
transmit-receive lters and time-variant channel propagation is denoted
by
l,n,n
where n and : stand for the indices of the time-domain sample
and OFDM symbol, respectively.
Outline
Introduction
Time-variant multipath channels (cont.)
In the time-variant multipath channels, a number of basis expansion
models (BEMs) can be employed for reducing the number of channel
parameters while capturing the time-variation of the channels.
Using BEMs,
l,n,n
(after CP removal) can be approximated by
l,n,n
=
Q
g=1
/
n+
+n
s
,g
c
g,l
, | {0, ..., 1 1}, (3)
where
s
= +
denotes the OFDM symbol length after CP

insertion, n = 0, ..., 1, : = 0, ..., ` 1 and ` is the number
of both data and pilot OFDM symbols in a burst. /
n+
+n
s
,g
stand for the th basis function values of the used BEM.
g
samples

m N N
g
x
,
.
m N
x
, 1 m
x
, 0 m
x
, 1 m
x
, 2
.
m N N
g
x
, 1 m N N
g
x
,
.
m N
x
, 1
Outline
Introduction
In the time-variant multipath channels, a number of basis expansion
models (BEMs) can be employed for reducing the number of channel
parameters while capturing the time-variation of the channels.
Using BEMs,
l,n,n
(after CP removal) can be approximated by
l,n,n
=
Q
g=1
/
n+
+n
s
,g
c
g,l
, | {0, ..., 1 1}, (3)
where
s
= +
denotes the OFDM symbol length after CP

insertion, n = 0, ..., 1, : = 0, ..., ` 1 and ` is the number
of both data and pilot OFDM symbols in a burst. /
n+
+n
s
,g
stand for the th basis function values of the used BEM.
g
samples

m N N
g
x
,
.
m N
x
, 1 m
x
, 0 m
x
, 1 m
x
, 2
.
m N N
g
x
, 1 m N N
g
x
,
.
m N
x
, 1
Outline
Introduction
It is noted that the use of BEMs oers a signicant dimension
reduction in the time-variant channel representation, i.e., Q `.
For instance, in the current LTE system settings, three LTE time
slots contain 21 OFDM symbols with 128-FFT (the smallest FFT
size used in the LTE settings) and the resulting number of the
time-variant channel parameters corresponding to one channel tap
gain
l,n,n
in (3) will be ` = 128 21 = 2688.
By using the discrete prolate spheroidal (DPS)-BEM as shown in (3),
the number of the basis functions Q can vary from 3 (for low user
speeds, e.g., about 10km/h) to 5 (for moderate user speeds, e.g.,
about 100km/h) under the required MSE of the DPS-BEM-based
channel approximation below 10
10
as shown in Fig. 1.
Outline
Introduction
gain
l,n,n
in (3) will be ` = 128 21 = 2688.
10
as shown in Fig. 1.
Outline
Introduction
gain
l,n,n
in (3) will be ` = 128 21 = 2688.
10
as shown in Fig. 1.
Outline
Introduction
Numerical results of BEM-based channel tting
1 2 3 4 5 6 7 8 9 10 11 12 13 14
10
30
10
25
10
20
10
15
10
10
10
5
10
0
Number of used basis functions
M
S
E

o
f

D
P
S
B
E
M
b
a
s
e
d

c
h
a
n
n
e
l

f
i
t
t
i
n
g

1 km/h
5 km/h
10 km/h
20 km/h
30 km/h
40 km/h
50 km/h
60 km/h
70 km/h
80 km/h
90 km/h
100 km/h
120 km/h
140 km/h
160 km/h
180 km/h
200 km/h
Figure 1: Normalized MSE of DPS-BEM-based approximation (tting) of
time-varying channels generated by Jakes model under dierent mobile speeds.
Outline
Introduction
Over the above doubly-selective channels, after CP removal, the nth
received sample in the :th OFDM symbol can be represented by
j
n,n
=
J1
l=0
l,n,n
r
nl,n
+.
n,n
, (4)
where n = 0, ..., 1 and .
n,n
is the additive white Gaussian noise
(AWGN) with variance
o
.
m N N
g
x
,
.
m N
x
, 1 m
x
, 0 m
x
, 1 m
x
, 2
.
m N N
g
x
, 1 m N N
g
x
,
.
m N
x
, 1
CP portion
m N
g
h
, 0 ,
.
m
h
, 0 , 1 m
h
, 0 , 0
Channel impulse response
m m m N m m N N m N m
x h x h x h y
g g
, 0 , 0 , 0 , 1 , 0 , 1 , , 0 , , 0

Outline
Introduction
Received OFDM signals in the frequency domain
At the receiver, after performing CP removal and FFT, the received
samples in the frequency domain can be determined by
Y
|,n
=
1
n=0
j
n,n
c
2t|n
= H
|,n
A
|,n
+j
|,n
+7
|,n
, (5)
where H
|,n
=
1
1
n=0
J1
l=0

l,n,n
c
2t|l
,
7
|,n
=
1
1
n=0
.
n,n
c
(
2!r
.
)
and j
|,n
=
1
1
.=0,. =|
A
.,n
1
n=0
(
J1
l=0

l,n,n
c
2t.l
)
c
2tn(.|)
.
It is noted that the received samples in the time domain are:
j
n,n
=
J1
l=0
l,n,n
r
nl,n
+.
n,n
.
Outline
Introduction
Y
|,n
=
1
n=0
j
n,n
c
2t|n
= H
|,n
A
|,n
+j
|,n
+7
|,n
, (5)
where H
|,n
=
1
1
n=0
J1
l=0

l,n,n
c
2t|l
,
7
|,n
=
1
1
n=0
.
n,n
c
(
2!r
.
)
and j
|,n
=
1
1
.=0,. =|
A
.,n
1
n=0
(
J1
l=0

l,n,n
c
2t.l
)
c
2tn(.|)
.
j
n,n
=
J1
l=0
l,n,n
r
nl,n
+.
n,n
.
Outline
Introduction
Y
|,n
=
1
n=0
j
n,n
c
2t|n
= H
|,n
A
|,n
+j
|,n
+7
|,n
, (5)
where H
|,n
=
1
1
n=0
J1
l=0

l,n,n
c
2t|l
,
7
|,n
=
1
1
n=0
.
n,n
c
(
2!r
.
)
and j
|,n
=
1
1
.=0,. =|
A
.,n
1
n=0
(
J1
l=0

l,n,n
c
2t.l
)
c
2tn(.|)
.
j
n,n
=
J1
l=0
l,n,n
r
nl,n
+.
n,n
.
Outline
Introduction
Received OFDM signals over block-fading channel
Y
|,n
=
1
n=0
j
n,n
c
2t|n
= H
|,n
A
|,n
+ 0 +7
|,n
, (6)
where H
|,n
=
J1
l=0

l
c
2t|l
.
j
n,n
=
J1
l=0
l
r
nl,n
+ .
n,n
.
Burst structure
S/P
IFFT Insert CP
DAC RF
Clk
Osc
RF
LO
Pilotinsertion
Conv.
Encoder
P/S
c
i
MQAM
mapping
Information
bits,u
i
RF
LO
Clk
Osc
MQAM
Demapper
RF ADC S/P decoder
-1
FFT CP
Information
bits,u
i
Outline
Introduction
Y
|,n
=
1
n=0
j
n,n
c
2t|n
= H
|,n
A
|,n
+ 0 +7
|,n
, (6)
where H
|,n
=
J1
l=0

l
c
2t|l
.
j
n,n
=
J1
l=0
l
r
nl,n
+ .
n,n
.
Burst structure
S/P
IFFT Insert CP
DAC RF
Clk
Osc
RF
LO
Pilotinsertion
Conv.
Encoder
P/S
c
i
MQAM
mapping
Information
bits,u
i
RF
LO
Clk
Osc
MQAM
Demapper
RF ADC S/P decoder
-1
FFT CP
Information
bits,u
i
Outline
Introduction
Y
|,n
=
1
n=0
j
n,n
c
2t|n
= H
|,n
A
|,n
+ 0 +7
|,n
, (6)
where H
|,n
=
J1
l=0

l
c
2t|l
.
j
n,n
=
J1
l=0
l
r
nl,n
+ .
n,n
.
Burst structure
S/P
IFFT Insert CP
DAC RF
Clk
Osc
RF
LO
Pilotinsertion
Conv.
Encoder
P/S
c
i
MQAM
mapping
Information
bits,u
i
RF
LO
Clk
Osc
MQAM
Demapper
RF ADC S/P decoder
-1
FFT CP
Information
bits,u
i
Outline
Introduction
Under maximum likelihood (ML) detection, the transmitted QAM
symbols can be recovered in the frequency domain by
A
|,n
= arg min
!r
Y
|,n
H
|,n
A
|,n
2
. (7)
Based on the detected QAM symbols

A
|,n
, the transmitted data
bits can be recovered accordingly by channel decoders.
It is noted that the ML symbol detection process (7) needs to know
channel frequency response (CFR) H
|,n
before providing a detected
version of the transmitted symbol

A
|,n
Possible research problem
In OFDM-based transmissions (e.g., WiFi, WiMAX, LTE systems) over
time-variant multipath channels, the problem of estimating channel
responses (i.e., H
|,n
) is of importance in research and industry as well.
Outline
Introduction
Under maximum likelihood (ML) detection, the transmitted QAM
symbols can be recovered in the frequency domain by
A
|,n
= arg min
!r
Y
|,n
H
|,n
A
|,n
2
. (7)
Based on the detected QAM symbols

A
|,n
, the transmitted data
bits can be recovered accordingly by channel decoders.
It is noted that the ML symbol detection process (7) needs to know
channel frequency response (CFR) H
|,n
before providing a detected
version of the transmitted symbol

A
|,n
Possible research problem
In OFDM-based transmissions (e.g., WiFi, WiMAX, LTE systems) over
time-variant multipath channels, the problem of estimating channel
responses (i.e., H
|,n
) is of importance in research and industry as well.
Outline
Introduction
Downlink multiuser MIMO-OFDM transmissions in
WiMAX systems (IEEE 802.16e)
M
U
-
M
I
M
O

e
n
c
o
d
e
r
Base station
User 4
FFT Decoder
I
F
F
T
User 3
FFT Decoder
User 2
FFT Decoder
User 1
FFT Decoder
Outline
Introduction
In the presence of CFO, after CP removal, the nth received sample
in the :th OFDM symbol can be represented by
j
n,n
= c
2z
.
(n+
+n
s
)
J1
l=0
l,n,n
r
nl,n
+.
n,n
, (8)
where:
= 0, ..., 1
and = denote the absolute and normalized CFOs,
respectively.
is the sampling period of the system.
As observed in (8), the presence of CFO introduces a time-domain
phase rotation that will translate into ICI in the frequency domain.
In addition, the time-variation of the multipath channels also induces
ICI in the frequency domain.
Outline
Introduction
Carrier frequency oset (CFO) estimation (cont.)
Consequently, the presence of both CFO and doubly selective
channels would incur a signicant ICI power at OFDM receivers,
giving rise to a considerable irreducible error oor in the receiver
performance.
For CFO compensation and reliable coherent data
detection/decoding, the CFO and CIR estimates are indispensible at
OFDM receivers.
To obtain the estimates, CFO and channel estimates can be
obtained either jointly or separately by existing estimation
techniques (e.g., RLS, ML, MAP,...).
Outline
Introduction
Doubly selective channel estimation: Introduction
The last decade has witnessed numerous intensive studies in
employing OFDM for broadband communication systems to exploit
its high spectral eciency and robustness against multipath
(frequency-selective) fading channels.
In the current literature, most of these studies have assumed
frequency-selective channels to be time-invariant (i.e., quasi-static or
block-fading) within a transmission burst.
This channel assumption can be used in a wireless system with
stationary and/or low-speed users.
Outline
Introduction
Doubly selective channel estimation: Introduction (cont.)
In a wireless network with rapidly moving nodes (e.g., users in cars
and trains in 4G-LTE systems), the resultant time-selectivity of the
channel impulse response (CIR) introduces a large number of
channel parameters (much greater than that of
quasi-static/block-fading channels).
In addition, the time-variation of the channel leads to a loss of
subcarrier orthogonality, resulting in inter-carrier interference (ICI) in
OFDM receivers.
Under such a scenario, the assumption of quasi-static fading
channels becomes inappropriate.
As a result, time- and frequency-selective (doubly selective) channels
should be considered in the wireless system investigation and
analysis.
Outline
Introduction
Independent Fading Paths
Receiver diversity techniques
Transmitter Diversity
Section 3.5: Diversity techniques
Mobile Communications - Chapter 3: Physical-layer transmissions Section 3.5: Diversity techniques 1
Outline
Introduction
1
Introduction
2
Space diversity
Frequency diversity
Time diversity
3
Maximal Ratio Combining (MRC)
Equal-Gain Combining (EGC)
Selection combining (SC)
Threshold Combining (TC)
4
Channel Known at Transmitter
Channel Unknown at Transmitter
Outline
Introduction
Introduction
As observed in Section 3.2, Rayleigh fading induces a very large
power penalty on the performance of modulation over wireless
channels.
One of the most powerful techniques to mitigate the eects of fading
is to use diversity-combining of independently fading signal paths.
Diversity-combining exploits the fact that independent signal paths
have a low probability of experiencing deep fades simultaneously.
These independent paths are combined in some ways such that the
fading of the resultant signal is reduced.
Diversity techniques that mitigate the eect of multipath fading are
called microdiversity.
Diversity to mitigate the eects of shadowing from buildings and
objects is called macrodiversity. Macrodiversity is generally
implemented by combining signals received by several base stations
or access points.
Outline
Introduction
Space diversity
Frequency diversity
Time diversity
Space diversity
There are many ways of achieving independent fading paths in a
wireless system.
One method is to use multiple transmit or receive antennas, also
called an antenna array, where the elements of the array are
separated in distance. This type of diversity is referred to as space
diversity.
Note that with receiver space diversity, independent fading paths are
generated without an increase in transmit signal power or bandwidth.
Coherent combining of the diversity signals leads to an increase in
SNR at the receiver over the SNR that would be obtained with just
a single receive antenna.
Space diversity also requires that the separation between antennas is
large enough so that the fading amplitudes corresponding to each
antenna are approximately independent.
Outline
Introduction
Space diversity
Frequency diversity
Time diversity
Frequency diversity
Frequency diversity is achieved by transmitting the same narrowband
signal at dierent carrier frequencies.
This technique requires additional transmit power to send the signal
over multiple frequency bands.
Spread spectrum techniques are sometimes described as providing
frequency diversity since the channel gain varies across the
bandwidth of the transmitted signal.
However, this is not equivalent to sending the same information
signal over indepedently fading paths.
Outline
Introduction
Space diversity
Frequency diversity
Time diversity
Time diversity
Time diversity is achieved by transmitting the same signal at
dierent times.
Time diversity does not require increased transmit power, but it does
decrease the data rate since data is repeated in the diversity time
slots rather than sending new data in these time slots.
Time diversity can also be achieved through coding and interleaving.
Time diversity cannot be used for stationary wireless applications,
since fading gains are highly correlated over time.
Outline
Introduction
1
1 1
T j
e a h
2
2 2
T j
e a h
M
j
M M
e a h
T
x
X
X
X
1
1
T j
e g

2
2
T j
e g

M
j
M
e g
T
De-
mod
Outline
Introduction
Maximal Ratio Combining (cont.)
In receiver diversity the independent fading paths associated with
multiple receive antennas are combined to obtain a resultant signal
that is then passed through a standard demodulator.
Under the use of ` receive antennas over at-fading (single
channel-tap, i.e., 1 = 1) channels, the received signals are
j
.
=
.
r + n
.
, i = 1, ..., ` (1)
where
.
=
.,1
+ ,
.,J
= o
.
c
0
and n
.
(0,
0
).
Weight each branch with p
.
c
0
: Co-phasing. If not co-phasing,

then what happens ?
Combine signals from these ` receive antennas, one have
j =
1
.=1
p
.
c
0
j
.
=
(
1
.=1
p
.
o
.
)
r +
1
.=1
p
.
c
0
n
.
. (2)
Outline
Introduction
Maximal Ratio Combining (cont.)
After combining the signals, the resultant SNR is
SNR =
(
1
.=1
p
.
o
.
)
2
1
.=1
p
2
.
, (3)
One needs to nd {p
.
}
1
.=1
to maximize SNR ?.
The solution to the simple optimization problem can be obtained by
taking partial derivatives of (3) or using the Swartz inequality. In
particular, the solution is
p
:
= o
.
,
0
(4)
and the resultant combined SNR
is
=1
o
2
0
=
1
.=1

.
. (5)
The
increases linearly with the number of diversity branches `.

Outline
Introduction
Maximal Ratio Combining: An example of 2 Rx-antennas
1
1 1
T j
e a h
2
2 2
T j
e a h
Channel
estimator
*
1
h
*
1
h
X X
*
Channel
estimator
*
2
h
2
h
Maximum
likelihood
detector
1 1 1
n x h y
1
n
x
2
n
2 2 2
n x h y
Interference
+ noise
Interference
+ noise
y
Anten-1
Anten-2
x
Outline
Introduction
MRC: Probability of error in symbol detection
The detection performance of a diversity system, whether it uses
space diversity or another form of diversity, in terms of probability of
symbol error 1
c
for demodulation in AWGN with SNR
can be
determined by
1
c
=

0
1
c
()j
;
()d. (6)
We can obtain a simple upper bound on the average probability of
error by applying the Cherno bound Q(r) c
i
2
2
to the Q
function.
Recall that for static channel gains with MRC, we can approximate
the probability of error as
1
c
= c
1
Q
(
a
1
)
c
1
c
o
2
= c
1
c
o
(;
1
+...+;
)2
.
(7)
Outline
Introduction
MRC: Probability of error in symbol detection (cont.)
Integrating over the chi-squared distribution for
yields
1
c
c
1
1
.=1
1
1 + a
1
.
,2
. (8)
In the limit of high SNR and assuming that the
.
s are identically
distributed with
.
= , one will have
1
c
c
1
(
a
1
2
)
1
. (9)
The distribution of the combined SNR j
;
() leads to a decrease in
1
c
due to diversity combining.
The resultant performance advantage is called the diversity gain.
Outline
Introduction
Diversity order
For some diversity systems, their averaged probability of error can be
expressed in the form
1
c
= c
1
, (10)
where c is a constant depending on the specic modulation and
coding, is the averaged received SNR per branch and ` is called
the diversity order of the system.
The diversity order indicates how the slope of the average probability
of error as a function of averaged SNR changes with diversity.
Recall that a general approximation for average error probability in
Rayleigh fading with no diversity is 1
c
= c
1
,(2a
1
). This
expression has a diversity order of one, consistent with a single
receive antenna.
The maximum diversity order of a system with ` antennas is `,
and when the diversity order equals ` the system is said to achieve
full diversity order.
Outline
Introduction
Diversity order: Numerical results of MRC
0 5 10 15 20 25 30
10
6
10
5
10
4
10
3
10
2
10
1
10
0
b
(dB)
P
b
M = 1
M = 2
M = 4
M = 8
M = 10
Outline
Introduction
1
1 1
T j
e a h
2
2 2
T j
e a h
M
j
M M
e a h
T
x
X
X
X
1
T j
e
2
T j
e
M
j
e
T
De-
mod
Outline
Introduction
Equal-Gain Combining (cont.)
MRC requires knowledge of the time-varying SNR on each branch,
which can be very dicult to measure.
A simpler technique is equal-gain combining, which co-phases the
signals on each branch and then combines them with equal
weighting, i.e., p
.
= c
0
.
The SNR of the combiner output, assuming the same noise PSD
0
in each branch, is then given by
=
1
0
`
(
1
.=1
)
2
, (11)
Outline
Introduction
1
1 1
T j
e a h
2
2 2
T j
e a h
M
j
M M
e a h
T
x
De-
mod
Measure
SNR
Measure
SNR
Measure
SNR
Outline
Introduction
Selection combining (cont.)
In selection combining (SC), the combiner outputs the signal on
the branch with the highest SNR.
Since only one branch is used at a time, SC often requires just one
receiver that is switched into the active antenna branch.
A dedicated receiver on each antenna branch may be needed for
systems that transmit continuously in order to simultaneously and
continuously monitor SNR on each branch.
Since only one branch output is used, co-phasing of multiple
branches is not required.
As a result, this technique can be used with either coherent or
dierential modulation.
Outline
Introduction
Selection combining (cont.)
Under SC implementation, the pdf of
is
j
;
() =
`
(
1 c
;;
)
11
c
;;
(12)
As a result, the averaged output SNR of the combiner in Rayleigh
fading is
1
.=1
1
.
. (13)
The average SNR gain increases with M, but not linearly.
The biggest gain is obtained by going from no diversity (` = 1) to
two-branch diversity (` = 2).
Outline
Introduction
SC: averaged probability of error in BPSK detection
0 5 10 15 20 25 30
10
6
10
5
10
4
10
3
10
2
10
1
10
0
b
(dB)
P
b
M = 1
M = 2
M = 4
M = 8
M = 10
Outline
Introduction
1
1 1
T j
e a h
2
2 2
T j
e a h
M
j
M M
e a h
T
x
De-
mod
Compare
SNR
T
J
Outline
Introduction
Threshold Combining (cont.)
SC for wireless systems transmitting continuously may require a
dedicated receiver on each branch to continuously monitor branch
SNR.
A simpler type of combining, called threshold combining, avoids
the need for a dedicated receiver on each branch by scanning each
of the branches in sequential order and outputting the rst signal
with SNR above a given threshold
T
.
As in SC, since only one branch output is used at a time, co-phasing
is not required.
Thus, this technique can be used with either coherent or dierential
(noncoherent) modulation.
There are several criteria the combiner can use to decide which
branch to switch to.
The simplest criterion is to switch randomly to another branch.
Outline
Introduction
Transmitter Diversity: Introduction
In transmit diversity, there are multiple transmit antennas with the
transmit power divided among these antennas.
Transmit diversity is desirable in cellular systems where more space,
power, and processing capability is available on the transmit side
rather than the receive side.
Transmit diversity design depends on whether or not the complex
channel gain is known at the transmitter or not.
When this gain is known, the system is very similar to receiver
diversity.
However, without this channel knowledge, transmit diversity gain
requires a combination of space and time diversity via a novel
technique called the Alamouti scheme.
Outline
Introduction
Channel Known at Transmitter: Transmission model
1
1 1
T j
e a h
2
2 2
T j
e a h
M
j
M M
e a h
T
De-mod
Channel
estimator
x
2
2
T j
e g

X
X
X
1
1
T j
e g

M
j
M
e g
T
Modulation
Coded bits
Base Station
Limited feedback link of channel state information (CSI)
Outline
Introduction
Channel known at transmitter: detailed implementations
Consider a transmit diversity system with ` transmit antennas and
one receive antenna.
Assume the path gain associated with the ith transmit antenna
given by
.
= o
.
c
0
is known at the transmitter via limited feedback

links from mobile terminals.
This is referred to as having channel side information (CSI) at the
transmitter or CSIT.
Let r denote the transmitted signal with total energy per symbol 1
s
This signal is multiplied by a complex gain p
.
c
0
, 0 p
.
1 and
sent through the ith transmit antenna.
Due to the average total energy constraint 1
s
, the weights p
.
c
0
must satisfy
1
.=1
p
2
.
= 1
Outline
Introduction
Channel known at transmitter (cont.)
The weighted signals transmitted over all antennas are added via
signal superposition at the receive antenna, which leads to a
received signal given by
j =
1
.=1
p
.
o
.
r + n, n (0,
0
) . (14)
One can obtain the weights p
.
that achieve the maximum SNR:
p
.
=
o
=1
o
2
, (15)
and the resultant SNR is
=
1
s
0
1
.=1
o
2
.
=
1
.=1
.
, (16)
where
=
2
/
0
equal to the branch SNR between the th transmit
antenna and the receive antenna.
Thus, we see that transmit diversity when the channel gains are known at
the transmitter is very similar to receiver diversity with MRC: the received
Outline
Introduction
Channel unknown at transmitter: the Alamouti scheme
We now consider the same model as in the previous subsection but
assume that the transmitter no longer knows the channel gains
.
= o
.
c
0
, so there is no CSIT.
In this case it is not obvious how to obtain diversity gain. Consider,
for example, a naive strategy whereby for a two-antenna system we
divide the transmit energy equally between the two antennas.
Thus, the transmit signal on antenna i will be r
.
=
.5r where r is
the transmit signal with energy per symbol 1
s
.
Assume two antennas have complex Gaussian channel gains
{
.
= o
.
c
0
}
2
.=1
with zero-mean and unit variant (
0
= 1).
The received signal is
j =
.5(
1
+
2
)r + n. (17)
Outline
Introduction
The Alamouti scheme (cont.)
Note that
1
+
2
is the sum of two complex Gaussian random
variables, and is thus a complex Gaussian as well with mean equal to
the sum of means (zero) and variance equal to the sum of variances.
Thus
.5(
1
+
2
) is a complex Gaussian random variable with
zero-mean and unit-variance (1), so the received signal has the
same distribution as if we had just used one antenna with the
full energy per symbol.
In other words, we have obtained no performance advantage from
the two antennas, since we could not divide our energy intelligently
between them or obtain coherent combining through co-phasing.
Transmit diversity gain can be obtained even in the absence of
channel information with an appropriate scheme to exploit the
antennas.
Outline
Introduction
A particularly simple and prevalent scheme for this diversity that
combines both space and time diversity was developed by Alamouti.
Alamoutis scheme is designed for a digital communication system
with two-antenna transmit diversity.
The scheme works over two symbol periods where it is assumed
that the channel gain is constant over this time duration.
Over the rst symbol period two dierent symbols :
1
and :
2
each
with energy 1
s
,2 are transmitted simultaneously from antennas 1
and 2, respectively.
Over the next symbol period, symbol :
2
is transmitted from
antenna 1 and symbol :
1
is transmitted from antenna 2, each with
symbol energy 1
s
,2.
Outline
Introduction
Assume complex channel gains
{
.
= o
.
c
0
}
2
.=1
between the ith
transmit antenna and the receive antenna.
The received symbol over the rst symbol period is
j
1
=
1
:
1
+
2
:
2
+ n
1
, (18)
and the received symbol over the second symbol period is
j
2
=
1
:
2
+
2
:
1
+ n
2
, (19)
where {n
.
}
2
.=1
is the AWGN noise sample at the receiver associated
with the ith symbol transmission. We assume the noise sample has
zero-mean and power of
0
.
Outline
Introduction
The receiver uses these sequentially received symbols to form the
vector y = [j
1
j
2
]
T
given by
y =
[

1

2
1
] [
:
1
:
2
]
+
[
n
1
n
2
]
= H
.
s + n, (20)
where H
.
=
[

1

2
1
]
, s = [:
1
:
2
]
T
and n = [n
1
n
2
]
T
.
Let us dene the new vector z = H
J
.
y. The structure of H
.
implies that
H
J
.
H
.
=
(
2
+
2
2
)
I
2
(21)
is diagonal and thus
z = [.
1
.
2
]
T
=
(
2
+
2
2
)
I
2
s + n, (22)
where n = H
J
.
n is a complex Gaussian noise vector with mean zero
and covariance matrix 1
(
n n
J
)
=
(
2
+
2
2
)
I
2
0
.
Outline
Introduction
The diagonal nature of z eectively decouples the two symbol
transmissions, so that each component of z corresponds to one of
the transmitted symbols:
.
.
=
(
2
+
2
2
)
:
.
+ n
.
, i = 1, 2. (23)
The received SNR thus corresponds to the SNR for .
.
given by
.
=
(
2
+
2
2
)
1
s
2
0
, (24)
where the factor of 2 comes from the fact that :
.
is transmitted
using half the total symbol energy 1
s
.
The received SNR is thus equal to the sum of SNRs on each branch,
identical to the case of transmit diversity with MRC assuming that
the channel gains are known at the transmitter.
Thus, the Alamouti scheme achieves a diversity order of 2, the
maximum possible for a two-antenna transmit system, despite the
fact that channel knowledge is not available at the transmitter.
Outline
Introduction
The Alamouti scheme: An example of 2 Tx-antennas
1
1 1
T j
e a h
2
2 2
T j
e a h
Channel
estimator
Combiner
Maximum likelihood detector
*
2
1
s
s
1
y
Antenna 1
Antenna 2
1
s
*
1
2
s
s
2
1
n
n
Interference
+ noise
1
h
2
h
2
y
2
s
1
h
2
h
Rx-antenna
Outline
Introduction
The Alamouti scheme: BER results of BPSK
Outline
Introduction
Possible problems to be considered in theses
In Alamoutis scheme, wireless channels are assumed to be at- and
block-fading.
Doubly selective channels can be considered in Alamoutis scheme
by using OFDM and BEMs.
The study results can be employed in LTE downlink transmissions
with mobile terminals.
Outline
Introduction
Precoding
Scheduling (user selection)
Section 3.6: Space Division Multiple Access (SDMA)
Email: nlhung@dut.udn.vn; nnguyenlehung@yahoo.com
Mobile Communications - Chapter 3: Physical-layer transmissions Section 3.6: Space Division Multiple Access (SDMA) 1
Outline
Introduction
Precoding
1
Introduction
SDMA and OFDM
Multiuser transmission
2
Precoding
Precoding classication
An example of linear precoding
Power allocation in ZF precoding
Possible research problems
3
Exhaustive selection
Greedy selection
Outline
Introduction
Precoding
SDMA and OFDM
SDMA with OFDM
The integration of multi-antenna and OFDM techniques has
provided remarkable diversity and capacity gains in broadband
wireless communications.
In multiuser (MU) transmissions, the use of multiantenna array at
the base station (BS) enables simultaneous transmission of multiple
data streams to multiple users by exploiting spatial separations
among users.
A
B
S
/
e
N
B

A
M
S
/
U
E

(
a
)

IFFT
SU-MIMO
precoder
A
B
S
/
e
N
B

A
M
S
/
U
E
1

(
b
)

IFFT
MU-MIMO
precoder
A
M
S
/
U
E
2

Outline
Introduction
Precoding
SDMA and OFDM
A simple example of multiuser (MU) transmission
1 , 1
h
2 , 1
h
M
h
, 1
Base Station
1
s
Modulation
Coded bits
of user 1
2
s
Modulation
Coded bits
of user 2
1 , 2
h
2 , 2
h
M
h
, 2
Antenna 1
Antenna M
De-mod
Channel
estimator
User 2
De-mod
Channel
estimator
User 1
1
y
2
y
j
1
= :
1
1
n=1
1,n
+:
2
1
n=1
1,n
+.
1
, and j
2
= :
2
1
n=1
2,n
+:
1
1
n=1
2,n
+.
2
Outline
Introduction
Precoding
In the so-called space division multiple access (SDMA), multiuser
diversity is the primary factor that increases signicantly the system
sum-rate (throughput).
As a result, an appropriate multiuser encoding technique (at the BS)
is indispensable to attain the considerable sum-rate gain in SDMA.
It is well-known that dirty paper coding (DPC) is an optimal
multiuser encoding strategy that achieves the capacity limit of MU
broadcast (BC) channels but at the cost of extremely high
computation burden as the number of users is large.
Recent studies have introduced several suboptimal multiuser
encoding techniques with lower complexity (relative to DPC) that
can be categorized into:
nonlinear precoding such as: vector perturbation, Tomlinson
Harashima techniques
linear precoding such as: minimum mean squared error (MMSE),
zero-forcing.
Outline
Introduction
Precoding
Multiuser transmission techniques
Broadband communications
LTE (4G) system
Broadband communications
(high data rate and reliability)
Diversity
Time Freq.
Signal
Space
Multi-
user
Space
Multipath channel
Modeling
CSI feedback
Analog Digital
Vector
quantization
g
Quasi-static Time-variant
BEMs AR
LBG
Grassmannian
Random
Scheduling Precoding
Exhaustive
search
Greed or iterative
search
Linear
methods
Non-linear
methods
Codebook-
based ones
MMSE BD DPC THP PU
2
RC
Random
user selection
VP
Outline
Introduction
Precoding
1 , 1
h
2 , 1
h
M
h
, 1
Base Station
Feedback link of
channel state information (CSI)
1
s
X
X
X
1 , 1
w
Modulation
Coded bits
of user 1
2 , 1
w
M
w
, 1
2
s
X
X
X
1 , 2
w
Modulation
Coded bits
of user 2
2 , 2
w
M
w
, 2
1 , 2
h
2 , 2
h
M
h
, 2
Antenna 1
Antenna M
De-mod
Channel
estimator
User 2
De-mod
Channel
estimator
User 1
1
y
2
y
1
= s
1
f
r=1
r
1r
1r
+s
2
f
r=1
r
2r
1r
+z
1
and
2
= s
2
f
r=1
r
2r
2r
+s
1
f
r=1
r
1r
2r
+z
2
Outline
Introduction
Precoding
Inter-user interference
The received signals at user-n can be determined by
j
u
= :
u
1
n=1
n
u,n
u,n
+:
u
n=1
n
u
,n
u,n
+.
u
, n, n
{1, 2},
(1)
where :
u
1
n=1
n
u
,n
u,n
is called as inter-user interference
that would signicantly degrade the performance of the system.
Precoding design is to nd the weighting coecients {n
u,n
}
2
u=1
that satisfy the following condition
1
n=1
n
u
,n
u,n
= 0 with n, n
{1, 2} (2)
to eliminate the inter-user interference :
1
n=1
n
u
,n
u,n
.
The above technique is called as zero-forcing (ZF) precoding.
The problem of nding the weighting coecients {n
u,n
}
2
u=1
can be
easily solved by expressing received signals in a vector form.
Outline
Introduction
Precoding
Zero forcing (ZF) precoding formulation
In the presence of two users, the previous equations become
[
j
1
j
2
]
=
[

1,1
. . .
1,1
2,1
. . .
2,1
]
n
1,1
n
2,1
.
.
.
.
.
.
n
1,1
n
2,1
[
:
1
:
2
]
+
[
.
1
.
2
]
.
In the presence of l users, the received signal can be expressed by:
y = HWs +z, (3)
where y =
j
1
.
.
.
j
, H =
1,1
. . .
1,
.
.
.
.
.
.
.
.
.
,1
. . .
,
, s =
1
.
.
.
W =
n
1,1
. . . n
,1
.
.
. . . .
.
.
.
n
1,
. . . n
,
= [w
1
, . . . , w
] with
w
= [n
,1
, . . . , n
,
]
, and z = [.
1
, . . . , .
.
Outline
Introduction
Precoding
Zero-forcing precoding formulation (cont.)
To eliminate inter-user interference, precoding matrix W can be
determined by
W = H
J
(
HH
J
)
1
H
(4)
so that
y = HWs +z = s +z. (5)
With precoding, the received signal can be written by
y = Hx +z, (6)
where x = [r
1
, . . . , r
1
]
T
= Ws are the transmitted signals in a
vector form at ` antennas in the base station.
Under the power constraint of 1
max
at the BS, one has
[
1
n=1
r
n
2
]
=
[
x
2
]
1
max
, (7)
Outline
Introduction
Precoding
The power constraint (7) is equivalent to
I
u=1
`
u
1
u
1
max
. (8)
where `
u
=
[
(
HH
J
)
1
]
u,u
and :
u
=
1
u
:
u
After ZF precoding, the received signals at l users are given by
y =
j
1
.
.
.
j
I
1
1
:
1
.
.
.
1
I
:
I
.
1
.
.
.
.
I
(9)
Hence, the resultant sum-rate of the multiuser system is
C = max
1
r
:
!
r=1
X
r
1
r
1
max
I
u=1
log
2
(1 +1
u
) (bps/Hz) (10)
Outline
Introduction
Precoding
Power allocation in ZF precoding (cont.)
The optimal power allocation [1
u
, n {1, ..., l}] in (10) can be
easily determined by the following waterlling process
1
u
= (j,`
u
1)
+
(11)
where r
+
denotes max(r, 0), and the water level j is chosen to
satisfy
I
u=1
(j `
u
)
+
= 1
max
. (12)
Given a set of selected users = {1, ..., l}, the above precoding
process attempts to eliminate the inter-user interference and
maximize the system sum-rate.
The problem of how to perform user selection (nding the set
= {1, ..., l}) with a reasonable complexity for maximizing the
system sum-rate will be addressed in the next section.
Outline
Introduction
Precoding
Precoding in LTE downlink transmissions
Data bits
of user 1
Channel
encoder
Interleaver
Layer
mapper
MQAM
mapper
MQAM
mapper
Precoding
OFDMA
modulator
OFDMA
modulator
Precoding matrix
generator
Recovered data bits Channel decoder
Channel
Estimator
OFDMA
Demodulator
BER evaluator
of user 1
OFDMA
Demodulator
Channel State
Information (CSI)
MIMO
demapper
Limited feedback
link
User 1
Base Station (BS)
Data bits
of user N
Channel
encoder
Interleaver
Layer
mapper
MQAM
mapper
MQAM
mapper
W
X
Y= W*X
BER evaluator
of user N
Multipath fading
channel
User N
Outline
Introduction
Precoding
Greedy selection
Given a precoding technique, scheduling (user selection) is to nd a
set of users among all active users to maximize the system sum-rate.
Obviously, the simple optimal method for user selection is exhaustive
search but its complexity is impractically high as the number of users
is large.
Outline
Introduction
Precoding
Greedy selection
Greedy selection
Greedy user selection algorithm
1
Initialization:
0
= {1, 2, ...,
I
} is the set of all available users
indices
0
= {} is the set of selected users initially assigned to a null set.
j = 0 stands for the number of selected users, initially set to zero.
C
0
= 0 is the system sum-rate of selected users, initially set to zero.
2
Repetition: Assuming that selecting user n in the set
q
maximizes the resulting sum-rate of the system called C
max
.
j = j + 1
If C
max
< C
1
or j .
or j .
go to Step 3 otherwise do:
=
max
=
1
{} (select one more user)
=
1
{} (ignore user- in later consideration)
Go to Step 2.
3
Stop the user selection process and compute the ZF weighting
vectors based on the composite channel matrix of selected users.

Mobile Communication Slide

Загружено:

Сведения о документе

Авторское право

Доступные форматы

Поделиться этим документом

Поделиться или встроить документ

Параметры публикации

Этот документ был вам полезен?

Это неприемлемый материал?

Авторское право:

Доступные форматы

Mobile Communication Slide

Загружено:

Авторское право:

Доступные форматы

Abstract of the course

Content of the course

is the transmitted power,

are the transmitter and receiver antenna gains,

) + 0.8 for medium or small city

denote the mean envelop and mean squared levels of

stands for standard deviation.

2,T cos (2)

2,T sin (2)

(t), then the receiver can

Mobile communications-Chapter 3: Physical-layer transmissions Section 3.1: Digital modulations 44

is a Gaussian random variable with

(), and therefore 1

, dened as the probability that

, averaged over the distribution of

, one will have

() by making the change

()d = j(:)d:. (14)

for BPSK in Rayleigh Fading

for MQAM in Rayleigh Fading and AWGN.

are two codewords

must form another

is equal to the weight of C

C = R+e = C+e +e. (17)

C = C, i.e., the corrupted codeword is correctly decoded. The

, ..., 0, ..., 1},

denotes the CP length.

, ..., 0, ..., 1},

denotes the CP length.

denotes the OFDM symbol length after CP

denotes the OFDM symbol length after CP

: Co-phasing. If not co-phasing,

increases linearly with the number of diversity branches `.

is known at the transmitter via limited feedback

go to Step 3 otherwise do:

{} (select one more user)

Вам также может понравиться