Modeling, Analysis, and Design of Multi-tier and Cognitive Cellular Wireless Networks - Prof Ekram HossainSlides set#1

Modeling, Analysis, and Design of Multi-tier and Cognitive Cellular Wireless Networks
Modeling, Analysis, and Design of Multi-tier and
Cognitive Cellular Wireless Networks
Ekram Hossain
Department of Electrical and Computer Engineering
University of Manitoba, Winnipeg, Canada
http://home.cc.umanitoba.ca/hossaina
Institut Technology Telcom (IT Telkom)
27 August 2013
1/48

Wireless Communications, Networks, and Services
Research Group at U. of Manitoba
2/48

Wireless Communications, Networks, and Services
Research Group at U. of Manitoba
Current research interests:
Cognitive radio and dynamic spectrum access
Hierarchical cellular wireless networks (small cell networks)
Green cellular radio systems
Applied game theory and network economics
Current group members:
3 PDF, 8 Ph.D. students, 2 M.Sc. students
3/48

Outline
1 Introduction
2 Challenges in Modeling, Analysis, and Design of HetNets
3 Preliminaries: Stochastic Geometry and Poisson Point Process
4 Categorization of Performance Evaluation Techniques
4/48

Introduction
Evolution of the population of wireless devices
Number of connected devices
2020
50b
40b
30b
20b
10b
2010 2015
5/48

Introduction
Evolution of the population of wireless devices
Global Mobile Data Trac Forecast Report presented by
Cisco predicts 2.4 exabytes mobile data trac per month for
the year 2013.
M2M communications and IoT (Internet of Things)
Three evolution phases of user population:
1 connected consumer electronics phase (smart phones, tablets,
computers, IPTVs)
2 connected industry phase (sensor networks, industry and
buildings automation, surveillance, and eHealth applications)
3 connected everything phase (IoT phase)
A signi

cant part of this trac will be carried through cellular
networks.
6/48

Introduction
Multi-tier cellular wireless networks
Improvement of cell coverage, network capacity, and better
quality-of-service (QoS) provisioning are some of the major
challenges for next generation cellular networks.
Universal frequency reuse and make transmitters and receivers
closer
Hierarchical layering of cells (referred to as HetNets), an
ecient solution to improve cell coverage and network
capacity.
Adopted in the evolving Long Term Evolution
(LTE)/LTE-Advanced (LTE-A) systems
3GPP Release-8 (LTE), 3GPP Release 10 onwards
(LTE-Advanced)
7/48

Introduction
LTE/LTE-A HetNet
Long-Term Evolution (LTE) and LTE-Advanced systems are
designed to support high-speed packet-switched services in 4G
cellular wireless networks.
The cells or radio base stations in LTE/LTE-A can be
classi

ed as: i) macrocell base station (referred as MeNB),
and ii) small cells (e.g., microcells, picocells, femtocells).
Small cell is an umbrella term for low-power radio access
nodes that operate in both licensed and unlicensed spectrum
and have a range of 10 meter to several hundred meters.
Small cells will improve the cell coverage and area
spectral-eciency (capacity per unit area).
8/48

Introduction
LTE/LTE-A HetNet
MeNB-A UE 1
Macrocell Base
Station A
(MeNB-A)
HeNB-A3 UE 1 HeNB-A3
HeNB-A2 HeNB-A2 UE 1
HeNB-A1 UE 1
Relay Node
MeNB-A UE 2
Picocell
PC-A1
RN-A1 UE 1
RN-A1
MeNB-A UE 3
PC-A1 UE 1
PC-A1 UE 2
HeNB-A1
X2
Un
MeNB-B
HeNB-B2 UE 1
S1
HeNB-B2
HeNB-B1 UE 1
MeNB-B UE 2
PC-B1
RN-B1 UE 1
Relay Node
RN-B1
MeNB-B UE 1
PC-B1 UE 1
PC-B1 UE 2
HeNB-B1
X2
Un
MeNB-C
MeNB-C UE 1
HeNB-C3 UE 1
MeNB-C UE 2
RN-C1 UE 1
MeNB-C UE 3
Relay Node
RN-C1
PC-C1 UE 2
MeNB-C UE 4
HeNB-C3
HeNB-C2 UE 1
HeNB-C2
HeNB-C1 UE 1
HeNB-C1
X2
Un
Picocell
PC-C1
PC-C1 UE 1
LTE Evolved
Packet Core
HeNB Gateway
MME / S-GW
HeNB Gateway
X2
X2
X2
S1
S1
S1
S1 S1
Internet
9/48

Introduction
Comparison among dierent radio base stations in
LTE/LTE-A
Attributes MeNB Picocell HeNB Wi-Fi
BS Installation Mobile Operator Mobile Operator Customer Customer
Site
Acquisition
Mobile Operator Mobile Operator Customer Customer
Transmission
Range
300-2000 m 40-100 m 10-30 m 100-200 m
Transmission
Power
40 W (approx.) 200 mW- 2 W 10-100 mW 100-200 mW
Band License Licensed band Licensed band Licensed band Unlicensed
band
System
Bandwidth
5, 10, 15, 20 MHz
(with CA up to 100
MHz)
5, 10, 15, 20 MHz
(with CA up to
100 MHz)
5, 10, 15, 20
MHz (with CA up
to 100 MHz)
5, 10, 20 MHz
Transmission
Rate
up to 1 Gbps up to 300 Mbps 100 Mbps-1
Gbps
up to 600
Mbps
Cost $ 60,000/yr $ 10,000/yr $ 200/yr $ 100-200/yr
Power
Consumption
High Moderate Low Low
10/48

Introduction
Motivations for small cells
High data rate and improved quality-of-services to subscribers
Eliminate coverage holes in macrocell footprint
Extended battery life of mobile phones
Macrocell load reduced (hence more resources for macrocell
users)
Mitigate spectrum underutilization problem
11/48

Challenges in Modeling, Analysis, and Design of HetNets
HetNet characteristics
Introduction of small cells results in a substantial shift in the
cellular network architecture with features such as
topological randomness
high variability in the speci

cations of the network elements
unbalanced uplink-downlink association
trac ooading and load balancing
very dense deployment
12/48

Topological randomness
Modeling a Heterogeneous Cellular Network (HCN)
25
15
!
5
(a) (b) (c)
Fig. 2: Example of different macrocell only models. Traditional grid networks remain the most popular, but 4G systems have
smaller and more irregular cell sizes, and perhaps are just as well modeled by a totally random BS placement.
5
J. G. Andrews, H. Claussen, M. Dohler, S. Rangan, and M. C. Reed, Femtocells: Past, present, and future, IEEE
Journal on Selected Areas in Communications, Special Issue on Femtocell Networks, April 2012.
Traditional grid model Actual 4G macrocells today Completely random BSs
to the femtocell user is assumed to be only from the various
macrocells, which in a fairly sparse femtocell deployment, is
probably accurate. In the uplink as well, the strong interference
is bound to come from nearby mobiles transmitting at high
power up to the macro base station, so the model may be
reasonable. The main limitation of this model vs. Model 1 is
that the performance of downlink macrocell users – who may
experience strong femtocell interference depending on their
position – cannot be accurately characterized.
The third model, which appears to be the most recent, is
to allow both the macrocells and femtocells to be randomly
placed. This is the approach of three papers in this special
issue [61]–[63], and to the best of our knowledge, these
are the first full-length works to propose such an approach
(earlier versions being [64], [65]. Both of these papers are
for the downlink only and an extension to the uplink would
be desirable. An appealing aspect of this approach is that the
randomness actually allows significantly improved tractability
and the SINR distribution can be found explicitly. This may
allow the fundamental impact of different PHY and MAC
designs to be evaluated theoretically in the future.
in this section we turn our attention to some of the new
challenges that arise in femtocell deployments. To motivate
future research and an appreciation for the disruptive potential
of femtocells, we now overview the broader challenges of fem-tocells,
focusing on both technical and economic/regulatory
issues.
A. Technical Challenges
1) Interference Coordination: Perhaps the most significant
and widely-discussed challenge for femtocell deployments is
the possibility of stronger, less predictable, and more varied
interference, as shown in Fig 3. This occurs predominantly
when femtocells are deployed in the same spectrum as the
legacy (outdoor) wireless network, but can also occur even
when femtocells are in a different but adjacent frequency band
due to out-of-band radiation, particularly in dense deploy-ments.
As discussed in the previous section, the introduction
of femtocells fundamentally alters the cellular topology by
creating an underlay of small cells, with largely random
placements and possible restrictions on access to certain BSs.
!
20
10
0
T diti l id d l 0 5 10 15 20 25
cells
Zoom
w/ femtoc
m w/ picoc
Zoom
cells too
13/48

Eect of network geometry
SINR is one of the main performance metrics in wireless
communications:
SINR(y) =
Pt (x0)Ah0 kx0 yk
N +
P
xi2 I
Pt (xi )Ahi kxi yk
Network geometry along with propagation environment aects
SINR.
SINR impacts network performance metrics such as
outage probability, Pout = P(SINR

)
coverage probability, Pc = 1 Pout
bandwidth normalized average rate, E[ln(1 + SINR)]
network capacity (or throughput), C = (1 Pout ), subject to
Pout , = no. of active links per unit area
14/48

User association
User association, spectrum access methods, etc. aect
network geometry (and hence SINR) and performance of
resource allocation methods
In a single-tier network with all BSs having the same transmit
power, a user associates to the nearest BS (for which the
average RSS is also the highest in the downlink).
15/48

User association
Dierent BSs having dierent transmit powers.
With the strongest RSS or SINR-based association, the BS
may not necessarily be the closest one.
Distance to the BS depends on relative transmit powers and
propagation conditions.
Example: In

g., r is larger than rs , but
r (Ps=Pm)1= rs .
rs r (Ps/Pm)1/ƞ
r
rmr
r
rm r (Pm/Ps)1/ƞ
rsr
Macro-cell User Scenario Small-cell User Scenario
Highest RSS Connectivity
16/48

Unbalanced uplink-downlink association
In downlink, a user may associate with a macro BS, while in
the uplink, it may associate with a small cell BS.
Downlink
Uplink
17/48

Trac ooading and load balancing
Biasing can be used in multi-tier cellular networks to ooad users
from one network tier to another tier.
Biasing is known as range extension in the 3GPP standard.
Instead of associating to the network entity oering the highest
signal power, a user associates to a small cell if
PsTr
s Pmr
m ; where T 1:
i.e., if rm

Pm
PsT
1

rs .
Without biasing, rm

Pm
Ps
1

rs , that is, biasing will decrease the
minimum distance between a small cell user and interfering MBSs.
18/48

Multi-tier cognitive cellular network
Each network element performs spectrum sensing to access
the spectrum.
Cognitive spectrum access aects the locations and density of
interferers.
re
19/48

Challenges in HetNet modeling, analysis, and design
Traditional grid-based model fails to capture the basic HetNet
characteristics.
New modeling/design paradigms are required.
Need design methods that account for the topological randomness
Consider universal frequency reuse (which is essential for high
spectral eciency).
Network functionalities and their optimization techniques have to be
revisited and adapted to the HetNet characteristics.
Centralized control for HetNets is infeasible.
Innovative distributed network management is required.
20/48

Stochastic geometry for modeling HetNets
Stochastic geometry is a powerful tool used to study and analyze
networks with random topologies.
Stochastic geometry has been successfully adapted to model ad hoc
wireless networks from more than three decades.
Stochastic geometry has recently been used to model and analyze
single-tier cellular networks and HetNets.
Stochastic point process is used to abstract the network model.
Stochastic geometry analysis provides statistical and spatial
averages for the performance metrics.
Stochastic
geometry
analysis
EŽĚĞƐ͛distribution
MAC layer behavior
Physical layer characteristics
Distribution of simultaneous
active nodes
Outage Probability
Network Capacity
21/48

Preliminaries: Stochastic Geometry and Poisson Point Process
Point process
A stochastic point process is a type of random process for which
any one realization consists of a set of isolated points either in
time or geographical space, or in even more general spaces.
22/48

Applications of point processes
Modeling and analysis of spatial/temporal data
Diverse disciplines: forestry, plant ecology, epidemiology,
geography, seismology, materials science, astronomy,
economics
Frequently used as models for random events in time, e.g.,
arrival of customers in a queue (queueing theory), impulses in
a neuron (computational neuroscience), particles in a Geiger
counter, location of users in a wireless/mobile network,
searches on the web
23/48

Formulation of a point process
by observing the arrival or inter-arrival time
by counting the number of points
by counting the number of points within a speci

Constructing a new point process
By transforming or changing an existing point process
Transformation operations include:
mapping (scaling, translation, rotation, projection, etc. )
superposition
clustering
thinning
marking
25/48

Major point processes
Stochastic point process is used to abstract the network
topology.
Four main point processes used in the literature for modeling
wireless networks:
Poisson point process (PPP)
binomial point process (BPP)
hard core point process (HCPP)
Poisson cluster process (PCP)
PPP is the simplest and most widely used point process.
26/48

PPP, HCPP, and PCP
0 2 4 6 8 10 12 14 16 18 20
20
18
16
14
12
10
8
6
4
2
0
0 2 4 6 8 10 12 14 16 18 20
20
18
16
14
12
10
8
6
4
2
0
0 2 4 6 8 10 12 14 16 18 20
20
18
16
14
12
10
8
6
4
2
0
27/48

Poison Point Process (PPP)
PPP provides tractable results that help understanding the
relationship between the performance metrics and the design
parameters.
PPP can model random network with randomized channel access.
Provides tight bound for networks with planned deployment and
networks with coordinated spectrum access
Most of the available literature assume that the nodes are
distributed according to a PPP.
Results obtained using PPP are accurate (within 1-2 dB) with those
obtained (by measurements) for legacy cellular networks as well as
HetNets.
28/48

PPP
Let = fxi ; i = 1; 2; 3; :::g be a point process in Rd with
intensity d , then is a PPP i
for any compact set A Rd , the number of points in A is a
Poisson random variable
numbers of points existing within disjoint sets are independent.
Number of points inside any bounded region A 2 Rd is given
by
P
n
N(~A) = k
o
=

d~A
k
ed~A
k!
29/48

PPP
Slivnyak's theorem: the statistics seen from a PPP is independent
of the test location.
Campbell's theorem (valid for a general point process): Let
f : Rd ! [0;1) be a function over a PP and (B) is the
intensity of the PP. Then the average of the sum of the function
E

X
xi2
f (xi )
#
=
Z
Rd
f (x)(dx)
i.e., when the transmitting nodes form a point process and f (x)
represents path-loss, the average interference seen at the origin.
Example: For a PPP with density , E

P
f (xi )
xi2
#
=
R
Rd f (x)dx,
and when f (x) = jjxjj (i.e., singular path-loss model), for
d = 2, E(I) = E

P
f (xi )
xi2
#
= 2
R1
0 rrdr = 2
h
r 2
2
i1
0
=
1: 30/48

PPP
E[I] = 1 for a PPP with a singular path-loss model.
A consequence of the path-loss law and the property of the PPP
that nodes can be arbitrarily close
Probability generating functional (PGFL): the average of a
product of a function over the point process
PGFL for PPP:
E

Y
f (xi )
xi2
#
= exp

Z
Rd

:
(1 f (x)) (dx)
APPGFL is very useful to evaluate the Laplace transform of the sum
x2 f (x).
31/48

Poisson

eld of interferers and aggregate interference
Laplace transform of the pdf of interference for a PPP
network:
LI(s) =e

(Ps)
2
E

h
2 L(1 2
e )
;Pshr

r 2
e E

1ePshr

e

where h can follow any distribution.
For Rayleigh fading,
LI(t) =e

(Ps)
2
Eh

h
2
L(1 2
e )
;sPhr

Psr2
e
Ps+r

e

:
For = 4,
LI(s) =e
q

p
Ps
arctan
Ps

r2
e
!
:
In general, the Laplace transform cannot be inverted to
obtain the pdf of the interference.
32/48

Modeling steps
1 Abstract the network into a convenient point process.
2 Identify the network geometry w.r.t. the test receiver based
on the network characteristics.
3 Identify the point process for the interference sources and
derive its parameters.
4 Derive the Laplace transform (LT), moment generating
function (MGF) or the characteristic function (CF) of the pdf
aggregate interference.
Note that MGF of I(t) = LI(s) and CF of I(t) =
LI(j!), where j =
p
1 and ! is a real-valued parameter.
33/48

Categorization of Performance Evaluation Techniques
Five techniques for performance evaluation
There are FIVE main techniques to overcome the problem due
to the non-existence of pdf of the aggregate interference.
1 Assume Rayleigh fading and obtain the exact SINR statistics
2 Obtain tight bounds
3 Generate moments/cumulants and approximate the pdf
4 Plancherel-Parseval theorem
5 Inversion
34/48

Technique #1: Rayleigh fading assumption
With Rayleigh fading on the useful link, for
interference-limited networks, the cdf of SINR (i.e., outage
probability) for a receiver at a distance r from its transmitter
is evaluated as FSINR() = 1 LI(s)js=c
FSINR() = P fSINR g
= P
(
PtAh0r
N + I

)
= 1 P
(
h0
(N + I)r
PtA
)
= 1 E

exp

(N + I)r
PtA
!#
= 1 exp

Nr
PtA
!
E

exp

Ir
PtA
!#
= 1 exp (Nc) LI(s)js=c ; where c =
r
PtA
: (1)
35/48

Technique #1: Rayleigh fading assumption
Coverage probability, Pc = 1 FSINR()
= exp (Nc) LI(s)js=c.
With Rayleigh fading, the coverage probability in an
interference-limited network is
Pc =
Z
r
fR(r )LI(c)dr : (2)
The average transmission rate is
E[ln (1 + SINR)] =
Z 1
0
P fln (1 + SINR) tg dt
=
Z 1
0
P

SINR

et 1

dt
=
Z 1
0
eNc(et1)LI

c

et 1

dt: (3)
36/48

Technique #2: Dominant interferers by region bounds or
nearest n interferers
Tight lower bound on the cdf of SINR
can be obtained by looking at the
vulnerability region.
Laplace transform of the
interference distribution is not
required.
For deterministic channel gains, the
vulnerability radius is given by
1
rv =

r .
rv
r
x
Vulnerability Circle Bx(rv)
Tight lower bound on the cdf of SINR can also be obtained by
considering the strongest n interferers.
Upper bound can be obtained by Markov inequality, Chebyshev's
inequality, Jensen's inequality, or Cherno bound.
37/48

Technique #2: Dominant interferers by region bounds
For example, in CSMA networks, transmitters contend to
access the spectrum.
Contention-based access creates protection regions for the
receivers.
Protection regions are centred around transmitters.
Vulnerability region is crescent shaped.
38/48

Technique #2: Dominant interferers by region bounds
Divide the aggregate interference I into interferences from two
disjoint regions I1 and I2.
According to the PPP, I1 and I2 are independent, and the outage
probability is obtained as
P

S
I

= P

I
S

= P

I1 + I2
S

= P

I1
S

+ P

I2
S

+ P

I1 + I2
S

jI1
S

; I2
S

| {z }
=0
= P

I1
S

+ P

I2
S

P

I1
S

= P f Bx (rv )6= g
39/48

Modeling, Analysis, and Design of Multi-tier and Cognitive Cellular Wireless Networks - Prof Ekram HossainSlides set#1

More Related Content

What's hot (20)

Viewers also liked (15)

Similar to Modeling, Analysis, and Design of Multi-tier and Cognitive Cellular Wireless Networks - Prof Ekram HossainSlides set#1 (20)

More from indonesiabelajar (7)

Recently uploaded (20)

Modeling, Analysis, and Design of Multi-tier and Cognitive Cellular Wireless Networks - Prof Ekram HossainSlides set#1