The Capacity of

Wireless Networks
Danss Course, Sunday, 23/11/03
 Wireless ad hoc network
• No wired backbone
• No centralized control
• Nodes may cooperate in routing each other’s data
packets
• At the Network Layer – problems are in routing,
mobility of nodes and power constraints
• At the MAC layer – problems with protocols such as
TDMA, FDMA,CDMA
• At the Physical layer – problems in power control
 Lecture Minutes
Arbitrary networks
1. Two models: protocol and physical
2. An upper bound on transport capacity
3. Constructive lower bound on transport capacity
Random networks
1. Two models: protocol and physical
2. Constructive lower bound on throughput capacity
Conclusions
 Arbitrary Networks
• n nodes are arbitrary located in a unit area disc
• Each node is can transmit at W bits/sec over the channel
• Destination is arbitrary
• Rate is arbitrary
• Transmission range is arbitrary
• Will later add some assumptions on the network
• When does a transmission received successfully ?
Allowing for two possible models for successful
reception over one hop: The protocol model and the
Physical model
 Protocol Model
• Let Xi denote the location of a node
• A transmission is successfully received by Xj if:
Xk  X j  1     X i  X j

For every other node Xk simultaneously transmitting

•  is the guarding zone specified by the protocol

xk xi
 1     r
r
xl xj

1     r
 Physical Model
• Let  X k ; k  

Be a subset of nodes simultaneously transmitting

• Let Pk be the power level chosen at node Xk
• Transmission from node Xi is successfully received at node Xj
if:
Pi

Xi  X j
  2
Pk
N  k i
kT Xk  X

j
 Transport Capacity of Arbitrary
Networks
• Network transport one bit-meter when one bit transported one
meter toward its destination
• Main result 1:

Under the Protocol Model the transport capacity is

 (W n ) bit - meters/sec
• Main result 2:
Under the Physical Model, cW n bit - meters/sec is feasible

 1
While c' Wn  bit - meters/sec is not
 Arbitrary Network – upper bound
on transport capacity
Assumptions:
• There are n nodes arbitrarily located in a disk of unit
area on the plane
• The network transport nT bits over T seconds, i.e.
each node generate bits at rate 
• The average distance between source and destination
of a bit is L
• Transmissions are slotted into synchronized slots of
length  sec
 Theorem
• In the protocol model, the transport capacity nL is bounded
as follows:
8 1
nL  W n bit - meters/sec
 

• In the physical model,

1
 1
 2  2  1
nL    W n  bit - meters/sec
   
 Arbitrary Network – constructive
lower bound
• There is a placement of nodes and an assignment of traffic
patterns such that the network can achieve under protocol
model W n
bit - meters/sec
1  2 n  8

1 1
• Proof – define r := 1  2 n
 2
4
Place transmitters at locations:
(j(1  2 )r  r, k (1  2  )r) and (j(1  2  )r, k (1  2 )r  r) where j  k is even
Place receivers at locations:
(j(1  2 )r  r, k (1  2 )r) and (j(1  2 )r, k (1  2 )r  r) where j  k is odd
 A constructive lower bound on
capacity of arbitrary network

r

)) (( r)+1(>
)) (( )) ((
r r
)) ((
 Random Networks
• n nodes are randomly located on S2 (the surface of a sphere of
area 1sq m) or in a disk of area 1sq m in the plane
• Each node has randomly chosen destination to send (n)
bits/sec
• All transmissions employ the same nominal range or power
• Two models: Protocol and Physical
 Protocol Model
• Let Xi denote the location of a node and r the common range
• A transmission is successfully received by Xj if:

1. Xi  X j  r

2. Xk  X j  1     r

For every other Xk simultaneously transmitting

 Physical Model
• Let  X k ; k  

Be a subset of nodes simultaneously transmitting

• Let P be the common power level
• Transmission from node Xi is successfully received at node Xj
if:
P

Xi  X j

P
N  k i
kT Xk  X

j
 Throughput Capacity of Random
Networks
• Main result 1:
Under the Protocol Model the order of the throughput capacity

W
 ( n)   ( ) bit/sec
n log n
• Main result 2:
cW
Under the Physical Model,  (n)  bit/sec is feasible
n log n
c' W
While  ( n)  bit/sec is not
n
 Random Networks: A constructive
lower bound on capacity
We will show a scheme such that each source-destination pair
can be guaranteed a channel of capacity cW
bit/sec
(1  ) 2 n log n
With probability approaching 1 as n  
Steps
• Define the Voronoi tessellation
• Bound the number of interfering neighbors of a Voronoi cell
• Bound the length of an all-cell transmission schedule
• Define the routes of a packet on the Voronoi tessellation
• Prove that each cell contains at least one node
• Calculate the expected routes that pass through a cell and infer the
expected traffic of each node
 Spatial tessellation
• Let {a1,a2,….ap} be a set of p points on S2
• The Voronoi cell V(ai) is the set of all points which are closer
to ai than of the other aj’s i.e.:


V (a i ) : x  S 2 : x  a i  Min1 j  p x  a j 
• Point ai is called the generator of the Voronoi cell V(ai)
A Voronoi tessellation of S
2
 Tessellation properties
• For each >0, There is a Voronoi tessellation such that Each
cell contains a disk of radius  and is contained in a disk of
radius 2
We will use a Voronoi tessellation for which :
1. Every Voronoi cell contains a disk of area 100logn/n . Let (n) be its
radius
2. Every Voronoi cell is contained in a disk of radius 2(n)
 Adjacency and interference
• Adjacent cells are two cells that share a common point.
• We will choose the range of transmission r(n) so that:

r (n)  8  (n)
With this range, every node in a cell is within a distance r(n)
from every node in its own cell or adjacent cell

8(n)
2(n)
 A bound on the number of
Interfering cells
• Two cells are interfering neighbors if there is a point in one cell
which is within a distance of (2+)r(n) of some point in the
other cell
• Lemma – Every cell in Vn has no more than c1 interfering cells.
c1 grows no faster than linearly in (1+)2
Proof – if V’ is an interfering neighbor of V, then V’ and similarly
every other interfering neighbor, must be contained within a
common large disk D of radius 6(n)+ (2+)r(n)
 A bound on the length of an all-cell
inclusive transmission schedule
• Lemma - In the protocol model, there is a schedule for
transmitting packets such that in every (1+c1) slots, each cell in
Vn gets one slot in which to transmit
Proof – A graph of degree no more than c1 can have its vertices
colored by using no more than (1+c1) colors.
So color the graph such that no two interfering neighbors have the
same color, so in each slot all the nodes with the same color
transmit
• There is a schedule also for the physical model….
 The routes of packets
• Source – destination pairs – let Yi be a randomly
chosen location such that Xi and Yi are independent.
The destination Xdest(i) is chosen as the node Xj which is
closest to Yi
• Corollary: The random sequence {Li} = {straight line
connecting Xi and Yi } is i.i.d.
• Routes of packets will be choose to approximate these
straight line segments
• Final destination will be one hop away from Yi , with
high probability
 Each cell contains at least one node
Definition 1:
Let F be a set of subset. A finite set of points A is said to be
shattered by F if for every subset B of A there is a set F in F
A F  B
such that
Definition 2:
The VC-dimension of F , denoted by VC- dim(F ) , is defined as
the supremum of the sizes of all finite sets that can be shattered
by F
 Vapnic-Chervonenkis Theorem
If F is a set of finite VC dimension d and {Xi}is a sequence of
i.i.d. random variables with common probability distribution
P, then for every  > 0,

N
1
Pr(sup
N  I(X
i 1
i  F )  P( F )   )  1  

Whenever
 8d 16e 4 2
N  Max  log , log 
   
VCdim of the set of disks in R 2

~
x2

x2

~
x1 ~
x3
x1 x3

x4

x1  x 3  180  ~
x4
 A cell contains at least one node
Let F denote the class of disks of area 100logn/n.
So VCdim(F) is 3. Let V be a cell contained in a disk D. Hence:
 Number of nodes in D 100 log n 
Pr  sup    (n)   1   (n)
 n n 
Whenever
 24 16e 4 2 
n  Max  log , log 
  ( n )  ( n )  ( n )  ( n ) 
This is satisfied when
50 log n
 (n)   (n) 
n
 Pr  Number of nodes in V  50logn   1   (n)
 Mean number of routes served by
each cell
• First calculate the probability that a line Li or great circle
intersect a cell V
Lemma: for every line Li and cell V
log n
Pr  line Li intersect V   c
n
• So the expected number of lines Li that intersect a cell is
bounded as:
E  Number of lines {Li } intersect V   c n log n

• The same as for great circles !

 Actual traffic served by each cell
• We bounded the mean number of routes passing
through each cell. However, we need to bound the
actual random number of routes served by each cell !!
• Remember the sequence {Xi ,Yi} is i.i.d.
• Therefore , we can appeal to uniform convergence
• We will show that each great circle that intersect a
disc D, can be mapped to a point on the band F(D)
that is equidistant from the center of D
• Then we can bound the VCdim of the band and so of
the great circles
 Transforming great circles
intersecting disks into points lying in
equatorial bands

C F(D)
 Lower bound on throughput
capacity
• Because of uniform convergence, we obtain:
There is a sequence  (n)  0 such that
 
Pr sup Number of lines Li intersect V   c nlogn  1   (n)

since each line Li carries traffic of rate  (n) 

 
Pr sup Traffic needed to be carried by V   c (n) nlogn  1   (n)
 Lower bound on throughput
capacity
• We have shown that there is a schedule for transmitting
packets such that in every (1+c1) slots, each cell can transmit.
• Thus the rate at which each cell transmit is W/(1+c1) bits/sec
• On the other hand, the rate a cell needs to transmit is less than:
c (n) nlogn
• So with high probability, and because c1 is grow linearly with
(1+)2 we have:
W
c(n) nlogn  
1  C1
cW
 ( n) 
1    2 nlogn
 Conclusions
• Designers may want to consider designing networks with
small number of nodes
• Communication with nearby nodes at constant bit rates can be
provided in a dense clusters of nodes, since the source –
1
destination distance shrink asO( )
n
? Questions