EE-5108: ADVANCED COMMUNICATION

NETWORKS

Dr. Haroon Rasheed

PhD (Telecommunication)

Optimization Techniques Lecture 6, October 3, 2012. 1/11

 Introduction

Introduction
The course is intended to be an introduction to optimization and
is by no mean intended to be comprehensive.
I hope that the course will serve as a starting point for students
who may apply optimization in their future works or researches.
Main Objectives:
Study fundamentals of optimization theory.
General introduction to basic concepts.
Example applications to communications and networks.

Optimization Techniques Lecture 1, December 1, 2011. 4/13

1 Application: Power Allocation in Multicarrier Communications
Shannon Formula in Multicarrier Communications
Power-allocation in multicarrier communications
KKT conditions in Optimal Power-allocation
Observations
Waterfilling Algorithm
Capacities of Frequency Selective Channels

2 Reference Books

Optimization Techniques Lecture 6, October 3, 2012. 2/11

 Application: Power Allocation in Multicarrier Communications

Power Allocation in Multicarrier Communications

In multicarrier system, the bandwidth of a communication channel is divided into several
narrow bands which we shall refer to as subchannels.
Since the bandwidth is narrow in each subchannel, the subchannel frequency response
is approximately flat.
As a result we can send bit streams in parallel over these subchannels.
For data transmission, different subchannels may have different qualities measured in
term of the noise power levels, as shown below.

Optimization Techniques Lecture 6, October 3, 2012. 3/11

 Application: Power Allocation in Multicarrier Communications Shannon Formula in Multicarrier Communications

Shannon Formula in Multicarrier Communications

In the power allocation problem, we have a limited amount of total transmit power to be
allocated to different subchannels

We shall assume additive white Gaussian noise (AWGN) for each subchannel.

Let N be the number of subchannels. Denote the noise levels on these subchannels by
n1 , n2 ....nN .

If we allocate transmit signal powers p1 , p2 ....pN for the N subchannels, then from Shan-
non capacity formula from information theory tells us that subchannel i can support the
transmission of data at the bit rate up to
 
p
W log2 1 + i bits per second(bps)
ni

where W is the subchannel bandwidth in Hz.

In order to maximize the total bit rate over all subchannels. We shall see that the
KKT conditions for optimality can be used to obtain an optimal solution.

Optimization Techniques Lecture 6, October 3, 2012. 4/11

 Application: Power Allocation in Multicarrier Communications Power-allocation in multicarrier communications

Power-allocation in multicarrier communications

The power allocation problem is a convex optimization problem.

Let p = (p1 , p2 ....pN ), define the following functions

N  
p
log 1 + i
X
C(p) = −
ni
i=1
N
X
g0 (x) = pi − P
i=1
gi (x) = −pi , i  {1, 2, ...N}

Optimization Techniques Lecture 6, October 3, 2012. 5/11

 Application: Power Allocation in Multicarrier Communications KKT conditions in Optimal Power-allocation

KKT conditions in Optimal Power-allocation

The Lagrangian of the above problem is,

N
X
Λ(p,λ) = C(p) + λ0 g0 (p) + λi gi (p)
i=1

From the Lagrangian optimality condition of KKT,

N
X
∇C(p∗ ) + λ∗0 ∇g0 (p∗ ) + λ∗i ∇gi (p∗ )
i=1
∂Λ(p∗ , λ∗ ) 1
=− ∗ + λ∗0 − λ∗i
∂pi pi + ni
1
pi∗ + ni =
λ∗0 − λ∗i

From the complimentary slackness condition of KKT,

λ∗i gi (p∗ ) = −λ∗i pi∗ = 0

1
pi∗ > 0 =⇒ pi∗ + ni =
λ∗0

Optimization Techniques Lecture 6, October 3, 2012. 6/11

 Application: Power Allocation in Multicarrier Communications Observations

Observations
From above analysis we have two distinct observations

Observation 1: The values of signal plus noise powers are equal for all subchannels in
which nonzero powers are allocated.

In practice, pi∗ ≥ 0 and ni∗ > 0, implying λ∗0 > 0

From complementary slackness

 
N
X
λ∗0 g0 (p∗ ) = λ∗0  pi∗ − P  = 0
i=1

PN
Observation 2: All available power is allocated. we conclude from i=1 pi∗ = P

Optimization Techniques Lecture 6, October 3, 2012. 7/11

 Application: Power Allocation in Multicarrier Communications Waterfilling Algorithm

Waterfilling Algorithm
Observations 1 and 2 give rise to a procedure commonly referred to as waterfilling
algorithm
Since it is analogous to pouring all available water into a container whose depths in
different parts are specified by the parameters n1 , n2 ....nN .

Note that better subchannels are allocated more powers.

Example: Given that (n1 , n2 , n3 , n4 ) = (3, 2, 1, 4) and P = 10, allocate optimal power in
each subchannel.

Optimization Techniques Lecture 6, October 3, 2012. 8/11

 Application: Power Allocation in Multicarrier Communications Capacities of Frequency Selective Channels

Capacities of Frequency Selective Channels

Consider a frequency-selective channel in which the channel gains in the N subchannels
are h1 , h2 ..., hN in addition to different noise levels n1 , n2 ..., nN .
This scenario is an approximation of a filtered AWGN channel in practice.
The power allocation problem is now to choose the power levels p1 , p2 ..., pN for the N
subchannels to maximize the total capacity over all subchannels.
If we allocate power pi , then subchannels i can support
 
hp
W log2 1 + i i bps
ni
ni
By defining the modified noise powers ñi = hi
, i{1, , ...N},, we have the same problem
before, i.e.,
N  
X pi
minimize − log2 1+
ñi
i=1
N
X
subject to pi ≤ P, −p ≤ 0 for all i
i=1

By making N sufficiently large for good approximation, waterfilling is used to compute the
channel capacity for frequency selective channels.

Optimization Techniques Lecture 6, October 3, 2012. 9/11

 Reference Books

Lecture Source and Reference Books

Introduction to linear optimization by Dirmitris Birtsimas

Convex Optimization by Stephen Boyed and Lieven Vanden-
berghe
Lecture Notes from my PhD Advanced studies course Optimiza-
tion in Communication and Network

Optimization Techniques Lecture 6, October 3, 2012. 11/11