1860

IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 19, NO. 10, OCTOBER 2001

Joint Power Control and Intracell Scheduling of

DS-CDMA Nonreal Time Data
Fredrik Berggren, Student Member, IEEE, Seong-Lyun Kim, Member, IEEE, Riku Jntti, and
Jens Zander, Member, IEEE

AbstractThe performance of DS-CDMA systems depends on

the success in managing interference arising from both intercell
and intracell transmissions. Interference management in terms of
power control for real time data services like voice has been widely
studied and shown to be a crucial component for the functionality of such systems. In this work we consider the problem of supporting downlink nonreal time data services, where in addition to
power control, there is also the possibility of controlling the interference by means of transmission scheduling. One such decentralized schedule is to use time division so that users transmit in a
one-by-one fashion within each cell. We show that this has merits in
terms of saving energy and increasing system capacity. We combine
this form of intracell scheduling with a suggested distributed power
control algorithm for the intercell interference management. We
address its rate of convergence and show that the algorithm converges to a power allocation that supports the nonreal time data
users, using the minimum required power while meeting requirements on average data rate. Numerical results indicate a big potential of increased capacity and that a significant amount of energy
can be saved with the proposed transmission scheme.
Index TermsCode division multiaccess, land mobile radio cellular systems, power control, scheduling.

I. INTRODUCTION

S THE demand for wireless wideband access is expected

to increase, forthcoming personal communication systems will have to support real time and nonreal time circuitand packet-switched services. The limited amount of spectrum
along with the expected use of multimedia services make
efficient sharing of the scarce bandwidth necessary. In practice, spectrum sharing introduces interference, degrading the
ability of reliable communications. Therefore, some form of
interference control mechanism is needed. The development
of efficient radio resource control for reducing unnecessary
interference is essential for the provisioning of wireless data
services of future wireless personal communication systems. As

Manuscript received December 1, 2000; revised June 1, 2001. This work was
supported in part by grants from VINNOVA and the Academy of Finland. Parts
of this work were presented at the IEEE MMT 2000, Dec. 36, Duck Key, FL,
USA.
F. Berggren and J. Zander are with Radio Communication Systems,
Department of Signals, Sensors and Systems, Royal Institute of Technology,
S-100 44, Stockholm, Sweden (e-mail: fredrik.berggren@radio.kth.se;
jens.zander@radio.kth.se).
S.-L. Kim is with the Radio Resource Management and Optimization Laboratory, Information and Communications University, Yusong, Taejon 305-600,
Korea (e-mail: slkim@icu.ac.kr).
R. Jntti is with the Control Engineering Laboratory, Department of Automation and Systems Technology, Helsinki University of Technology, FIN-02015
Hut, Finland (e-mail: riku.jantti@hut.fi).
Publisher Item Identifier S 0733-8716(01)08482-7.

the characteristics of wireless data are different from those of

speech, e.g., variable data rates and different quality of service
requirements are to be provided, the problem of efficient radio
resource management becomes even more intricate. In this
work we focus on interference control mechanisms suitable for
wireless data services.
In the DS-code division multiacess (CDMA) cellular system,
an adopted multiple access method for future personal communication systems [1], the total received interference can be divided into intracell interference, originating from simultaneous
transmissions within the cell, and intercell interference, caused
by transmitters in other cells. The quality of a connection is
often assumed to depend on the signal-to-interference ratio at
the receiver. To alleviate the impact of intercell and intracell interference, transmitter power has turned out to be an important
controllable resource for balancing the desired signal and interference powers at the receiver, rendering increased capacity
and lower energy consumption. Much work has been performed
in the area of designing quality-based power control algorithms
for finding the minimum power allocation when users require
fixed data rates at a given target signal-to-interference ratio.
Typically this direction sees applications for real time services
like voice (see [2], [3] and references therein). The iterative
power control algorithms suggested for this problem direct the
powers to a fixed point where every user has minimized its
power. The optimal power solution for this problem has been
shown to be Pareto optimal [4], i.e., minimizing each users
transmitter power also minimizes the total sum power. Integrated over time, this solution will minimize the energy consumption. However, the requirement on a fixed data rate and
thus signal-to-interference ratio prohibits any interruption of the
data stream; therefore, the degree of freedom to coordinate the
transmissions in time is limited. Herein we consider the problem
of supporting a given set of nonreal time service users with a certain required average data rate with minimum energy consumption. As for the fixed data rate case, we will obtain this via a minimum power solution. With average data rate, we here mean that
a least amount of data has to be delivered over a specified time
interval, a more relaxed constraint than a fixed data rate. This
increases the options of adaptively planning the transmissions
in a more efficient way. For such a planning scheme to be practical, it should be able to quickly adapt to changing propagation
and traffic conditions and preferably be distributed. One scheme
which is distributed in the sense that it does not require any coordination between base stations is to plan the transmissions independently in a cell-wise manner. Thus, in addition to power
control, by utilizing the possibility of scheduling the transmissions within the cell, the intracell interference can be efficiently

07338716/01$10.00 2001 IEEE

BERGGREN et al.: JOINT POWER CONTROL AND INTRACELL SCHEDULING

avoided. Since DS-CDMA systems are considered to be interference limited, less interference allows for reduced transmission power, in turn increasing the capacity and lowering the energy consumption. Hence it suggests that if the same amount of
data can be delivered using this type of scheduling, the impact
will be higher energy efficiency since no power is wasted on
combating the intracell interference.
A fundamental issue for these nonreal time services is the
multiple access scheme, whether simultaneous transmissions
are beneficial or not [5][7]. Although different models are considered, the results of those works to some extent exhibit a time
division solution for maximizing the system throughput. In [8]
and [9], the concept of utilizing a hybrid of TDMA and CDMA
was outlined. This combined access method was shown to take
advantage of the high intracell capacity of TDMA and the intercell interference suppression ability of CDMA. In [10], it was
found that throughput in the DS-CDMA downlink is maximized
when each base station transmits to at most one user at a time
and uses maximum power. Schemes which coordinate transmissions between cells were also suggested for a cellular highway
system. Intercell coordination for hexagonal systems was further considered in [11]. Recently, a similar concept to [10],
where simultaneous transmissions within the cell are avoided,
has been proposed for supporting high data rates (HDR) in the
downlink for a system based on current CDMA physical layer
structure [12]. In the HDR-CDMA system, due to the vastly
different requirements, high-data-rate users are separated from
low-rate voice users by different RF carriers. The high-data-rate
users are scheduled over time slots, where the slot lengths depend on channel conditions and transmission is executed with
a constant maximum transmit power. Aspects of power control
and scheduling were investigated in [6], where the problem of
minimizing the time span for emptying all users data buffers
was addressed. A discrete time model was assumed and the results showed that the optimal solution may require that time division must sometimes be induced. In [13] it was found that
under some conditions, the same amount of data could be delivered in the uplink with less energy by utilizing time division. A
protocol that automatically in a distributed fashion induces time
division based on buffer backlog was suggested to reduce the
energy consumption. In [14], it has been also found that higher
throughput in the DS-CDMA uplink can be obtained by scheduling delay tolerant users, and this gain does not necessarily require more average transmission power.
In this work, we consider efficient use of the downlink radio
resources by exploiting the delay tolerance of nonreal time
service users. Particularly, we elaborate on one-by-one scheduling (TDMA) for the downlink transmission of wireless data
in a DS-CDMA system. A crucial component is quality-based
power control but rather than just focusing on minimum power
solutions [2][4], we will direct this work toward minimum
energy solutions as well, a more natural description seen over
a time interval. The main focus herein is to offer a requested
average data rate with minimum energy consumption by using
intracell scheduling, which is a form of a multiplexing scheme.
We show that intracell scheduling in terms of time division
within the cells merits higher energy efficiency and system
capacity than continuous transmission at a fixed data rate. This

1861

means that either less energy is consumed for transmitting the

same amount of data, compared with a continuous transmission
scheme, or in an increased capacity (transmission rate, number
of admitted users) for the same transmitting powers. In these
types of systems, the base station transmitter has to support
nonreal time data services as well as real time services, like
voice. Since there is often a limit on the total power that can be
transmitted by the base station, it is crucial that every service
uses as small a transmit power as possible. By minimizing the
power levels for the nonreal time services, interference can
be reduced and also more of the base stations total available
transmitter power is left for other services. Hence, more nonreal
time service users can be supported without occupying excess
transmit power resources from real time services. With strong
resemblance to the continuous transmission problem, the
minimum power allocation can be found from a linear equation
system but expressed in terms of the base station powers. For
finding the optimal powers, a distributed power control algorithm is suggested and proven to drive the transmit power to
the minimum power level that guarantees the average data rate
to the nonreal time users. It is found that this power solution
minimizes the energy consumption. Moreover, it is shown that
the power control algorithm can converge at a faster rate in
the average data rate case than for the continuous data rate
case. Based on the spectral radius, i.e., the maximum modulus
eigenvalue, of the power control iteration matrix, we identify
a capacity gain of the proposed power control/scheduling
concept.
The rest of the paper is organized as follows. In Section II, the
system model is described and a general problem definition is
given. Section III contains investigations of intracell scheduling,
which is shown to be a necessary condition for solving the given
problem. In Section IV, a distributed power control algorithm to
be used together with intracell scheduling is suggested and its
convergence properties are scrutinized. Numerical results are
contained in Section V and the conclusions end the paper in
Section VI.
II. MINIMUM ENERGY PROBLEM
A. System Model
Consider the downlink of a DS-CDMA cellular system with
base stations where nonreal time service users can access
a common frequency channel. Let user be connected to base
:
contain the
station and the set
users connected to base station . Consider a time instant where
the link gain between base station and user is stationary and
. Denote by
the
given by
vector of individual downlink powers dedicated to the users. If
we assume user to be connected to base station , the signal-tointerference ratio for user can be written as
SIR

where
user and

(1)

is the intercell interference for

is the thermal noise. The quantity
is the

1862

IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 19, NO. 10, OCTOBER 2001

normalized cross correlation between and at the receiver of

user , that is the effective fraction of the received signal power
from transmitter that contributes to the interference experienced by user . Ideally, the spreading sequences can be chosen
for
but in reality some posiorthogonal, i.e.,
tive correlation will occur due to multipath propagation. In the
analysis, we will exclude the case of no intracell interference,
which would make intracell coordination unnecessary. For the
, we refer to as the orthogonality factor of
case when
the system. We assume that transmission rates are adapted by
varying the processing gain and that the effective data rate of
user is given by

by . Consider the following problem of finding the transmission powers during that minimize the total energy consumed:

(4)

(5)

s.t.

(6)
(7)

SIR

(8)

(2)
The instantaneous data rate is assumed to follow as

is the spreading bandwidth and

a required bitIn (2),
is
energy-to-interference-spectral-density ratio. Therefore,
achieved with a tolerable maximum bit error probability which
is assumed to be a monotonously decreasing function of , the
. Due to hardware constraints, the output power
desired
of the power amplifier is limited. In the downlink, the power
constraint is typically related to the sum of the base stations
power rather than the powers allocated to the individual users.
to be the
vector of base station
Let us define
. Therefore, if the cells have the
powers where
maximum power constraints given by a vector , any feasible
. The minimum
power assignment must then fulfill
at the quality
power assignment for achieving the data rates
is obtained by putting (1) and (2) on a matrix
specified by
form. Hence, solving the linear equation system

SIR

(9)

which is the same linear relation as in (2). Thus, we seek the optimum stationary base station power allocation which delivers
the data within the interval with minimum energy. Replacing
, the minimum energy problem
constraint (6) with
reduces to the minimum power assignment problem for continuous transmission considered in [2][4] and [15], which has its
solution given by (3). Constraint (6) also restricts the problem
formulation from the use of intercell coordination by requiring
all base stations to transmit with constant powers over the whole
scheduling interval.
III. INTRACELL SCHEDULING

(3)
where

is an
identity matrix,
matrix with elements

denotes an

and
is an
vector with
gives the componentwise smallest power vector. In this notation,
the indexes may correspond to the same physical base station.
The power control algorithms in [2][4] and [15] solve (3) in a
distributed fashion.
B. Problem Definition
Assume now that there exists a specified scheduling interval,
, over which a nonreal time service user requires a
bits have to
minimum average data rate . That is, at least
be received over the scheduling interval at the quality specified

As the service requirement is only specified in terms of an

average data rate over a certain interval, described by constraint
(5), it opens possibilities of a more flexible use of the spectrum,
e.g., to schedule transmissions. One possible strategy to provide
the average data rate is continuous transmission with the rate
, i.e., reducing the problem to (3). Another way would be to
transmit with a higher rate but for a shorter time. This may not
come for free though. When a user utilizes a higher data rate, a
high transmission power might be needed to compensate for the
decreased processing gain, which causes increased interference
for other users. Alternatively, a user can still transmit at a higher
rate using the same power, if the experienced interference can
be decreased. Our approach is to decrease the interference by
proper transmission scheduling. To elaborate on this further, let
us use the following definitions similar to those in [5].
is inDefinition 1: A rate vector
stantaneously achievable if there exists a feasible power vector
such that
SIR
for all
.
be an instantaneously
Let
achievable rate vector, where is a discrete index. Then an
average rate is defined as a linear combination of such vectors.
is
Definition 2: A rate vector
achievable on average if it can be expressed as
where
and
.

BERGGREN et al.: JOINT POWER CONTROL AND INTRACELL SCHEDULING

To interpret Definition 2, suppose

bits are transmitted
for user during an interval of length , using a sequence of
. Since
,
represents the
rate vectors
is used.
fraction of the scheduling interval the rate vector
Now, let us consider rate vectors , such that not more than one
user per cell has nonzero rate simultaneously. Multiplexing with
these vectors, we can define intracell one-by-one scheduling as
follows.
Definition 3: If transmission is performed using a sequence
, where for every such rate vector
and
of rate vectors
base station , there exists at most one component
for all
, then it is referred to as (intracell) one-by-one
scheduling.
Further let

denote the fraction of the scheduling interval during which user

is served in the one-by-one schedule. Define for all users
(10)

1863

tion is that
have

, it implies that

and we

(13)
and all users in cell can be scheduled
Hence
within .
Proposition 1 says that capacity is not lost for scheduling
, we can
the nonreal time services. In fact, since
choose either to increase the transmission rates or admit more
nonreal time service users. Alternatively, the transmit power
can be lowered so that more of the base stations power is left
for other services, e.g., real time traffic like voice. Given a
rate vector , to determine if it is instantaneously achievable
is rather difficult with limited information of the link gains.
applies,
However, in the case where the assumption
the following proposition gives a necessary condition.
, a necessary condition
Proposition 2: If
to be instantaneously achievable is
for a rate vector
for all cells .
, then the power can be solved from
Proof: Let
(1) and (2)

(11)
and therefore
denotes the intracell-to-intercell interference ratio
where
for user . We can then prove the following proposition.
is instantaProposition 1: If
neously achievable, it is achievable in average by one-by-one
scheduling.
Proof: We prove the proposition by showing the achievability of one-by-one scheduling when the total power that is
used for all the users in the cell is instead assigned to one single
user at a time in the cell. Consider an arbitrary cell . When
to one single user in cell , the
assigning the total power
generated intercell interference remains the same as if the cell
power were divided on several users. Thus other cells are not
influenced by the intracell scheduling in cell . Therefore, it is
sufficient to prove the proposition for one single cell. For the
proposition to be true, there must exist s such that for every
user
(12)

Clearly the total power must be positive and limited in all cells,
from which the condition follows.
A more conservative condition applicable to a single cell
can be found by including the maximum power constraint
. Following the proof above, this can equivalently be expressed as
(14)
Therefore, if and the maximum achievable signal-to-interfer, can be measured for all users
ence ratio,
in the cell, (14) can serve as an admission criteria in cell for a
.
given intercell interference level
A. Energy Efficiency

where

. Using (10) and (11), we get

Then the proposition is true for cell if the average rate can
be achieved for all users in the cell within the interval by
. Since our assumpone-by-one scheduling, i.e.,

A major benefit from using one-by-one scheduling is that it

requires less energy for delivering the same amount of data as
for continuous transmission, which we state below.
Corollary 1: For any single cell, one-by-one scheduling requires the least energy of all schedules for providing an average
data rate.
Proof: Assume that the minimum energy solution requires
that at least two users transmit simultaneously between the times

1864

IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 19, NO. 10, OCTOBER 2001

and ,
and that the sum power of the
is
users is . Then the proof of Proposition 1 gives that if
assigned to one single user in the cell, the same data amount
can be delivered in
units by
transmitted in
one-by-one scheduling. Since the energy consumption in
of one-by-one scheduling is
, simultaneous
transmissions cannot minimize the consumed energy.
Corollary 1 says that if no coordination between cells is considered, which by constraint (6) is our assumption, one-by-one
scheduling is necessary in all cells for minimizing the energy
consumption. In other words, the solution of (4)(8) will use
time division within the cells. The energy consumed in a cell
when we assign
by one-by-one scheduling is
to a single user. For a single cell , let us define the relative
energy efficiency of one-by-one scheduling compared to continuous transmission

Fig. 1. An example of two users transmitting in a one-by-one fashion using the

power P (solid). By utilizing the whole scheduling interval T , the power can be
decreased to P (dashed), where  = t =t ,  = 1  , and P T = P t .

where the last inequality follows from Proposition 1. For user

the relative energy efficiency is

This means that the whole system will use relatively less energy
, from the defias will every user in it. Assuming that
, the relative energy efficiency is approximately
nition of
. This can intuitively
given by
be understood as follows: If there is much intracell interference
high) and less intercell interference (
low),
(
there is a large possible gain from one-by-one scheduling in
terms of improved energy efficiency.
, it is possible to lower the transmission
Since
power and compensate the lower data rate by prolonging the
duration of the transmission so that users are scheduled over
the whole interval . For that purpose, we may normalize the
s such that
which gives
.
with , we must consequently
From (12), when replacing
for maintaining the same
adjust the power to
average data rate. Therefore
and the energy consumption remains the same, thus and
are unchanged. This is exemplified in Fig. 1.

Corollary 2: For any single cell, the minimum time span

for transmitting the data of all users is achieved by one-by-one
and is given by
.
scheduling using
Proof: Assume that the minimum time span solution requires that at least two users transmit simultaneously between
and that the sum power of
the times and ,
the users is . Then the proof of Proposition 1 gives that if
is assigned to one single user in the cell, the same data amount
can be delivered in
units by
transmitted in
one-by-one scheduling. This contradicts the assumption, thus
one-by-one scheduling can meet any time span achievable by
simultaneous transmission. From (2) it follows that the rate is
maximized and thus time span minimized when using maximum
.
power
IV. POWER CONTROL
Now, let us investigate suitable power control for one-by-one
in
scheduling. Since Proposition 1 says that
all cells when using the same total transmission power as for
continuous transmission, it is possible to deliver data with
one-by-one scheduling but using less transmission power. The
for all
lowest such power assignment will yield
cells , which is found by utilizing the whole scheduling interval for transmission. For that purpose we use (12) expressed
in the base station powers to solve for a state such that

B. Minimum Time Span

For some services, the quality of service guarantee can be
in the form of a maximum acceptable message delay. The
problem of minimizing the maximum message delay given a
certain amount of data in each users buffer was studied in [6]
and referred to as the minimum time span problem. Therein, an
analysis was carried out but practical algorithms for achieving
the optimal schedule were not suggested. Consider any single
bits
cell in the system and the case where each user has
to transmit and let us apply our proposed schedule. Then we
can state that one-by-one scheduling minimizes the total time
it takes for all users in the cell to transmit their data.

for all

cells. With

tion system in

and evaluating for

, this is a linear equawe get

BERGGREN et al.: JOINT POWER CONTROL AND INTRACELL SCHEDULING

Proceeding in the same manner for the other rows, we can write
the problem in matrix form as
(15)
identity matrix,
where is an
matrix with elements

and

is a

in an

iteration

vector with

Now, for the purpose of solving (15) in a distributed fashion,

consider the following iterative power control procedure:
(16)

1865

given by
which is the solution to
.
Remark: It is also possible to prove convergence of
by
verifying that it is a standard interference function [3], which in
addition guarantees its asynchronous convergence.
It should be also pointed out that the algorithm will converge
to a fixed point even if is not achievable on average. In that
.
case, at least one cell will use the maximum power
This phenomenon could be utilized as an incentive for congestion control by disconnecting users or decreasing some users
data rates. The power control algorithm (17) also exhibits the
property that if all cells are supported, the powers decrease monotonously and all cells stay supported. This is stated below
denotes the value at iteration .
where
where
Corollary 3: If there exists an iteration
for all cells , then

and

or equivalently for each cell

Taking into account the power constraint, let us define the mapping

for all and

.
Proof: We can equivalently express the power control algorithm as

(17)
operator works componentwise. The following
where the
result guarantees that the optimal power allocation will be
found.
, the mapping
Proposition 3: For any initial
converges with geometric rate to a power vector that supports
for all cells , if such a power allocation is feasible.
Proof: Consider the weighted maximum norm
and the consistent weighted matrix norm
where
is a nonsingular matrix. It can be
shown that is irreducible and nonnegative and therefore has
. By
a positive real eigenvalue equal to its spectral radius
[16, Th. 3.7], a necessary condition for a feasible solution of (15)
. Let
be the PerronFrobeto exist is that
and choose
nius eigenvector of , that is
, then

Therefore

so if
and the fixed point is within the feasible power
range, is a pseudo-contraction mapping with respect to the
weighted maximum norm and converges to a unique fixed point

Let
and assume that
which gives
cell we have

in all cells ,
for all . Therefore, for any

By mathematical induction, the relation holds for

.
The proof of Proposition 3 illustrates how the rate of convergence is strongly related to the spectral radius of the iteration
matrix. It is easy to see that the asymptotic rate of convergence
. Reference [17, Th. 2-3.4]
is equal to the spectral radius
says that all matrix norms are greater than or equal to the spectral
radius. Thus, it follows that the matrix norm used in the proof of
Proposition 3 is the smallest possible one which justifies its use
as a convergence measure. Now, let us compare with the continuous transmission problem (3). A related way to distributively
find the minimum power allocation, neglecting the power consuggested
straints, is to use a mapping
in [15]. In a similar manner, it can be shown that this algorithm
. Rate of converwill asymptotically converge at the rate
gence is a means to compare how capable iterative algorithms

1866

IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 19, NO. 10, OCTOBER 2001

are of quickly bringing the powers to desirable levels. The issue

of comparing rates of convergence can therefore be reduced to
relating the different spectral radii of and . If we define a
matrix with elements
new

we have that
with equality only if there exists
exactly one user per cell. This is due to [17, Th. 2.1-16], which
states that the spectral radius becomes larger when the magnitude of the matrix elements increases. The matrix can be interpreted as the iteration matrix of a problem where no intracell
interference exists, for example by fully orthogonal spreading
sequences at the receiver but it is also useful for describing intracell time division. We can relate its spectral properties to
by the following proposition.
Proposition 4: The spectral radius fulfills
where the inequality becomes equality only if there is exiff is instanactly one user per cell. In particular,
taneously achievable.
be an eigenvector such that
.
Proof: Let
Then for any

we obtain for row

which contradicts the assumption that

is not an eigenvalue
of . Therefore, and have the same nonzero eigenvalues.
is the maximum modulus eigenvalue of , it folSince
lows that it is also the spectral radius of and the corresponding
PerronFrobenius eigenvector can be found from (19). A necessary condition for to be instantaneously achievable is that
(3) has a feasible solution. By [16, Th. 3.7], (3) has a unique and
. Therefore, we can conclude that
positive solution iff
iff is instantaneously achievable.
Hence, if
is instantaneously achievable, (17) will always
converge to a fixed point that will support an average data rate. If
the mobiles connect to the base station with lowest attenuation,
a condition for determining if is achievable in average sense
is given by the following result.
, a sufficient condition for (15) to
Corollary 4: When
is that
have a positive solution

(18)

Define a vector

with
(19)

so that with

, (18) becomes

Therefore we have shown that is an eigenvector of

with
diseigenvalue . By Lemma 1 (see the Appendix), has
tinct eigenvalues which therefore also are eigenvalues of . It is
clear that if zero is an eigenvalue of , it is also an eigenvalue of
. We next show that does not have any other nonzero eigenvalues than has. If has exactly distinct eigenvalues then
distinct eigenvalues, we must
it is trivial, but if has
show that does not have any additional eigenvalues. Assume
is an eigenvalue of but not of . Let
that
be the eigenvector of corresponding to . If we choose

for all cells .

Proof: From [16, Corollary 1, pp. 17], the spectral radius,
in this case is less than or equal to the maximum modulus
we have
for all
row sum of . When
and from [16, Th. 3.7], the spectral radius must be strictly less
than unity if the powers should be positive. This gives

So, if
for all cells , the corollary
follows.
It should be pointed out that Corollary 4 could be very con. This is likely to occur when
servative, especially if
is large, implying that Corollary 4 has its practical applicability limited to low values of .

BERGGREN et al.: JOINT POWER CONTROL AND INTRACELL SCHEDULING

Fig. 2.

1867

The relative system energy efficiency for different orthogonality factors  as function of the user data rate.

Proposition 4 shows that the power control for the scheduled

system (15) directs the powers faster to the fixed point than for
(3). In an ever-changing radio environment, the increased rate of
convergence relates to how quickly the powers reach desirable
levels. The spectral radius is important from a rate of convergence point of view but it has been also identified as a measure strongly related to congestion and capacity [18]. For exmay always be found so that
.
ample, an
For upholding the same congestion, spectral radius in this case,
all users in the one-by-one scheduled system may therefore intimes. Thus capacity
crease their data rates
in this sense is not less by this transmission scheme, which also
was indicated by Proposition 1.
Since the power vector obtained from (15) is the smallest possible for supporting the average data rate and the necessary condition from Corollary 1 of one-by-one scheduling is fulfilled, we
can conclude that this cell-site power allocation is the optimum
and solves (4). The practical version of the suggested scheduling and power control procedure could be to let one user in
bits have been
each cell independently transmit until its
delivered, then user is shut off until all other users in the cell
have delivered their data. There are, however, different orders in
which the users in each cell could be scheduled, which we will
leave for further work.
V. NUMERICAL RESULTS
Numerical evaluation is performed by system simulations of
a two-dimensional hexagonal system, in which a wrap-around
technique is used. In each cell, an omnidirectional antenna is
located at the center. We let ten users be dispersed uniformly
over each cells area and connected to the base station with
lowest attenuation. The cell radius is taken to be 1 km and the
thermal noise level is 150 dBW. The link gain is modeled

as a distance-dependent path loss of fourth power and independent log-normal shadowing with a standard deviation of 8
dB. The spreading bandwidth is set to
MHz and
the required bit-energy-to-noise-spectral-density ratio
dB for all users. The same data rate
is used by all
users. In Fig. 2, the result of 1000 independently generated
matrices is depicted for different orthogonality factors . The
figure shows the overall system relative energy efficiency of the
suggested power control and scheduling concept for the cases
where the rates are instantaneously achievable. That is, we plot
, where
is the power vector
for supporting instantaneously, which can be solved for from
.
(3). The base station powers are given by
The plot shows that for high data rates and orthogonality factors,
the same amount of data can be delivered with approximately
four times less energy if we are to use the proposed transmission
scheme. For high data rates, the gain strongly varies with the orthogonality factor. Fig. 3 shows the rate of convergence for the
power control algorithm, measured as the normalized Euclidean
, in
distance to the fixed point,
each iteration for two different data rates. The initial power
and
vector is chosen from a uniform distribution,
is set to 5 W. It is seen, for example, that for
kb/s at 20
iterations, the normalized distance is about 10 , which in turn
shows that the algorithm converges fast and suggests that it is
able to quickly react to changes in interference. As the data rate
is decreased by half to 5 kb/s, the convergence is faster, which
is expected since the spectral radius of the iteration matrix decreases by the same portion. The capacity increase is illustrated
is plotted as function of the orby Fig. 4 where
thogonality factor. The relation seems highly linear with a maximum gain of approximately 2.7 times when the orthogonality
factor is unity. If all users have the same quality targets and
data rates, the linear behavior can quantitatively be explained

1868

IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 19, NO. 10, OCTOBER 2001

Fig. 3.

The convergence rate measured as normalized Euclidean distance to the fixed point for the two cases of R = 10 and 5 kb/s.

Fig. 4.

The relative capacity gain as function of the orthogonality factor  .

by comparing the signal-to-interference ratios for different orthogonality factors

SIR
SIR

studying Figs. 2 and 4. For example, if

and the data
rate is larger than 9 kb/s, the relative energy efficiency is larger
than 2.7, the relative capacity gain, whereas for lower data rates
there is a larger gain in the relative capacity than for the relative
energy efficiency. For small values of the orthogonality factor,
, the relative capacity gain is always larger than the
e.g.,
relative energy efficiency.
VI. CONCLUDING REMARKS

which is linear in . The relation between the relative energy

efficiency and capacity gain can be found by simultaneously

In this work, we considered the problem of delivering data

to users with a required average data rate for a downlink

BERGGREN et al.: JOINT POWER CONTROL AND INTRACELL SCHEDULING

DS-CDMA system. It is observed that having a quality of service constraint based on an average rather than an instantaneous
data rate opens possibilities of flexible transmission scheduling.
We proposed a form of combined intracell scheduling and
power control, where only one user in the cell is allowed to
transmit at a time. With such a scheduling, we showed that
the same amount of data can be delivered with less energy.
For the purpose of supporting the average data rate with as
low transmit powers as possible, we combined the scheduling
with a suggested convergent distributed power control algorithm adjusting cell-site powers. The power control problem
was reduced to a linear equation system. By comparing the
different spectral radii of the iteration matrices, it was found
that the system capacity was increased from the proposed
scheduling/power control scheme as compared to the normal
power control problem. This is due to a lower spectral radius
of the matrix describing the scheduled system, which also
means that the power control can converge faster. Numerical
results indicate that with the proposed concept the transmit
powers can be significantly decreased. This leaves more power
resources free, which opens possibilities for supporting more
nonreal time traffic. The issue of determining the order in
which intracell transmissions should be performed was not
addressed here. In addition, it is expected that there will be
gains which are not captured in this work. Preferably one could
start transmitting to users with good channels. For example,
in each cell , if the users are scheduled in the order of the
, it is likely that if is sufficiently long, a
factors
time-varying channel may improve the channel conditions of
the worst mobiles. Future work includes other models than (2),
e.g., the nonlinear relations described in [13] and also intercell
coordination considerations. Further, upper bounds on the data
rates, discrete time scale, and powers should be investigated.
APPENDIX
Lemma 1:
has less than or equal to
distinct eigenvalues.
. As
Proof: Consider the transposed matrix
in
are equal for a given it implies that
all rows
distinct nonzero eigenvalues. Using the
it can have up to
well-known property that shares the same set of eigenvalues
, the Lemma follows.
as of
ACKNOWLEDGMENT
The authors are grateful to the anonymous reviewers for their
helpful and constructive comments.
REFERENCES
[1] E. Dahlman, P. Beming, J. Knutsson, F. Ovesj, M. Persson, and C.
Roobol, WCDMAThe radio interface for future mobile multimedia
communications, IEEE Trans. Vehic. Technol., vol. 47, pp. 11051118,
1998.
[2] R. Jntti and S.-L. Kim, Second-order power control with asymptotically fast convergence, IEEE J. Selected Areas Commun., vol. 18, pp.
447457, Mar. 2000.
[3] R. D. Yates, A framework for uplink power control in cellular radio
systems, IEEE J. Selected Areas Commun., vol. 13, pp. 13411347,
July 1995.

1869

[4] D. Mitra, An asynchronous distributed algorithm for power control in

cellular radio systems, in Proc. Fifth WINLAB Workshop Third Generation Wireless Information Networks, 1993, pp. 249259.
[5] J. Zander, Performance bounds for joint power control and link adaptation for NRT bearers in centralized (bunched) wireless networks, in
Proc. IEEE PIMRC99, 1999.
[6] R. Jntti and S.-L. Kim, Transmission rate scheduling for the nonreal time data in DS-CDMA systems, IEEE Commun. Lett., vol. 5, pp.
200202, 2001.
[7] S.-J. Oh and K. M. Wasserman, Optimality of greedy power control
and variable spreading gain in multi-class CDMA mobile networks, in
Proc. ACM/IEEE MobiCom99, pp. 102112.
[8] M. L. Honig and U. Madhow, Hybrid intra-cell TDMA/inter-cell
CDMA with inter-cell interference suppression for wireless networks,
in Proc. IEEE Vehicular Technol. Conf., 1993, pp. 309312.
[9] M. A. Arad and A. Leon-Garcia, Scheduled CDMA: A hybrid multiple
access for wireless ATM networks, in Proc. PIMRC96, pp. 913917.
[10] A. Bedekar, S. Borst, K. Ramanan, P. Whiting, and E. Yeh, Downlink
scheduling in CDMA data networks, in Proc. IEEE GLOBECOM99,
pp. 26532657.
[11] L. Maileander, H. Huang, and H. Viswanathan, Simple inter-cell coordination schemes for a high speed CDMA packet downlink, in Proc.
IEEE VTC00 Spring, pp. 18451848.
[12] P. Bender, P. Black, M. Grob, R. Padovani, N. Sindhushyana, and A.
Viterbi, CDMA/HDR: A bandwidth-efficient high-speed wireless data
service for nomadic users, IEEE Commun. Mag., vol. 38, pp. 7077,
2000.
[13] J. M. Rulnick and N. Bambos, Power-induced time division on asynchronous channels, ACM/Baltzer Wireless Networks J., vol. 5, no. 2,
pp. 7180, 1999.
[14] S. Ramakrishna and J. M. Holtzman, A scheme for throughput
maximization in a dual-class CDMA system, IEEE J. Selected Areas
Commun., vol. 16, pp. 830844, June 1998.
[15] G. J. Foschini and Z. Miljanic, A simple distributed autonomous power
control algorithm and its convergence, IEEE Trans. Vehic. Technol.,
vol. 42, pp. 641646, 1993.
[16] R. S. Varga, Matrix Iterative Analysis. Englewood Cliffs, NJ: PrenticeHall, 1962.
[17] D. M. Young, Iterative Solutions of Large Linear Systems. New York:
Academic, 1971.
[18] S. V. Hanly, Congestion measures in DS-CDMA networks, IEEE
Trans. Commun., vol. 47, pp. 426437, Mar. 1999.

Fredrik Berggren (S01) received the M.Sc. degree

in applied physics and electrical engineering from
Linkping University, Sweden, in 1998 and the Lic.
Tech. degree from the Royal Institute of Technology
(KTH), Stockholm, Sweden, in 2001. Since 1999 he
has been with the Radio Communication Systems
Group at the Royal Institute of Technology, where
he currently is working toward the Ph.D. degree.
His research interests include transmitter power
control and quality of service provisioning in cellular
radio systems.

Seong-Lyun Kim (S93M95) received the B.S. degree in economics from Seoul National University in
1988 and the M.S. and Ph.D. degrees in operations research (with application to wireless communications)
from KAIST in 1990 and 1994, respectively.
He has been with Information and Communications University (ICU), Korea, since September,
2000, heading the Radio Resource Management and
Optimization Laboratory. Before joining ICU, he
was a Faculty Member at the Radio Communication
Systems Group, Department of Signals, Sensors
and Systems, Royal Institute of Technology (KTH), Stockholm, Sweden.
His research and education focus include radio resource management and
economics of wireless multimedia. He has recently coauthored a book (with
J. Zander) Radio Resource Management in Wireless Networks (New York:
Artech, 2001).
Dr. Kim is currently an Associate Editor for the IEEE TRANSACTIONS ON
VEHICULAR TECHNOLOGY.

1870

IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 19, NO. 10, OCTOBER 2001

Riku Jntti received the M.Sc. degree (with distinction) in electrical engineering and D.Sc. degree in automation and systems technology from Helsinki University of Technology (HUT), Finland, in 1997 and
2001, respectively.
From 1998 to 1999, he was with the Radio
Communication Systems Group, Royal Institute of
Technology (KTH), Stockholm, Sweden, as a guest
postgraduate student. He is currently working as an
Acting Professor of Communications Engineering
in the University of Vaasa, Finland, and as a
Researcher at HUT. His research interests include transmitter power control
and multiquality of service radio resource management.

Jens Zander (S81M84) received the M.Sc. and

a Ph.D. degrees, both in electrical engineering, from
Linkping University, Sweden, in 1979 and 1985, respectively.
Between 1985 and 1989, he was a Partner and
Executive Vice President of SECTRA, a telecommunications company now listed on the Stockholm
exchange. In 1989 he was appointed Full Professor
to the chair in Radio Communication Systems at
the Royal Institute of Technology in Stockholm,
Sweden, and has been the Manager of the Radio
Communication Systems Laboratory at this university since. Since 1992, he
has also been a Senior Scientific Advisor to Swedens Defense Research
Institute (FOI). He is Cofounder and Scientific director of Wireless@KTH, the
center for Wireless Systems at KTH launched in 2001 and he is on the board
of directors in several Swedish companies in the wireless sector. In addition,
he has published four textbooks in wireless communications and networking,
the last being Radio Resource Management in Wireless Networks (New York:
Artech, 2001), and more than 100 papers in scientific journals and conferences.
His research interests include wireless resource management and architectures
for future wireless access infrastructures.
Dr. Zander is a member of the Swedish Academy of Engineering Sciences
and a Chairman of the Swedish IEEE VT/COM chapter. In 1992 he received the
IEEE VTS Jack Neubauer award for the best systems paper of the year. He is an
area editor for the Wireless Networks and Wireless Personal Communications
journals.