17th Telecommunications forum TELFOR 2009

Serbia, Belgrade, November 24-26, 2009.

Scheduling Optimization in HSDPA Networks

Simulating Maximum Terminal Capabilities
Toni Janevski, Senior Member, IEEE, and Kire Jakimoski

In UMTS, the bearers do not set any absolute quality

guarantees (such can never be given in a wireless
transmission) in terms of data rate for interactive and
background traffic classes. The introduction of minimum
service guarantees for users is a relevant factor, and it is
taken into consideration in the performance evaluation of
the different HSDPA schedulers. The service guarantees
interact with the notion of fairness and the level of
satisfaction among users. Very unfair scheduling
mechanisms can lead to the starvation of the least
favourable users in highly loaded networks, and as
described in [1], the starvation of users could have
negative effects on the performance of higher layer
protocols, like TCP. These concepts and their effect on the
HSDPA performance are thus important for our
investigation.
A number of different scheduling algorithms have been
proposed to address these issues [2-6], all with their
respective advantages and trade-offs. Several simulation
experiments were conducted in this paper for evaluating
the performance of different algorithms. We evaluate the
system performance of different groups of schedulers
using CAT 10 UEs (offering up to 14.4 Mbps theoretical
bit rate) in a heavy load scenario with FTP traffic in terms
of system throughput and fairness. Section II discusses
scheduling in HSDPA, and the main performance
measures related to scheduling. Section III presents the
simulation results of the scheduling algorithms. In our
simulations experiments, the performance of CAT 10 UEs
is determined under 14 scheduling schemes. The results
are given in average throughput and delay and overall cell
throughput. Finally, Section IV contains summary of the
work done.

Abstract In HSDPA, a single logical channel is shared

amongst multiple contending users. Besides better link
utilization, scheduling disciplines seek to achieve fair
allocation of this shared resource. However, these two
criteria can potentially be in conflict. This paper investigates
these two criteria performances in a heavy load scenario
using FTP traffic with achievable maximum data rate of 14.4
Mbps, simulating different scheduling algorithms in order to
find the best solution for this scenario. Comparison is also
made with previous given simulation results with achievable
maximum data rate of 3.6 Mbps for three schedulers.
Simulation results show that Proportional Fair Time
scheduling algorithm provides the best results comparing
link utilization and fairness among the users, under heavy
load conditions.
Keywords HSDPA, Scheduling algorithm, Scheduling
optimization, Simulation, Terminal Category.

I. INTRODUCTION

SDPA is a 3.5G wireless system standardized as a set

of technological advancements to UMTS in order to
improve network capacity and increases the peak data
rates up to 14.4 Mbps for downlink packet traffic [1] -[4].
HSDPA utilizes a common downlink shared channel
known as high speed downlink shared channel (HSDSCH), and employs fast link adaptation for downlink
data transfer to mobiles, based on adaptive modulation and
coding (AMC), hybrid automatic repeat request (HARQ)
and a shorter minimum allocation time (transmission time
interval, TTI) of 2ms. In addition to these physical layer
features, the packet scheduling functionality is moved
from the centralized radio network controller (RNC) to the
base station (Node B), where it is embedded in a new
MAC entity known as MAC-hs.
Packet Scheduling functionality plays a key role in
HSDPA. The features included in HSDPA and the new
location of the scheduler in the Node-B open new
possibilities for the design of this functionality for the
evolution of WCDMA. The main goal of the Packet
Scheduler is to maximize the network throughput while
satisfying the Quality of Service (QoS) requirements of
the users.

II. SCHEDULING IN HSDPA AND PERFORMANCE

MEASURES
The process of scheduling refers to the process of
allocation of transmitter time and power (at Node-B) to
the randomly time-varying mobile data connections
(mobile users, UE). Scheduling decisions control the
allocation of resources amongst users, and this allocation
determines the overall performance of a system. In
HSDPA, the packet scheduler is moved from the Radio
Network Controller (RNC) to Node B, with the resulting
advantage of fast link adaptation techniques. The idea is to
enable scheduling such that, if required, most of the cell
capacity may be allocated to one user, when its conditions
are favorable. In the optimum scenario, scheduling should
be able to track fast fading of users.
In HSDPA, the User Equipment (UE) sends its
downlink channel quality feedback to Node B in the form

Dr. Toni Janevski is Professor at the Faculty of Electrical Engineering

and Information Technologies, Ss. Cyril and Methodius University,
Karpos 2 bb, 1000 Skopje, Macedonia, (e-mail: [email protected]).
Kire Jakimoski is Ph.D. student at the Faculty of Electrical
Engineering and Information Technologies, Ss. Cyril and Methodius
University, Karpos 2 bb, 1000 Skopje, Macedonia, (e-mail:
[email protected]).

206

of the Channel Quality Indicator (CQI). The packet

scheduling algorithm should take into account the radio
channel conditions (based on CQI value), UE capabilities,
resource availability, buffer status, and the QoS
requirement of different users.
High resource utilization is one of the main aims of
scheduling. The best strategy to maximize link utilization
in a wireless network is to schedule users who have the
best channel condition. But, if Node B always serves users
with good channel condition, then users in poor channel
condition would be starved. This may cause a degradation
of performance of a higher layer protocol, such as TCP.
Hence, it is important not to investigate the performance
of a scheduling algorithm in terms of only the total system
throughput or link utilization.
The throughput at UE depends upon the scheduling
scheme employed at Node-B. In [7] the total cell
throughput of a HSDPA system having a total of Nu users
with a mean bit rate of Ri is given by:

Nu
T = E Ri ,
i =1

also considering Node B waiting times is the key insight

for aiming at short-term fairness.
III. PERFORMANCE EVALUATION
In this section, we evaluate the performance of our
proposed scheduling algorithm by means of simulation
with the help of Network Simulator ns-2 [9] and its
Enhanced UMTS Radio Access Network Extensions
(EURANE) [10].
A. Simulation Mode

(1)

Fig. 1. Topology used in simulation

In this paper we simulate heavy load conditions.
Simulation topology and the bandwidth and link delays for
each wired link used are shown in Fig. 1. Simulations of
the highest possible throughput assigning Terminal
Category 10 offering the highest bit rates (theoretical 14.4
Mbps) has been done with previous modification of CQI
code.

where for Round Robin, Ri is given by:

Ri =

1
Nu

W ( N log 2( M ) ) m ,i
SF
N s ,i

(2)

In (2) W is a chip rate, SF is spreading factor, Ns,i is the

number of transmissions for user i due to HARQ, M is the
modulation order, is code rate and km varies with the
position of UE. For maximum C/I scheduling, the
expression for Ri is the same as in (2) but multiplied to the
probability that a Transmission Time Interval (TTI) is
allocated to user i, for which no simple or closed for
formula has been derived.
The performance evaluation of a scheduling algorithm
must be based on two metrics:
Link utilization
User level fairness long-term and short-term
Link utilization can be measured by the total system
throughput. For ensuring user level fairness, users should
be scheduled taking into account their QoS requirements.
For a quantitative measure of long-term fairness, Jain et al
[8] have proposed the fairness index by

( x )
N

i =1 i
N
2
i =1 i

N x

B. Traffic Model
In simulations as a traffic source FTP traffic generator
is used within TCP Agent, which is standard FTP
generator of NS-2. This kind of traffic belongs to
background class applications. This class presents the
most delay latency tolerance since the destination does not
expect the data within a certain time. Typical examples of
this class are e-mail, file transfer protocol (FTP), short
messages (SMS), and multimedia messages (MMS). FTP
is one of the most popular and widely used Internet
applications besides Hypertext Transfer protocol (HTTP),
email, etc. These Internet applications rely on two
common protocols, namely, Transmission Control
Protocol and the Internet Protocol (TCP/IP), to reliably
transport data across heterogeneous networks. QoS
requirements of this service class are: one way delay no
limit; bit error rate between 4*10-5 and 6*10-8; delay
variation no limit; use of retransmission mechanism
MAC-hs, RLC; transport layer TCP.

xi 0 i

(3)

C. Simulation Experiments
This subsection presents the simulation results of the
algorithms using Category 10 terminals (Table 1) offering
the highest bit rates (theoretical 14.4 Mbps) Comparison is
also made with previous results for three schedulers using
Category 5 terminals (Table 1) offering 3.6 Mbps bit rate.
Mobiles are considered as pedestrian moving at equal
distances from the Node-B (base station).

where xi is the performance measure of concern for user i,

which may be in terms of delay, or throughput, or fraction
of demand served, and N is the number of users. In this
paper, xi is calculated based on the fraction of demand
served, i.e. Ri/RTi, where Ri is the average throughput
achieved by user i, and RTi is the maximum bit rate (MBR)
requirement of the user. This fairness index lies between 0
and 1; as the variance of xi values increases, the index
approaches to 0.
For short time-scale fairness, the waiting time of MAChs PDUs should also be considered while providing longterm fairness. Thus, providing long-term fairness while

207

TABLE 1: HSDPA TERMINAL CAPABILITY CATEGORIES [11]

1
2
3
4
5
6
7
8
9
10
11
12

Maximum
number of
parallel codes
per HS-DSCH

Minimum
inter-TTI
interval

Transport
channel bits
per TTI

Achievable
maximum
data rate
(Mbps)

5
5
5
5
5
5
10
10
15
15
5
5

3
3
2
2
1
1
1
1
1
1
2
1

7298
7298
7298
7298
7298
7298
14411
14411
20251
27952
3630
3630

1.2
1.2
1.8
1.8
3.6
3.6
7.2
7.2
10.2
14.4
0.9
1.8

700
Round Robin scheduling
Max C/I scheduling

600

Fair Channel Dependent Scheduling

Average throughput [kbps]

Cate
gory

500

400

300

200
100

0
0

50

100

150

200
250
300
350
400
Distance from the Node B [meters]

450

500

550

Fig. 2. Average throughput of the simulated mobile users

5
Round Robin scheduling
Max C/I scheduling
Fair Channel Dependent Scheduling

Average delay [s]

In Fig. 2 and Fig. 3 are presented results for average

throughput and average delay for the RR, max C/I and
FCDS packet schedulers.
The simulation environment of this results is the same
with the one in Fig.1.c,d in [12], except that here is
assigned CAT 10 UEs instead of CAT 5 UEs. Average
throughput is normally, as expected increased for all three
scheduling algorithms. But, what we want to stress from
the results is that increasing of the average throughput acts
differently at the three scheduling algorithms when the
distance from the Node-B is analyzed. For RR scheduler,
as mobile users are closer to the Node-B, they will
experience more increased throughput, changing the CAT
5 to CAT 10 and users that are at the periphery of the cell
will not experience big difference of throughput.
The effect of increasing the achievable maximum data
rate from 3.6 to 14.4 is opposite for the mobile users if
they use Max C/I scheduler. Users that are closer to the
Node-B will not experience big difference at increasing of
the throughput, and those that are further from the Node-B
will gain more increasing of the throughput. FCDS
scheduler has the worst results from increasing the
achievable maximum data rate from 3.6 to 14.4 (CAT 5 to
CAT 10). Only users closer to the periphery will improve
a little their throughput performance.
Increasing the achievable maximum data rate from CAT
5 UEs to CAT 10 UEs, caused better fairness of the users
using max C/I scheduler, worst fairness using RR
scheduler and approximately the same fairness using
FCDS scheduler. Improvement of the fairness of the max
C/I schedulers is explained with the better increasing of
throughput for the users that are further distanced from the
Node-B, and the explanation is vice versa for the RR
scheduler.
Average delay of the users has similar attitude
comparing the results in Fig.1.d in [12] with the results in
Fig. 3. The results of average delay are better with CAT
10 UEs at round robin and max C/I and worse at FCDS
scheduler compared with CAT 5. The best results of
average delay give the RR scheduler. Average delay is not
satisfying a part of UEs that are using FCDS and max C/I.

0
0

50

100

150

200
250
300
350
400
Distance from the Node B [meters]

450

500

550

Fig. 3. Average delay of the simulated mobile users

700

Round Robin
Max C/I
FCDS
Proportional Fair Time
Fair Throughput
Proportional Fair Throughput
Prioritized Fair Throughput
Mean C/I
Prioritized Max C/I
Fair Time
Prioritized Dif f erentiated Serv ices
Prioritized Ray leigh Peak
Weighted Dif f erentiated Serv ices
Weighted Ray leigh Peak

Average throughput [kbps]

600
500

400
300

200
100

0
0

50

100

150

200
250
300
350
400
Distance from the Node B [meters]

450

500

550

Fig. 4. Average throughput of all 14 packet schedulers

using CAT 10 UEs
7

Round Robin
Max C/I
FCDS
Fair Throughput
Proportional Fair Throughput
Prioritized Fair Throughput
Proportional Fair Time
Mean C/I
Prioritized Max C/I
Fair Time
Prioritized Dif f erentiated Serv ices
Prioritized Ray leigh Peak
Weighted Dif f erentiated Serv ices
Weighted Ray leigh Peak

Average delay [s]

2
1

0
0

50

100

150

200
250
300
350
400
Distance from the Node B [meters]

450

500

550

Fig. 5. Average delay of all 14 packet schedulers using

CAT 10 UEs

208

D. Performances of all 14 scheduling algorithms using

CAT 10 UEs
Three groups of scheduling algorithms which have
similar average throughput results are detected in Fig. 4:
Round Robin, Fair Time, Weighted Rayleigh Peak,
Prioritized Differentiated Services, Prioritized Rayleigh
Peak, Weighted Differentiated Services scheduling; max
C/I, prioritized max C/I scheduling; Fair Throughput,
Proportional Fair Throughput, Prioritized Fair Throughput
scheduling.
The first group where there are 6 scheduling algorithms
produces fair time management and division within users,
but doesnt care about the users with low CQI. The users
near the edge of the cell have low average throughput. The
second group with two scheduling algorithms (max C/I
and prioritized max C/I) and mean C/I gives huge amount
of network resources for those UEs placed closer to NodeB, but users from further parts are not satisfied. Because
of its lack of fairness, these three scheduling algorithms
cannot be used in real HSDPA networks. The third group
of fair throughput algorithms gives the best results of
fairness of the users, but overall cell throughput together
with FCDS, as can bee seen in Fig. 6 isnt so optimal.
Fig. 5 presents the results of the average delay of all 14
simulated scheduling algorithms for the simulated
scenario. The worst results in average delay have FCDS,
mean C/I, max C/I and Prioritized max C/I scheduling
algorithms, because approximately half of the users that
are further from the Node B using these algorithms have
very high average delay values. The best results in average
delay, considering all 20 users gives the Proportional Fair
Time scheduling algorithm. The curve of this algorithm, as
we can see in Fig. 5 has the most stable low average delay
results for all simulated users.
Simulations results show that Proportional Fair Time is
the best solutions for the simulated environment with
achievable maximum data rate of 14.4 Mbps (CAT 10
UEs), giving the best cell resources utilization. Using this
scheduling algorithm, all users in the cell can feel equal.
Finally, the overall cell throughput is utilized optimally
using Proportional Fair Time scheduling algorithm, as can
be seen in Fig. 6. Proportional Fair Time scheduling
attempts to give all users/flows the same probability of
being scheduled while keeping attention to the
instantaneous quality received by using a Relative
Instantaneous Channel Quality (RICQ) defined as the
ratio between current and average user SIR. With this
method users are scheduled only during constructive
fades, thereby raising both the overall cell throughput and
the user data rates while keeping fairness among users.
The flow having the higher RICQ value is scheduled first.

8.000

Cell throughput [kbps]

7.000
6.000
5.000
4.000
3.000
2.000
1.000

Fa
Fa
ir
Ti
ir
m
Pr
Th
e
op
ro
or
u
gh
ti o
pu
na
Pr
t
lF
op
ai
rT
or
ti o
im
na
e
M
lF
ea
ai
n
rT
C
/I
hr
P
Pr
ou
rio
io
gh
ri t
ri t
iz
pu
iz
ed
ed
t
M
Fa
Pr
ax
ir
io
C
Th
ri t
/
I
iz
ro
ed
ug
Pr
D
hp
io
iff
ri t
ut
.S
iz
er
ed
vi
W
R
ce
ei
ay
s
gh
le
te
ig
d
h
W
Pe
D
ei
iff
gh
ak
.S
te
e
d
rv
R
i
ce
ay
s
le
ig
h
P
ea
k

/I

S
D
FC

M
ax

Scheduling algorithms

Fig. 6. Overall cell throughput for different Scheduling

Algorithms
Furthermore, we have also tested different HSDPA
scheduling algorithms and recommended the most
efficient scheduling technique for achievable maximum
data rate of 14.4 Mbps (CAT 10 UEs) in a heavily loaded
scenario in a pedestrian environment, using non-real time
service (FTP traffic). We conclude that the Proportional
Fair Time scheduling algorithm ensures fairness among
users in the given scenario, according to their QoS
requirements, while seeking to maximize link utilization.
REFERENCES
[1]

Stevens R. "TCP/IP Illustrated". Volume 1. The Protocols. Addison

Wesley, 1994.
[2] G. Aniba and S. Aissa, Adaptive Proportional Fairness for Packet
Scheduling in HSDPA, IEEE Communication Society, Globecom,
2004.
[3] Li-Chun Wang and Ming-Chi Chen, Comparisons of linkadaptation based Scheduling algorithms for the WCDMA system
with high-speed downlink packet access, Canadian Journal of
Electrical and Computer Engineering (CJECE), Vol. 29:109 116,
2004.
[4] Hua Fu and Dong In Kim, Analysis of Throughput and Fairness
with Downlink Scheduling in WCDMA Networks, IEEE
Transactions on Wireless Communications, Vol.5, Nr.8, August,
2006.
[5] Kinda Khawam, The Modified Proportional Fair Scheduler, 17th
Annual IEEE International Symposium on Personal, Indoor and
Mobile Radio Communications, 2006.
[6] T. Kolding, Link and system performance aspects of proportional
fair scheduling in wcdma/hsdpa, in Vehicular Technology
conference, 2003. VTC 2003-Fall. 2003 IEEE 58th, vol.3, pp. 17171722.
[7] M. Assaad and Dj. Zeghlache, Cross-Layer Design in HSDPA
System to Reduce the TCP Effect, IEEE Journal on
Selected Areas in Communications, vol. 24, no. 3, pp. 614-625,
March 2006.
[8] Rajendra K Jain, Dah-Ming W Chiu, and William R Hawe. A
Quantitative Measure of Fairness and Discrimination for Resource
Allocation in Shared Systems. 1984.
[9] Network Simulator ns-allinone 2.30 and its documentation is
available on http://www.isi.edu/nsnam/dist/.
[10] Seacorn, Deliverable d3.2v2, end-to-end network model for
enhanced
umts,
Available
at
http://www.tiwmc.nl/eurane/D32v2Seacorn.pdf.gz,
http://www.tiwmc.nl/eurane/eurane_user_guide_1_6.pdf and EURANE code
available at http://www.ti-wmc.nl/eurane.
[11] John Wiley and Sons HSDPA HSUPA for UMTS Jun.2006.
[12] T. Janevski and K. Jakimoski, Comparative Analysis of Packet
Scheduling Schemes for HSDPA Cellular Networks, Telfor 2008,
Serbia, Belgrade, 25-27 November, 2008.

IV. CONCLUSION
In this paper, we have evaluated many scheduling
techniques in a heavy load scenario in a single-service
case using achievable maximum data rate of 14.4 Mbps
(CAT 10 UEs).
In the first case, we have addressed a comparison study
between higher and lower terminal capabilities (CAT 10
and CAT 5 UEs) for three schedulers simulated in
previous papers and have analyzed the difference.

209