Saunder June 7, 2010 19:40 K10322C007

Chapter 7
Radio Resource
Optimization and
Scheduling Techniques
for HSPA and LTE
Advanced Technologies
Tarek Bejaoui, Anis Masmoudi, and Nidal Nasser
Contents
7.1 Introduction ................................................................................. 266
7.1.1 Objective and Context ........................................................ 266
7.1.2 Radio Resource Management for Advanced Wireless Systems ... 267
7.2 Radio Resource Management for Evolved UMTS Networks ............... 268
7.2.1 MAC Sublayer ................................................................... 269
7.2.2 RLC Sublayer .................................................................... 270
7.2.3 PDCP Sublayer .................................................................. 270
7.2.4 RRC Sublayer .................................................................... 270
7.3 Overview of Packet Scheduling in HSPA and Beyond ........................ 271
7.4 Overview on Packet Scheduling in LTE ............................................ 273
7.4.1 Radio Admission Control .................................................... 273
7.4.2 Uplink Packet Scheduling .................................................... 274
7.4.3 Downlink Packet Scheduling ............................................... 275
7.4.4 Time and Frequency Domains Packet Scheduling .................. 277
7.4.5 Scheduling and Persistent Scheduling ................................... 277
265
Saunder June 7, 2010 19:40 K10322C007
266 Evolved Cellular Network Planning and Optimization
7.5 HSPA and LTE Scheduling Techniques Models ................................ 277
7.5.1 Fair Resource Scheduling Technique Protocol ........................ 278
7.5.2 Fair Throughput Scheduling Technique ................................ 279
7.5.3 Maximum CIR Scheme (Max C/I) ....................................... 280
7.5.4 Fair Channel-Dependent Scheduling (FCDS) Protocol ........... 283
7.5.5 Score-based Scheduling ....................................................... 284
7.6 New Optimized Scheduling Techniques for Multiple Services Case ..... 285
7.6.1 Concept of Cross-Layer Design ............................................ 285
7.6.2 Description of the Introduced Techniques (Protocols) ............ 286
7.6.2.1 Prioritized Differentiated Services Scheduling ........... 286
7.6.2.2 Prioritized Rayleigh Peak Scheduling ...................... 286
7.6.2.3 Weighted Differentiated Services Scheduling ........... 287
7.6.2.4 Weighted Rayleigh Peak Scheduling ....................... 287
7.6.2.5 ThreshOld-based Priority (TOP) Scheduling
Algorithm ............................................................ 287
7.6.3 Scheduling Techniques Optimization According
Services Profiles and Requirements ....................................... 289
7.6.3.1 Prioritized Differentiated Services Scheduling
Technique ........................................................... 289
7.6.3.2 Weighted Differentiated Services Scheduling
Technique Optimization According to Services
Profiles and Requirements ..................................... 291
7.7 Summary and Open Problems ......................................................... 295
References .............................................................................................. 296
7.1 Introduction
In the near future, a broad range of multimedia applications with guaranteed quality
of service (QoS) is expected to be provided by new evolved UMTS networks. The
3
rd
Generation Partnership Project (3GPP) has standardized in Release 6 and 7 new
cellular-based systems denoted respectively as high-speed packet access (HSPA) and
evolved high-speed packet access (eHSPAor HSPA+). Another area of focus defined
in 3GPP Release 8 is the introduction of a new OFDM-based technology through
the long-term evolution (LTE) work item, often referred to as the evolved UMTS
terrestrial radio access network (EUTRAN). It is the next generation cellular wireless
standard that is considered the prominent path to the 4G cellular wireless system.
7.1.1 Objective and Context
The next-generation cellular systems rely on newtechnologies. They make it possible
to bring improved support and performance for constrained services, thanks to
important newadditions suchas enhancedreceivers, multiple-input multiple-output,
Saunder June 7, 2010 19:40 K10322C007
Radio Resource Optimization and Scheduling Techniques 267
continuous packet connectivity, higher order modulations, fast cell selection, and
fast packet scheduling.
The long-termevolution network is introduced through the definition of the new
flatter-IPcore network, and will improve performance by providing higher data rates,
reduced latency, and improved spectral efficiency. The focus was on enhancement
of the radio-access technology (UTRA) and the optimization and simplification of
the radio access network (UTRAN) as well.
The EUTRAN uses a simplified single node architecture consisting of the
EUTRAN NodeB denoted eNB, which communicates with the evolved packet core
(EPC), the mobility management entity (MME), and the user plane entity (UPE).
In the EUTRAN, this eNB supports all the functions in a typical radio network such
as radio bearer control, connection mobility management, admission control and
scheduling, dynamic resource allocation, inter-cell interference coordination, load
balancing, and inter-radio access technology functions. The access stratum resides
then completely at this node.
Radio resource management (RRM) is therefore one of the key design features of
HSPA, HSPA+, and LTE. The objective of the RRM techniques is to optimize the
use of radio resources while fulfilling the quality requirements of the largest possible
number of users. The most representative of these techniques is the packet scheduling.
A packet scheduler controls the allocation of channels to users within the system
coverage area by deciding which user should transmit during a given time interval.
7.1.2 Radio Resource Management for Advanced
Wireless Systems
The next-generation cellular networks are expected to support a broad spectrum
of multimedia services with guaranteed QoS. The resource access protocol that
defines how the wireless medium is shared among contending users, is there-
fore a pioneer element on which depends the overall performance of these net-
works. In this context, an efficient resource allocation scheme should handle a
wide range of information bit rates as well as various types of real-time and
non-real-time services with different traffic characteristics and QoS guarantees. In
addition, the protocol must operate under different constraints of mobility, dynamic
traffic load variations, and highly sensitive wireless links. Under these constraints,
QoS provisioning becomes a challenging task and difficult to ensure. Effective man-
agement of the limited radio resources is therefore important to enhance the net-
work performance. Some cross-layered radio resource management algorithms are
then designed and proposed to optimally adapt to channel conditions and specific
applications requirements. Their purpose is to solve the issue of the lack of built-in
mechanisms for protocol layers that makes it very difficult to provide guaranteed
QoS for multimedia applications.
The packet scheduling constitutes one of the fundamental RRM techniques
for QoS provisioning to the evolved UMTS networks. It controls the allocation of
Saunder June 7, 2010 19:40 K10322C007
268 Evolved Cellular Network Planning and Optimization
channels to users who have data to transfer within the coverage area of the system
and, to a large extent, it determines the overall behavior of the system.
In this chapter, we will investigate some of the radio resource management
features and discuss the various QoS requirements in evolved UMTS networks,
and some of the solutions proposed for effective management of the limited radio
resources to enhance their performance. In this context, the chapter concentrates on
the packets scheduling schemes proposed for QoS provisioning in such networks. It
gives an accurate analytical modeling of some of these protocols like Fair Resource,
Fair Throughput, Proportional Fair algorithms, and the maximum CIR scheme,
which were proposed for HSPA systems. These scheduling protocols that are defined
today as conventional are not suitable for aggregatedmultiple services withdifferent
profiles and required QoS parameters (data, voice, video, etc.).
Indeed, many of these proposed protocols focus on different layers separately.
Each layer communicates with its peer using a set of rules and conventions collec-
tively known as layered protocol, and should perform its own defined functions,
without knowledge of details on the services implementation in the other layers. In
implementing protocols in these layers, control is passed from one layer to the next.
The interactions between layers are controlled, each layer has the property that it
only uses the functions of the layer below, and only exports functionality to the layer
above. In wireless networks, wireless channels and networks are dynamic in behav-
ior, such as temporal and spatial changes quality and user distribution. Furthermore,
meeting the end-to-end performance requirements of demanding applications is ex-
tremely challenging without interaction between protocol layers. The conventional
layered protocol architecture is inflexible and unable to adapt to such dynamically
changing network behaviors, since the various protocol layers can only communicate
with each other in a strict and primitive manner. In such a case, the layers are most
often designed to operate under worst conditions, rather than adapting to changing
conditions. This eventually leads to inefficient use of spectrum and energy.
Thus, new optimized packet scheduling techniques adapted for multiple services
(multi-class of real-time and non-real-time services) in the next-generation evolved
UMTS networks are presented. Some novel approaches based on cross-layered radio
resource management protocols that attempt to focus the radio channel conditions
are then explored.
7.2 Radio Resource Management for Evolved
UMTS Networks
Several RRMfunctions are defined for the evolved UMTS networks HSPA/HSPA+
and LTE. In LTE, these functions are assigned to eNB(s) and mapped over the layers
1, 2, and 3. They include the radio bearer control (RBC), the radio admission control
(RAC), the connection mobility management (CMM), the dynamic resource allo-
cation (DRA) or packet scheduling, the inter-cell interference coordination (ICIC),
and load balancing (LB).
Saunder June 7, 2010 19:40 K10322C007
Radio Resource Optimization and Scheduling Techniques 269
Compared to HSPA, LTE introduces new functionalities in base stations like
the radio link control layer (RLC), radio resource control (RRC), and the functions
defined for the packet data convergence protocol (PDCP) as ciphering and header
compression. The medium access control (MAC) layer functionality is similar to
HSPA operation and remains in the eNB as shown in Figure 7.1 [1].
It depicts the architecture for downlink (as for uplink) of the layer 2 of the
radio access protocol in the eNodeB, which is constituted by the PDCP/RLC/MAC
sublayers supporting the radio resource management.
An overview on services and functions provided by each sublayer are presented
in the following sections [1].
7.2.1 MAC Sublayer
The MACsublayer is a protocol layer that arbitrates and controls access to the shared
transmissionmedium. It runs inboththe UEandthe eNBandhas different behaviors
when running in each, generally giving commands in the eNB and responding to
them in the UE.
Thus, the main functions of the MAC sublayer includes mapping between log-
ical channels and transport channels, multiplexing/demultiplexing of RLC packet
data units (PDUs) belonging to one or different radio bearers into/from transport
blocks (TB) delivered to/fromthe physical layer on transport channels, traffic volume
measurement reporting, error correction through HARQ, priority handling be-
tween logical channels of one UE, priority handling between UEs by means of
dynamic scheduling, MBMS service identification, and transport format selection
and padding.
Logical channels
Scheduling/Priority handling
Segm.
ARQ etc
Segm.
ARQ etc
Segm.
ARQ etc
Segm.
ARQ etc
ROHC
Security
ROHC
Radio bearers
Security
ROHC
Security
ROHC
Security
. . .
. . .
PDCP
RLC
MAC
Transport channels
HARQ
MCCH CCCH BCCH PCCH
Multiplexing UE
n
HARQ
Multiplexing UE
1
MTCH
Figure 7.1 Layer 2 structure for DL.
Saunder June 7, 2010 19:40 K10322C007
270 Evolved Cellular Network Planning and Optimization
Those marked with circles at the interface between sublayers in Figure 7.1, are
the service access points (SAPs) for peer-to-peer communication. The SAP between
the physical layer and the MAC sublayer provides the transport channels. The SAPs
between the MAC sublayer and the RLC sublayer provide the logical channels.
7.2.2 RLC Sublayer
The main services and functions of the RLC sublayer includes the transfer of upper
layer PDUs supporting acknowledged mode (AM) or unacknowledged mode (UM),
transparent mode (TM) data transfer, error correction through automatic repeat
request (ARQ), segmentation according to the size of the TB, re-segmentation of
PDUs that need to be retransmitted, concatenation of SDUs for the same radio
bearer, in-sequence delivery of upper layer PDUs except at handover, duplicate
detection, protocol error detection and recovery, and SDU discard and reset.
7.2.3 PDCP Sublayer
The PDCP sublayer performs both user plane and control plane main functions.
The PDCP sublayer functions in the user plane include header compression and
decompression (ROHC only), transfer of user data, in-sequence delivery of upper
layer PDUs at handover for RLC AM, duplicate detection of lower layer SDUs at
handover for RLC AM, retransmission of PDCP SDUs at handover for RLC AM,
ciphering and timer-based SDU discard in uplink. The PDCP sublayer functions in
the control plane include ciphering and integrity protection and transfer of control
plane data.
7.2.4 RRC Sublayer
The radio resource control (RRC) protocol is being used to configure and control the
radio resource betweenthe eNBand the user equipment. The RRCsublayer performs
the following control plane main functions: broadcast of system information related
to access stratum (AS) and non-access stratum (NAS), paging, establishment, main-
tenance and release of an RRCconnection between the UE and EUTRAN, signaling
radio bearer management, security handling, mobility management, including UE
measurement reporting and configuration, active mode handover, idle mode mobil-
ity control, MBMS notification services and radio bearer management for MBMS,
QoS management, and NAS direct message transfer to/from NAS from/to UE.
The RRCspecifications defined for LTE are slightly different fromthose defined
for legacy 3G-RNC systems like HSPA. The following describes a few:
The number of RRC states: two states in LTE and five in 3G-RNC system.
The number of signaling radio bearers: The LTE has three signaling radio
bearers and 3G-RNC system has four.
Saunder June 7, 2010 19:40 K10322C007
Radio Resource Optimization and Scheduling Techniques 271
MAC entity: Only one MAC entity is defined for LTE, whereas in 3G-RNC
systems there are four different MAC entities based on different types of trans-
port channels, and then less signaling is involved.
As there is no common transport channel defined in LTE, the radio bearer
mapping is much simpler.
No RRC connection mobility is defined in LTE, like cell update and ura
update.
Instead of having two domain identities (CS and PS domains) as in 3G-RNC
systems, only the PS domain identity is specified, with less complexity and
signaling overhead.
As there is only the PS domain in LTE, there is no signaling connection release
procedure.
In LTE, a limited number of most frequently transmitted parameters is in-
cluded in the MIB, and the scheduling information that mainly indicates when
the SI messages are transmitted is contained in the SIB type I, whereas in 3G-
RNC systems, MIB includes both the frequently transmitted parameters and
the scheduling information.
Only one type of paging is required for LTE and two types are required in
3G-RNC systems.
In case of reconfiguration, only one reconfiguration message is used in LTE
to reconfigure all logical, transport, and physical channels, and then fewer
signaling messages are exchanged.
In LTE, the latency of the RRC connection establishment is reduced since no
NBAP protocol is used.
In LTE, there is no need to specify the RRC state in an RRC message.
In LTE, there is no need to define activation time. This leads to a significant
reduction in the latency during establishment and reconfiguration of radio
bearers.
In LTE, only a shared channel is defined, and there is no need to define
the downlink transport channel configuration in the RRC reconfiguration
message. This will reduce the signaling message size effectively. All DL-SCH
transport channel information is broadcast in system information.
7.3 Overview of Packet Scheduling
in HSPA and Beyond
One of the most important features of HSPA is packet scheduling. The main goal
of packet scheduling is to maximize the system throughput while satisfying the
QoS requirements of the users. The packet scheduler determines which user the
shared channel transmission should be assigned to at a given time. In HSDPA, the
packet scheduler can exploit the short-term variations in the radio conditions of
different users by selecting those with favorable instantaneous channel conditions
Saunder June 7, 2010 19:40 K10322C007
272 Evolved Cellular Network Planning and Optimization
for transmission, which is illustrated in Figure 7.2. This idea is based on the fact
that good channel conditions allow for higher data rates (R) by using a higher
order modulation and coding schemes [2], which results in increasing the system
throughput.
In order to quickly obtain up-to-date information on the channel conditions of
different users, the functionality of the packet scheduler has been moved from the
radio network controller (RNC) in UMTS to the medium access control high-speed
(MAC-hs) sublayer at the NodeB [2], as shown in Figure 7.3. The MAC-hs is a new
sublayer that is added to the MAClayer at the NodeB in HSDPA in order to execute
the packet scheduling algorithm. In addition, the time transmission interval (TTI)
(i.e., the time between two convective transmissions) has been reduced from10 ms in
UMTS Release 99 to 2 ms inRelease 5 that includes HSDPA. This is because it allows
the packet scheduler to better exploit the varying channel conditions of different users
in its scheduling decisions and to increase the granularity in the scheduling process.
It should be noted that favoring users with good channel conditions may prevent
those with bad channel conditions from being served and may, therefore, result in
starvation. A good design of a scheduling algorithm should take into account not
only maximization of the systemthroughput through service differentiation, but also
being fair to users who use the same service and pay the same amount of money.
That is, scheduling algorithms should balance the trade-off between maximizing
throughput and fairness.
The packet scheduler for HSDPA implemented at the MAC-hs layer of NodeB
works as follows (see Figure 7.3). Every TTI, each user regularly informs the NodeB
of his channel quality condition by sending a report known as a channel quality
indicator (CQI) in the uplink to the NodeB. The CQI contains information about
the instantaneous channel quality of the user. This information includes the size
UE
3
Served UE
R
i
(t)
Time
UE
1
UE
3
Channel quality
Scheduling
Channel quality
UE
1
UE
2
UE
3
UE
3
UE
2
UE
1
UE
2
UE
1
Channel quality
Figure 7.2 Exploiting the user channel quality for scheduling decisions.
Saunder June 7, 2010 19:40 K10322C007
Radio Resource Optimization and Scheduling Techniques 273
Node-B
UE
RNS
RLC
MAC
PHY
RLC
PHY
MAC-hs
MAC
CN
RNC
Figure 7.3 The MAC-hs at the NodeB in HSDPA.
of the transport block that the NodeB should send to the user, the number of
simultaneous channel codes, and the type of modulation and coding schemes that
the user can support. NodeBthen would select the appropriate mobile user according
to the adopted scheduling discipline and send data to the selected user at the specified
rates. The user is able to measure his current channel conditions by measuring the
power of the received signal fromthe NodeBand then using a set of models described
in [3], determine his current supportable data rates (i.e., the rates that he can receive
data from the NodeB given his current channel condition). Therefore, users with
good channel conditions will enjoy potentially higher supportable data rates by using
higher modulation and coding rates, whereas users with bad channel conditions will
experience lower data rates instead of adjusting their transmission power.
7.4 Overview on Packet Scheduling in LTE
The packet scheduling constitutes one of the RRMfunctions defined for LTE. In this
chapter, the focus will only be on the scheduling and admission control functions.
The radio admission control, the QoS management, and the persistant scheduling
functions are defined for layer 3.
7.4.1 Radio Admission Control
In order to decide about the acceptance of the requests for newevolved packet system
(EPS) bearers in the cell, the admission control algorithm takes into consideration
Saunder June 7, 2010 19:40 K10322C007
274 Evolved Cellular Network Planning and Optimization
several parameters like the resource availability in the cell, the priority level, and the
required QoS by the new EPS bearer, as well as the currently provided QoS to the
active sessions.
This algorithm managing the admission control is not specified by 3GPP but
is specific to each eNB vendor. In general, a new request is only granted if it is
estimated that QoS for the new EPS bearer can be fulfilled, while guaranteeing an
acceptable service to the existing in-progress sessions in the cell having the same or
higher priority.
In LTE, a set of associated QoS parameters defines the QoS profile of each EPS
bearer. This set consists of an allocation retention priority (ARP), an uplink and
downlink guaranteed bit rate (GBR), a prioritized bit rate (PBR), and a QoS class
identifier (QCI) [4, 5]. These parameters belonging to existing bearers could be
modified dynamically, and it is possible to consider simultaneously different services
by activating parallel bearers with different QoS profiles.
The ARP parameter defines the level of priority required for the admission
control decision. It is an integer that ranges between 1 and 16. The GBR parameter
is specified only for EPS GBR bearers. For non-GBR bearers, an aggregate MBR
(AMBR) is specified. PBR is a QoS parameter specified for the uplink per bearer
introduced to avoid the uplink scheduling starvation problem that may occur for
UE with multiple bearers.
As for QCI, 3GPP specifications define a mapping table for nine values and
their corresponding typical services (cf. Table 7.1). This parameter includes other
parameters like the layer 2 packet delay budget, packet loss rate, and scheduling
priority.
To ensure high spectral efficiency in the LTE cell while providing the required
QoS, much more focus should be on dynamic packet scheduling and link adaptation.
The link adaptation is performed to adapt the selection of modulation and
channel coding schemes to current channel conditions on the basis of CQI feedback
from the users in the cell. This leads to the definition of the data rate and the error
probabilities of each link.
In this chapter, only packet scheduling techniques for LTE will be discussed.
7.4.2 Uplink Packet Scheduling
On the uplink, LTE uses an approach called SC-FDMA, which is somewhat similar
to OFDMA but has a 2-to-6-dB peak-to-average ratio of the signal advantage over
the OFDMA.
An uplink packet scheduler has to share the available radio resources between
users while taking into account requirements and limitations imposed by other RRM
functionalities.
The LTE uplink is a constrained link due to single-carrier FDMA transmission.
It limits both frequency and multi-user diversity. Therefore, the packet scheduler
has to fulfill the hard QoS requirements by users having data to transmit over an
Saunder June 7, 2010 19:40 K10322C007
Radio Resource Optimization and Scheduling Techniques 275
Table 7.1 LTE QCI (QoS Class Identier), as Dened by 3GPP TS 23.207
Resource Packet Delay Packet Error
QCI Type Priority Budget Loss Rate Example Services
1 GBR 2 100 ms 10
2
Conversational voice
2 GBR 4 150 ms 10
3
Conversational video
(live streaming)
3 GBR 3 50 ms 10
3
Real-time gaming
4 GBR 5 300 ms 10
6
Non-Convers. video
(buffered streaming)
5 GBR 1 100 ms 10
6
IMS signaling
6 GBR 6 300 ms 10
6
Video (buffered streaming)
7 Non-GBR 7 100 ms 10
3
Voice, live streaming,
interactive gaming
8 Non-GBR 8 300 ms 10
6
``Premium bearer''
for video (buffered)
9 Non-GBR 9 300 ms 10
6
``Default bearer'' for video
interface typically characterized by high interference variability. On the basis of
information conveyed by the buffer status reports (BSRs), the scheduler can handle
the prioritization between these users. Since synchronous HARQ is used for LTE
uplink, this packet scheduler interacts closely with the HARQ manager, and then
user equipment (UE) must be scheduled if an earlier transmission has failed. Its
power capabilities must be considered when, for example, packet scheduler allocates
the uplink transmission bandwidth to a specific UE. As multi-antenna transmission
techniques are used in LTE, enhancing system performance and service capabilities,
the uplink packet scheduler can simultaneously allocate resources to several users.
However, in LTE uplink, users cannot be scheduled for transmission on a physical
uplink shared channel (PUSCH) unless they are listening to the L1/L2 control
channel.
Compared to eNB, these users are low-power devices and consequently they
cannot be allocated a high transmission bandwidth to compensate the effects of
radio environment conditions, especially in macro-cellular network configuration.
7.4.3 Downlink Packet Scheduling
LTE implements OFDM in the downlink. Its basic principle is to split a high-rate
data stream into a number of parallel low-rate data streams, each narrowband signal
Saunder June 7, 2010 19:40 K10322C007
276 Evolved Cellular Network Planning and Optimization
carried by a subcarrier. They are generated in the frequency domain and combined
to form the broadband stream while using the inverse fast Fourier transform (IFFT)
algorithm. To avoid any performance degradation in high-speed conditions, the
subcarriers have a 15-kHz spacing from each other, maintained regardless of the
overall channel bandwidth. In LTE, the number of subcarriers ranges from 75 in a
1.25-MHz channel to 1200 in a 20-MHz channel. Thanks to OFDMA, different
users could be assigned different subcarriers over time. Over bothtime and frequency,
a minimum resource block that the system can assign to a user transmission consists
of 12 subcarriers over 14 symbols, as shown in Figure 7.4.
In the downlink, the dynamic packet scheduler performs scheduling decisions
every TTI and allocates to the users both physical resource blocks (PRBs) and selected
modulation and coding schemes. They are signaled to the scheduled users on the
PDCCH. In LTE, an active user with an EPS bearer has several data flows. It has
a control plane data flow for the RRC protocol and one or multiple user plane
data flows for EPS bearers; each of them are uniquely identified with a 5-bit logical
channel identification (LCID) field. On the basis of the scheduled transport block
size (TBS) for a particular user, the medium access control protocol decides the
amount of data sent from each LCID.
Even though a user has several data flows, the scheduling decisions are carried
out on a per-user basis. As in uplink, the packet scheduler interacts closely with the
HARQmanager since it is responsible for scheduling retransmissions. Asynchronous
adaptive HARQis supported, and then the scheduler dynamically schedules pending
HARQ retransmissions in time and frequency domain. However, it is not allowed
to send at the same time a new and pending HARQ transmission to each scheduler
user.
User 1
User 2
F
r
e
q
u
e
n
c
y
User 3
User 4
Time
Minimum resource block consists of 14 symbols
and 12 subcarriers
Figure 7.4 LTE OFDMA downlink resource assignment in frequency and time.
Saunder June 7, 2010 19:40 K10322C007
Radio Resource Optimization and Scheduling Techniques 277
On the basis of CQI feedback from terminals operating within the LTE cell,
the downlink packet scheduler is informed through the link adaptation about the
supported modulation and coding scheme for a user depending on the selected set
of PRB. As in HSDPA, an outer loop link adaptation algorithm could be applied to
control the block error rate of the first transmissions.
7.4.4 Time and Frequency Domains Packet Scheduling
To improve the LTE systemcapacity, 3GPP has proposed efficient techniques called
time domain packet scheduling (TDPS) and frequency domain packet scheduling
(FDPS). In case of frequency fast fading, the TDPS can provide multiuser diversity
gains that depend on the amount and speed of the fading. In LTE, TDPS gains are
relatively low due to the typically used large bandwidths and both the mobile and
base stations antenna diversity capability. These gains can also be affected by high
mobile speed and multipath propagation. Indeed, on the basis of the CQI feedback,
the packet scheduler selects only the pool of PRB having the highest channel quality
on which the buffered data in the eNB will be transmitted.
With the FDPS technique that exploits frequency selective power variations on
either the useful signal or the interference, users are scheduled on the PRB with
high channel quality. The PRB where users are experiencing deep fades are avoided.
A high FDPS gain is achieved when the effective coherence bandwidth of the radio
channel is less than a system bandwidth equal to or larger than 5 MHz. The main
drawbacks of this technique are high scheduler complexity and increased signaling
overhead in both the uplink and downlink.
7.4.5 Scheduling and Persistent Scheduling
In LTE, transmissions can be carried out with higher data rates while the schedul-
ing of the resources is performed on the basis of the channel conditions knowl-
edge. In addition to dynamic scheduling applied both to the uplink and downlink,
LTE supports persistent scheduling for which radio resources are allocated to a user
for a given set of subframes. This is due to the fact that for services with small
payloads and regular packet arrivals, the control signaling required for dynamic
scheduling might be disproportionately large relative to the amount of user data
transmitted.
7.5 HSPA and LTE Scheduling Techniques Models
In this section, some HSPA scheduling techniques models proposed in literature are
presented. Some of them could be extended to LTE.
Saunder June 7, 2010 19:40 K10322C007
278 Evolved Cellular Network Planning and Optimization
7.5.1 Fair Resource Scheduling Technique Protocol
The maximum bit rate R
max,i
ensured by user i is as follows:
R
max,i
=
_

_
min
_
TBS
i
TTI
delay
,
TBS
i
TTI
delay
n
cap
n
Tot
_
(without codes
multiplexing)
min
_
TBS
i
TTI
delay
,
TBS
i
TTI
delay
n
cap
n
Tot
_
max
_
15

k
i =1
n
i
, 1
_
(with codes
multiplexing)
(7.1)
where n
i
denotes the number of codes referring to the appropriate transport block
size TBS
i
(by AMC) for a given user i within the cell according to the quality of its
link to the NodeB (tables correspondence [3] according to the terminal category),
n
Tot
is the total number of users within the served cell, and n
cap
is the number of
users included in the capacity-limited cell (for which the number of assigned codes
is exactly equal to 15). The summation in Equation 7.1 is applied to the numbers
of codes referring to the users of the served cell, and k is the number of users in the
cell. The n
cap
/n
Tot
ratio is equal to R
cap
2
/r
2
in the case of a uniform traffic (in terms
of user density), where R
cap
is the size of the capacity-limited cell, and r is the size
of the served cell.
Therefore, the bit rate (R
ass
)
FR
guaranteed by the Fair Resource technique
protocol at the cell border is established by replacing in Equation 7.1 the coverage
bit rate R
cov,i
=
TBS
i
TTI
delay
with that at the cell border min
i
_
TBS
i
TTI
delay
_
as follows:
(R
ens
)
FR
=
_

_
min
i
_
TBS
i
TTI
delay
_
min
_
n
cap
n
Tot
, 1
_
(without codes
multiplexing)
min
i
_
TBS
i
TTI
delay
_
min
_
n
cap
n
Tot
, 1
_
max
_
15

k
i =1
n
i
, 1
_
(with codes
multiplexing)
(7.2)
Saunder June 7, 2010 19:40 K10322C007
Radio Resource Optimization and Scheduling Techniques 279
In this context, the expression of (R
ens
)
FR
includes both the coverage bit rate
of the HSPA cell and a term of capacity limitation (number of HSPA codes of the
cell) that is, the minimum between the coverage-limited bit rate and that limited
by capacity without codes multiplexing. If we try to increase the number of users
while keeping the same guaranteed bit rate (R
ens
)
FR
, we should reduce the cell size in
order to be at a higher border TBS
i
: Its the cell breathing phenomenon, such as for
a basic UMTS (WCDMA) network, but ensured in HSPA through adaptation in
modulation and coding (AMC) instead of power control in basic UMTS (Rel 99).
The fact of specifying a minimum(guaranteed) bit rate and the cell size (and thus the
minimum transport block size or the coverage-limited bit rate) limits the maximum
number of users served (subscriber density) with their respective codes n
i
. Inversely,
if the minimum guaranteed bit rate and subscriber density are given, then the cell
size should be well determined (dimensioning). Moreover, at a given cell size, the
ensured bit rate (R
ass
)
FR
is provided by Equation 7.2.
Since Fair Resource protocol tries to share the bandwidth and the available
resources equally among the users while maximizing the cell bit rate versus Fair
Throughput protocol, we can adopt this protocol for non-real-time (NRT) services.
In fact, this method provides a good compromise between the fairness of users of
different services (Web browsing, FTP, etc.) and the maximization of the global
bit rate within the cell. Besides, NRT services dont require a minimum bit rate
(the Fair Resource method doesnt ensure any guaranteed bit rate to the different
users).
7.5.2 Fair Throughput Scheduling Technique
The maximum ensured bit rate by each of the users of a cell by applying the Fair
Throughput scheduling technique protocol (without codes multiplexing) is givenby
the expression (Equation 7.19). We realized in Chapter 6 that the Fair Throughput
protocol isnt, in any case, limited by coverage.
We also provided in Chapter 6 the maximumensured bit rate per user with codes
multiplexing (given by Equation 7.20). We conclude that the maximum ensured
bit rate by applying Fair Throughput (R
ens
)
F T
corresponds exactly to the maximum
number of available codes in HSPA (equal to 15 for the category 10 of mobile
terminals), so by applying this scheduling protocol, the balanced bit rate is always
capacity-limited (limited by the number of codes or physical channels).
Since Fair Throughput protocol tries to offer, if possible, the same bit rate for
all the users, we can set the number of users and the cell size so as to guarantee a
given bit rate. Therefore, this protocol is adapted for real-time (RT) services with
guaranteed bit rates [such as for constant bit rate (CBR) users] more than the Fair
Resource protocol, which doesnt guarantee a given bit rate, especially for users far
away from the NodeB.
Saunder June 7, 2010 19:40 K10322C007
280 Evolved Cellular Network Planning and Optimization
7.5.3 Maximum CIR Scheme (Max C/I)
In Max C/I scheduling, the channel is allocated in each TTI (transmit time interval)
to the user having the best SINR (signal-to-interference-and-noise ratio), in other
words the best channel quality. This scheduler maximizes the cell capacity but does
not guarantee any QoS to the user. Users at the border of the cell always have poor
channel conditions (due to the attenuation, interference, and absence of fast power
control) and experience low bit rate [6].
The user bit rate achieved by this scheduler depends upon the wireless channel
model. In this paragraph, we estimate the cell throughput and user bit rate in the
case of uncorrelated and correlated Rayleigh fading channels [6].
In order to estimate the cell capacity and user bit rate, the probability that the
shared channel is allocated to a given user (e.g., user i ), denoted by Pr(i ), should be
evaluated. Pr(i ) can be written as:
Pr(i ) = Prob(SINR
i
> SINR
j
for j = 1 . . N
u
and j = i )
=
N
u

j =i
Prob(SINR
i
> SINR
j
) (7.3)
where N
u
is the total number of the users within the cell. The expression SINR
i
>
SINR
j
can be expressed as:
_
N
T

l
i
=1
|
l
i
,i
|
2
_
X
i
>
_
_
N
T

l
j
=1
|
l
j
, j
|
2
_
_
X
j
(7.4)
where X
i
is given by Equation 7.5 for user i :
X
i
=
10
bs
i
/
10

l =i
_
P
l
_
d
l
d
i
_

10
bs
l
/10
_
(7.5)
where s
i
corresponds to log-normal shadowing with zero mean and standard
deviation (
2
between 8 and 12 dB). The shadowing loss s is correlated be-
tween the BSs [7]. This effect is usually modeled by considering the shadowing as
a sum (in dB) of a component common to all base stations s
c
and a component s
si
specific to base station i noted BS
i
. The shadowing loss expression is given as:
s
i
= as
c
+bs
si
(7.6)
Saunder June 7, 2010 19:40 K10322C007
Radio Resource Optimization and Scheduling Techniques 281
where a
2
+b
2
= 1. The mean and variances of the log-normal variables are:
_
E(s
i
) = E(s
c
) = E(s
si
) = 0
Var (s
i
) = Var (s
c
) = Var (s
si
) =
2
E(s
si
s
sk
) = 0 if i = k (7.7)

l
i
,i
is the complex path gain between the user i and its serving NodeB, N
T
is the
number of resolvable multipath components, and d
i
is the distance between the user
i and its serving NodeB.
The distribution function of the expression v =
_
N
T
l =1
|
l
|
2
_
X
i
can be approx-
imated by:
pdf (v) =
N
T

l =1
(
l
)
N
T
2

N
T
r =l
(
l

r
)

2
f
v
e

(10 log v10 log

l

f
)
2
2
2
f
(7.8)
where
l
= E(|
l
|
2
),
f
= C +
X
, and
2
f
=
2
(2.2) +
2
X
. C = 0.5772 is
the Euler constant and (2.2) =
2
/6 is the Riemann-Zeta function.
The probability Prob(SIR
i
> SIR
j
) is then given by:
Prob(SIR
i
> SIR
j
) =
N
T

l
i
=1
N
T

l
j
=1
(
l
i
,i
)
N
T
2

N
T
r
i
=l
i
(
l
i
,i

r
i
,i
)
(
l
j
, j
)
N
T
2

N
T
r
j
=l
j
(
l
j
, j

r
j
, j
)

_
_
Q
_
_

(10 log
l
i
,i
+
f,i
10 log
l
j
, j

f, j
)
_

2
f,i
+
2
f, j
_
_
_
_
(7.9)
Consequently, the bit rate of user i is given by:
R
i
= Pr(i )

CQI
R
CQI
p
CQI ,i
N
s
= Pr(i )

CQI
TBS
CQI
TTI
delay
p
CQI ,i
N
s
(7.10)
where R
CQI
is the instantaneous bit rate referring to CQI (channel quality indicator),
and p
CQI ,i
is the (discrete) probability given by Equation 7.37 referring to CQI in the
position of user i , N
s
is the number of corresponding HARQtransmissions, TBS
CQI
is the transport block size referring to CQI according to tables correspondence [3]
versus terminal category, and TTI
delay
is the transmit time interval [equal to 2 ms in
the case of HSDPA (high-speed downlink packet access)].
Saunder June 7, 2010 19:40 K10322C007
282 Evolved Cellular Network Planning and Optimization
In the case of correlated Rayleigh fading, the user bit rate can be estimated using
the same method described earlier (i.e., in the case of uncorrelated Rayleigh fading).
Consequently, the probability that the channel is allocated to user i is given by:
Pr (i ) = Prob(SINR
i
> SINR
j
for j = 1 . . N
u
and j = i )
=
N
u

j =i
Prob(SINR
i
> SINR
j
)
=
N
u

j =i
1

N
T
l
i
=1

l
i
,i
1

N
T
l
j
=1

l
j
, j
N
T

l
i
=1
N
T

l
j
=1
1

N
T
r
i
=l
i
_
1

r
i
,i

l
i
,i
_
1

N
T
r
j
=l
j
_
1

r
j
, j

l
j
, j
_

_
_
Q
_
_

(10 log
l
i
,i
+
f,i
10 log
l
j
, j

f, j
)
_

2
f,i
+
2
f, j
_
_
_
_
(7.11)
where
l
, l = 1, . . . , N
T
are the eigenvalues of the matrix DC. D and C are the
N
T
N
T
path power and covariance matrices given by:
D =
_
_
_
_
_
_
_
_
_
_

1
0 0
0
2
0


0 0
N
T
_
_
_
_
_
_
_
_
_
_
(7.12)
C =
_
_
_
_
_
_
_
_
_
_
1

12

1N
T

21
1

2N
T

N
T
1

N
T
2
1
_
_
_
_
_
_
_
_
_
_
(7.13)
where
l ,l
is the correlation parameter between the paths l and l .
Saunder June 7, 2010 19:40 K10322C007
Radio Resource Optimization and Scheduling Techniques 283
Consequently, the bit rate of user i is given by Equation 7.10 where Pr(i ) is
given by Equation 7.11.
7.5.4 Fair Channel-Dependent Scheduling
(FCDS) Protocol
The fair channel-dependent scheduling (FCDS) protocol introducedin[810] forms
a trade-off between the two other extremes: low power use (system capacity) and
fairness. In practice, the signal is fluctuating around a mean value that displays slow
trends as well. This underlying slow fluctuation accounts for the distance from the
base station.
The time scale of the so-called fading variations in the signal itself, due to multi-
path reception and/or shadow fading, is much smaller than that of the variations
of local mean. The scheduling is done on the basis of the relative power (i.e., the
instantaneous power relative to its ownrecent history). So, the transmissionlevel of all
mobile terminals is first translated with respect to their local means, and subsequently
normalized with their local standard deviations. A transmission is scheduled to the
UE that has the lowest value for the relative power.
The idea of a relative power, introduced earlier, needs the definition of a local
mean, with local referring to the recent time history. Exponential smoothing weighs
past observations with exponentially decreasing weights in order to update the value
for the local mean. It takes the local mean of the previous period and adjusts it up or
down based on what actually occurred in that period. By the choice of a weighting
factor, this procedure can be made sensitive to a small or gradual drift in the process.
This method is simple and therefore has low data storage requirements and data
processing since only the actual (instantaneous) value and the old local mean value
are needed to update the new local mean value. Comparing with, for example,
moving averaging, the low storage and higher weights on more recent samples are
two properties in favor of the FCDS method. The performance of these algorithms
is considered with the parameters presented in Table 7.2.
Note that t either refers to the physical time unit or the corresponding integer
index.
The local mean, introduced earlier, as well as the variance, are updated with each
time unit according to the following algorithm [11]:
_

t
=
1
P
t
+(1
1
)
t1
v
t
=
2
(P
t

t
)
2
+(1
2
) v
t1
(7.14)
In other words, the new local mean (or variance) is a weighted average of the
instantaneous contribution and the old mean (or variance). In the rest of this study,
the parameter refers to both
1
and
2
when not specified any further. As we
Saunder June 7, 2010 19:40 K10322C007
284 Evolved Cellular Network Planning and Optimization
Table 7.2 Variables Used in FCDS
P
t
(Instantaneous) transmission power at time t

t
Local mean of P
t
based on the time interval [t
0
,t]
v
t
Local variance of P
t
based on the time interval [t
0
,t]

t
Local standard deviation, defined by
2
t
= v
t

1
Smoothing coefficient w.r.t. the local mean

2
Smoothing coefficient w.r.t. the local variance
will see later, extreme values of (0, respectively 1) lead to the extreme variants in
scheduling: C/I-based scheduling, respectively Round Robin scheduling.
The criterion that determines the optimal mobile node for the next downlink
transmission at time t is formulated as follows, where the superscript i is used to
denote the situation at node i :
min
i
__
P
i
t

i
t
_
/
_
v
i
t
_
(7.15)
Here, the data point, P
t
, is translated with respect to the local mean,
t
, and is next
normalized with the corresponding local standard deviation,
t
. This subsequently
translated and normalized transmission power is referred to as the scaled power, P
s
.
At each time t, the value for P
s
is compared for all nodes i and the most optimal
node is selected for the downlink transmission.
7.5.5 Score-based Scheduling
The score-based (SB) scheduler, proposed by Bonald in [11], consists of allocating
the channel to the user having the maximumtransmission rate relative to its past rate
statistics [6]. This algorithm can be explained as follows: Let us consider a HSPA
system with two active users. Let r
1, v
and r
2, v
where v = 1, . . . , n be the past
transmission rates for each user (even when the TTI is not attributed to this user)
observed over a window size n. The idea is to classify the past rates of each user in
decreasing order and to give a rank for each rate (e.g., rank 1 for the highest rate).
During the TTI n + 1, if the possible rate of user 1 r
1, n+1
is classified in rank 1
relative to his own rate statistics and the rate of user 2 r
2, n+1
is classified in rank 3
relative to his own rate statistics, in this case the channel is allocated to user 1 even if
r
2, n+1
> r
1, n+1
. This algorithm has the advantage of not suffering from asymmetric
fading and data rate constraints, which is not the case with the proportional fair
algorithm.
Saunder June 7, 2010 19:40 K10322C007
Radio Resource Optimization and Scheduling Techniques 285
7.6 New Optimized Scheduling Techniques
for Multiple Services Case
By using multiple services having different characteristics (some requiring a guar-
anteed bit rate such as streaming, voice, and so on at the opposite of interactive
services), new scheduling protocols should be used. The following scheduling meth-
ods adapted for the multiple services case were first introduced by A. Masmoudi et al.
[12] and Nasser and Bejaoui [13], and implemented in NS-2 Eurane simulators and
tested with multiple services. For some of them, the cross-layer design is applied to
optimize the use of bandwidth and improve their performance.
7.6.1 Concept of Cross-Layer Design
In next-generation wireless networks, the mechanisms and protocols at the different
layers in the protocol stack will interact dynamically with each other to provide guar-
anteed services. This is central to cross-layer design. The cross-layer design concept
(cf. Figure 7.5) does away with the rigid structure of the layered protocol architec-
ture that has served extremely well in the development and implementation of both
past and current communication systems. Such architectures, for which each layer
is responsible to serve only the higher layer, had the advantage to exhibit a high de-
gree of modularity, which allows an easy replacement and theoretically an arbitrary
combination of protocols.
In cross-layer architecture, parameters have to be exchanged interactively to over-
come the robust characteristics and constraints of multimedia traffic and wireless
Application layer
(QoS requirements)
Transport layer
(Receiver window size)
Network layer
(Topology information)
MAC layer
(Bit error rate)
Physical layer
(Signal strength)
Cross-layer interactions
Protocol interactions
C
r
o
s
s
-
l
a
y
e
r

c
o
n
t
r
o
l

p
l
a
n
e

Figure 7.5 The general concept of cross-layer design.
Saunder June 7, 2010 19:40 K10322C007
286 Evolved Cellular Network Planning and Optimization
networks. Thus, active exploration of the various synergies of exchanging informa-
tion between lower layers and the upper layer is emphasized. Their main purpose is to
help improve the end-to-end performance given network resources by meeting high
data rates, higher performance gains, and QoS requirements for various real-time
and non-real-time applications.
However, to preserve the layered structure as much as possible, the interlayer de-
pendencies introduced by the cross-layer design should be kept to a minimum. Thus,
cross-layering should be viewed as an enhancement or a complement, not an alter-
native, to layered design. Its ultimate goal is then to preserve the key characteristics
of a layered architecture and to allow for performance improvements.
In practice, the purpose of using the cross-layer design is either to optimize the
use of bandwidth through RRM algorithms in cellular networks, as in the wireless
local area networks (WLANs), or to enhance the path selection process and achieve
minimum energy consumption in ad-hoc and wireless sensor networks.
For HSPA, the cross-layer design is applied to improve the performance of
scheduling techniques and to achieve bit-level, packet-level, and call-level QoS. To
achieve a superior performance, the scheduling and control mechanisms can benefit
from information coming from different layers as from the PHY one about the state
of the wireless channel.
7.6.2 Description of the Introduced Techniques (Protocols)
7.6.2.1 Prioritized Differentiated Services Scheduling
This scheduling method differentiates between services requiring guaranteed bit rate
(such as CBR services, video streaming, voice, etc.) and non-guaranteed bit rate ones
(such as interactive Web browsing, FTP, email, etc.). It introduces a parameter spec-
ifying the maximum priority degree for which service is considered as a guaranteed
bit rate and are scheduled at a Fair Throughput manner, while the remaining others
(non-guaranteed bit rate flows) are scheduled at Fair Time fashion. Priority degrees
among all services are taken into account.
7.6.2.2 Prioritized Rayleigh Peak Scheduling
This method attempts to schedule guaranteed bit rate services first only in Rayleigh
peak instances [i.e., if their quality is good enough or, in terms of implementation,
provided that their CQI is greater than or equal to a specified CQI threshold value).
In this case, Fair Throughput scheduling is applied to guaranteed bit rate flows
having an acceptable CQI value. Otherwise, if CQI is less than this CQI threshold,
flows are treated at Fair Resource scheduling.
The basic Rayleigh peak scheduling protocol as simulated has the same concept
as the Proportional Fair scheduling (Proportional Fair Throughput or Proportional
Fair Resource) [5] except by applying another method consisting of a specified CQI
threshold to schedule and serve mobiles using the guaranteed bit rate services so as to
Saunder June 7, 2010 19:40 K10322C007
Radio Resource Optimization and Scheduling Techniques 287
disadvantage the packets whose channel is suffering fromdeep Rayleigh peak fading.
In prioritized Rayleigh peak scheduling, we apply the same principle as the Rayleigh
peak scheduling while complying with the different services differentiation priority
degrees.
7.6.2.3 Weighted Differentiated Services Scheduling
It is similar to the prioritized differentiated services scheduling except that instead
of taking an absolute priority degree for the guaranteed bit rate services, and in
order to increase the resources reserved to non-guaranteed bit rate services, this
method assigns weights both to guaranteed and non-guaranteed bit rate services to
balance the bandwidth dedicated for each of them being scheduled respectively in
Fair Throughput and Fair Resource manners.
This method thus consists in differentiating guaranteed bit rate services to non-
guaranteed bit rate ones through two weight coefficients for each one in order
to keep a fixed proportion of radio resources guaranteed for NRT services often
disadvantaged in terms of priority degrees in both previous protocols.
7.6.2.4 Weighted Rayleigh Peak Scheduling
It has the same principle as the previous method but adds the condition for CQI to
be above a pre-determined threshold (at Rayleigh peaks) for guaranteed bit rate flows
before deciding to serve them according to Fair Throughput scheduling. The aim
of this method is to avoid wasting resources and dedicating them to bad TTI links
(especially at Fair Throughput). If CQI is less than the threshold, non-guaranteed
bit rate flows are scheduled at Fair Resource fashion.
Hence, it is a hybrid method (between Fair Throughput and Fair Resource)
according to the service type, while complying with a cyclic weighted allocation
between RT and NRT services. It has the same basis as the weighted differentiated
services scheduling while applying the CQI threshold rule as defined in the Rayleigh
peak scheduling.
7.6.2.5 ThreshOld-based Priority (TOP) Scheduling Algorithm
Only a few cross-layered resource allocation protocols have been proposed for the
HSPA system. The most pioneering of such cross-layered protocols was developed
by Nasser and Bejaoui [13]. This scheduling protocol that aims at enhancing the
average cell throughput is called threshold-based priority (TOP) and serves users
to access packets in the downlink. It is proposed to provide priority scheduling
between non-real-time services of different QoS classes and fairness between users
within the same class. Within each queue allowed respectively to the non-real-time
interactive (class-1) and background (class-2) traffic classes, packet data units (PDUs)
are supposed to be waiting for service in a first-in-first-out manner, and the service
discipline is considered to be preemptive resume. According to the TOP scheme, the
Saunder June 7, 2010 19:40 K10322C007
288 Evolved Cellular Network Planning and Optimization
service is made by considering two steps. The first one is dedicated to the selection
of a class to be served. Thus, after a specific user is served, the next user to be selected
for service is a class-2 user if the number of PDUs of class-2 in buffer C denoted m
2
is greater than the threshold T
2
and m
1
of those of class-1, less than the threshold T
1
.
Otherwise, service is allocated to a class-1 user if present in the system. The second
step is for the selection of which user of the selected class to be served. In the selected
class, the user with the highest priority is selected for transmission where the priority
for user i at time t is calculated as follows:
P
i
(t) =
_
CQI if S
i
(t) R
CQI W otherwise
(7.16)
where CQI is the channel quality indicator for user i that represents the current
channel condition for this user, S
i
(t) average throughput for user i up to time
t, R is the predefined minimum throughput (e.g., 64 kbps) and W = R/S
i
(t).
TOP prioritizes users based on their radio channel quality. However, it increases the
priority of those with average throughput below a certain threshold by W and hence
increases their chance of getting served.
The CQIused to determine the rate at which the user can support from the
NodeBis mappedusing the signal-to-noise ratio (SNR)

according to the following

equation [3]:
CQI =
_

_
0 SNR 16
_
SNR
1, 02
+16, 62
_
16 < SNR < 14
30 14 SNR
(7.17)
The choice of the channel model to be considered is important in evaluating
the effectiveness of the protocol. It describes how much the radio signal attenuates
on its way from the NodeB to the user and therefore describes how the channel
condition of the user changes with time depending on the environment of the user
and his speed. The propagation model used in the performance evaluation of TOP
consists of five parts: path loss, shadowing, multipath fading, intra-cell interference,
and inter-cell interference. The path loss is calculated as follows:
L(d ) = 137, 4 +10 log
10
(d ) (7.18)
where d is the distance from the UE to the NodeB in kilometers, is the path loss
exponent and is equal to 3.52. Shadowing is modeled through a log-normal distri-
bution and a correlation distance with a mean value of 0 dB. The multipath fading

Signal-to-noise ratio (SNR) is the signal strength relative to background noise.

Saunder June 7, 2010 19:40 K10322C007
Radio Resource Optimization and Scheduling Techniques 289
corresponds to 3GPP channel models for pedestrian and vehicle A environments.
The intra-cell and inter-cell interferences are assumed to be constants and are set
equal to 30 dBm and 70 dBm, respectively. Then at the user side, the SNR is
extracted from the received signal from the NodeB to determine how strong the
signal is according to the following formula:
SNR = P
x
L
Total
10 log
10
_
10
I
intra
L
Total
10
+10
I
inter
10
_
= P
x
10 log
10
_
10
I
intra
10
+10
I
inter+
L
Total
10
_
(7.19)
where P
x
is the transmitted code power in dBm, L
Total
is the sum of the path loss,
shadowing, and multipath fading in dB, I
intra
and I
inter
are the inter-cell and intra-cell
interferences, respectively, in dBm.
7.6.3 Scheduling Techniques Optimization According Services
Proles and Requirements
7.6.3.1 Prioritized Differentiated Services Scheduling Technique
Consider two services 1 and 2 with respective nominal guaranteed bit rates R
min
1
and R
min
2
and with respective served number of users N
1
(referring to transport
block sizes TBS
i
and to number of codes n
i
) and N
2
(referring to transport block
sizes TBS
i
and to number of codes n

i
) since the cell size is already fixed. Some
of the TBS
i
s and the TBS
i
s can be equal if the traffic is not uniform. Service 1 is
assumed to have the priority over service 2, and each of the services thus uses the
Fair Throughput. Assuming that both nominal services bit rate R
min
1
and R
min
2
are
CBR type, having the highest priority degrees versus the other services types, using
the Fair Throughput protocol, and assuming N
1
and N
2
as their respective number
of users, then the number of available codes n

remaining for the CBR service with

priority degree 2 is:
n

= max
_
0 ; 15 R
min
1
TTI
delay

N
1

j =1
n
j
TBS
j
_
(7.20)
where n
j
is the number of codes referringaccording to Table 6.1 from 3GPP
standard [3]to the CQI of user j with service 1 (in the cell), and whose appropriate
transport block size is TBS
j
. The expression (Equation 7.20) is valid with or without
codes multiplexing.
The maximumbit rate ensured by multiplexing all the available HSPA codes (15
codes with spreading factor 16) is that given by Equation 7.20 by applying the Fair
Throughput protocol only to the users of the CBR service with priority degree 1.
Saunder June 7, 2010 19:40 K10322C007
290 Evolved Cellular Network Planning and Optimization
The bit rate R
min
1
isnt reached by the users except if it doesnt exceed the maximum
bit rate. However, if
R
min
1

15
TTI
delay

N
1
j =1
n
j
TBS
j
,
then the number n

of the remaining codes for the service 2 is null, and the guaranteed
bit rate for the CBR service with priority degree 1 is lower than the required bit rate
R
min
1
(while using all the 15 available codes). So we should reduce the cell size until
having a minimum bit rate equal to R
min
1
.
For the service with priority degree 2, the maximum ensured bit rate per user
(by using the n

remaining codes out of the radio resources) can be determined like

in Equation 6.20 according to:
R

ens
=
n

TTI
delay

N
2
j =1
n

j
TBS

j
(7.21)
where n

j
and TBS

j
are respectively the number of codes and the transport block size
referring, according to the terminal category (from 3GPP standard [3], cf. Table 7.1
for terminal category 10) to the CQI of user j with service 2.
The number of available codes remaining for NRT services can be determined
with the same manner as in Equation 7.20 as follows:
n

= max
_
0 ; n

R
min
2
TTI
delay

N
2

j =1
n

j
TBS

j
_
(7.22)
with the same notations for n

j
and TBS

j
as in Equation 7.20 while considering the
users of service with priority degree 2.
By generalizing, the bit rate per user i of the NRT services (the remaining
resources and bandwidth being equally shared among the users) is provided, by
applying the Fair Resource scheduling protocol, as in Equation 7.1, by:
R

i
=
_

_
min
_
TBS

i
TTI
delay
,
TBS

i
TTI
delay
n

cap
n

Tot
_
(without codes
multiplexing)
min
_
TBS

i
TTI
delay
,
TBS

i
TTI
delay
n

cap
n

Tot
_
max
_
n

i
n

i
, 1
_
(with codes
multiplexing)
(7.23)
Saunder June 7, 2010 19:40 K10322C007
Radio Resource Optimization and Scheduling Techniques 291
where n

i
and TBS

i
are respectively the number of codes and the transport block size
referring to the CQI of the NRT user i according to the adequate terminal category
from the standard [3] (cf. Table 7.1 for category 10), n

Tot
is the total number of
NRT users in the served cell, and n

cap
is the number of users included in the cell
whose size is capacity-limited (the number of codes n

allocated to it is exactly equal

to that remaining for NRT services).
So the guaranteed bit rate R

ens
for the users of interactive services is that present
at the cell border (the smallest TBS
i
size). It is given similarly to Equation 7.2 by:
R

ens
=
_

_
min
i
_
TBS

i
TTI
delay
_
min
_
n

cap
n

Tot
, 1
_
(without codes
multiplexing)
min
i
_
TBS

i
TTI
delay
_
min
_
n

cap
n

Tot
, 1
_
max
_
n

i
n

i
, 1
_
(with codes
multiplexing)
(7.24)
R

ens
refers to the guaranteed bit rate at the cell border. The summation in each of
the expressions (Equations 7.23 and 7.24) applies to users of all the NRT services in
the cell.
Therefore, the fact of specifying a minimum bit rate for all the users limits the
maximum cell size (or the allowed maximum path loss), thus the coverage of NRT
services in HSPA translates the fact that the bit rate in each point of the cell is above
a threshold, whereas their capacity is according to the Best Effort policy (with a
fairness among the users according to the quality of their link and the number of
HSPA codes available for NRT services).
7.6.3.2 Weighted Differentiated Services Scheduling Technique:
Optimization According to Services Profiles
and Requirements
The objective of this paragraph is to optimize the weighting coefficients of the
Weighted Differentiated Services Scheduling protocol (called also Weighted
Round Robin or WRR) in order to adapt it according to the different profiles
of multiple services (NRT or CBR, etc.).
Some NRT services consume much radio resources often at the expense of other
services in the multiservice case. This is mainly due to their high edge bit rate
occupying the channel if only for a short time (such as the FTP service which hardly
leaves enough bandwidth for the other services with lower edge bit rate such as the
Web Browsing service). The solution is to balance the packets scheduling versus the
edge bit rate and the load of each service such that the users have an equal probability
Saunder June 7, 2010 19:40 K10322C007
292 Evolved Cellular Network Planning and Optimization
to serve the minimum bit rate of each service. This solution consists in assigning
to the user of a given service i a weighting coefficient w
(i )
1
taking into account the
effective target load (the minimum) versus the edge bit rate [14]. The expression of
this weight coefficient is given by the following equation:
w
(i )
1
=

(i )
R
(i )
min
R
(i )
max

s
k=1

(k)
R
(k)
min
R
(k)
max
(7.25)
where s is the number of NRT services with respective edge bit rates R
(2.1)
max
,
R
(2.2)
max
, . . . , R
(s)
max
(maximum source bit rates), R
(2.1)
min
, R
(2.2)
min
, . . . , R
(s )
min
are the re-
spective required minimum bit rates, and
(i )
is the planned or effective number of
users of service i (per area unit). Hence the service weight takes into account the
real need of this service while avoiding the bad impact of the high edge bit rate
services considered as great consumers of radio resources. This weight coefficient
selected according to the Best Effort strategy to balance the load of the different
servicesis suitable if all the services are NRT or interactive. Weighting the services
differently is the basis of this scheduling protocol.
On the other hand, if services are all RT (or having a required guaranteed bit
rate for their QoS), we have seen that the Fair Throughput protocol is the most
adapted in this case since it tries to guarantee a constant bit rate for all the users.
This protocol is equivalent to assigning high weights to the users far fromthe NodeB
or having the worst CQI values and weights lower than the most advantaged ones
in terms of quality and position. More formally, and according to Section 7.5.2
earlier about Fair Throughput protocol, it is as if we use the weighted differentiated
services scheduling protocol with a weight associated to each of the users inversely
proportional to its referring transport block size TBS
i
(i.e., the weight w
(i )
21
associated
to the user with block size TBS
i
is:
w
(i )
21
=
1
TBS
i

N
k=1
1
TBS
k
(7.26)
where N is the number of mobiles using HSPA within the cell, and TBS
k
is the
block size referring to user k. The expression (Equation 7.26) is valid if the bit rate
required by the RT service is above the minimum bit rate guaranteed by the Fair
Throughput for each user independently of the available number of codes (given by
Equation 7.19). In contrast, if the required bit rate is above it, then all the users will
be satisfied without time multiplexing according to the Fair Throughput protocol
or another technique.
Yet, the different types of RT services shouldnt have the same priority degrees.
In fact, the voice service for example, should have more priority than streaming
services due to the importance of this service and in order to maximize the number
Saunder June 7, 2010 19:40 K10322C007
Radio Resource Optimization and Scheduling Techniques 293
of served users. Moreover, if we choose to optimize the weights assigned to RT
services so as to maximize the number of served users, we fall on exhaustive (hard)
priority degrees avoiding the services requiring a more guaranteed bit rate. Otherwise,
the resources wont be assigned to these last services (like the streaming) except if
all the more priority services (such as the voice) are satisfied and if HSPA resources
are available. This exhaustive priority may not satisfy low priority services or not
provide them the required bit rates. For that, we give non-exhaustive weights for
the priority degrees in order to obtain a compromise between the priorities required
by different services and the satisfaction of users from different services profiles. As
an example, we can choose uniform weights as follows: Given s RT services with
decreasing priority degrees 1, 2, . . . , s , then the weight w
(i )
22
associated to service
with priority degree i is:
w
(i )
22
=
s +1 i
_
s (s +1)
2
_ =
2(s +1 i )
s (s +1)
(7.27)
So the global weight w
(i )
2
(RT service) will be:
w
(i )
2
=
w
(i )
21
w
(i )
22

s
k=1
w
(k)
21
w
(k)
22
(7.28)
Moreover, lets nowexamine the case of aggregated services with different natures
(at guaranteed and non-guaranteed bit rates: NRT, CBR, etc.). In fact, if there are
together s
1
RT services and s
2
NRT services, the more natural is to assign the RT
services (at guaranteed required bit rates) priorities higher than NRT ones by giving
favor to those whose required bit rate is the lowest (such as voice). For NRT services,
interactive services are assigned priorities higher than those in background, then
these priority degrees are converted into weights w
(i )
22
according to a form function
(uniform as in Equation 7.27 or other law). For weighting coefficients w
(i )
1
, they
are calculated as for RT services in Equation 7.25 except by including also in the
summation of the denominator the terms of RT services. For these latter (e.g., CBR
services), the traffic source transmit packets at the same cadence, and the nominal
required bit rate is equal to the maximum edge bit rate from the source. Hence the
coefficients w
(i )
1
are obtained by Equation 7.25 by assuming the equal bit rates R
(i )
min
and R
(i )
max
of RT services. Consequently, in the multiservice case (RT and NRT), the
global weight coefficient w
(i )
associated to a NRT service i will be:
w
(i )
=
w
(i )
1
w
(i )
21
w
(i )
22

s
1
+s
2
k=1
w
(k)
1
w
(k)
21
w
(k)
22
(7.29)
where w
(i )
22
=
1
s
2
for all NRT services since we decided to apply the Fair Resource
protocol for this type of service. In contrast, for RT services, we will also take into
Saunder June 7, 2010 19:40 K10322C007
294 Evolved Cellular Network Planning and Optimization
account coefficients w
(i )
21
from Equation 7.26 applied only to RT users in order to
obtainas possibleequal bit rates for all these users not exceeding the source bit
rate (partial Fair Throughput), so the global coefficient associated to a RT service
i will be given by Equation 7.29 except by assuming coefficients w
(i )
21
according to
Equation 7.26.
For the global weighting coefficient of RT services users (CBR at guaranteed
bit rates), we can replace the coefficient w
(i )
22
referring to priority degrees (given by
Equation 7.27) by a coefficient proportional to the required bit rate of the concerned
CBR service, as follows:
w
(i )
22
=
R
(i )
min

s
1
k=1
R
(k)
min
(7.30)
where R
(i )
min
is the required bit rate of service i , and s
1
is the number of CBRservices at
guaranteed bit rates. We can also define w
(i )
22
as inversely proportional to the required
bit rate of the concerned CBR service i .
The coefficient w
(i )
1
can be ignored (taken equal to 1) in the case of guaranteed
bit rate services since the load or the area density of users is already included in
the coefficient expression w
(i )
21
(according to Equation 7.26) referring to the bit rate
balancing according to Fair Throughput protocol. For the users of NRT services, we
can either take into account the priority degrees of different services in the coefficient
w
(i )
22
or introduce some fairness among the users (elimination of priority levels): the
global coefficient w
(i )
becomes thus, in this case, equal to w
(i )
1
. The expression of the
global coefficient w
(i )
is written therefore according to Equation 7.29, except with
the sub-coefficients in Table 7.3. In the notations of Table 7.3, two different users
of the same service i have the same weight coefficients.
The combination of the sub-coefficients according to Equation 7.29 has the
disadvantage of giving the same importance to the different factors (materialized by
Table 7.3 Values of Weighting Sub-coefcients for Multiple Services
(Different Proles)
Service of User i Is at
Guaranteed Bit Rate Service of User i Is at Non-guaranteed Bit Rate
w
(i )
1
Constant equal to 1 Equation 7.25
w
(i )
21
Equation 7.26 Constant equal to 1/s
2
w
(i )
22
Equation 7.30
_

_
Constant equal to 1/s
2
if fairness among users
Equation 7.27 else (taking priority degrees)
Saunder June 7, 2010 19:40 K10322C007
Radio Resource Optimization and Scheduling Techniques 295
the sub-coefficients) so as to result in the global weight coefficient. In order to take
into account the weight dedicated to each factor (load, priority, etc.), we can rewrite
the global weight coefficient linearly as follows (instead of Equation 7.29):
w
(i )
=
1
w
(i )
1
+
21
w
(i )
21
+
22
w
(i )
22
(7.31)
where
1
,
21
, and
22
are the respective weights referring to sub-coefficients w
(i )
1
,
w
(i )
21
, and w
(i )
22
. Thus, we can give more importance to any of the factors influencing
services differentiation by decreasing the weights referring to negligible criteria, for
example, in order not to take into account the priorities within the coefficient w
(i )
22
referring to the users of non-guaranteed bit rate services, we can take a null coefficient

22
, etc.
In the case of the WRRprotocol, to determine the maximumbit rate R
ass
ensured
for each of the WRR users independently of the number of available HSPA physical
channels, we proceed similar to the Fair Throughput protocol. In order to determine
the maximumbit rate ensured per user in WRR (by using the totality of the available
HSPA codes for this scheduling protocol, WRR is studied in this paragraph), we
find similarly as in the Fair Throughput protocol that:
(R
ens
)
WRR
=
15 TBS
i
w
(i )
TTI
delay

j
n
j
w
( j )
(7.32)
Note that the index j in the summation of Equation 7.32 refers to the user and
not to the service as for the other notations of weight coefficient indexes in this
paragraph. Thus, two different users i and j of the same service should have the
same weights (w
(i )
= w
( j )
).
7.7 Summary and Open Problems
HSPA and LTE have been introduced in the new UMTS standard to provide high
data rate services in wireless cellular networks. In this chapter, some scheduling
protocols of literature used in HSPA- and LTE-based networks are presented. Since
aggregation of multiple services is so important, this chapter also provides some
optimized scheduling techniques adapted for the multiservice case as well as an
overview of the optimization of the cross-layer architecture design.
As future problems are studied, comparison between scheduling protocols should
be made onthe basis of whichone(s) are better for eachservice profile or requirements.
Performance can be evaluated in terms of throughput, delay, and retransmissions.
The impact of intra-cellular interference and multiuser diversity on the total HSPA
capacity can also be studied and examined as open issues.
Saunder June 7, 2010 19:40 K10322C007
296 Evolved Cellular Network Planning and Optimization
References
[1] 3GPP TS 36300 900, Evolved Universal Terrestrial radio Access (E-UTRA) and Evolved
Universal Terrestrial Radio Access Network (E-UTRAN); Overall description, stage 2,
Release 9 (V9.0.0), June 2009.
[2] 3GPP TS 25.308, High Speed Downlink Packet Access (HSDPA); Overall Description,
Release 5, March 2003.
[3] T. E. Kolding, K. Pedersen, J. Wigard, F. Frederiksen, and P. E. Mogensen, Performance
Aspects of WCDMA Systems with High Speed Downlink Packet Access (HSDPA), VTC,
vol. 1, pp. 477481, September 2002, Online: http://nds2.ir.nokia.com/downloads/.
[4] 3GPP TS 25.214, Physical Layer Procedures (FDD), 3GPP Release 6, December 2003.
[5] M. Assaad, Cross Layer Study in HSDPA System, Ph.D. thesis, Ecole Nationale des
T el ecommunications, Paris, France, March 2006.
[6] K. S. Gilhousen, I. M. Jacobs, R. Padovani, A. J. Viterbi, L. A. Weaver, and C. E. Wheatley,
On the Capacity of a Cellular CDMA System, Vehicular Technology, IEEE Transactions
on, vol. 40 Issue: 2, May 1991.
[7] I.C.C. de Bruin, G. Heijenk, M. El Zarki, and J. Lei Zan, Fair Channel-Dependent
Scheduling inCDMASystems, 12
th
ISTSummit onMobile andWireless Communications
Summit 2003, Aveiro, Portugal, June 1518, 2003, pp. 737741.
[8] L. Zan, G. Heijenk, and M. El Zarki, Fair and Power-Efficient Channel-
Dependent Scheduling for CDMA Packet Networks, Proceedings of International Con-
ference on Wireless Networks (ICWN), June 2003, URL: http://www.home.cs.utwente.
nl/heijenk/.
[9] L. Zan, G. Heijenk, and M. El Zarki, AReal-Time Traffic Scheduling AlgorithminCDMA
Packet Networks, Proceedings of 14th IEEE International Symposium on Personal, Indoor,
and Mobile Radio Communications PIMRC 2003, September 2003, Beijing, China.
[10] S. W. Roberts, Control Chart Test Based on Geometric Moving Averages, Technometrics,
1, 1959, pp. 239250.
[11] Thomas bonald, A Score-based Opportunistic Scheduler for Fading Radio Channels,
Proceedings of European Wireless, 2004.
[12] A. Masmoudi, D. Zeghlache, and S. Tabbane, Resource and Scheduling Optimization in
HSDPA-based UMTS Networks, Proceedings of World Wireless Congress (WWC2005), San
Francisco, California, May, 2427, 2005.
[13] N. Nasser and T. Bejaoui, User Satisfaction-based Scheduling Algorithm for High-Speed
Wireless Networks, ACMInternational Wireless Communications andMobile Computing
Conference (IWCMC), Honolulu, Hawaii, August 2007, pp. 164169.
[14] T. Bonald and A. Prouti` ere, Wireless Downlink Data Channels: User Performance and
Cell Dimensioning, MobiCom03, San Diego, California, September 1419, 20.