Bachelor ThesisHSDPA CQI Mapping Optimization

Based on Real Network Layouts

Supervisor: Prof. Dr.-Ing. Markus Rupp

Assistant: Dipl.-Ing. Martin Wrulich

by

Marıa Elsa Feliz Fernandez

Institute of Communications and Radio-Frequency Engineering

November 2007 - October 2008

Acknowledgements

I would like to thank all those people who have somehow supported me dur-ing my work in this Bachelor Thesis and also during all my career in Madridand Vienna. First I want to express my gratitude to Professor Markus Ruppfor his supervision and warm wellcome, and also to all the staff of the Insti-tute who did my time in the university more comfortable.

I am specially grateful to my assistant Martin Wrulich for his constant sup-port, guidance and patience in the development of this thesis, for the interesthe expressed since the first day and the general support in my abroad expe-rience.

I am deeply grateful to my parents, who have always expressed their love,support and interest and have given me the opportunity to study abroad,and also for visiting Vienna with me. I want specially thank to my brotherfor helping me everytime he could, and all my friends in Madrid and Leon.

I want to thank also the friends that I met in Vienna who did my life thereduring six months really special and happy and showed their interest in mywork, specially Corinna, Alberto, Cristina and the turkish girls. I thank theirlovely frienship and support.

I want to dedicate a special mention to Borja, for coming with me to theadventure of having an Erasmus experience in Vienna, for his constant sup-port and love during this years, for his care and patience and for giving mecourage in the difficults moments since we started our careers. Without hissupport and help this project would have never been possible.

i

Abstract

In this bachelor thesis, the SISO HSDPA simulator developed for MobilkomAustria AG shall be extended in order to handle real network layout data.The Mobilkom Austria AG will provide measured path-loss matrices of aHSDPA cluster. This data has to be converted in a suitable form to beanalyzable in the simulator. Furthermore, a memory efficient loading of thedata has to be implemented. Based on this real network data, an optimizationof the CQI mapping of HSDPA mobiles shall be performed in order to find themapping which maximizes the overall cell throughput. This mapping can beused to implement a suitable CQI ”re-mapping” at the Node-B, granting anoptimum HSDPA network performance. The source-code has to be developedin MATLAB to ensure simple debugging and feature extendability.

ii

Contents

1 Introduction 11.1 Third Generation Services . . . . . . . . . . . . . . . . . . . . 11.2 Technology . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.3 Work environment . . . . . . . . . . . . . . . . . . . . . . . . 21.4 Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31.5 Structure of the thesis . . . . . . . . . . . . . . . . . . . . . . 3

2 HSDPA Principles 52.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52.2 HSDPA Standardization . . . . . . . . . . . . . . . . . . . . . 72.3 HSDPA vs Release 4 DCH . . . . . . . . . . . . . . . . . . . . 8

2.3.1 Radio resource management architecture . . . . . . . . 102.4 HSDPA operation principle . . . . . . . . . . . . . . . . . . . 112.5 HSDPA channels . . . . . . . . . . . . . . . . . . . . . . . . . 13

2.5.1 HSDPA new channels . . . . . . . . . . . . . . . . . . . 132.5.2 High-speed dedicated physical control channel . . . . . 20

3 HSDPA Simulator 223.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223.2 System model . . . . . . . . . . . . . . . . . . . . . . . . . . . 233.3 Simulation process . . . . . . . . . . . . . . . . . . . . . . . . 24

3.3.1 Load settings . . . . . . . . . . . . . . . . . . . . . . . 243.3.2 Precalculations . . . . . . . . . . . . . . . . . . . . . . 273.3.3 Simulation loop . . . . . . . . . . . . . . . . . . . . . . 303.3.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

4 Real Network Layouts 364.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 364.2 Files structure and information . . . . . . . . . . . . . . . . . 384.3 Network setting based on real layout data . . . . . . . . . . . 39

iii

5 CQI Mapping Optimizations 455.1 CQI Basis in HSDPA . . . . . . . . . . . . . . . . . . . . . . . 455.2 CQI mapping proposal . . . . . . . . . . . . . . . . . . . . . . 49

5.2.1 CQI Table . . . . . . . . . . . . . . . . . . . . . . . . . 505.3 CQI optimizations in the Simulator . . . . . . . . . . . . . . . 50

5.3.1 Source Code Enhancements . . . . . . . . . . . . . . . 525.3.2 Consequences . . . . . . . . . . . . . . . . . . . . . . . 535.3.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . 56

6 Miscellaneous 596.1 New scenarios . . . . . . . . . . . . . . . . . . . . . . . . . . . 596.2 More efficient implementation of the

pathloss generation . . . . . . . . . . . . . . . . . . . . . . . . 616.3 Further enhancements . . . . . . . . . . . . . . . . . . . . . . 62

6.3.1 Outage users . . . . . . . . . . . . . . . . . . . . . . . 626.3.2 Throughput figures . . . . . . . . . . . . . . . . . . . . 64

7 Conclusions 687.1 Simulator enhancement . . . . . . . . . . . . . . . . . . . . . . 697.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 707.3 Future enhancements . . . . . . . . . . . . . . . . . . . . . . . 72

iv

List of Figures

1.1 View coverage map about HSDPA deployment in the world, [1]. 2

2.1 Estimated cell throughput per sector, [2]. . . . . . . . . . . . . 62.2 Downlink data rates, [2]. . . . . . . . . . . . . . . . . . . . . . 72.3 Fundamental properties of the DCH and HS-DSCH, [2]. . . . . 102.4 RRM architecture, [2]. . . . . . . . . . . . . . . . . . . . . . . 112.5 HSDPA Node B scheduling principle, [2]. . . . . . . . . . . . . 122.6 HSDPA operation channels. . . . . . . . . . . . . . . . . . . . 132.7 HS-DSCH channel coding chain, [3]. . . . . . . . . . . . . . . . 152.8 16QAM and 4QAM constellations, [2]. . . . . . . . . . . . . . 152.9 Relative timing between HS-SCCH and HS-DSCH, [2]. . . . . 19

3.1 Three main steps in the simulator: load settings, precalcula-tions (i.e Node-B and users positions in order to prepare thenetwork) and simulation loop to obtain the HSDPA data rate. 25

3.2 Network layout with 7 and 19 base stations. . . . . . . . . . . 283.3 Example of the grid positions generation in the serving cell. . 283.4 Example of the users position in the serving cell. . . . . . . . . 293.5 Overview of the basic steps in the simulator. . . . . . . . . . . 303.6 Average data rates with RNC power control of the HS-DSCH. 34

4.1 Information extracted from the header of the files. . . . . . . . 384.2 Read pathloss from data files given by Mobilkom AG. . . . . . 394.3 Node-Bs positions and respective prediction files. . . . . . . . 414.4 Provisional grid, main BS and sector shape. . . . . . . . . . . 424.5 Last step: user positions located randomly and uniformly

within the limits of the generated sector (green points). . . . . 434.6 Diagram of the overall handling real layouts data process. . . . 44

5.1 High-Speed Dedicated Physical Control Channel that carriesthe uplink. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

v

5.2 HS-DSCH link adaptation principle: (1) the UE reports low-quality channel information and the Node B allocates a lowbit rate; (2) the UE reports high-quality channel informationand the Node B allocates a high bit rate, [2]. . . . . . . . . . . 47

5.3 Block diagram showing the received signal at the HSDPA userand report of the CQI to the serving HS-DSCH cell, [4]. . . . . 48

5.4 CQI mapping. . . . . . . . . . . . . . . . . . . . . . . . . . . . 535.5 Coarse view of the simulation including slope and shift. . . . . 545.6 Different values of the slope for the CQI mapping. . . . . . . . 555.7 Different values of the shift for the CQI mapping. . . . . . . . 555.8 Throughput as function of the slope value with shift = 0. . . . 575.9 Throughput as function of the shift value with slope = 1. . . . 575.10 Throughput as function of the slope and shift. . . . . . . . . . 58

6.1 Four possible user positions configuration. . . . . . . . . . . . 606.2 Snapshot and exhaustive snapshot scenarios. . . . . . . . . . . 606.3 Fixed angle and fixed distance scenarios. . . . . . . . . . . . . 616.4 Network with 19 base stations and 3 sectors model. . . . . . . 626.5 Variation of the user pathloss with the distance. . . . . . . . . 636.6 Variation of the user pathloss with the angle. . . . . . . . . . . 636.7 HSDPA outage users as function of the power. . . . . . . . . . 646.8 Throughput as a function of the angle with two fixed BS - user

positions distances (50 meters in the figure of the left and 250meters in the right). . . . . . . . . . . . . . . . . . . . . . . . 66

6.9 Throughput as a function of the distances with two fixed an-gles between the users and the BS (80o in the figure of the leftand 40o in the right). . . . . . . . . . . . . . . . . . . . . . . . 67

vi

Abbreviations

16QAM - 16-Quadrature Amplitude Modulation3G - Third Generation3GPP - Third Generation Partnership ProjectAMC - Adaptive Modulation and CodingARP - Allocation and Retention PriorityARQ - Automatic Repeat RequestAWGN - Additive White Gaussian NoiseBLER - Block Error RateBS - Base StationBTS - Base Transceiver StationCDMA - Code Division Multiple AccessCmCH-PI - Common Transport Channel Priority IndicatorCPICH - Common Pilot ChannelCQI - Channel Quality IndicatorCSI - Channel State InformationDEM - Digital Elevation ModelsFCS - Fast Cell SelectionFCSS - Fast Cell Site SelectionFP - Frame ProtocolGGSN - Gateway GPRS Support NodeGSM - Global System for Mobile CommunicationsHARQ - Hybrid Automatic Repeat RequestHSDPA - High-Speed Downlink Packet AccessHS-DPCCH - Dedicated High-Speed Physical Control ChannelHS-DSCH - High-Speed Dedicated Shared ChannelHSPA - High-Speed Packet AccessHS-PDSCH - High-Speed Physical Downlink Shared ChannelHS-SSCH - High-Speed Shared Control ChannelsHSUPA - High-Speed Uplink Packet AccessIR - Incremental RedundancyITU - International Telecommunication Union

vii

Max C/I - Maximum Carrier to InterferenceMCS - Modulation and Coding SchemeMIMO - Multiple Input Multiple OutputMS - Mobile StationPDP - Power Delay ProfilePDU - Protocol Data UnitPF - Proportional FairQoS - Quality of ServiceRLC - Radio Link ControlRNC - Radio Network ControlRR - Round RobinRRM - Radio Resource ManagementSAW - Stop And WaitSF - Spreading FactorSGSN - Serving GPRS Support NodeSINR - Signal to Noise and Interference RatioSISO - Single Input Single OutputSNR - Signal to Noise RatioSPI - Scheduling Priority IndicatorTCP - Transmission Control ProtocolTBS - Transport Block SizeTTI - Transmit Time IntervalUE - User EquipmentUMTS - Universal Mobile Telecommunications SystemUTRAN - UMTS Terrestrial Radios Access NetworkWCDMA - Wideband Code Division Multiple AccessWSS - Widesense Stationary

viii

Chapter 1

Introduction

1.1 Third Generation Services

During the last decades, the mobile communication market evolution has ledto demands for higher data rates and larger system capacity. To successfullysatisfy these requirements, Third Generation systems must increase theirspectral efficiency and support high user data rates, especially on the down-link direction of the communication path due to its heavier load. For thispurpose, the 3GPP has standardized in Release 5 a new technology calledHigh Speed Downlink Packet Access (HSDPA) that represents an evolutionof the WCDMA radio interface. These technological enhancements can allowoperators to enable new high data rate services, improve the QoS of alreadyexisting services, and achieve a lower cost per delivered data bit.

Consumers are expected to acquire mobile data services if their contentsadd value to the consumer’s life by satisfying a concrete necessity or require-ment. From the end user’s interest, the value provided by the service contentscontribute to his cost-effectiveness, time-efficiency, or simple entertainment;for instance, rich content services like video telephony, audio/video clips, andmap based information, or fast Internet access for business users.

1.2 Technology

High-Speed Downlink Packet Access, or also known as HSDPA, is a mobiletelephone protocol in the High-Speed Packet Access (HSPA) family of thirdgeneration (3G) technologies designed to reduce the latency of the link andincrease data transfer rates and the capacity of such networks through thetransfer of data using a cellular phone. HSDPA is associated with the vari-

1

ous Universal Mobile Telecommunications System (UMTS) networks. Cur-rent HSDPA deployments support down-link speeds of 1.8, 3.6, 7.2 and 14.4Mbps. The first phase of HSDPA has been specified in the 3rd GenerationPartnership Project (3GPP) Release 5. The second phase of HSDPA is spec-ified in the 3GPP Release 7 and has been named HSPA Evolved or alsoHSPA+; it can achieve data rates of up to 25 Mbps, [1].

As a difference with other WCDMA channels, the High-Speed DownlinkShared Channel (HS-DSCH) lacks two basic features - fast power controland variable spreading factor. Instead, it presents an improved downlinkperformance by using adaptive modulation and coding (AMC), fast packetscheduling and fast retransmissions at the base station, known as hybrid au-tomatic repeat-request (HARQ), together with a shorter 2-ms TransmissionTime Interval (TTI). Figure 1.1 shows the coverage map of deployed HSDPAtechnology around the world.

Figure 1.1: View coverage map about HSDPA deployment in the world, [1].

1.3 Work environment

Due to the importance of the HSDPA technology, a SISO-HSDPA Systemlevel simulator was developed in a collaboration of the Institute of Com-munications and Radio Frequency Engineering and Mobilkom Austria AG.

2

The source-code is based on MATLAB and it simulates a mixed network inwhich both UMTS and HSDPA traffics are present. MATLAB is a high-leveltechnical computing language and interactive environment for algorithm de-velopment, data visualization, data analysis, and numeric computation. It isan extended engineering tool and has enhanced the tradicional languages. Itallows solving technical computing problems and a wide range of applicationslike signal and image processing, communications, easy matrix manipulation,plotting of functions and data, implementation of algorithms, creation of userinterfaces, and interfacing with programs in other languages. The extensionof the initial simulator is also developed in MATLAB, to ensure extendabilityand simple debugging.

1.4 Objectives

Let me briefly sketch the motivation of this bachelor thesis. The first goalof the simulator was to evaluate the HSDPA throughput performance inthe mixed traffic network; and based on this groundwork this thesis shoulddevelop two enhancements, namely:

• Extend the SISO HSDPA Simulator in order to handle realnetwork layout data: this is the main functionality on which thework has been focused; the Mobilkom Austria AG provided measuredpath-loss matrices of a HSDPA cluster, including parameters like an-tenna gain patterns, the height of the antennas and the Nodes-B posi-tions. This data has to be converted in a suitable form to be analyzablein the simulator.

• CQI mapping optimizations: an optimization of the CQI mappingof HSDPA mobiles shall be performed in order to find the mappingwhich maximizes the overall cell throughput. This optimization canresult in a new mapping at the Node-B that grant an optimum HSDPAnetwork performance.

1.5 Structure of the thesis

This thesis report is organized as follows:

• Chapter 1: a short introduction about the technology and work en-vironment, objectives and outline of the thesis is described.

• Chapter 2: provides a general overview of the HSDPA technology,like basis and key features.

3

• Chapter 3: this chapter describes the initial SISO-HSDPA Simulator,including the system model and simulation process.

• Chapter 4: the conversions of the real data matrices into useful infor-mation for the simulator, the overall process and necessary new func-tionalities and modifications in the initial source-code for handling thenew real data layout are explained in this section.

• Chapter 5: this chapter outlines the CQI basis in HSDPA, the currentCQI mapping and the mapping optimizations investigated in the thesis,as well as the new functions including in the original simulator for thispurpose.

• Chapter 6: includes a description of some investigations and newfunctionalities for the enhancement of the simulator that do not fitinto the previous chapters.

• Chapter 7: summarises the main conclusions of the bachelor thesis.

4

Chapter 2

HSDPA Principles

This chapter covers high-speed downlink packet access (HSDPA) principlesfor wide-band code division multiple access (WCDMA) - the new key fea-ture included in Release 5 specifications and enhanced further in Release 6specifications. HSDPA has been designed to increase downlink packet datathroughput, compared to the rates provided by the Release 4 (also calledRelease 99) WCDMA specifications by means of fast physical layer (L1)retransmission and transmission combining as well as fast link adaptationcontrolled by the Node B.

2.1 Introduction

HSDPA, also called 3.5G, is the evolution of the third generation (3G) andis considered the previous step before the fourth generation (4G), the futureHigh-Speed Mobile Network. HSDPA and High-Speed Uplink Packet Access(HSUPA) are the components of the High-Speed Packet Access (HSPA) fam-ily. HSPA is an upgrade of the network infrastructure and it is part of theWCDMA 3G network. As an enhancement of UMTS, HSDPA was designedto improve the quality of service, increase the peak data rates (currentlyspeeds supported by HSDPA are 1.8, 3.6, 7.2 and 14.4 Mbps). Also com-pared to UMTS, the spectral efficiency is significantly increased, and thisallows more users being able to use high data rates on a single carrier. Thefundamental techniques used in HSDPA to achieve this improvements areAdaptive Modulation and Coding (AMC), extensive multi-code operationand a fast and spectrally efficient retransmission strategy. The assignment ofthe HS-DSCH (High-Speed Downlink Shared Channel) among the users on

5

a TTI basis (1 TTI = 2 ms) is coordinated by a fast scheduler. Higher cellcapacity and higher spectral efficiency are required to provide these higherdata rates and new services with the current base station sites. Figure 2.1illustrates the estimated cell capacity per sector per 5MHz with WCDMA,with basic HSPA and with enhanced HSPA in the macro-cell environment.

Figure 2.1: Estimated cell throughput per sector, [2].

HSDPA is able to satisfy the most demanding multimedia applications suchas email attachments, Power Point presentations or web pages. An HSDPA3.6 Mbps network can provide a 3MB music file in 8.3 seconds and a 5 MBvideo clip in 13.9 seconds. Speeds achieved by HSDPA reach 14.4 Mbps butcurrently most network operators provide speeds up to 3.6 Mbps, with therollout of 7.2 Mbps quickly growing. It is important to note that the to-tal available downlink speed within one sector is split among all the activeusers. Also, HSDPA can coexist with Release 4 in the same frequency bandof 5 MHz. In Austria four HSDPA operators are giving service, MobilkomAustria, Hutchison 3 Austria and ONE Austria serving HSDPA data ratesof 7.2 Mbps, and Mobilkom Austria serving HSUPA (with a data rate of 1.4Mbps). Currently only Telstra (in Australia) is serving HSDPA data ratesof 14.4 Mbps. There are 185 commercial HSDPA networks in 92 differentcountries; the current deployment of HSPA networks in the world and the

6

HSDPA data rates supported are shown in Table 2.1.

Data rate Networks0-3.6 Mbps 563.6-7.2 Mpbs 917.2-14.4 Mpbs 38

Table 2.1: Current HSDPA commercial networks and data rates, [1].

2.2 HSDPA Standardization

High-speed downlink packet access (HSDPA) was standardized as part of3GPP Release 5 with the first specification version in March 2002. High-speed uplink packet access (HSUPA) was part of 3GPP Release 6 with thefirst specification version in December 2004. HSDPA and HSUPA togetherare called ’high-speed packet access’(HSPA). The first commercial HSDPAnetworks were available at the end of 2005, as we can see in Figure 2.2, andmany improvements have been introduced in the Release 6, 7, and 8.

Figure 2.2: Downlink data rates, [2].

7

HSPA is deployed on top of the WCDMA network. Both of them can shareall the network elements in the core network and in the radio network includ-ing base stations, Radio Network Controller (RNC), Serving GPRS SupportNode (SGSN) and Gateway GPRS Support Node (GGSN). WCDMA andHSPA are also sharing the base station sites, antennas and antenna lines.

3GPP creates the technical content of the specifications, based around workitems, though small changes can be introduced directly as ”change requests”against specification, but it is the organizational partners that actually pub-lish the work. In addition to the organizational partners, there are alsomarket representation partners, such as the UMTS Forum, part of 3GPP.With bigger items a feasibility study is done usually before rushing in tomaking actual changes to the specifications, [2].

A feasibility study for HSDPA was started in March 2000 in line with 3GPPprinciples, having at least four supporting companies. Motorola and Nokiasupporting the start of the work from the vendor side and BT/Cellnet, T-Mobile and NTT DoCoMo from the operator side. The study was finalizedfor the TSG RAN plenary for March 2001 and there were issues studied toimprove the downlink packet data transmission over Release 4 specifications.Physical layer retransmissions and BTS-based scheduling were studied as wellas adaptive coding and modulation. The study also included some investiga-tions for multi-antenna transmission and reception technology, titled MIMO(Multiple Input Multiple Output), and also Fast Cell Selection (FCS), [2].

2.3 HSDPA vs Release 4 DCH

In Release 4 specifications basically exist three different methods for down-link packet data operation: dedicated channel (DCH), forward access channel(FACH) and downlink shared channel (DSCH). The most interesting com-parison is between Release 4 and HSDPA dedicated channel; the FACH isused either for small data volumes or when setting up the connection andduring state transfers. In connection with HSDPA, the FACH is used tocarry the signalling when the terminal has moved. However the DSCH hasbeen replaced with the high-speed DSCH of HSDPA.

The Release 4 based DCH is the key part of the system, and Release 5

8

HSDPA is always operated with the DCH running in parallel. If the serviceis only for packet data, then at least the signalling radio bearer (SRB) iscarried on the DCH.

In case the service is circuit-switched, then the service always runs on theDCH. In Release 5, uplink user data always go on the DCH (when HSDPA isactive), whereas in Release 6 an alternative is provided by the Enhanced DCH(E-DCH) with the introduction of high-speed uplink packet access (HSUPA).In the case of multiple services, the reserved capacity is the sum of the peakdata rate of the services. The main functionality for the DCH is the fastpower control in addition to encoding the data packet provided by the RNC.Furthermore, soft handover is supported for the DCH. As a difference withRelease 4, HSDPA introduces some methods for improving downlink packetdata in terms of capacity and bit rates. The key differences between theHS-DSCH (HSDPA dedicated channel) and the Release 4 DCH-based packetdata operation are as follows:

• Lack of fast power control. Instead, link adaptation selects the suitablecombination of codes, coding rates and modulation to be used.

• Support of higher order modulation than the DCH. With 16-QuadratureAmplitude Modulation (16QAM) the number of bits carried per symbolis doubled in favourable conditions compared to the quadrature phaseshift keying (4QAM) in Release 4.

• User allocation with base station based scheduling every 2ms, includingfast physical layer signalling. With DCH the higher layer signallingfrom the RNC allocates semi-permanent code (and a spreading factor)to be used. The transmission time interval (TTI) is also longer withthe DCH, allowing values such as 10, 20, 40 or 80 ms. (The longest islimited in the specific case of small data rates that have a spreadingfactor of 512).

• Use of physical layer retransmissions and retransmission combining,while with the DCH - if retransmissions are used - they are based onRLC level retransmissions.

• Lack of soft handover. Data are sent from one serving HS-DSCH cellonly.

• Lack of physical layer control information on the HS-PDSCH. This iscarried instead on the HS-SCCH for HSDPA use and on the associatedDCH (uplink power control, etc).

9

• Multicode operation with a fixed spreading factor. Only spreadingfactor 16 is used, while with the DCH the spreading factor could be astatic parameter between 4 and 512.

• The DCH may use both turbo-coding or convolutional coding, while inHSDPA only turbo-coding is used. This was motivated by the fact thatturbo codes make it possible to increase data rate without increasingthe power of a transmission, or they can be used to decrease the amountof power used to transmit at a certain data rate.

• No discontinuous transmission (DTX) on the slot level. The HS-PDSCHis either fully transmitted or not transmitted at all during the 2-ms TTI.

The main differences are summarised in Figure 2.3:

Figure 2.3: Fundamental properties of the DCH and HS-DSCH, [2].

2.3.1 Radio resource management architecture

The radio resource management (RRM) functionality with HSDPA and HSUPAis significantly changed compared to Release 4. In Release 4 the schedulingcontrol was purely based in the radio network controller (RNC) while in thebase station (BTS or Node B in 3GPP terminology) mainly a power con-trol related functionality (fast closed loop power control) was located. InRelease 4 if there were two RNCs involved in the connection, the schedul-ing was distributed. The serving RNC (SRNC) - the one being connectedto the core network for that connection - would handle the scheduling for

10

dedicated channels (DCHs) and the one actually being connected to the basetransceiver station (BTS) would handle the common channels.

Due to the BTS based scheduling, the overall RRM architecture changed.The SRNC will still retain control of handovers and is the one which willdecide the suitable mapping for quality of service (QoS) parameters. WithHSDPA the situation is simplified because, as there are no soft handovers forHSDPA data, the utilization of the Iur interface can be avoided by perform-ing SRNC relocation, when the serving high-speed downlink shared channel(HS-DSCH) cell is under a different controlling RNC (CRNC). Thus, just asingle RNC could be enough for the typical HSDPA scenario, [2]. Figure 2.4shows the new RRM architecture.

Figure 2.4: RRM architecture, [2].

2.4 HSDPA operation principle

HSDPA is based on a fast Node B scheduling where the Node B estimatesthe channel quality of each active HSDPA user on the basis of the physical

11

layer feedback received in the uplink. Scheduling and link adaptation arethen conducted on a fast pace depending on the scheduling algorithm andthe user prioritization scheme. The general HSDPA operation principle isshown in Figure 2.5.

Figure 2.5: HSDPA Node B scheduling principle, [2].

The other new key technology is physical layer retransmission. In Release4 when the data has not been received correctly, is necessary to retrans-mit it again from the RNC. In Release 4 there is no difference in physicallayer operation, regardless if the packet is a retransmission or a new packet.With HSDPA the packet is first received in the buffer in the BTS. The BTSkeeps the packet in the buffer even if has sent it to the user and, in caseof packet decoding failure, retransmission automatically takes places fromthe base station without RNC involvement. So, the terminal can combinethe transmissions, capturing the energy of both. Using a radio link control(RLC)-acknowledged mode of operation, RLC layer acknowledgement is pro-vided in the RLC layer as would be done for Release 4 based operation.

12

2.5 HSDPA channels

2.5.1 HSDPA new channels

Several new channels have been introduced for HSDPA operation. For userdata there is the high-speed downlink shared channel (HS-DSCH) and thecorresponding physical channel. For the associated signalling needs there aretwo channels: high-speed dedicated physical control channel (HS-DPCCH)in the uplink direction and high-speed shared control channel (HS-SCCH)in the downlink. In addition to the basic HSDPA channel covered in Re-lease 5 specifications, there is now a new channel in Release 6 specifications -the fractional dedicated physical channel (F-DPCH) - to cover for operationwhen all downlink traffic is carried on the HS-DSCH. The channels neededfor HSDPA operation are shown in Figure 2.6.

Figure 2.6: HSDPA operation channels.

High-speed downlink shared channel

The HS-DSCH is the transport channel that carries the actual user data.In the physical layer the HS-DSCH is mapped onto the high-speed physical

13

downlink shared channel (HS-PDSCH).

An important property of the HS-DSCH is that it can dynamically allocatethe resource. When the Node-B decides which user is going to be served,the data is sent continuously during the 2-ms TTI, so there is no discontinu-ous transmission (DTX) on the slot level like with the DCH. With DTX thedownlink interference generated is reduced, but it keeps the code resource oc-cupied according to the highest data rate possible on the DCH, because thecode resource reservation is not changed when moving to a lower data rate;(the only way to reduce resource consumption is to reconfigure the radio link,but this takes time in reconfiguring the data rate to a new smaller value, andthen a new reconfiguration to upgrade the data rate again). As a differenceto DTX, with HS-DSCH, once there are no more data to be transmitted forthat user, there is no transmission on the HS-DSCH again for the same user,but the resources in the according 2-ms are allocated to another user. Let’ssee the important technical apects:

• Adaptative Modulation and Coding:

To cope with the dynamic range of the signal-to-noise ratio (Es/No)at the UE, HSDPA adapts the modulation, the coding rate and num-ber of channelization codes to the instantaneous radio conditions. Thecombination of the first two mechanisms is denominated Adaptive Mod-ulation and Coding (AMC).

The channel coding is simpler than the corresponding DCH one, be-cause in the HS-DSCH there is no need to deal with DTX or compressedmode, and there is only one transport channel active at a time becausefewer steps in multiplexing/de-multiplexing are needed. The HS-DSCHchannel coding chain is illustrated in Figure 2.7.

– 16QAM :

While the DCH only uses 4QAM modulation, the HS-DSCH mayadditionally use the higher order modulation 16QAM. HS-DSCHincorporates this modulation to increase the peak data rates forusers served under favourable radio conditions. Support of 4QAMis mandatory for the mobile, despite the support of 16QAM is op-tional for the network and the UE. The inclusion of this highorder modulation introduces some complexity challenges for thereceiver terminal, which needs to estimate the relative amplitudeof the received symbols, whereas it only requires the detection of

14

Figure 2.7: HS-DSCH channel coding chain, [3].

the signal phase in the 4QAM case. The turbo encoder is in chargeof the data protection. The 16QAM constellation rearrangementdepends on the transmission number, because the symbols in theconstellation do not have the same error probability. The 16QAMand 4QAM constellations are shown in Figure 2.8.

Figure 2.8: 16QAM and 4QAM constellations, [2].

15

QAM is a digital modulation that transports data by changingthe amplitude of two carrier signals. These two waves, generallysinusoidal, are in the same frecuency but with a phase differenceof ninety degrees; both signal paths - I and Q - carry information.It is used for the data transmission with a high speed by channelswith restricted bandwidth. By having more constellation points -16 instead of 4 - now 4 bits can be carried per symbol instead of2 bits per symbol with 4QAM.

In reception the use of higher order modulation like 16QAM in-troduces additional decision boundaries, as shown in Figure 2.8.Signal quality needs to be better when using 16QAM instead of4QAM. Because of this, with 16QAM it is not sufficient to figureout the phase correctly but also the amplitude needs to be esti-mated for more accurate phase estimate.

The channel coefficients can be estimated from the common pi-lot channel (CPICH), which directly gives phase information. Theoffset of the HS-DSCH data channel to the CPICH however has tobe signalled in order to estimate the amplitude information. Thissuggests that at the base station - during the 2-ms transmission -power changes should be avoided.

In the system there can be other traffic that is consuming codespace as well - such as circuit switched speech or video calls - whichcannot be mapped on HSDPA. Thus, radio resource managementwill then determine the available code space for the scheduler atthe BTS, [2].

– Bit scrambling:

The bit scrambling functionality was introduced to avoid longsequences repeating the same symbol, as long sequences of ’0s’or ’1s’. These could occur with some type of content, and espe-cially when not using ciphering at higher layers. In such a casethe terminal would have difficulties with HS-DSCH power levelestimation and, thus, physical layer scrambling operation was in-troduced. Operation is the same for all users and is purely forensuring good signal properties for demodulation, [2].

16

– HS-DSCH Link Adaptation

HSDPA utilizes link adaptation techniques to substitute powercontrol and variable spreading factor. The HS-DSCH link-adaptationalgorithm at the Node-B is very dynamic, and adjusts the transmitbit rate on the HS-DSCH every 2-ms TTI. It is based on the phys-ical layer CQI being provided by the terminal. Various sourcescontribute to the time-variant SINR at the user even though theHS-DSCH transmit power is assumed to be constant. The totaltransmit power from the serving HS-DSCH cell is time variant dueto the transmission of the power controlled DCHs; the downlinkradio channel is time variant if the user equipment is somehowmoving; and finally, the experienced inter-cell interference at theuser position is also time variant. For the purpose of HS-DSCHlink adaptation, the user therefore periodically sends a CQI to theserving HS-DSCH cell on the uplink high-speed dedicated physicalcontrol channel (HS-DPCCH), [4].

Using link adaptation, the network will also gain from the lim-itation of power control dynamics in the downlink. As signalsin the downlink cannot use a too large dynamic range to avoidthe near-far problem between signals from the same source, thedownlink power control dynamics is also limited. While in the up-link a 71-dB or more dynamic range is used, in the downlink onlyaround 10 to 15 dBs can be utilized. The exact number dependson the implementation, channel environment and spreading fac-tors applied. This means that for users close to the base stationthe power level transmitted is higher than necessary for reliablesignal detection. Using link adaptation, there is a difference of afew decibels in the signal strength, just by changing from 4QAMto 16QAM; and by playing with the coding rates and the numberof codes the total dynamic range can reach 30 dB.

• Hybrid ARQ :

HSDPA incorporates a physical layer retransmission functionality thatadds robustness against link adaptation errors and improves the per-formance significantly.

The Hybrid ARQ functionality consists of a two stage matching func-tionality which allows tuning two different retransmission types. These

17

two Hybrid ARQ strategies are: (1) identical retransmissions (alsocalled soft combining) or (2) non-identical retransmissions (or so-called’incremental redundancy’). The Hybrid ARQ technique is fundamen-tally different from the WCDMA retransmissions because the UE de-coder combines the soft information of multiple transmissions of atransport block at bit level. Let us go a little bit more into detail:

Soft combining : as proposed in [5] every retransmission is simply areplica of the coded first transmission. The same bits after rate match-ing operation are sent, for every retransmission of the same packet.The decoder at the receiver combines these multiple replicas of thetransmitted packet weighted by the received SNR prior to decoding(so called ”soft combining”). This technique requires some memory onthe mobile terminal, which must store the soft information of unsuc-cessfully decoded transmissions. The delay in the retransmissions andmemory required shell be as small as possible.

Incremental Redundancy (IR): it requires even more memory in thereceiver user equipment capabilities. The retransmissions include ad-ditional redundant information that is incrementally transmitted if thedecoding fails on the first attempt. That causes the effective coding rateto be increased with the number of retransmissions. Incremental Re-dundancy can be further classified in Partial IR and Full IR. Partial IRincludes the systematic bits in every coded word, which implies thatevery retransmission is self-decodable, whereas Full IR only includesparity bits, and therefore its retransmissions are not self-decodable.

If due to a signalling error that could fill the buffer with undesireddata, due to a low coverage, or due to a change of the serving HS-DSCH cell, the number of physical layer retransmissions exceeds themaximum or the retransmissions fail, the radio link layer will handlefurther retransmissions.

High-speed shared control channel (HS-SCCH)

The HSDPA concept includes a Shared Control Channel (HS-SCCH) to sig-nal the users when they are going to be served as well as the necessaryinformation for the decoding process. Compared with the HS-DSCH, theHS-SCCH has two slots offset, as shown if Figure 2.9. This enables the HS-SCCH to carry time-critical signalling information which allows the terminalto demodulate the correct codes. A spreading factor of 128 allows 40 bits

18

per slot to be carried (with 4QAM modulation). The phase reference doesnot change when using HS-DSCH due to the lack of pilots or power controlbits on the HS-SCCH.

Figure 2.9: Relative timing between HS-SCCH and HS-DSCH, [2].

The HS-SCCH carries the following information, [6]:

• UE Id Mask: to identify the user to be served in the next TTI.

• Transport Format Related Information: specifies the set of channeliza-tion codes, and the modulation. The actual coding rate is derived fromthe transport block size and other transport format parameters [2].

• Hybrid ARQ Related Information: such as if the next transmissionis new or related to an earlier transmitted packet, and if it shouldbe combined, the associated ARQ process, and information about theredundancy version, [7].

This control information solely applies to the UE to be served in the nextTTI, which permits this signalling channel to be a shared one. The RNCcan specify the recommended power of the HS-SCCH (offset relative to thepilots bits of the associated DPCH), [8].

The timing between the HS-SCCH and the HS-DSCH allows the terminalto have one slot time to receive the information which codes have to de-spread and with which to modulate. For the remaining parameters, a slotprocessing time is needed before a new 2-ms TTI starts.

19

When HSDPA is operated using the time multiplexing principle, then onlyone HS- SCCH can be configured. In this case only one user receives data ata time. When there is code multiplexing, then more than one HS-SCCH isneeded. A single terminal may consider at most four HS-SCCHs; the systemitself could configure even more. The use of code multi-plexing is not nec-essarily needed either when the carrier is shared with DCH traffic, or whenthere is a desire to have HSDPA data users operating with reasonable datarates -in the order of 384 kbps or more. In general, the data rate availablefor each user in different cases will depend on power allocation, the environ-ment and the type of terminal being used. The channel coding is one-thirdconvolutional coding (as turbo-coding does not make sense with such a smallamount of information). In the second part there is a cyclic redundancycheck (CRC) to make sure that there is no corruption of the information. Asignalling error with, say, an HARQ process number would cause problemsas it would cause buffer corruption; thus, a 16-bit CRC is used to ensuresufficient reliability, [2].

2.5.2 High-speed dedicated physical control channel

An uplink High Speed Dedicated Physical Control Channel (HS-DPCCH)carries the necessary control information in the uplink, namely, the ARQacknowledgements, and the Channel Quality Indicator (CQI) reports. TheCQI reports are deeply described in Section 5. To aid the power control oper-ation of the HS-DPCCH an associated Dedicated Physical Channel (DPCH)is run for every user. This information from the terminal to the base stationallows for the link adaptation and physical layer retransmissions.

According to [8], the RNC may set the maximum transmission power onall the codes of the HS-DSCH and HS-SCCH channels in the cell. Likewise,the RNC determines the maximum number of channelization codes to beused by the HS-DSCH channel.

By keeping the existing uplink DPCCH and DPDCH unchanged the ac-tive set can also accommodate Release 4 based base stations. The initialuplink DPCCH transmit power is set by higher layers. Subsequently the up-link transmit power control procedure simultaneously controls the power of aDPCCH and its corresponding DPDCHs (if present). The relative transmitpower offset between DPCCH and DPDCHs is determined by the network.

Any change in the uplink DPCCH transmit power shall take place immedi-ately before the start of the pilot field on the DPCCH. The change in DPCCH

20

power with respect to its previous value is derived by the UE. The previousvalue of DPCCH power shall be that used in the previous slot, except in theevent of an interruption in transmission due to the use of compressed mode,when the previous value shall be that used in the last slot before the trans-mission gap. During the operation of the uplink power control procedure theUE transmit power shall not exceed a maximum allowed value which is thelower out of the maximum output power of the terminal power class and avalue which may be set by higher layer signalling. Uplink power control shallbe performed while the UE transmit power is below the maximum allowedoutput power, [9].

As already mentioned, the uplink feedback information is carried on the HS-DPCCH. The HARQ feedback informs the base station whether the packetwas decoded correctly or not. The CQI, respectively, tells the base stationscheduler the data rate the terminal expects to be able to receive at a givenpoint in time.

Fractional DPCCH

For Release 6, further optimization took place for the situation where onlypacket services are active in the downlink other than the signalling radiobearer (SRB). In such a case, especially with lower data rates, the downlinkDCH introduces too much overhead and can also consume too much codespace if looking for a large number of users using a low data rate service (likeVoIP). The solution was to use an F-DPCH, which is basically a stripped-down version of DPCH that handles the power control.

The code resource is time-shared, thus several users can share the same codespace for power control information. Each user sees only the channel whichhas one symbol per slot for transmission power control (TPC) informationand assumes there is no transmission in the rest of the symbols. With severalusers, the network configures each user having the same code but differentframe timing and, thus, users can be transmitted on the single code source.Up to ten users can share one SF 256 code, thus reducing code space uti-lization for the associated DCH for users with all services mapped to theHS-DSCH,[2].

21

Chapter 3

HSDPA Simulator

In this chapter the initial simulator on which this bachelor thesis is basedand all the work that was developed is explained. It contains the informationabout the initial simulator created by M.Wrulich, et al., in which I had toimplement some new functionalities for the enhancement of the simulator.The chapter is organized as follows: Section 3.1 briefly sketchs the purposeand context of the simulator; a short description of the system-level modelfor the investigation is included in Section 3.2 and Section 3.3 explains thestructure of the simulator.

3.1 Introduction

Let me briefly introduce the work environment and goals of the initial SISOHSDPA Simulator. In the program, written in MATLAB, a mixed UMTSand HSDPA network is simulated. As described in Chapter 2, one of theadvantages in a WCDMA network is that HSDPA can coexist within anexisting 5MHz band of Release 4, allowing for sharing the power amplifierand spreading codes at the Node-B between the HS-DSCH and the Release-4 dedicated channels. Even if HSDPA is widely installed, a mixed carrieroperation is more cost-efficient for cells that are not fully loaded, [10]. Oneof the goals of the initial simulator was to deduce the optimum Node-Bpower split within the mixed scenario, by means of snapshot based networksimulations, in order to maximize the overall cell throughput. The obtainedresults can be used for the cell operation planning by the network operators.

22

3.2 System model

According to [10], the HSDPA performance is done with a system-level si-mulator in which the following aspects are modelled:

• Channel modeling: the channel coefficient used for the simulationallows us to model the radio propagation and it considers the macro-scale pathloss, that depends on the the distance between the base sta-tion and the user equipment, the shadow fading and the fast fadingwith multiple paths and no time correlation.

• HSDPA modeling: this modeling represents the HSDPA transmis-sion performance in an accurate way by using a simplified system leveldescription in which they have modelled the channel quality observedby the user equipment and the bit error/decoding performance. Themodels are the so-called link-measurement model and link-performancemodel respectively. In the link-measurement model, the signal-to-noise-and-interference ration (SINR) is evaluated after Rake-combining anddespreading for each user equipment in the cell. The link-performancemodel aims at an analytical approximation of the block error ratio(BLER), where it is assumed that the scheduler in the Node B decidesto serve a specific user with an MCS as specified by the CQI mappingtable [11]. Furthermore, it is assumed that the desired HSDPA useralways gets the full available transmission power and there is enoughdata to transmit (full buffer assumption).

• Release 4 modeling: the Release 4 traffic is only coarsly modelled,since the main goal of the simulator was the prediction of the achievableuser data rates with HSDPA within the existing Release 4 network. In[10] further information and the estimation of the number of DCH usersthat can be served can be found.

• Power split: the total intra-cell transmission power depends on thetransmission power of the DCH and the HSDPA traffic, and in eachcell, the total available transmit power is shared between DCH andHSDPA users. The power of HSDPA can be allocated in the Basestation downlink power budget by means of two possibilities:

– By sending Node-B application part (NBAP) messages to the basestation, the RNC can dynamically allocate the HSDPA power.This is kept at a fixed level by the base station, and the DCHpower varies accordingly to the fast closed loop power control.

23

– The base station is allowed to allocate all unused power for HS-DPA, instead of sending NBAP messages.

The total intra-cell transmission power is calculated as folows:

Pintra = PDCH + PHS−DSCH + Pother (3.1)

Pother incorporates the power from the common pilot channel and otherneeded common channels. This explains the fact that the total intra-cell power depends on the DCH and HSDPA transmission power.

3.3 Simulation process

After explaining briefly the system model of the simulator, now the initialSISO HSDPA Simulator itself is going to be treated in detail. Figure 3.1illustrates the three main steps:

• Load settings: before starting the process, the function ”load settings”is called. This settings file allows for the specification of the simulation,i.e. the kind of network, channel and user equipment.

• Precalculations: after the settings are loaded and before the simula-tion starts, there are some precalculations which allow us creating somenecessary elements like user and Node-B positions, all the pathlossesfor every users and PDP for serving links.

• Simulation loop: the last step is the simulation loop, which calcu-lates different average data rates by means of multiple independentsnapshots.

3.3.1 Load settings

The settings are divided in: network, channel, user equipment and simulator,and by changing the parameters we can specify the kind of simulation.

• Network settings: the mixed traffic network is modelled accordingto the parameters:

24

Figure 3.1: Three main steps in the simulator: load settings, precalcula-tions (i.e Node-B and users positions in order to prepare the network) andsimulation loop to obtain the HSDPA data rate.

– R’99: contains the needed parameteres for Release ’99, like thebandwith (5 MHz), the chiprate, the UMTS load in percent andthe UMTS required Eb/No for the requested UMTS DCH bearer.

– Node-B: here, some variables like the distance between Node-Bs,the power level of each Node-B (the maximum power, the CPICHpower and the common power), and the power distribution of theNode-B are specified.

– Power distribution: determines the power distribution amongthe neighboring Node-Bs, thus specifying the intercell-interferencestructure.

– HSDPA: in this part the HSDPA network is specified, thus thenumber of HSDPA users, the spreading factor of HSDPA trans-missions (fixed at 16), the number of codes, the absolute HSDPA

25

power and the TTI value (usually 2 ms) are assigned.

– MAC-hs: this is used in an enhanced version of the simulator forscheduling variables.

– Other: variables like the grid density, which determines the num-ber of grid points for user positioning within the cell, the G factorof the network or the other cells interferences are chosen here.

– Network structure: the number of base stations (7 or 19), thenumber of sectors for each base station (1 or 3) and the antennagain pattern are specified.

• Channel settings: the channel modeling is done through three dif-ferent fadings that can attenuate the signals in the communicationsbetween the base station and the users.

– Deterministic fading: first the model (COST231, Berger, fixed,exponent, tr25848 or none) is selected, and accordinglyly the nec-essary variables like frequency or antenna height are assigned.

– Shadow fading: there are two possibilities for the shadow fad-ing: the lognormal model or the lognormal moving model, andaccordinglyly different parameters can be selected.

– Fast fading: it can be modeled with a Rayleigh model or it canbe omitted.

– Power Delay Profile: the oversampling factor, the model (pedes-trian A or B, vehicular A or B, or none) and the chiprate areselected here.

• User Equipment settings: the parameters needed to model the userequipment are defined here.

– General: the user category class and different noise powers seenin the receiver, like the receiver noise figure or the thermal noisedensity are defined.

– Movement: contains the setting for speed of the user.

– Receiver: the reciver type is specified, and in case of a Rakereceiver the number of fingers is also determined.

– Traffic: will be used in further developements of the simulator.

• Simulator settings: now, some parameters of the simulator are de-fined, although they do not belong to the communication scheme itself.

26

– Simulation type: the kind of simulation can be chosen here. Itis possible to simulate all the possible users positions (also calledgrid positions) uniformly, or it can be done by means of multiplesimulations based on multiple snapshot, in which we simulate asmany positions as number of users we have set.

– Link performance model: the COST290 model, where a simplelink performance model based on cost 290 or no model can be used,but a model is needed to performance the link and simulate a realsituation with a BLER value different to zero.

– R’99 datarate model: a simple data rate model described in[10].

– Power distribution: settings for the step-size of the power loopin the simulator.

– Display/save results: backup options.

3.3.2 Precalculations

As already said, when the parameters which determine the kind of simulationand the variables are loaded, some precalculations have to be done to preparethe network before the simulation loop starts. The Node-B positions, userpositions, PDP for links in serving site, and the pathlosses for every users arecreated here. Some of these calculations have been enhanced in the improvedversion of the simulator developed in this thesis.

1. Node-B positions: the network consists of one central hexagonalcell and six or eighteen hexagons more around it, depending on thedesired number of interfering Node-Bs. The base stations are situatedin the center of each hexagon, and the serving Node-B is in the centralhexagon, so it is in the middle of the network. This configuration isdescribed as cellular layout 2 in [12]. Both configurations are shown inFigure 3.2.

2. Users positions: the simulator determines the users postions by usinga grid of points inside the main sector (it is considered the first sectorin the serving node), as we can see in Figure 3.3. We can observe inthis figure that some part in the bottom of the sector is missing. Itcorresponds to a 10 square meters area around the base station whichis considered as a limit where no users are positioned.

The users are positioned randomly in the grid positions. The number ofgrid points and users is detetermined in the HSDPA traffic simulationsettings. This is shown in Figure 3.4.

27

Figure 3.2: Network layout with 7 and 19 base stations.

Figure 3.3: Example of the grid positions generation in the serving cell.

3. Power Delay Profile (PDP): gives the intensity of a signal receivedthrough a multipath channel as a function of time delay. The timedelay is the difference in travel time between multipath arrivals. Inthis part, the PDP profile of a given ITU model and for all links in theserving site is generated.

28

Figure 3.4: Example of the users position in the serving cell.

4. Users pathloss: in this part, the pathlosses from all Node-Bs in thesimulated network structure to the given users are generated. As brieflyexplained in the channel modeling and channel settings, the radio prop-agation model used in the simulator considers three different elements:macro-scale pathloss, shadow fading, and small-scale fading with mul-tiple paths and no correlation in time, since this initial simulator issnapshot based with no correlations in between, [10]. The channel co-efficient which models the radio propragation between the base stationand the user equipment can thus be written as:

h(τ) = d · s ·L∑

t=1

√pl · fl · δ(τ − τl) (3.2)

where d denotes the deterministic pathloss, s the shadow fading, pl

and τl are the relative power and delay of the multipath components,fl represents L independent Rayleigh fading processes at fixed timeslots and δ denotes the Dirac function.

The macro-scale pathloss depends on the distance between the BS andthe UE, which is modelled accordingly to the COST231 model [13],and it depends also on the antenna gain pattern if a sectorized modelis used. The shadow fading is modelled by a lognormal random variablewith zero mean and σs = 8 dB, with no correlation in time. [10]

29

5. Noise power: settings needed to evaluate the noise power level.

3.3.3 Simulation loop

After the settings are loaded and the precalculations are done, the systemis prepared to start the simulation of multiple independent snapshots. Aconfiguration with three sectors model has been used, and the simulationsare done in the first sector of the serving Node-B, the so-called target sector.The results are getting after the average of all individual calculations.

Figure 3.5 illustrates an overview of the HSDPA calculations explained be-low, where we can observe that the SINR evaluation is the first step, thenthe CQI is mapped as function of the SINR, and using both, the block errorrate (BLER) and the transport block size (TBS), the HSDPA and Release 4data rate are estimated.

Figure 3.5: Overview of the basic steps in the simulator.

• HSDPA System Level Modelling:

Let us now go a little bit into detail onto the HSDPA system-levelmodeling.

30

1. SINR:

The Signal-to-noise-interference ratio (SINR) is used to evaluatethe channel quality as observed by the receiver, where a standardsingle antenna Rake is used because it is the most common inthe SISO HSDPA terminals. The SINR is calculated after Rake-combining and despreading for every users in the sector, accordingto the expression:

SINRu =

NF∑i=1

SF · PHS−DSCH

γ· |hi|2

Pintra,residual + Pinter + Pnoise

(3.3)

where u is ther user, SF represents the spreading factor, PHS−DSCH

denotes the power used for the HS-DSCH, Pintra,residual is theresidual intracell interference in the downlink, Pinter denotes thetransmitted interfering power from the neighbouring base stations,Pnoise is the noise power seen at the receiver, and γ represents thenumber of assigned spreading codes. The residual intracell inter-ference arriving at the receiver from the serving base station isgiven by [14], Pintra,residual = Pintra ·

∑Ll=1 |hl|2, where L is the to-

tal number of taps of the current realisation, denoted by hl, andPintra denotes the total power transmitted in the serving cell, [10].

The structure of the Rake reciver is implicitly shown in Equation3.3, since in the numerator the useful signal power is added up,which is cancelled out from the interference power in the denom-inator for all the NF available fingers consecutively. The receiverweights and the location of the fingers can be chosen perfectlysince perfect channel state information (CSI) is assumed at thereceiver. Accordingly, only the squared absolute values of thechannel coefficients (for each tap), |hi|2 occur in the equation. Itis also assumed that the transmission power of the HS-DSCH isdivided equally among all HS-PDSCH.

31

2. SINR to CQI:

The next step in the simulator is to calculate the CQI value for agiven SINR. This is done via a linear mapping, i.e.

CQI =

0, SINR ≤ −3.5

SINR[dB] + 3.5, −3.5 < SINR < 26.5

30, SINR ≥ 26.5

(3.4)

The CQI values (ranging from 0 to 30) are used by the link adap-tation algorithm at the Node-B. Each value represents a specificcombination of the transport block size (TBS), the number ofcodes and the modulation type. Thus, each value indicates themaximum TBS that can be correctly received with 90% probabi-lity.

3. CQI to TBS:

The bit-error/decoding performance, also called link-performancemodel, can be described once the channel quality is known. Thesimulator uses the link performance modelling for the transportformats of each mobile category class, given by the range of pos-sible CQI values. The tables for each category, used to determinethe number of codes as a function of the CQI and the TransportBlock Size (TBS) can be found in [11]. In the simulator we usedthe UE categories 1 to 6, and the table is shown in Chapter 5.

4. BLER:

The Block Error Ratio (BLER) is calculated according to an an-alytical model specified in [15]. Due to the snapshot based sim-ulation approach, no HARQ retransmissions gains are modelled.The BLER is considered directly in the evaluation of the through-put. According to [16], the BLER under AWGN conditions andutilizing a standard Rake receiver together with turbo coding, cananalitically be well approximated by:

32

BLER = [102SINR−1.03CQI+5.26√

3−log10(CQI) + 1]−10.7 (3.5)

5. HSDPA data rate:

After all the steps done in the simulation, the HSDPA user datarate can be calculated. It is important to note that the TBS de-notes the maximum amount of data that can be transmitted viathe network in one TTI of 2 ms to the UE without exceeding aBLER of 0.1 in average, and accordingly the HSDPA user datarate is calculated with Equation 3.6,

Ru = TBS · 1

2ms· (1−BLER) (3.6)

which is consecutively averaged over fading realizations and overall the individual snapshot simulations to get the average HSDPAcell and user data rate.

• Release 4:

As explained before, the investigations conducted so far were focusedon the prediction of the achievable HSDPA user data rates in depen-dence of a given Release 4 DCH load in the cell, and because of thatthe Release 4 traffic is modelled only coarsly. This prevents an exactevaluation, and the possibility of predicting the R’4 cell throughput atthe actual power level is very limited. The simulator just roughly esti-mates the total DCH cell throughput in order to be able to predict theoverall cell throughput as function of the power distribution. Furtherinformation can be found in [10].

3.3.4 Results

A short description of the simulation results concludes the inicial simulatorexplanation. The investigation identifies the optimun power setting in orderto maximize the total cell throughput.

33

Figure 3.6 shows the achievable average cell throughput on DCH and HS-DPA, and the total cell throughput (the sum of the throughput on DCH andHSDPA) versus the amount of power that is allocated for HSDPA traffic.The simulation parameters are as follows: Release 4 load of 20%, Node-Bdistance of 0.5 km, Pedestrian A model, 10 codes used for HS-DSCH and theuser category 6. The HSDPA cell throughput increases when more poweris allocated to HSDPA, while the DCH throughput decreases as there willbe less power for the transmission of these channels. It is observed that theHSDPA power allocation that maximizes the total cell throughput is around4 W, which results in the total cell throughput of 735 kbps, with around600 kbps being carried on HSDPA with PHS−DSCH = 6.078 W. Note thatthe HSDPA user data rate starts softly decreasing when the power reaches 6W because an increase in power results also in an increment of the intercellinterference. The R’99 cell data rate is always decreasing, due to the contin-uous increasing interference seen by the transmissions.

Figure 3.6: Average data rates with RNC power control of the HS-DSCH.

34

In practice, the optimal HSDPA power setting also depends on the offeredtraffic in the cell and the mixture of DCH- and HSDPA-capable UEs. Thegain in the cell throughput from introducing HSDPA mainly comes from ahigher spectral efficiency for the HS-DSCH over the DCH by using HybridARQ and adaptive modulation and coding, multiuser diversity gain fromusing fast PF scheduling, and better utilization of the available cell trans-mission power. Further information about the optimizations can be found in[10].

35

Chapter 4

Real Network Layouts

One of the main goals of the work was to extend the SISO HSDPA Simulatorin order to handle real network layout data that the Mobilkom Austria AGprovided by means of measured pathloss matrices of a HSDPA cluster, andparameters like the height of the antennas and the Nodes-B positions. Itwas necessary to convert this data in a useful form to be analyzable in thesimulator. Furthermore, the source-code was enabled to use the data filesafter the necessary modifications. The conversion of the real data matricesinto suitable information for the simulator, the overall process and necessarynew functionalities and modifications in the initial source-code for handlingthe new real data layout are described in this chapter.

4.1 Introduction

This chapter leads with the use of geographical information in mobile ra-dio communications from a propagation perspective. Some interest in theCOST231 project has focused on the types, resolution and accuracy of digi-tal terrain databases required for propagation modelling. Despite it is notpossible in this thesis to list the contents of the files (the accurate location,tilt and azimuth values of the antennas) since this is confidential informa-tion, a brief description from [17] about acquisition of geographical data isdepicted below.

Traditionally, geographical information has been obtained from paper maps.In the last decade increasing use has been made of high resolution remotesensing (aerial and satellite) for acquisition and of digital sto-rage and dis-tribution methods. The generation of Digital Elevation Models (DEM) andthe efficient and accurate extraction of radial data from them is reviewed in

36

[18].

Indoor propagation modelling possess the heaviest reliance on high resolu-tion geographical information. For urban propagation, it is essential to haveaccurate information at least about the average height of individual build-ings, when modelling larger cells or performing interference calculations andwhen terminals are operating close to roof-top height. The incorporationof information about clutter, particularly vegetation, is very important sincepropagation characteristics are quite sensitive to scatterers around terminals.

Aerial stereo photography provides a means of obtaining quite accurate dataon the heights and outlines (resolution of the order 1m) of building and ter-rain features, the location of vegetation, etc. The wealth of data can, inprinciple, be extracted from these sources. However, the extraction of datais a quite intensive labour.

Information about the building cladding, windows, etc. is more difficultto obtain. It appears that accurate geographical information of this naturemust be obtained using video cameras to capture data, for example. Wallproperties are particularly important for estimating building penetration.For outdoor propagation, in practice we may only need this level of detailaround potential BTS sites.

Considering the influence of database information on prediction accuracyit is noted in [19] that prediction errors can be attributed to database inac-curacies arising from the omission of vegetation data and the poor resolutionof terrain height data. Probably one of the most important causes of theeffect of database errors in field estimations is the effect of database inaccu-racies on model evaluation. Complex models which visually correspond tothe measured data often display a large error standard-deviation with res-pect to empirical models because of spatial offsets. These offsets can ariseif there are small database or measurement location errors. The existenceof these offsets between predictions and measurements is not necessarily anindication of a poor model. This problem is typically addressed by sepa-rately comparing the locally averaged model prediction and the statistics ofthe faster variations separately with the measured data.

37

4.2 Files structure and information

The package provided by Mobilkom AG consists on twelve ”prediction files”;each one corresponding to a specific Base Station situated in the city centerof Vienna. All of them represent a map of a central area of the city, with therespective Node-B situated in the center, and the generated pathloss aroundit for different distances.

The files are divided in header and pathlosses values. Each value of thepathloss that is read in the file corresponds to a pixel in the map. Establish-ing this correspondence is possible by using the information in the header ofthe files, since Mobilkom AG also provided the necessary information to man-age it, such as the lenght in bytes of all the interesting parameters. Figure4.1 shows the useful information that we can extract from the header.

Figure 4.1: Information extracted from the header of the files.

By using this information we are able to include the real pathloss in thesimulator as well as the Node-B position, grid position and users positions ac-cordingly to the extracted parameters. Besides the mentioned files, there aretwo files more including information about the base stations identifier andstreets where are situated, and the antennas identification (class, tilt, azi-muth and power). These files are not used so far in the simulation, although

38

it is possible to read them and extract information by means of specific func-tions that can be enabled in the load settings.

Once we have read all the information, we are able to represent the realpathloss extracted from the files. Figure 4.2 shows the read pathloss valuesfrom one of the prediction files. This figure gives an idea about the kind ofinformation attached in the files. By using the new implemented functions,we are able to convert ASCII data into suitable information that we can plot.The figure represents the variation of the pathloss with the distance in thecity of Vienna, with the Base Station located in the center of the map. Asobserved with the different colors in the figure, in the surroundings of theBase Station the pathloss is low and increasing with the distance.

Figure 4.2: Read pathloss from data files given by Mobilkom AG.

4.3 Network setting based on real layout data

The overall simulation process for setting a suitable network in the case thatthe real data is used is explained below. To achieve a global source code

39

which involves both functionalities (handling real data layout or theoricalpathloss calculation), the structure of the initial MATLAB code has beenkept. Besides some necessary new functions, the functionalities already ex-isting but adapted for the new purpose have been integrated in the originalsource code. Thus, the simulation mode can easily be chosen just by settingthe according variable at the beginning. Let us now list the main steps ofthe initial procedures of the simulator:

1. Reading data: as mentioned before, the option for reading the datafiles can be enabled in the load settings function. In such case, thefunction xf read out data works on the prediction files and extracts theheader of the files, with the information before explained, and the in-formation about the pathloss. In the case that reading also the two leftfiles is desirable, this function calls also read bts csv and read site csvin order to extract their respective information.

For the overall process, each parameter is stored in a structure af-ter calling these functions in order to be used next by the necessary,pertinent functions during all the simulation.

2. Node-B positions: following the same process than the initial sim-ulator, the function which locates the Node-Bs is called first. In thisfunction, the base stations are situated accordingly to the stored in-formation extracted from the pedestrian files, or it can be done in theoriginal way (hexagonal layout). Figure 4.3 shows the positions of thetwelve Node-Bs belong to the twelve prediction files. Notice that somecoordenates are taken by two stations; this is because the position isthe same but not the height.

3. Provisional grid positions: the next step is to generate the gridpositions where the users will be situated. As occured with the Node-Bs, the new grid is calculated in the original function accordingly to theactive option. If the real data are being used, we generate a provisionalgrid around the center of the global map, where the strongest Node-B(which is the one providing the lowest pathloss possible) is situated.This is explained graphically in next steps.

4. Pathloss: in the initial simulator, the pathloss was theorically calcu-lated distinguishing between deterministic pathloss as function of theUE-NodeB distance and as function of the antenna gain pattern (cal-culated accordingly to the COST 231 Hata model [13]), and stochasticpathloss, based on shadow and fast fading.

40

Figure 4.3: Node-Bs positions and respective prediction files.

In this step of the simulation, only the deterministic pathloss as func-tion of the antenna gain is calculated since the rest of the contributionsare included in the prediction files. It is generated for each one of thethree sectors as in the first version of the simulator.

If the pathloss is set to be read-in from the prediction files, we ex-tract it from the prediction files by using a function which extractsthe information according to the structure of the files. This functionreads and stores the pathloss for every pixels in each one of the twelvevirtual maps. Next, we extract the pathlosses corresponding to theprovisional grid created before, and it is summed up to the antennagain (in dB) calculated in order to obtain a final pathloss in all thegrid positions. Finally a 3D-matrix with the final pathloss for everygrid position, sector, and Node-B is generated.

5. Find main BS and sector: once we have the pathloss matrix, thenext step is to find the main sector and Node-B, which means the Node-B and sector where the lowest pathloss occured (within the region of

41

interest), and that way we can set a Node-B and pathloss informationfor the simulation process.

For this purpose, there is a function which reads the value of everyposition in the 3D-matrix, and extracts for every pixel which one ofthe twelve files (Nodes-B) and sector generates the minimum pathloss.This information is kept in a new matrix and then we read every po-sition of it and we count which Node-B and sector is repeated moretimes in that matrix (these will be the main BS and sector for thesimulation).

6. Generate final grid positions: this step fixes the grid positions thatcan be generated. The new grid consists of those positions of the pro-visional grid, where the lowest pathloss was generated by the main BS(target sector) simultaneously.

Figure 4.4 shows the situation so far, where the provisional grid inblue, the main Node-B and sector and finally the definitive grid ingreen (which defines the sector shape) are illustrated.

Figure 4.4: Provisional grid, main BS and sector shape.

7. Generate users positions: this is the last step in the new simulationprocess. The final goal of the process is setting up a complete network

42

suitable for simulation. Once the grid positions are defined, users arelocated randomly (uniformly) using the same functions as utilized inthe initial version. An example in which a zoom of the previous figureafter the users are generated (plotted in black) is shown in Figure 4.5.

Figure 4.5: Last step: user positions located randomly and uniformly withinthe limits of the generated sector (green points).

For a better visualization, Figure 4.6 shows a general overview of the networksetting process when the real data layouts are handled, summarizing the stepsbefore explained in a diagram.

43

Figure 4.6: Diagram of the overall handling real layouts data process.

44

Chapter 5

CQI Mapping Optimizations

This chapter explains more deeply the CQI mapping in HSDPA performance.Section 5.1 includes an introduction and the CQI basis. In Section 5.2 themapping proposal by the 3GPP specifications is explained, and Section 5.3describes the enhancements in the SISO HSDPA Simulator, where the objectof this part of the work and the achieved results are explained.

5.1 CQI Basis in HSDPA

The Channel Quality Indicator (CQI) is the key indicator for HSDPA down-link channel quality. The CQI value that the terminal reports is a functionof: multipath environment, terminal receiver type, ratio of the interferenceof the own base station compared with others and expected BTS HSDPApower availability. Thus, the value reported does not just correspond to thesignal to interference ratio (SIR) that the terminal is experiencing.

When a UE reports a particular CQI, it is reporting that under the cur-rent radio conditions (including the current power settings), the UE is ableto receive the reported CQI and lower CQI’s, at single-transmission BLERno greater than 0.1. User equipments report CQI and HARQ-ACK feed-back information to the Node-B through the uplink HS-DPCCH. The clearbenefit of the approach is that the solution defined will automatically ac-commodate the various possible receiver implementations and environmentvariations and, thus, gives an indication of the best data rates needed by theterminal to cope with the environment in question. This removes the needfrom the network end to have to consider the delay profile characteristicsof the cell/sector, [2]. Possible implementatios of the CQI evolution can befound in [20].

45

The HS-DPCCH carries uplink feedback signalling related to the downlinkHS-DSCH transmission. This signalling consists of HARQ-ACK and CQI asshown in Figure 5.1. Each 2-ms subframe, like those of the downlink physi-cal channels, consists of three slots, each with 2560 chips. The HARQ-ACKis carried in the first slot of the HS-DPCCH subframe and the CQI in thesecond and third slots. A system parameter k controls the CQI transmissionfrequency, and also control repetition for both slots is controlled by a sep-arate parameter. Repetition over multiple 2-ms periods is needed in somecases, like cell edge operation when the available power would not ensuresufficient quality for feedback reception, [2].

Figure 5.1: High-Speed Dedicated Physical Control Channel that carries theuplink.

Furthermore, we can explain some cases in which the CQI is used for theperformance of HSDPA, say:

• Link adaptation: as already said, the CQI is used for the dynamicHS-DSCH link adaptation, in addition to the scheduling decision. Theuser equipment sends every 2-ms TTI a CQI to the serving HS-DSCHcell on the uplink HS-DPCCH. By means of the CQI, the maximum

46

TBS that can be received correctly with at least 90% probability isindicated. The range of CQI values is 0-30, and each step correspondsappoximately to a 1-dB step in HS-DSCH SINR, [9]. The HS-DSCHlink adaptation algorithm at the Node B adjusts the transmit bit rateon the HS-DSCH every TTI when a user is scheduled for transmission.Ideally, the HS-DSCH transmit bit rate should be adjusted as a functionof the per-TTI HS-DSCH signal-to-interference plus noise ratio (SINR)experienced at the user end. Figure 5.2 illustrates the general principleHS-DSCH link adaptation, [2].

Figure 5.2: HS-DSCH link adaptation principle: (1) the UE reports low-quality channel information and the Node B allocates a low bit rate; (2) theUE reports high-quality channel information and the Node B allocates a highbit rate, [2].

As shown in Figure 5.3, several factors contribute to variance of HS-DSCH SINR although the HS-DSCH transmit power is supossed con-stant. Due to the transmission of power controlled DCHs, the totaltransmit power from the serving HS-DSCH cell is time variant, likethe downlink radio channel in case than the user is moving, and theother cell interference experienced at the user end is also time variant.The signaling delay on the uplink is used for the case that the SINRat the user has changed from the time when the CQI was estimateduntil the actual transmission on the HS-DSCH. In such a case, the useof the recommended transmission format by the CQI report for the

47

HS-DSCH transmission does not necessarily guarantee that the BLEPis lower than 0.1, [4].

Figure 5.3: Block diagram showing the received signal at the HSDPA userand report of the CQI to the serving HS-DSCH cell, [4].

A simple link adaptation algorithm would directly follow the CQI valuesreported by the UE, but there may be a need to adjust the UE-reportedCQI by adding an offset because the HS-DSCH transmit power fromthe Node B to the user might be different from the assumed HS-DSCHtransmit power by the UE at the time it derived the CQI report; be-sides, the UE assumes a received HS-PDSCH power level PHS−PDSCH

derived from the observed P/S-CPICH power level PCPICH accordingto PHS−PDSCH = PCPICH + G, where G is a power offset parametersignalled to the UE via RRC signalling from the RNC. The effect offeedback delays on link adaptation performance has previously been ad-dressed in [21], [22]. In order to further adjust the CQI index receivedfrom a user before applying it for adjustment of the HS-DSCH transmis-sion format, these studies indicate a need for an outer loop HS-DSCHlink adaptation algorithm, which can be based on ACKs/NACKs frompast transmissions. The algorithm adjusts the offset values to ascer-tain the average targetted retransmission probability. If too many re-transmissions occur, an unnecessary delay is added, whereas too few

48

retransmissions indicate that the transport block sizes used are notlarge enough, unnecessarily lowering throughput, [2].

• HS-SCCH power control: sufficient power should be allocated to trans-mission of the HS-SCCH to ensure reliable reception quality of the HS-SCCH, which is important since the transport block on the HS-DSCHcan only be decoded if the HS-SCCH has been correctly received. Fur-thermore, reduce HS-SCCH transmission power is also desirable, inorder to decrease the interference levels in the network. Hence, it isgenerally recommended to control the HS-SCCH power every TTI. The3GPP specifications do not explicitly specify any power control mecha-nism for the HS-SCCH. The HS-SCCH transmit power can be adjustedrelative to the transmit power of the associated downlink DPCCH, oras a function of the CQI report received from the user. This is possibleif there is an internal table at the Node B expressing a power offset be-tween each CQI index and the required HS-SCCH power. In both casesit is possible to implement a pseudo closed loop power control schemefor the HS-SCCH, relying on either feedback information from the userabout the reception quality of the associated DPCH or the HS-DSCH(CQI). Common to both approaches is the Node B need for a prioriknowledge of a power offset parameter before it can adjust HS-SCCHtransmit power as a function of either DPCCH power or CQI.

5.2 CQI mapping proposal

The CQI from the user expresses the recommended transmission format,signaling via a CQI index n ε [0-30], where the CQI integer index number ncan be regarded as a pointer to a vector, i.e.,

fCQI(n) = [An,Mn, Xn,4n] ∀ n ε [0− 30] (5.1)

where An is the recommended number of bits in a TTI, (TBS), Mn denotesthe recommended number of HS-PDSCH codes, Xn ε [4QAM, 16QAM] speci-fies if the user recommends the Node-B to use 4QAM or 16QAM modulation,and 4n is a recommended power offset for the nth CQI index. Notice thatthe CQI mapping tables for function fCQI(n) specified by 3GPP in [23], areorganized, so that An ≤ An+1. The CQI index estimated by the user satisfies

n = arg maxn{An|BLEP < 0.1} (5.2)

in the 2-ms interval that ends one slot before the CQI is sent, where BLEP isthe per-TTI block error probability. Note that the CQI index in (5.2) depends

49

on the experienced HS-DSCH user SINR since the BLEP for different valuesof An depends on the SINR. The received CQI index n is adjusted by thescaling factor A, so CQI index n is used instead. The latter one is possible byassuming that the Node-B has a table with a priori knowledge of the relativeHS-DSCH SINR difference in decibels between the entries in the CQI table,which is denoted by fSINR(n). Note that the CQI tables have the followingapproximate property: fSINR(n) ' 1·n dB ∀ n ε [0−30], i.e., the resolution ofthe CQI mapping table is approximately 1 dB. Given this a priori knowledgeand scaling factor A, the used CQI index for HS-DSCH link adaptation isobtained as follows, [4]

n = arg minn{fSINR(n)− (fSINR(n)− An)} (5.3)

5.2.1 CQI Table

The CQI is based on the 5-bit CQI mapping table shown in Table 5.1. Foreach CQI, the table lists an associated turbo-encoder information word sizeexpressed as Transport Block Size (TBS), modulation type and number ofHS-PDSCH. These parameters describe the reference, single-transmissionmodulation and coding scheme associated with each CQI value. It is as-sumed that the reference turbo-codeword comprises all of the systematicbits, plus the parity bits identified by assuming that the number of soft met-ric locations in the UE is at least equal to the number of unpunctured codedbits before rate matching. The CQI table has been designed to uniformlyand linearly quantize the UE SNR observation space, and offers Node-B andUE manufacturers flexibility in respectively implementing the Scheduler andthe UE signal processing required to generate the CQI report, [24].

5.3 CQI optimizations in the Simulator

One of the main goals of the work is to optimizate the CQI mapping in orderto maximize the overall cell throughput. In the initial version of the simu-lator, the CQI mapping was implemented accordingly to Equation 3.4. Asmentioned, for the purpose of HS-DSCH link adaptation, the UE periodi-cally sends a CQI to the serving HS-DSCH cell on the uplink HS-DPCCH.The CQI indicates the maximum transport block size that can be receivedcorrectly with at least 90% probability. This information is signalled via aCQI index in the range from 0 to 30, where each step corresponds appox-imately to a 1-dB step in HS-DSCH SINR [9]. A simple link adaptationalgorithm would directly follow the CQI values reported by the UE. How-ever, there may be a need to adjust the UE-reported CQI by adding an offset.

50

CQI value Transport Block Size Nr of HS-PDSCH Modulation0 N/A 4QAM1 137 1 4QAM2 173 1 4QAM3 233 1 4QAM4 317 1 4QAM5 377 1 4QAM6 461 1 4QAM7 650 2 4QAM8 792 2 4QAM9 931 2 4QAM10 1262 3 4QAM11 1483 3 4QAM12 1742 3 4QAM13 2279 4 4QAM14 2583 4 4QAM15 3319 5 4QAM16 3565 5 16QAM17 4189 5 16QAM18 4664 5 16QAM19 5287 5 16QAM20 5887 5 16QAM21 6554 5 16QAM22 7168 5 16QAM23 7168 5 16QAM24 7168 5 16QAM25 7168 5 16QAM26 7168 5 16QAM27 7168 5 16QAM28 7168 5 16QAM29 7168 5 16QAM30 7168 5 16QAM

Table 5.1: CQI mapping table for UE categories 1 to 6, [11].

The CQI mapping is a very important issue for the optimization of the net-work and HSDPA performace. A poorly implemented CQI algorithm on onemodel of UE can have a significant adverse impact on the efficient allocation

51

of network resources, which will affect all other UEs on the network. Whenthe terminal is close to the base station and assumes high HS-DSCH powerallocation (based on the value given by the network), a high CQI value isreported. Respectively, when the terminal is closer to the cell edge, thenthe reported CQI is much lower, especially if the expected HSDPA Node-Bpower allocation is low as well, [2].

The initial formula for the mapping used previously in the Simulator al-ready explained, has been modified. For the purpose of achieving the newmapping and visualization, we represent the CQI values as function of theSINR, and the possible modifications in the original formula are:

• Slope: by adding a scale factor is possible to change the slope of thestraight line. It is important to find an optimum slope, otherwise theresults could be very negative.

• Shift: the shift value is used like an offset, indicating in which valueof the SINR the CQI starts to have values different to cero. As in thecase of the slope, the selection of this offset is also important for theoptimization.

The new formula results as follows:

CQI = (SINR− shift) · slope+ 3.5, (5.4)

where the SINR (in dB), the slope and the shift are inputs of the func-tion and the output is the CQI value. The slope and shift values are chosenin the loading settings, and also a range of values is possible, in order toevaluate the different results. Note that the CQI values under 0 and over 30are mapping like 0 and 30 respectively. Figure 5.4 illustrates an example inwhich two possibilities are shown.

5.3.1 Source Code Enhancements

Concerning to the new CQI evaluation, some changes in the initial sourcecode have been necessary. Thus, the main function in the Simulator andthe function which calculates one SISO HSDPA network snapshot have beenmodificated in order to handle the new functionalities. Moreover, the newnecessary functions for the purpose of CQI mapping according to maximize

52

Figure 5.4: CQI mapping.

the overall cell throughput by changing the slope and shift have been imple-mented.

Figure 5.5 illustrates a coarse view of the simulation process in which thenew functionalities have been included. The first step is to load the sim-ulator settings, where it is possible to fix the range of values for the slopeand shift we want to make the simulation with. These variables are found inthe UE settings. Second, the precalculations are done by means of the mainfunction. Then the simulation loop starts with a determined value of theslope and the shift, the function for the SINR to CQI mapping is called andall the calculations are done, so finally there is a throughput value for thoseshift and slope values. The simulation is done once again and it lasts whilethe range of values is not finished. At the end of the process the throughputas function of the slope and shift is achieved and represented in a figure.

5.3.2 Consequences

For the purpose of the optimization of the network and HSDPA performace,the CQI needs to be mapped in the most efficient way. In the model of theUE, the implementation of a poorly CQI algorithm can result in an impor-

53

Figure 5.5: Coarse view of the simulation including slope and shift.

tant adverse impact on the efficient allocation of network resources, whichwill affect every user equipments on the network. Remember the object ofthe CQI reporting, when the terminal is close to the base station and assumeshigh HS-DSCH power allocation based on the value given by the network, ahigh CQI value is reported and respectively, when the UE is closer to the celledge and most of the interference comes from other cells, then the reportedCQI is much lower. Because of this great impact, during the work manydifferent values for the slope and shift have been used to evaluate the bestoption, by using the throughput evaluation. It is important to observe howthe choice of wrong values can lead to significant adverse consequences. Withthis object, two examples with different extreme slope and shift values areshown. Figure 5.6 illustrates the effect of choosing a very low or high valueof the slope. Both of the choices have unfavourable consequences in the CQIvalue assignment. In the case that the slope is very high, like showing thered line, the range fo SINR values for that the CQI is not zero or thirty isvery short; consequently, many users are getting a CQI value of zero, and inthe other side a very high CQI value is assigned for many users with a notvery good SINR. This is not optimum because the users are getting moreCQI than they are able to handle, their SINR is not so high so the BLERwill increase and the throughput decrease. On the other hand, the blue lineshows the effects of a very low slope, this is a more conservative situation,not many users are getting the highest CQI value although they present agood SINR value, and the CQI values are more distributed for every users.

54

Figure 5.6: Different values of the slope for the CQI mapping.

Figure 5.7: Different values of the shift for the CQI mapping.

In Figure 5.7 the effect of two extreme shift values is illustrated. As before,the choice is important for the mapping optimization and the consequences

55

are the same. A too low shift value results in users with a low SINR gettingCQI values different to zero, when it is impossible they receive any through-put because of the low signal quality; and respectively if the shift value istoo high, the users need to have also high SINR values to get a good channelquality indicator, this leads in the situation that user equipments with goodsignal qualities are loosing their capacity to obtain a good service.

5.3.3 Results

Once the source code had been enhanced and adapted for the new CQI map-ping, and the necessary functions were implemented, we may start with theinvestigations for the optimization of the final throughput. First of all it wasnecessary to find the optimum range of values for the slope and shift in whichwe can achieve high values of the throughput, and once the two ranges arefound, we have to find the optimum values.

Figures 5.8 and 5.9 illustrate two first evaluations of the effect that differentvalues for the slope and the shift have in the overall cell throughput achievedby the SISO HSDPA Simulator. For the evaluations, one of the two parame-ters have been fixed and the other one has a big range of values to observe theeffect of each single variable. The results show how important is the choice ofthese values, as explained in subsection 5.3.2. The throghput values in [bps]are calculated for different slopes and shifts, and we can observe a bit peakin the curve in values around 1 for the slope and 0 in case of the shift. Whenusing very distant values there is a big drop off in the evaluated throughput.

56

Figure 5.8: Throughput as function of the slope value with shift = 0.

Figure 5.9: Throughput as function of the shift value with slope = 1.

Finally, we obtain a 3-dimensional figure in which the throughput asfunction of the two parameters is represented. Thus, we can observe theeffect of the two variables changing during the simulation, and we can findan optimum value for the slope and shift which provide the optimum overallcell throughput. For this figure, after finding the suitable delimited range of

57

slope and shift values, the investigations have been done by using low stepsfrom the minimun to the maximum value, so that it is not probable to missany optimum couple of values.

Figure 5.10: Throughput as function of the slope and shift.

We concludes the chapter with Figure 5.10, which shows how the through-put is reaching different values and the maximum is achieved for a slope valueof 1 and a shift value of -0.4, according to the results that we observed in thetwo previously figures.

58

Chapter 6

Miscellaneous

This chapter includes other changes in the initial simulator and investigationsduring this thesis that really do not fit into the previous chapters. Section 6.1explains new scenarios for the positioning of the users. Section 6.2 includessome improvements for calculating the macro-scale pathloss by consideringthe three antenna radiations of every Nodes-B, and further enhancementssuch as the obtention of the outage users and plots of the average throughputare treated in 6.3.

6.1 New scenarios

In the new version of the simulator, it is possible to choose four differentsituations concerning the users positions. By using a new variable in theload settings function, the type of simulation is chosen, and accordingly theusers positions are established in the function xf User positions.m. Figure6.1 illustrates the four possibilities.

Figure 6.2 illustrates the first two scenarios in greater detail. Snapshot meansthe users are uniformly and randomly situated according to the grid posi-tions, whereas exhaustive snapshot refers the user positions are unformlysituated but fixed at all possible grid positions. It is importat to rememberthat the users are situated in the main sector of the serving Node-B.

In Figure 6.3 we can observe an explample with the remaining two possi-bilities. Fixed angle and fixed distance denote that the angle between theuser and the radiation directions or the distance to the main Node-B, re-spectively, are fixed in the load settings. In this explamples, the angle isfixed at 90o, which means that the users are situated in the same direction

59

Figure 6.1: Four possible user positions configuration.

Figure 6.2: Snapshot and exhaustive snapshot scenarios.

than the main radiation, and the distance is changing. In the other scenario,the distance between the users and the Node-B is 100 meters, but the anglechanges.

60

Figure 6.3: Fixed angle and fixed distance scenarios.

6.2 More efficient implementation of the

pathloss generation

In the initial version of the simulator, the macro-scale pathloss for a givenuser was calculated by using the three antennas in the main Node-B andthe most influent antenna of the neighbors nodes (evaluated by geometricalmeans). In the new function, based on the initial function xf user pathloss v2,we calculate the macro-scale pathloss by considering the three antenna ra-diations of every Nodes-B, including the principal and the neighboring, inorder to sum the overall cell interference for each user. Depending on theangle between the user and the antenna, the signal can strongly reach theuser equipment or not. For implementing this, it was necessary to create afunction which calculates the antenna gain for each user for every node andevery sector.

Figure 6.4 illustrates the network scheme used in the simulation, althoughan scenario with 7 cells is also possible. We can observe the user position inthe target sector, the main base station in the center and the three radiationdirections (90o, 210o and 330o).

By utilizing the new scenarios it is possible to represent the user pathlossas function of the distance with fixed angle, and as function of the anglewith fixed distance. Figure 6.5 illustrates how the deterministic pathloss be-

61

Figure 6.4: Network with 19 base stations and 3 sectors model.

haves when increasing the distance between the user and the Node-B. Whenthe user is near the BS, there is a low pathloss, and it is increasing withthe distance. In Figure 6.6 the pathloss as function of the angle in degreesis represented. As before, if the UE is situated in such position that theantenna is radiating directly to it (which is the case when the angle is 0o),the macro-scale pathloss is very low, whereas when the angle is changing thepathloss increases.

6.3 Further enhancements

Some other functionalities have been implemented in order to have an en-hanced comprehension and visualization of some features developed for thesimulator, namely:

6.3.1 Outage users

A new implemented function allows us to obtain the percentage of outageHSDPA users for a fixed power, so that we can calculate the cumulative dis-

62

Figure 6.5: Variation of the user pathloss with the distance.

Figure 6.6: Variation of the user pathloss with the angle.

tribution function. Besides, it’s possible to plot the number of outage usersas function of the power. As explained in [24], if the UE cannot support theminimum CQI (CQI = 1) at BLER = 0.1, then the UE reports CQI=0 whichindicates an ”Out of range” (OOR) condition to the Node B.

As we can observe in Figure 6.7 when a low power is used, the number of

63

outage users is very high; then, as the power is increasing, there is a big drop-off, and finally for the highest values of the power, the intercell interferencesare also increasing and a small increment in the outage users is produced.

Figure 6.7: HSDPA outage users as function of the power.

6.3.2 Throughput figures

The key performance metric for HSDPA is data throughput, which is highlydependent on the multipath and interference environment experienced by theUE. As already said, adaptive modulation and coding allows HSDPA usersnear the base station to convert their high signal-to-noise ratio into higherdata throughput. Since the scheduler, attempting to maximize average cellthroughput, will in general favor such users, they significantly determine theoverall cell throughput [14]. In the simulator, the overall cell throughput iscalculated by

Throughput = TBS · (1−BLER), (6.1)

thus, the average rate of messages succesfully delivered over the channelis estimated by means of the multiplication of the transport block size andthe probability of having no errors. This throughput (in terms of bits thatwent through correctly) is consecutively averaged over 500 realizations per

64

user.

In the new version of the simulator, we can plot the throughput of everyusers by means of a new function. The new user positioning explained be-fore, allows us to represent the throughput over the different scenarios. Thus,we can observe the influence of the user positions, focusing our interest inthe fixed angle and fixed distance scenarios.

Figure 6.8 shows the average throughput as function of the angle when everyuser has the same distance to the serving Node-B. Accordingly the effect ofthe angle between the direction radiation and the user-position can be ob-served. The users positioned around a 90o angle are situated in the samedirection that the main radiation of the antenna, so the achieved through-put is the highest possible, while as the angle is increasing or decreasing thethroughput decrements until finally it is completely zero. Furthermore, wecan observe in the figure the calculated average throughput under two dif-ferent situations in order to apreciate the influence of the distance: picturein the left corresponds to an scenario in which a distance of 50 meters tothe BS is fixed, so due to the closeness the throughput reachs values of 2.5Mbps. Picture in the right represents the opposite situation, where the dis-tance is 250 meters so the throughput is much lower, reaching a maximumvalue around 80 Kbps.

Table 6.1 specifies some of the main simulation settings used for the gen-eration of Figures 6.8 and 6.9.

Simulation settingsNumber of R’99 users 5Number of HSDPA users 40Node-B distance 0.25 KmCodes used for HS-DSCH 10Maximum Node-B power 20 WHSDPA range power 0-14 W

Table 6.1: Simulation settings used in the generation of Figures 6.8 and 6.9

In Figure 6.9 the throughput of the users is illustrated for the fixed anglescenario. Thus, the throughput as function of the distance can be observed,

65

Figure 6.8: Throughput as a function of the angle with two fixed BS - userpositions distances (50 meters in the figure of the left and 250 meters in theright).

and as before, the influence of the distance between the users and the mainNode-B is evaluated. We can observe how the throughput is higher aroundthe small distances and lower for langer distances. Here we also evaluate thethroughput of two different scenarios. The picture in the left side has beenobtained by using a fixed angle of 80o, so it is possible to achieve high valueslike a throughput of 2.5 Mbps, as before. As a difference, in the right sidethe angle used is 40o, so the antenna does not radiate with a proper directionand it is only possible to achieve maximum throughput values around 100Kbps.

66

Figure 6.9: Throughput as a function of the distances with two fixed anglesbetween the users and the BS (80o in the figure of the left and 40o in theright).

67

Chapter 7

Conclusions

This bachelor thesis has investigated the HSDPA throughput performance ina mixed traffic network. HSDPA represents an evolution of UMTS/WCDMA,as a Third Generation mobile telephony communications protocol. The goalof 3G systems is to provide users with high quality video and images forperson-to-person communication according to the new multimedia services.With the ever increasing commercial application of 3G worldwide, subscribersare showing a huge interest in high-speed data service. As a result, an expo-nential increase in data services and traffic is occurring. HSDPA can greatlyimprove the efficiency and data transmission rate of the radio network anddecrease the transmission delay and cost. Besides, HSDPA technology allowsmore subscribers to enjoy high-speed data service at the same time. In sum,HSDPA means better service quality for subscribers.

For our purpose we focused the investigation in two enhancements. First,the extension of the SISO HSDPA Simulator in order to handle real networklayout data. Second, the CQI mapping optimizations in order to maximizethe overall cell throughput.

This concluding chapter a summarises this bachelor thesis as follows: InSection 7.1 a brief summary of the HSDPA simulator, especially the newfunctionalities implemented during the thesis, is presented. Section 7.2 in-cludes a summary with the results obtained with the new developed func-tionalities, and Section 7.3 concludes this bachelor thesis describing somepossible future enhancements in the simulator.

68

7.1 Simulator enhancement

This thesis has consisted in the development of new functionalities in theMATLAB simulator created by the Institute of Communications and Radio-Frequency of the Vienna University of Technology. These functionalities arenecessary for the investigation of the main enhancements explained before.The initial purpose of the simulator was to evaluate the performance of amixed network in which both R’99 and HSDPA traffics are simulated, inorder to find the optimum parameters to obtain the maximum overall cellthroughput.

One of the most important extensions in the SISO HSDPA simulator was theinplementation of new functions that allow us for the handling of real networklayout data that the Mobilkom Austria AG provided by means of measuredpath-loss matrices of an HSDPA cluster. This data have been converted inan useful form to be analyzable in the simulator, and the source-code hasbeen enabled to use the data files after the necessary modifications.

The second extension concerns the CQI values reported by the user equip-ment. It consisted in the optimization of the CQI mapping in order to max-imize the overall cell throughput. This mapping is important since a poorlyalgorithm on one model of UE can have an adverse effect on the efficientallocation of network resources. The initial formula used previously in thesimulator has been modified. For the purpose of achieving the new mappingand visualization, we have represented the CQI values as function of theSINR, with two new parameters: a scale factor to change the slope of themapping, and a shift value used like an offset.

Finally, some other changes in the initial simulator have been implemented,namely:

• The possibility of choosing four different situations concerning the userspositions: snapshot, fixed distance, fixed angle and exhaustive snap-shot. The type of simulation is chosen in the settings and accordinglythe users positions are established. This enhances the simulation in thesense that it is possible to analize the user pathloss and throughput forevery scenarios.

• More efficient implementation of the macro-scale pathloss generation,by considering the three antenna radiations of every Nodes-B, includ-ing the principal and the neighboring, in order to sum the overall cellinterference for each user.

69

• HSDPA outage users calculation as function of the power.

• Analysis of the throughput of every users over the different scenariosexplained before, in order to observe the influence of the user positions.

7.2 Results

This section presents a summary of the results obtained with the new devel-oped functionalities in the simulator. First, table 7.1 shows the optimizationsin the new CQI mapping in order to maximize the overall cell throughput,by changing two new parameters: slope and shift. The tables present somevalues of the slope and shift that cause significative throughputs. As a con-clusion, the optimum pair of values are pointed out in yellow.

Shift = 0 Slope = 1

Slope Throughput Shift Throughput

0.5 0.6 Mbps -2 0.1 Mbps

0.75 1 Mbps -1 0.5 Mbps

1 1.5 Mbps -0.4 0.7 Mbps

1.2 0.1 Mbps 2 0.55 Mbps

Table 7.1: Variation of the throughput as function of the slope and shift.

Table 7.2 illustrates the percentage of outage HSDPA users as functionof the power. The number of outage users decreases with the increment ofthe power, except for the highest values of the power, when the intercell in-terferences are also increasing and a small increment in the outage users isproduced.

70

Power Outage users0.2 W 45%1 W 10%4 - 11 W 5%11 - 14 W 6%

Table 7.2: Variation of the outage users as function of the range of HSDPApower.

In Tables 7.3 and 7.4 some values of the macro-scale pathloss for differ-ent distances and angles between the user and the main Node-B are shown,and Table 7.5 presents some significative values of the throughput calculatedwithin different scenarios of the users possitions. We can observe the greatinfluence of the user position in the pathloss and throughput generation.

Angle Pathloss0o -95 dB30o -100 dB60o -110 dB100o-180o -133 dB

Table 7.3: Variation of the pathloss as function of the angle between the userand the Node-B.

Distance Pathloss20 m -55 dB100 m -78 dB250 m -93 dB500 m -103 dB

Table 7.4: Variation of the pathloss as function of the distance between theuser and the Node-B.

71

``````````````OrientationDistance

Close Medium Far

Good 2.5 Mbps 1.5 Mbps 90 KbpsMedium 2 Mbps 900 Kpbs 5 Kpbs

Bad 100 Kbps 5 Kpbs 0 Kpbs

Table 7.5: Variation of the throughput as function of the angle of orientationand distance between the user and the Node-B.

7.3 Future enhancements

During the development of this bachelor thesis only some possibilities of thenew simulator have been explored. A new CQI mapping and the handlingof real data network layouts were introduced but more investigations canbe developed, like some HSDPA technologies that still are not incorporated.Besides, a wide range of simulations is possible and the simulator is more com-plete and realistic thanks to the implementation of a time-based functionalityand the HSDPA scheduling developed in other thesis. Further system levelinvestigations on the simulator have already started, [25], [26], [27], [28], [29].

Besides these enhancements, let’s remember HSDPA Evolved or HSPA+,which is the next step in HSDPA and provides HSPA data rates up to 25Mbps on the downlink and 11 Mbps on the uplink with MIMO technologiesand higher order modulation. It also introduces an optional all-IP archi-tecture for the network where base stations are directly connected to theinternet. Australia’s Next G was the first network to enable some featuresof HSPA+ in June 2008, [30]. In this context, it could be interesting theinvestigation for adapting the simulator to new HSPA+ technologies.

72

Bibliography

[1] HSPA Mobile Broadband Today (http://www.gsmworld.com)

[2] Harri Holma and Antti Toskala. HSDPA/HSUPA for UMTS: High SpeedRadio Access for Mobile Communications. John Wiley & Sons. June2006.

[3] HSDPA Channel Coding Implementation Issues.(http://www.altera.com).

[4] Klaus Ingemann Pedersen, Member, IEEE, Frank Frederiksen, TroelsEmil Kolding, Member, IEEE, Tako Freerk Lootsma, and Preben E. Mo-gensen, Member, IEEE ,”Performance of High-Speed Downlink PacketAccess in Coexistence With Dedicated Channels”, in IEEE TransactionsOn Vehicular Technology, Vol. 56, No. 3, May 2007.

[5] Chase D. Code Combining - A Maximum - Likelihood Decoding Ap-proach for Combining an Arbitrary Number of Noise Packets. Commu-nications, IEEE Transactions on, Volume 33, May 1985. pp. 441-448.

[6] 3GPP. Technical Specification Group Radio Access Network. HighSpeed Downlink Packet Access(HSDPA); Overall Description. (3GPPTS 25.308 version 5.4.0).

[7] Pablo Amiegeiras Gutierrez. ”Packet Scheduling And Quality of Servicein HSDPA”. Ph. D. Thesis. October 2003.

[8] 3GPP. Technical Specification Group Radio Access Network. UTRANIub Interface NBAP signalling. (3GPP TR 25.433 version 5.5.0).

[9] 3GPP Technical Specification, TS 25.214, Physical Layer Procedures(FDD), version 6.4.0.

[10] Martin Wrulich, Werner Weiler and Markus Rupp. ”HSDPA Perfor-mance in a Mixed Traffic Network”. Proc. 2008 IEEE 67th VehicularTechnology Conference. Singapore. May 2008.

73

[11] 3GPP. Technical Specification Group Radio Access Network. Physicallayer procedures (FDD)(Release 7). (3GPP TS 25.214 version 7.6.0).

[12] Members of 3GPP, ”Technical specification group radio access network;multiple input multiple output in utra,” 3GPP, Tech. Rep. v7.0.0, Mar.2007.

[13] Members of COST 231. ”Digital mobile radio towards future generationsystems - COST 231 final report”. COST, Tech. Reo., 1999.

[14] A.Seeger, M.Sikora, and A.Klein, ”Variable orthogonality factor: a sim-ple interface between link and system level simulation for high speeddownlink packet access,” in Proc. of the IEEE Vehicular TechnologyConference 2003, Oct. 2003, pp. 2531-2534.

[15] F.Brouwer, I. de Bruin, J.Silva, N.Souto, F.Cercas, and A.Correia. ”Us-age of link-level performance indicators for HSDPA network-level simu-lations in E-UMTS”, in Proc. of the Eight Internationsl Symposium onSpread Spectrum Techniques and Applications, Sept. 2004.

[16] R.Litjens, ”HSDPA flow level performance and the impact of termi-nal mobility,” in Proc. of the Wireless Communications and NetworkingConference, Mar. 2005, pp. 1657-1663.

[17] Dieter J. Cichon and Thomas Kurner, ”Propagation Prediction Models”,Chapter 4, COST 1998.

[18] D. McGirl, P. C. Fannin, ”Digital Elevation Modelling,” COST 231TD(94)75, Prague, Czech Republic, 1994.

[19] B. E. Gschwendtner, ”Practical investigation using ray optical predictiontechniques in microcells,” COST 231 TD(94)127, 1994

[20] Qun Hou, Dexiu Huang, ”Channel Quality Indication (CQI) Applicationin HSDPA Simulation”, in IEEE Wireless Communications, Networkingand Mobile Computing, 2007. WiCom 2007. International Conferenceon, pp. 1200-1203, Sept.

[21] H. Vishwanathan (1999), ”Capacity of Markov channels with receiverCSI and delayed feedback”, IEEE Trans. on Information Theory, 45(2),pp. 761-771, March.

[22] A. Boariu (2003), ”Effect of delay commands on adaptive modulationscheme in fading channels”, IEEE Proc. Vehicular Technology Confer-ence (VTC), October.

74

[23] 3rd Generation Partnership Project, Physical Layer Procedures, (Mar.2004). TS 25.214, Version 6.1.0. [Online]. Available: www.3gpp.org

[24] TSG-RAN Working Group 4, ”Revised HSDPA CQI proposal,” 3GPP,Tech. Rep. R4-020612, Apr. 2002.

[25] M. Wrulich, C. Mehlfuhrer and M. Rupp. ”Interference Aware MMSEEqualization for MIMO TxAA”. Proc. of IEEE 3rd International Sym-posium on Communications, Control and Signal Processing (ISCCSP).1585-1589. 2008.

[26] C. Mehlfuhrer, M. Wrulich and M. Rupp. ”Intra-cell Interference AwareEqualization for TxAA HSDPA”. Proc. of IEEE 3rd International Sym-posium on Wireless Pervasive Computing. 406-409. 2008

[27] C. Mehlfuhrer, S. Caban, M. Wrulich and M. Rupp. ”Joint ThroughputOptimized CQI and Precoding Weight Calculation for MIMO HSDPA”.Conference Record of the Fourtysecond Asilomar on Signals, Systemsand Computers. Oct. 2008.

[28] M. Wrulich and M. Rupp. ”Efficient Link Measurement Model for Sys-tem Level Simulations of Alamouti Encoded MIMO HSDPA Transmis-sions”. International (ITG) Workshop on Smart Antennas (WSA). Feb.2008.

[29] M. Wrulich, S. Eder, I. Viering and M. Rupp. ”Efficient Link-to-SystemLevel Model for MIMO HSDPA”. Proc. of the 4th IEEE BroadbandWireless Access Workshop. 2008.

[30] Ericsson world-first in delivering innovative 3G technology to Telstra(http://www.ericsson.com)

75