Uplink Channel Dependent Scheduling for Future Cellular Systems

(1)

Institutionen för systemteknik

Department of Electrical Engineering

Examensarbete

Uplink Channel Dependent Scheduling for Future

Cellular Systems

Examensarbete utfört i Kommunikationssystem vid Tekniska högskolan i Linköping

av

Kristina Jersenius

LITH-ISY-EX--06/3905--SE

Linköping 2007

Department of Electrical Engineering Linköpings tekniska högskola

Linköpings universitet Linköpings universitet

(2)

(3)

Uplink Channel Dependent Scheduling for Future

Cellular Systems

Examensarbete utfört i Kommunikationssystem

vid Tekniska högskolan i Linköping

av

Kristina Jersenius

LITH-ISY-EX--06/3905--SE

Handledare: David Törnqvist

isy, Linköpings universitet Eva Englund

Ericsson Research

Examinator: Fredrik Gunnarsson

isy, Linköpings universitet

(4)

(5)

Avdelning, Institution

Division, Department

Division of Automatic Control Department of Electrical Engineering Linköpings universitet

SE-581 83 Linköping, Sweden

Datum Date 2007-01-12 Språk Language ¤ Svenska/Swedish ¤ Engelska/English ¤ £ Rapporttyp Report category ¤ Licentiatavhandling ¤ Examensarbete ¤ C-uppsats ¤ D-uppsats ¤ Övrig rapport ¤ £

URL för elektronisk version

http://www.control.isy.liu.se http://www.ep.liu.se/2006/3905 ISBN — ISRN LITH-ISY-EX--06/3905--SE

Serietitel och serienummer

Title of series, numbering

ISSN

—

Titel

Title

Kanalberoende upplänksschemaläggning för framtida cellulära system Uplink Channel Dependent Scheduling for Future Cellular Systems

Författare

Author

Kristina Jersenius

Sammanfattning

Abstract

One goal in the development of future cellular systems is to increase perfor-mance. Channel dependent scheduling can possibly contribute to a performance enhancement. It requires channel quality information and uplink channel knowl-edge is often incomplete. This master thesis work compares channel dependent scheduling and channel independent scheduling for a Single Carrier Frequency Division Multiple Access-based uplink in time domain and time and frequency domain assuming continuous channel quality information updates. It also evalu-ates different methods for providing channel quality information by investigating how the limited channel knowledge they supply affects the performance of channel dependent scheduling.

Single-cell simulations with perfect channel knowledge indicate small gains for channel dependent scheduling. Large gains are seen when performing frequency and time domain scheduling instead of only time domain scheduling. Limited chan-nel knowledge causes performance loss for chanchan-nel dependent scheduling. The per-formance is only slightly decreased if a method with sufficiently frequent providing of channel quality information updates is applied.

More realistic multi-cell simulations show large gains for channel dependent scheduling. It is possible that these results are influenced by link adaptation and scheduling problems due to non predictable interference when performing dy-namic scheduling. In the comparison between channel dependent and channel independent scheduling the channel dependent scheduling can benefit from the fact that the selected channel dependent scheduling algorithms result in a more static scheduling than the selected channel independent scheduling algorithms do.

Nyckelord

Keywords channel dependent scheduling, uplink, SC-FDMA, channel quality information, 3GPP Long Term Evolution

(6)

(7)

Abstract

One goal in the development of future cellular systems is to increase perfor-mance. Channel dependent scheduling can possibly contribute to a performance enhancement. It requires channel quality information and uplink channel knowl-edge is often incomplete. This master thesis work compares channel dependent scheduling and channel independent scheduling for a Single Carrier Frequency Division Multiple Access-based uplink in time domain and time and frequency domain assuming continuous channel quality information updates. It also evalu-ates different methods for providing channel quality information by investigating how the limited channel knowledge they supply affects the performance of channel dependent scheduling.

Single-cell simulations with perfect channel knowledge indicate small gains for channel dependent scheduling. Large gains are seen when performing frequency and time domain scheduling instead of only time domain scheduling. Limited channel knowledge causes performance loss for channel dependent scheduling. The performance is only slightly decreased if a method with sufficiently frequent pro-viding of channel quality information updates is applied.

More realistic multi-cell simulations show large gains for channel dependent scheduling. It is possible that these results are influenced by link adaptation and scheduling problems due to non predictable interference when performing dynamic scheduling. In the comparison between channel dependent and channel independent scheduling the channel dependent scheduling can benefit from the fact that the selected channel dependent scheduling algorithms result in a more static scheduling than the selected channel independent scheduling algorithms do.

(8)

(9)

Acknowledgments

I would like to thank all people working at Ericsson Research in Linköping for a great master thesis work experience. I am looking forward to continue working with you.

Most of all I thank my Ericsson supervisor Eva Englund for enthusiastically discussing my work with me, answering my questions and giving me good advice. I have enjoyed our discussions and learnt a lot from them. Special thanks also go to Pål Frenger, Per Magnusson, Ke Wang Helmersson and Niclas Wiberg for discussions, explanations and help with the simulator.

My examiner Fredrik Gunnarsson and my university supervisor David Törn-qvist are thanked for showing interest in my work and making useful comments on my report. Thanks also go to my opponent Johan Kjelsson for observant proof-reading and good comments which helped me to improve the report.

Finally, I thank my family and friends for supporting me. Linköping, January 2007

Kristina Jersenius

(10)

(11)

Abbreviations

3GPP The Third Generation Partnership Project

ARQ Automatic Repeat Request

BLER Block Error Rate

CDF Cumulative Distributive Function

CDMA Code Division Multiple Access

CDT Channel Dependent Time Domain Scheduling

CDFT Channel Dependent Frequency and Time Domain Scheduling

CQE Channel Quality Estimation

CQI Channel Quality Indicator

CSMA Carrier Sense Multiple Access

DA Dynamic Assignment

DFT Discrete Fourier Transform

E-UTRA Enhanced Universal Terrestrial Radio Access

FA Fixed Assignment

FDD Frequency Division Duplex

FDMA Frequency Division Multiple Access

FFT Fast Fourier Transform

FT Fair Throughput

GIR Gain to Interference Ratio

GSM Global System for Mobile Communications

GPRS General Packet Radio Service

HARQ Hybrid Automatic Repeat Request

IFFT Inverse Fast Fourier Transform

I-FDMA Interleaved SC-FDMA

IP Internet Protocol

L-FDMA Localized SC-FDMA

LTE Long Term Evolution

MAC Multiple Access Control

Max-SIR Maximum Signal to Interference Ratio

OFDM Orthogonal Frequency Division Multiplexing

PAPR Peak to Average Power Ratio

PF Proportional Fair

QoS Quality of Service

RLC Radio Link Controller

RR Round Robin

RRFT Round Robin Frequency and Time Domain Scheduling

(12)

Td Coherence Time

RTDMA Random Time Division Multiple Access

RU Resource Unit

SC-FDMA Single Carrier Frequency Division Multiple Access

SIR Signal to Interference Ratio

STATFT Static Frequency and Time Domain Scheduling

STATT Static Time Domain Scheduling

TCP Transmission Control Protocol

TDD Time Division Duplex

TDMA Time Division Multiple Access

(13)

Introduction

Over the last decades the mobile phone has gone from being more or less the equivalent of a walkie talkie to being a complex component being designed not only for speech but also for web browsing, messaging, video conversations, etc. At the same time the mobile telephone systems have been developed from the first generation systems to the third generation systems to support more demanding services.

The 3rd Generation Partnership Project (3GPP) is currently developing new concepts in order for the 3GPP radio access technologies to remain competitive in a long-term perspective. The objective with the Long Term Evolution (LTE) is to provide enhanced performance in terms of higher data rates, reduced delays, improved coverage (the percentage of a network service area which upholds a required communication quality) and capacity (the maximum amount of data that can be transmitted over a channel).

In the 3GPP LTE concept the physical layer is based on Orthogonal Frequency Division Multiplexing (OFDM) for the downlink, the communication link from base station to user terminal, and Single Carrier Frequency Division Multiple Access (SC-FDMA) for the uplink, the communication link from user terminal to base station. This physical layer allows users to transmit on different parts of the spectrum without interfering with each other, which means that channel dependent scheduling, a resource assignment based on the channel quality of these spectrum parts, can be performed. A scheduling of this kind can contribute to increased total throughput, the total amount of data per time unit that is delivered to all user terminals or all base stations in a network, in comparison to channel independent scheduling. In the downlink Channel Quality Indicator (CQI) reports, based on measurements on the downlink reference signals performed by the user terminals, are used for the scheduling. In the uplink the base station estimates the uplink CQI information from uplink reference signals.

(16)

1.1 Problem Statement and Definition

The assignment is to propose and evaluate some different methods for providing and using CQI information for SC-FDMA uplink channel dependent scheduling. The extreme cases where the scheduler uses no CQI information, channel indepen-dent scheduling, and where the scheduler uses perfect CQI information, channel dependent scheduling with complete CQI information, are examined. Both cases are compared to investigate whether there are any gains with channel dependent scheduling. If a gain is indicated it is considered how CQI information can be provided and how limited CQI knowledge affects the performance of the channel dependent scheduling.

In this master thesis the scheduling strategies are focused on channel quality knowledge. The implications of limited and delayed buffer knowledge are not in-cluded. Perfect knowledge of the user terminal transmit buffers is assumed. Other scheduling aspects such as Quality of Service (QoS) are not considered. Intercell interference management and link adaptation, which are linked to scheduling, are not investigated in detail. The 3GPP LTE concept presents two versions of the SC-FDMA transmission scheme, one resulting in localized transmission and the other in distributed transmission. This work only considers localized SC-FDMA. LTE should support the duplex arrangements of Frequency Division Duplex (FDD), where uplink and downlink are separated in frequency, and Time Division Duplex (TDD), where uplink and downlink are separated in time. For FDD the available CQI is limited to the one which can be derived from uplink reference signals. In TDD where the uplink and the downlink share the same frequency band it may be possible to use the downlink CQI reports. This study examines only FDD.

1.2 Previous Work

Channel dependent scheduling for the OFDM-based downlink has been thoroughly investigated and has shown performance gain. The SC-FDMA uplink channel de-pendent scheduling has not at all been examined to the same extent. There are mainly two factors which cause the uplink scheduling to be different from the downlink one. Firstly, for the localized SC-FDMA-based uplink resource blocks that are assigned to a user have to be contiguous in the frequency domain. This is not the case in the multi carrier downlink where a user can be assigned resource blocks distributed over the entire bandwidth. In other words, downlink schedul-ing algorithms have to be modified. Secondly, there is the lack of uplink channel knowledge because users typically only transmit reference signals when they trans-mit data, due to power litrans-mitations, as opposed to the downlink where the base station continuously transmits reference symbols. There are some 3GPP contri-butions that describe efforts that have been made to do uplink channel dependent scheduling and suggestions for how to obtain sufficient uplink CQI knowledge.

(17)

1.3 Method 3

1.3 Method

The following steps are taken to solve the problem stated in chapter 1.1:

• Literature study on scheduling in general and SC-FDMA uplink channel

dependent scheduling specifically

• Selecting and describing algorithms for uplink scheduling assuming perfect

channel knowledge and assuming no channel knowledge at all

• Dynamic simulations using a Java-based radio network simulator developed

at Ericsson Research to estimate possible performance gains of channel de-pendent scheduling

• Selecting and describing methods for providing and utilizing incomplete CQI

for scheduling

• Dynamic simulations of scheduling with incomplete channel knowledge and

analysis of the results

1.4 Thesis Outline

Chapter 2 contains a theoretical background describing 3GPP LTE, LTE uplink details including SC-FDMA and resource assignment. The third chapter is focused on downlink and uplink channel dependent scheduling.

In the fourth chapter the scheduling algorithms are explained in detail and the expected outcomes of using the algorithms are discussed. The simulation model and the selected simulation parameters are described in the fifth chapter. Chapter 6 includes a presentation and an analysis of the simulation results.

The last two chapters contain conclusions and a discussion about further stud-ies.

(18)

(19)

Chapter 2

Theoretical Background

This chapter introduces theory and concepts that are used in the remainder of the report. It describes the LTE uplink in detail and explains the concepts of resource assignment and channel dependent scheduling.

2.1 Resource Assignment

Resource assignment or scheduling decides how resources, time slots and frequency channels or subbands, are distributed among communicating stations, base sta-tions and user terminals, in a communication system.

2.1.1 Fixed and Dynamic Assignment

There are two types of resource assignment: Fixed Assignment (FA) and Dynamic Assignment (DA). In FA systems each communicating station or user is assigned its own channel. A user is never allowed to use the channel of another user even if it is not in use. For systems using FA there are high delays at low message arrival rates when the load of the system is low and there are many unused channels. On the other hand, there is a high maximum throughput at high arrival rates because of efficient bandwidth utilization. In DA systems there are more users than channels and channels are allocated to users who request to transmit. DA schemes show the opposite behavior of FA schemes. At low load the DA schemes result in low delays since most of the bandwidth is utilized. At high load the maximum throughput is lower than the maximum throughput of FA schemes because there are channels that are used for resource assignment communication instead of data transmission. FA schemes are suitable for systems with continuous traffic while it is better to apply a DA scheme when the traffic is bursty.

Multiple access is a word for how a common resource is shared among mul-tiple users. Jamalipour, Wada and Yamazato, [13], state that the first mulmul-tiple access communication systems used Frequency Division Multiple Access (FDMA), the division of the frequency band into frequency channels which are distributed among the users. FDMA was followed by Time Division Multiple Access (TDMA),

(20)

the division of time into time slots which are assigned to users. The basic ver-sions of TDMA and FDMA are examples of FA schemes and they have sufficient performance for voice services. However, as other services, such as web browsing, have been introduced and the number of users has increased DA schemes have become more common. Examples of possible DA schemes are the multiple access schemes of 2G (Global System for Mobile Communications, GSM) and 2.5G (Gen-eral Packet Radio Service, GPRS), TDMA combined with FDMA, and of LTE, OFDM for the downlink and SC-FDMA for the uplink.

There are contention-based, non-scheduled, and scheduled DA schemes. For a contention-based scheme there is no central scheduler that tells a user when it is allowed to use a certain resource. It is up to the user to decide when and over what channel it transmits. For a scheduled DA scheme there is a central scheduler, typically situated in the base station, that distributes resources to users. If the scheduler is situated in the base station it is straightforward to use scheduled DA schemes in the downlink since the base station has full knowledge of how much data it wants to send. In the uplink it is more complicated since the buffer knowledge is in the user terminals and has to be transmitted to the base station.

Contention-based Schemes

Simple contention-based DA schemes such as Random Time Division Multiple Access (RTDMA) schemes have one channel. It means that every station has access to the entire bandwidth when transmitting. It is up to every station to decide when it wants to transmit and this leads to a risk of conflict. A collision channel with feedback is therefore introduced. The transmission attempt of a station is answered by a feedback message telling the station if the attempt was successful or not. The first RTDMA scheme, the ALOHA-Algorithm, was developed in the 1960s according to Ahlin and Zander, [1]. In the ALOHA-Algorithm a station transmits its message as soon as it has one. If the transmission is successful it removes the message from its queue of messages and transmits the next one if there is one. If there is a collision the station waits for a random time interval and then tries to transmit the message again. The reason that the time interval is selected randomly is that if the stations wait for an equal amount of time there is a collision again.

Another contention-based DA scheme is Carrier Sense Multiple Access (CSMA) in which the stations measure signal level to detect transmission. If a station is about to transmit a message and it senses that another station is transmitting it postpones its transmission for some time. For a non-persistent CSMA scheme this time interval is selected randomly and is followed by a new sensing for other transmissions. In a persistent CSMA scheme the station waits until the end of the ongoing transmission of the other station and then transmits its own message.

Scheduled Schemes

More complex methods to resolve conflicts, Conflict Resolution Algorithms, than the primitive ALOHA and CSMA methods have been developed. There are also scheduled schemes that avoid collisions by using reservation packets. During a

(21)

2.2 The Long Term Evolution 7

reservation phase reservation packets containing station/user ID, pending messages and message size are sent using a simple contention-based assignment scheme such as ALOHA. Then a central controller, a scheduler, receives the reservation packets and decides which users should be allowed to send using what resources. The decision of the scheduler is sent forward to the users and a collision-free data phase follows. These reservation schemes are the basis of current resource allocation or scheduling methods. An alternative to using a contention-based DA scheme during the reservation phase in the systems of today is using an FA scheme.

2.1.2 Channel Dependent and Independent Scheduling

The main task of a scheduler is to distribute resources to the users. For an OFDM or SC-FDMA based scheduler the resources are time-frequency intervals consisting of a certain frequency and time amount. If the scheduling is performed in the time domain all resources are given to one user each scheduling time interval. A scheduling in frequency and time domain means that every scheduling time interval several users are allocated resources.

A channel independent scheduler does not consider channel quality. A typical example is the Round Robin (RR) scheduler which every scheduling time interval assigns all resources to the user which has waited the largest amount of time to transmit.

When a scheduler bases its resource assignment on channel conditions it is called channel dependent scheduling. Kwok, Yu-Kwong and V.K.N, [14], mention that the reason to do channel dependent scheduling is that a bad channel state gives a low throughput. There is channel dependent time domain scheduling (CDT), where one user with good overall channel conditions is given the entire frequency band every scheduling time interval, and channel dependent frequency and time domain scheduling (CDFT), where several users are allocated frequency domain subbands with good quality every scheduling time interval. For mobile users or users in a mobile environment the channel quality varies with frequency and time. A multiuser diversity gain can be obtained if resources are allocated to the user terminals with the best channel quality.

2.2 The Long Term Evolution

LTE is as mentioned in chapter 1 the 3GPP evolution of 3G. The 3GPP re-quirements for LTE described by Ekstrom, Furuskar, Karlsson, Meyer, Parkvall, Torsner and Wahlqvist, [6], are among others:

• Instantaneous peak-data rates of 100 Mbit/s in downlink and 50 Mbit/s in

uplink

• Improved user average throughput and improved throughput for cell-edge

users

(22)

• Spectrum flexibility in both uplink and downlink

• Handover to and from 3G and 2G systems

Spectrum flexibility includes the possibility to use the LTE-technology in spec-trum allocations of different sizes, from less than 5MHz up to 20 MHz, and the support of operation in both paired and unpaired spectrum. There are three pos-sible duplex arrangements: FDD, TDD and combined FDD/TDD, see Figure 2.1.

Figure 2.1. Duplex arrangements.

Most of the 3G concepts are based on Code Division Multiple Access (CDMA). In CDMA narrowband user information is spread into a wider spectrum. Each user is assigned its own code and all users can therefore use the entire frequency domain. OFDM for the downlink and SC-FDMA for the uplink are the multiple access schemes chosen for the 3G LTE partly because of their greater spectrum flexibility with possibility of wideband transmission bandwidths and the possibility of performing frequency domain link adaptation and scheduling.

It is important to note that the LTE uplink details in this report are 3GPP working assumptions and not specifications.

2.3 The LTE Uplink

The uplink is based on single-carrier TDMA combined with FDMA, SC-FDMA. SC-FDMA offers a low peak to average power ratio (PAPR) as opposed to OFDM which is used in the downlink. The low PAPR of the SC-FDMA uplink gives a

(23)

2.3 The LTE Uplink 9

power-efficient user to base station transmission and reduced mobile power con-sumption. The basis of the transmission scheme is DFT-Spread-OFDM, which can be seen as pre-coded OFDM where DFT plus possible spectrum shaping has been added to the original OFDM-scheme. 3GPP suggests two different types of SC-FDMA, L-FDMA (Localized SC-FDMA) resulting in localized transmission and I-FDMA (Interleaved SC-FDMA) resulting in distributed transmission. In L-FDMA a user is assigned consecutive subcarriers and in I-FDMA distributed equidistant subcarriers. I-FDMA has a lower PAPR than L-FDMA according to Lim, Hyung, Kyungjin and Goodman, [16]. See appendix A for more information about OFDM, DFT-Spread-OFDM and subcarriers.

A resource unit (RU) is the smallest frequency-time resource block of the up-link. It is 1 ms, a transmission time interval (TTI), in time and 180 kHz, 1 subband or 12 subcarriers times a subcarrier space of 15 kHz, in frequency. In each TTI there are two subframes. Each 0.5 ms subframe consists of six long blocks used for data and two short blocks used for reference symbols. An illustration of an RU can be seen in Figure 2.2. A cyclic prefix is inserted between the blocks. The purpose of the cyclic prefix is to enable efficient frequency-domain equalization, to reduce the impact of intersymbol interference, at the receiver side. Figure 2.3 shows the placement of an RU in the time and frequency domain with a total bandwidth of 100 subbands.

The two short blocks can be used for different purposes, demodulation/detection or uplink channel quality estimation (CQE). According to the 3GPP Technical Specification Report, [17], the uplink reference signal structure should allow for localized reference signals and distributed reference signals. Transmission of refer-ence signals may be achieved by using frequency division multiplexing or code divi-sion multiplexing of the reference signals of different users. Uplink CQE reference signals may occupy at least partly different spectrum than the data transmission. There are several possible alternative uplink reference structures for L-FDMA: one where all reference signals are localized, one where there are localized reference signals in one short block and distributed reference signals in the other and one where the transmission of localized reference signals every now and then is inter-rupted by a period where all users transmit distributed reference signals. The last alternative is called channel sounding.

2.3.1 Scheduling and Access

A mobile cannot start transmitting whenever it wants to. It has to ask for resources and a permission to transmit. This request is sent contention-based or using FA to the base station. The initial scheduling request includes the user terminal identity and possibly other scheduling information such as buffer status, priority, etc. In the base station the scheduler takes care of the mobile request and responds with a resource assignment. The resource assignment includes which RUs the user can use for its transmission. Orthogonal user signals are provided since an RU cannot be assigned to more than one user. The scheduling is typically performed per cell, a cellular communication network consists of cells, each TTI. See chapter 5.1 for a more detailed explanation of cell and cellular network.

(24)

Figure 2.2. A resource unit.

There are two types of random access procedure for uplink users: a non-synchronized random access procedure for a user terminal which is not uplink synchronized and a simplified procedure, synchronized random access procedure, for a user terminal which is time-aligned. The non-synchronized procedure includes the transmission of a preamble, which is sent to obtain uplink synchronization. A non-synchronized user terminal which wishes to transmit synchronizes to downlink transmissions and reads the Broadcast Control Channel, a downlink channel for broadcasting control information, to know when and at which frequency band it can send its preamble. The base station processes the preamble and answers the user terminal with a time alignment and scheduled resources or a time alignment and a resource allocation on which to send a scheduling request. A contention-based synchronized random access can be done on a regular basis, e.g. every second subframe.

(25)

2.3 The LTE Uplink 11

Figure 2.3. The placement of an RU in the time and frequency domain.

for user data of different QoS classes, user terminal buffers sizes, retransmissions waiting to be done and uplink CQI. Uplink scheduling is complicated since the scheduler, which is situated in the base station, does not automatically know what the users want to send, how much they want to send and what their channel quality is. There is a possibility of semi static scheduling, which means that the scheduler determines a scheduling that is valid for certain sequences of TTIs instead of only one TTI or that the scheduler sends a scheduling pattern that is valid for several TTIs.

2.3.2 Power Control

Power control is applied to compensate for differences in path gain, the received transmission power over the transmitted transmission power, between different users. Its task is to control the transmit power of simultaneously transmitting users to prevent that the base station experiences a large difference in received power between different users. Under perfect conditions this should not have to be done since the orthogonality of the user signals eliminates intra-cell interference. In the reality there is some interference because of imperfect transmitter and receiver implementations. Two alternative power control schemes are that the base station

(26)

measures the received power level and sends power control commands to the user terminals or that the user terminals measure downlink signal strength and control their own power.

2.3.3 Link Adaptation

The link adaptation consists of that the base station measures the uplink channel quality in order to decide modulation schemes, code block lengths and coding rates.

2.3.4 HARQ

The uplink Hybrid Automatic Repeat Request (HARQ) is based on an N-channel Stop-and-Wait protocol. In an Automatic Repeat Request (ARQ) system trans-mission errors are detected and retranstrans-missions are requested. In a HARQ system the ARQ is combined with error correction which means that some errors are not only detected but also corrected using error correcting codes. There are different types of HARQ systems. Type I HARQ tries to correct transmission errors and asks for a retransmission if the correction attempt fails. Type II HARQ does not only attempt to correct erroneous received blocks, but also combines retransmitted received blocks with received blocks from earlier transmission attempts to enhance the probability of a successful retransmission reception. A Stop-and-Wait proto-col means that one block at a time is sent over the channel. This transmission is followed by an ACK leading to a new transmission or a NACK leading to a retransmission. New transmissions are blocked while waiting for an ACK/NACK. The N-channel Stop-and-Wait protocol makes transmissions more efficient by al-lowing N simultaneous HARQ processes so that new data can be transmitted while waiting for ACK/NACK.

HARQ processes can be either synchronous or asynchronous. If the HARQ is synchronous the retransmissions of the HARQ processes must occur at known time instants. This affects the behavior of the scheduler since the scheduler has to make resource reservations for these retransmissions at the correct time instants. The synchronous HARQ can be either adaptive or non-adaptive. Non-adaptive HARQ means that the retransmission must have the same exact transmission for-mat (number of assigned bits, code rate and modulation scheme) and frequency allocation as at the first transmission attempt. For the adaptive HARQ the fre-quency allocation can be different. For asynchronous HARQ the timing of the retransmissions is not limited to certain time instants, but since retransmissions should be transmitted within a reasonable time the scheduler often prioritizes retransmissions. The advantage of the synchronous non-adaptive HARQ in com-parison to the adaptive synchronous and asynchronous HARQ is the reduction of downlink signalling since the base station does not have to send a new frequency allocation to the user terminal.

The uplink HARQ is likely to be synchronous. It has not been finally decided whether it will be adaptive or nonadaptive or if both should be supported.

(27)

2.4 E-UTRA (Enhanced Universal Terrestrial Radio Access) 13

2.4 E-UTRA (Enhanced Universal Terrestrial

Ra-dio Access)

UTRA, the air interface of a 3G network consists of three layers: the physical layer, the data layer and the network layer. The data layer is divided into two layers: the Radio Link Controller (RLC) and the MAC. The RLC layer offers services (Radio Bearers) to higher layers and provides ARQ retransmission ser-vices. The MAC layer offers services to the RLC layer. The tasks of the MAC layer include mapping the logical channels to transport channels, link adaptation, scheduling and providing the HARQ protocol. The logical channels are between RLC and MAC and the transport channels are between the MAC layer and the physical layer. Higher layers for web traffic is the Internet Protocol (IP) and the Transmission Control Protocol (TCP). The UTRA layers and channels and their connections are shown in Figure 2.4.

Figure 2.4. The layers of E-UTRA.

E-UTRA, the air interface of an evolved 3G network (a LTE network), includes a new physical layer. The new physical layer is based on OFDM for the downlink and SC-FDMA for the uplink.

(28)

(29)

Chapter 3

Channel Dependent

Scheduling

Channel dependent downlink scheduling has shown promising results. Channel dependent scheduling of the SC-FDMA uplink is different from the channel depen-dent scheduling of the OFDM downlink. This chapter describes channel dependepen-dent scheduling for both downlink and uplink. It explains why the uplink scheduling is more challenging and contains suggestions from previous work for solutions to the problem of deriving channel knowledge.

3.1 Downlink Scheduling

The channel dependent scheduler bases its decisions on CQI reports. The OFDM transmission scheme allows a nonconsecutive assignment of RUs to users, see Fig-ure 3.1.

3.1.1 CQI Reports

Periodically the user terminals provide CQI reports based on measurements of the gain-to-interference ratios (GIR) on downlink reference signals of known power. These CQI reports are transmitted on uplink control channels to the scheduler situated in the base station. The CQI report of a user may consist of one GIR measurement value per every subband. A higher GIR indicates a better subband. The power signal gain, G, multiplied with the transmitting power, P, of the in-tended base station results in the received power of the user. The signal power is changed when a signal travels from transmitting to receiving antenna because of distance attenuation, shadow fading, multipath fading and antenna gain (see chap-ter 5.2). The inchap-terference includes inchap-terfering signals, I, from other transmitting

base stations. I is defined as the total received power, Itot, from all base stations

subtracted by the received power from the intended base station,

(30)

Figure 3.1. Downlink scheduling.

I = Itot− P · G. (3.1)

GIR is defined as signal power gain over interference and noise, N,

GIR = G

I + N. (3.2)

The CQI report describes the channel quality experienced by the user in the latest measurement. This is an indication of the channel quality in the next TTI, but for example the interference might have changed somewhat in the next TTI causing the estimated CQI to be different from the true one.

If each GIR value component of a user CQI report is multiplied with the power assigned to the subband corresponding to that component it results in a signal-to-interference ratio (SIR) value per every subband,

SIR = P · G

I + N. (3.3)

3.1.2 Scheduling Algorithms

There are different downlink scheduling algorithms known from previous work: e.g. Maximum Signal to Interference Ratio (Max-SIR), Proportional Fair (PF) and Fair Throughput (FT). Max-SIR is an algorithm that maximizes total SIR every TTI. An algorithm that maximizes some measure of performance, e.g. total throughput, tends to make a scheduling that favors users with good channel conditions. Max-SIR is an example of such an algorithm. The algorithm is not fair to all users and

(31)

3.2 Uplink Scheduling 17

some users have much lower user throughput than others. PF and FT consider fairness. FT strives after throughput fairness, i.e. equal throughput for all users, by assigning the user with the lowest throughput the RU with the highest SIR. The goal of PF is to schedule users when their channel conditions are good compared to their average ones. A certain RU is assigned to the user who for that particular RU has the maximum RU-user specific throughput over average throughput. An adaptation of the PF algorithm is the Exponential Rule which acts as a PF as long as there are tolerable delays for all users, but prioritizes users with large delays as soon as there are any. There are also schedulers specifically designed for VoIP where delays also are of a larger importance.

It should be noted that finding the optimal scheduling according to some crite-rion , e.g. maximizing SIR, is a complex optimization problem. In the algorithms mentioned above sub-optimal scheduling is used to lower the complexity of the problem. The sub-optimal scheduling is based on the strategy of choosing the user-RU pair which has the largest criterion value and assigning the RU to the user. After that the next user-RU pair is considered in the same manner and so on until all RUs have been assigned. The complexity of the scheduling can be even further reduced by assigning every resource block its best user or turning it around and assigning every user its best resource.

3.1.3 Quality of Service Scheduling

The question of QoS can also be taken into consideration. QoS means that different traffic flows are divided into different QoS classes (e.g. web traffic, VoIP etc) which have different performance demands and are of different importance. One user can have many different traffic flows. It is the task of the scheduler to maximize quality within a QoS class while differentiating quality between different QoS classes. Laneri, [15], and Zhang, He and Chong, [22], describe efforts that have been made to develop algorithms that give a good overall performance at the same time as they are fair to the users and consider QoS differences. The algorithms start by assigning resource blocks to the QoS class with the highest priority and apply a fairness algorithm if necessary and then go on like that for all QoS classes. The optimization algorithms including fairness or not might look different for different QoS classes. The reason for doing QoS scheduling is that there can be users who have bad channel conditions but high-priority data to transmit.

3.2 Uplink Scheduling

There are three main reasons why uplink channel dependent scheduling is different from downlink channel dependent scheduling in an FDD framework:

• The difference between the reference signals in the downlink and the uplink

is that in the downlink the base station transmits reference symbols con-tinuously whereas in the uplink the mobiles only transmit reference signals when they transmit data. This is partly because the uplink is power-limited and mobile batteries have a hard time surviving continuous transmission of

(32)

reference signals and partly because a continuous transmission of reference signals from all users occupy resources that can be used for data transmission. The consequence is that there is not as much channel quality information available for the uplink as for the downlink.

• The downlink scheduling algorithms have to be adjusted to fit the uplink

since users can only be assigned RUs being contiguous in the frequency domain, see Figure 3.2. This extra condition added to the optimization problem makes it complex.

• In the downlink there is full buffer knowledge since the scheduler situated

in the base station knows exactly what the base station wishes to transmit. In the uplink the buffer knowledge exits in the user terminals and has to be transmitted to the base station.

Figure 3.2. Uplink scheduling.

The channel dependent scheduling in the TDD framework can seem easier because the uplink and the downlink sharing the same bandwidth makes it possible to predict the uplink CQI from measurements of the downlink CQI. However, it has to be taken into consideration that uplink interference is different from downlink interference. Another observation is that the interruption of downlink reference signal transmission by uplink transmission periods reduces the accuracy of downlink CQI estimation and uplink CQI prediction.

As QualComm, [7], mentions the gains of uplink channel dependent scheduling can probably never succeed the ones of downlink channel dependent scheduling due to the lack of complete channel knowledge. The costs of channel dependent scheduling in forms of signalling, parts of the bandwidth and time that could have been used for transmitting data are used for transmitting reference signals, must

(33)

also be taken into consideration when deciding if channel dependent scheduling is worth adopting.

3.2.1 Uplink CQI

For uplink CQI the CQI GIR value of a subband for a user is obtained by the base station by deriving the subband gain of the user, G, and the subband gain of the

scheduled user, Gsch, from uplink reference signals. The scheduled user is situated

in the same cell as the user and is the user who used the subband for transmission in the previous TTI. The signal power the base station receives from the cell where

the user is located is calculated by multiplying Gsch with the transmission power

of the scheduled user, Psch. Pschis known by the base station if it were assigned to

the scheduled user by the base station in the previous TTI. The uplink interference per subband, I, is estimated by subtracting the received signal power from the cell

of the user, Gsch· Psch, from the total received signal power from the users in all

other cells , Itot,

I = Itot− Gsch· Psch. (3.4)

The uplink GIR has the same definition as downlink GIR,

GIR = G

I + N. (3.5)

If it is assumed that user transmissions are orthogonal, an RU cannot be used by more than one user for transmission, the base station experienced the same interference regardless of which one of the users of the own cell the base station scheduler scheduled for transmission in the previous TTI. The assumption that the interference is the same as in the previous TTI prerequisites that the interference is approximately the same regardless of which users the neighboring cell schedulers schedule for transmission or that the neighboring cell schedulers schedule the same users as in the previous TTI. This is not probable. It makes a difference if a neighboring cell scheduler schedules a cell edge user or a user standing close to the base station and a dynamic scheduler varies its scheduling over time and possibly also frequency. The interference estimate should therefore have to be improved, e.g. by filtering it in the time domain or the frequency domain or both. An alternative to filtering interference is to reduce interference variations by controlling the intercell interference.

As mentioned above there are no continuous reference signals from all mobiles. The implication of only transmitting reference signals when transmitting data is that there is no updated CQI information for users who were not scheduled for transmission in the previous scheduling time period. According to Parkvall, [20], there could be a possibility of using uplink reference signals transmitted for other reasons, e.g. the reference signals provided when the long blocks are used for control signaling, CQI report transmissions and ACK/NACK transmissions. For new users who have never been transmitting CQI could perhaps be derived from resource requests or preambles. It may be possible to calculate the average gain,

(34)

the gain excluding multipath fading, over the entire bandwidth from downlink CQI reports.

Three possible reference signal structures are a localized reference signal struc-ture, a distributed reference signal structure and channel sounding. In the local-ized reference signal structure users who transmit data transmit reference signals on the part of the bandwidth where they transmit data. In the distributed ref-erence signal structure users who transmit data transmit refref-erence signals over the entire bandwidth. Channel sounding means that the transmission of local-ized reference signals is regularly interrupted by a period where all users transmit distributed reference signals. Common for the distributed and localized reference signal structures is that there is no CQI available for new users who have not begun transmission unless it is assumed that it is possible to derive CQI from scheduling requests, preambles or reference signals that have been transmitted when trans-mitting control signals. The CQI is only updated every scheduling time period for the users who are transmitting that time period. For channel sounding the channel information is updated for all users when there is a channel sounding time period.

For a distributed reference signal structure the channel gain information used for users who were not transmitting in the previous scheduling time period is from their latest transmission. This CQI delay can affect the behavior of the scheduling if the true channel quality is very different from the delayed one.

If the reference signals are narrowbanded updated channel gain information is not only restricted to the users currently transmitting, but also to the part of the bandwidth on which the users are currently transmitting. This limits the possibility of doing channel dependent scheduling in frequency and time domain. Perhaps an algorithm including frequency-hopping, the frequency allocation of the user is changed over time according to a pre-determined pattern, can be used instead. In this way the user has collected wideband CQI information after a while. Users who are not transmitting data are transmitting control signals and if frequency-hopping is used for control signalling as well these users also have CQI covering the entire bandwidth after a certain period of time. For a time domain scheduler a localized reference signal structure is equal to a distributed reference signal structure since the user transmission bandwidth is equal to the total bandwidth.

If the channel sounding reference structure is applied narrowband reference signal periods are interrupted by wideband reference signal periods. The wideband transmission reference signal periods make it possible to estimate CQI for the entire bandwidth for all users.

In a lot of the previous work done on uplink scheduling, e.g. in [16], the lack of channel knowledge is not dealt with. The work is concentrated on different scheduling algorithms and assumes instantaneous knowledge of the uplink channel conditions. There are however some suggestions about how to solve the problem of deriving CQI.

QualComm Europe, [8], suggests several different methods for obtaining CQI. They propose that one long block could be used for wideband CQI estimation instead of one of the short blocks making both short blocks available for

(35)

detec-3.2 Uplink Scheduling 21

tion/modulation. Another option is an alternative to channel sounding where one narrowband reference signal is sent for each subband periodically every T ms making the channel knowledge complete after N · T ms if there is a total of N subbands.

The reference signal structures of NTT DoCoMo, [4], [3], [19] and [5], are combined with total scheduling solutions. Different channel dependent scheduling methods are considered. In one of them the transmission bandwith is pre-assigned to each user based on the data rate of each user and traffic distribution of each cell. The transmission bandwidths of all users are of the same size and do not vary with CQI. If there are several users that have been pre-assigned the same part of the frequency domain the one with the best channel quality is assigned that part. This method is combined with a localized reference signal structure. Another method divides the frequency domain into multiple subbands and allows adaptive transmission bandwidth, i.e. the bandwidth varies with channel quality. On trial it applies a downlink channel dependent scheduling algorithm and resorts the subbands using a priority based ranking system if the first trial assignment is nonconsecutive. This method requires wideband reference signals for all users.

The drawbacks of wideband reference signal transmission are listed in the con-tributions [3] and [19]. It fills up the bandwidth with reference signalling, increases the user power consumption and gives increased CQI measurement error due to reduced signal power density e.g. for cell-edge users. Therefore it is proposed that also the reference signal transmission bandwidth is adaptive. First transmis-sion bandwith for each user is decided based on traffic size or data rate. Then the reference signal transmission bandwidth is determined based on transmission bandwidth, amount of traffic and transmission power limit. The reference signals are transmitted and based upon the CQI derived from them frequency domain parts are assigned to users in a consecutive way. To avoid CQI measurement er-rors for cell edge users contribution [5] proposes that path loss between user and base station should be considered when deciding the reference signal transmission bandwidth.

3.2.2 Scheduling Algorithms

The downlink scheduling algorithms mentioned in chapter 3.1.2 have to be mod-ified since uplink RUs have to be assigned consecutively. The consecutive assign-ment makes algorithms involving frequency multiplexing of users complex. To reduce the complexity of the uplink scheduling algorithms RUs can be grouped to-gether in frequency to form larger resource blocks to distribute among users. Lim, Hyung, Kyungjin and Goodman, [16], propose an approach for the SC-FDMA uplink where there are as many resource blocks as users and at every TTI each user is allocated a maximum of one resource block.

Nokia, [18], proposes a CDFT scheduling where the frequency domain is di-vided into four parts that are assigned to different users consecutively. The per-formance of this scheduling method is compared to a CDFT scheduling allow-ing nonconsecutive assignment and both total throughput and the 5th percentile user throughput show that the consecutive assignment condition limits the

(36)

perfor-mance. The simulations of NTT DoCoMo, [4], indicate that channel dependent scheduling with adaptive transmission bandwidth performs better than channel dependent scheduling with fixed transmission bandwidth, at least for small fixed transmission bandwidths. This comes at the cost of a more complicated algorithm. It can be discussed if the channel dependent scheduling should be performed in the time domain or in the frequency and time domain. Ekstrom, Furuskar, Karlsson, Meyer, Parkvall, Torsner and Wahlqvist, [6], recommend that primar-ily time domain scheduling is used, but for terminals with limited power, limited amount of data to transmit or both time and frequency domain scheduling is used. Time domain scheduling can potentially lead to inefficient bandwidth utilization in these cases. The user does not use all the bandwidth it has been assigned and if frequency and time domain scheduling were used instead it would be possible to assign the unused bandwidth to other users. Frequency and time domain schedul-ing algorithms are typically more complicated than time domain algorithms and frequency and time domain scheduling requires more downlink signalling since the base station has to signal frequency allocations to the users.

3.2.3 HARQ and Retransmission Effects

Synchronous HARQ has effects on the behavior of the scheduler. If a user has pending retransmissions it has to be assigned RUs regardless if it has good chan-nel quality or not. For adaptive HARQ it is enough that the user is assigned any frequency allocation. If the HARQ is nonadaptive however, the user has to be assigned exactly the same RUs as in the previous transmission attempt. Both cases limit the decision possibilities of the scheduler, especially the nonadaptive case since it leads to fragmentation of the frequency band. It is difficult to do a consecutive channel dependent time and frequency domain assignment of re-sources to users in a frequency band where parts here and there are occupied by retransmissions.

To avoid having a HARQ controlled scheduling it is important to keep the number of retransmissions low. This can be measured by the block error rate (BLER) which is defined as the number of incorrectly received blocks over the total number of received blocks,

BLER = Incorrectly received blocks

total number of received blocks. (3.6)

The link adaptation often decides modulation and code rate based on esti-mated SIR. According to Ruberg, [21], the BLER can be held at a selected target by using a SIR backoff. The SIR backoff decreases the value of estimated SIR. An overestimation of SIR causes a faulty link adaptation resulting in retransmissions due to difficulties of decoding at the receiver side. The SIR backoff factor can be static or dynamically adapted to the number of correctly and incorrectly received blocks. A code rate limit that is less than 1 also decreases the number of retrans-missions. If the code rate is close to 1 there is almost no coding and that also complicates the decoding at the receiver side.

(37)

3.2.4 Link Adaptation and Interference Estimate Effects

Link adaptation and scheduling are linked. Based on the quality of the resource a user has been assigned the link adaptation decides how many bits the user can transmit. If a user is assigned a high quality resource it is assigned many bits. For the link adaptation it is important that the estimated CQI of the resource is close to the true channel quality of the resource. If the estimated CQI is too high the link adaptation assigns more bits to the user than the receiver can manage to correctly receive and if it is too low the link adaptation assigns fewer bits to the user than it is possible for the receiver to receive.

The interference estimate described in chapter 3.2.1 affects the decision of the link adaptation. If the estimated interference is very different from the true one it causes the link adaptation to make incorrect modulation and coding rate selec-tions. This incorrect link adaptation can make it impossible for the receiver to decode the received information and it sends a NACK resulting in a retransmis-sion. A lot of retransmissions result in large delays and low throughput. These retransmissions are often caused by an overestimated channel quality, in this case an underestimation of the interference. The extra assigned bits which an overes-timated channel quality causes can compensate for a retransmission or two every now and then and result in a throughput which is the same if not even higher than it would have been if there were no retransmissions, but many retransmissions cause a cell throughput loss. For a type II HARQ every retransmitted block is combined with earlier received block and eventually it is possible to decode the block. If the channel quality is largely overestimated, a lot of retransmissions are required to enable a successful reception of the transmitted block. The retrans-missions of a HARQ process are typically only allowed for a certain time and after that the user has to start all over again. If this occurs often it gives rise to large delays. To underestimate the channel quality, in this case to overestimate the in-terference, also causes throughput loss since the link adaptation assigns fewer bits than it would have done if the channel quality estimation had been correct.

The channel dependent scheduling is also affected by the erroneous interference estimate. If the CQI input to the scheduler is incorrect the scheduler makes incorrect scheduling decisions.

3.2.5 Time Aspects

It is good if scheduling is done as often as possible since it is unnecessary to schedule users who have emptied their send buffers. However, it can be discussed how often the channel quality should be updated, is it every TTI or more seldom? The true CQI can be very different from the delayed one if CQI is updated infrequently and this affects the performance of the scheduling algorithm. The scheduling decisions become almost random if the CQI is updated more seldom than the channel changes. At the same time it is unnecessary to update the CQI faster than the channel changes. This can be worth thinking of when deciding how often there should be channel sounding periods in the uplink channel sounding reference signal structure. How often the channel gain needs to be updated can be found by sampling the channel gain with the smallest possible sample period, the scheduling

(38)

time period, and then investigate, e.g. utilizing the sample theorem, see appendix B, how much it is possible to reduce the sample period without changing the frequency content and appearance of the channel gain substantially. It should be kept in mind that the rate of change of the channel gain depends on the velocity of the user terminals. If channel sounding is done often enough the channel sounding reference signal structure could possibly provide CQI with sufficient accuracy for channel dependent scheduling.

(39)

Chapter 4

Uplink Scheduling

Algorithms

This chapter describes the uplink scheduling algorithms that are used in the sim-ulations. In the chapter it is stated what channel knowledge is used by the link adaptation in the different channel knowledge cases, how the algorithms are mod-ified in case of pending retransmissions and how the power is allocated. Expected outcome of using the different algorithms is also included.

4.1 No Channel Knowledge

4.1.1 Time Domain Scheduling

When there is a total lack of channel knowledge a time domain scheduler of a Round Robin type is applied. It assigns all RUs to the user who has waited the longest to transmit among the active users, i.e. users with data to transmit.

RR algorithm:

• Begin

1. Check which user that has waited the longest for allowance to transmit

and assign all RUs to this user

• End

An alternative channel independent time domain scheduler is a static time domain scheduler (STATT). It selects the user who has waited the longest for allowance to transmit and allows it to every TTI use all resources until it leaves the cell. When it leaves the cell a new user is selected.

STATT algorithm:

The user which is selected to transmit until it leaves the cell is called users.

(40)

• Begin

1. If users has left the cell

- Check which user has waited the longest for allowance to transmit,

assign all RUs to this user and make it into users

2. If users has not left the cell

- Assign all RUs to users

• End

4.1.2 Frequency and Time Domain Scheduling

An alternative channel independent scheduler is a time and frequency domain scheduler which every scheduling time period divides the frequency domain into resource unit groups (RUgroups), consisting of the same number of consecutive resource units, and assigns one group to every active user until there are no groups left. The number of groups can be equal to the number of active users as long as the number of active users does not exceed the total number of resource units. It is also possible to select a limit, L, to the number of groups which is smaller than the number of active users. If such a limit is selected the scheduler works as a Round Robin scheduler with frequency multiplexing of users. This means that the L users which has waited the longest for allowance to transmit are assigned one RUgroup each every scheduling time period.

Round Robin Frequency and Time domain scheduler (RRFT) algorithm: The number of active users is N and the number of available resource units is

Nr. The number of RUgroups is RUgroupsize and the RUgroup index is j. There

are bNr

N c RUs per every RUgroup. The maximum number of RUgroups is L.

• Begin 1. If N <= L - RUgroupsize = N 2. If N > L - RUgroupsize = L 3. For j = 1 to RUgroupsize

- Allocate RUgroup j to unassigned user with maximum transmission

waiting time

• End

The fact that there are bNr

N c RUs per every RU group is important when

selecting L. If L e.g. is set to 51 and Nr is 100 there are 49 unused subbands if N

is equal to 51. This makes the algorithm very bandwidth inefficient. If L is 5 there is a maximum of one unused subband if N is equal to 3 and if L is 10 there are 0-4

(41)

4.2 Perfect Channel Knowledge 27

unused subband depending on N. In this study there are 100 subbands and the selected limit is 10. Other ways of avoiding the bandwidth inefficiency would be to allow the users to be assigned different numbers of subbands or to only allow

the RUgroupsize to be equal to numbers for which Nris divisible.

A static alternative to the RRFT scheduler is the static frequency and time domain scheduler (STATFT) which from the start selects the L users who have waited the longest for allowance to transmit (or the number of users in the cell if the number of users in the cell is less than L) and assign them one RUgroup each. The L RUgroups are reserved for the selected L users until one of them leaves the cell. Then the L-1 remaining users and a new user, selected based on its assignment delay, are assigned L RUgroups. In this master thesis work L is set to 10.

STATFT algorithm:

The number of users is N and the number of available resource units is Nr.

The number of RUgroups is RUgroupsize and the RUgroup index is j. There are

bNr

N c RUs per every RUgroup. The maximum number of RUgroups is L and the

group of L selected users is called users.

• Begin

1. If one of the users in usershas left the cell

- If N <= L

- RUgroupsize = N

– If N > L

- RUgroupsize = L

– Find user with maximum transmission waiting time and make it a

part of users

- For j = 1 to RUgroupsize

- Allocate RUgroup j to unassigned user in users

2. If no one of the users in usershas left the cell

- Allocate each user in users its resource

• End

4.2 Perfect Channel Knowledge

When channel knowledge is available channel dependent scheduling is performed. Perfect channel knowledge means that the channel gain values are updated for all users and all subbands every scheduling time interval.

(42)

4.2.1 Channel Dependent Time Domain Scheduling

The following algorithm uses CDT scheduling. Every scheduling time interval the active user that has the largest average GIR is assigned all resource units.

CDT algorithm:

The user index is i and the RU index is k. The number of RUs is Nr.

• Begin

1. Get GIR vectors, for all users, containing GIR values per every RU:

(GIR)i= (. . . , GIRi,k, . . . )

2. Calculate average GIR per user:

(GIR)i,average= 1 Nr Nr X k=1 GIRi,k

3. Check which user has the largest average GIR and assign all RUs to

this user

• End

4.2.2 Channel Dependent Frequency and Time Domain

Schedul-ing

Time domain scheduling can lead to inefficient bandwidth utilization when user terminal buffers are limited. This problem can be dealt with by having both time and frequency domain scheduling. A channel dependent frequency and time domain (CDFT) scheduler is derived from the RRFT scheduler if the user-RUgroup assignment is done based on CQI instead of time. The user-RUgroup pair which has the largest GIR is found, the RUgroup is assigned to the user and then it continues in that way until all RUgroups have been assigned. This scheduling algorithm is basically the same as a downlink maximum SIR algorithm except that in a downlink maximum SIR algorithm a user can be assigned more than one resource block. This is not allowed in the uplink algorithm to prevent that RUs are assigned to users in a nonconsecutive manner.

CDFT algorithm (Max-GIR):

The user index is i and the RU index is k. The number of active users is N

and the number of available resource units is Nr. The number of RUgroups is

RUgroupsize and the RUgroup index is j. All RUgroups contain bNr

N c RUs. A

user can only be assigned one RUgroup. The maximum number of RUgroups is L.

(43)

4.3 Incomplete Channel Knowledge 29

1. If N <= L

- RUgroupsize = N

2. If N > L

- RUgroupsize = L

3. Get GIR vectors, for all users, containing GIR values per every RU:

(GIR)i= (. . . , GIRi,k, . . . )

4. Calculate the average GIR per RUgroup and user:

GIRi,j= 1 bNr Nc bNr_X_Nc·j k=(bNr/N c·(j−1))+1 GIRi,k

5. While there are unassigned RUgroups

- Find the unassigned user i - unallocated RUgroup j pair with

max-imum GIR and allocate RUgroup j to user i

• End

The same discussion about the selection of the RUgroupsize limit L that is made in 4.1.2 can be done for this scheduler as well. If there are different number of subbands per user a Max SIR-algorithm has to be used instead of a Max GIR-algorithm. There are also other factors that are important when selecting L. If many users are scheduled simultaneously every TTI it requires a lot of downlink signalling since the base station has to signal frequency allocations to many users. In this study L is 10.

4.3 Incomplete Channel Knowledge

4.3.1 Channel Sounding

Channel sounding means that the transmission of localized reference signals is on a regular basis interrupted by a period of transmission of reference signals that are distributed over the entire frequency domain. This means that channel gain knowledge for the entire bandwidth and all users can only be updated after a channel sounding period has occurred. The channel gain for a particular user who is scheduled for transmission in a certain part of the bandwidth could be updated for that frequency domain part. This is not done in this study. Here all users only get channel gain updates for the entire bandwidth every xth scheduling time period where x is found by investigating how far it is possible to down sample the gain without altering it substantially. The algorithms for channel dependent scheduling in chapter 4.2 are used with this CQI as input.

Uplink Channel Dependent Scheduling for Future Cellular Systems

Institutionen för systemteknik

Department of Electrical Engineering

Examensarbete

Uplink Channel Dependent Scheduling for Future

Cellular Systems

Uplink Channel Dependent Scheduling for Future

Cellular Systems

Examensarbete utfört i Kommunikationssystem

vid Tekniska högskolan i Linköping

av

Abstract

Acknowledgments

Abbreviations

Contents

Chapter 1

Introduction

1.1

Problem Statement and Definition

1.2

Previous Work

1.3

Method

1.4

Thesis Outline

Chapter 2

Theoretical Background

2.1

Resource Assignment

2.1.1

Fixed and Dynamic Assignment

2.1.2

Channel Dependent and Independent Scheduling

2.2

The Long Term Evolution

2.3

The LTE Uplink

2.3.1

Scheduling and Access

2.3.2

Power Control

2.3.3

Link Adaptation

2.3.4

HARQ

2.4

E-UTRA (Enhanced Universal Terrestrial

Ra-dio Access)

Chapter 3

Channel Dependent

Scheduling

3.1

Downlink Scheduling

3.1.1

CQI Reports

3.1.2

Scheduling Algorithms

3.1.3

Quality of Service Scheduling

3.2

Uplink Scheduling

3.2.1

Uplink CQI

3.2.2

Scheduling Algorithms

3.2.3

HARQ and Retransmission Effects

3.2.4

Link Adaptation and Interference Estimate Effects

3.2.5

Time Aspects

Chapter 4

Uplink Scheduling

Algorithms

4.1

No Channel Knowledge

4.1.1

Time Domain Scheduling

4.1.2

Frequency and Time Domain Scheduling