Institutionen för systemteknik
Department of Electrical Engineering
Examensarbete
Turbo Equalization for HSPA
Examensarbete utfört i Kommunikationssystem vid Tekniska högskolan i Linköping
av
Cagatay Konuskan
LiTH-ISY-EX–10/4301–SE Linköping 2010
Department of Electrical Engineering Linköpings tekniska högskola
Linköpings universitet Linköpings universitet
Turbo Equalization for HSPA
Examensarbete utfört i Kommunikationssystem
vid Tekniska högskolan i Linköping
av
Cagatay Konuskan
LiTH-ISY-EX–10/4301–SE
Handledare: Ning He
Radio Access Technologies, Ericsson AB Niklas Johansson
Radio Access Technologies, Ericsson AB Mikael Olofsson
isy, Linköpings universitet Examinator: Mikael Olofsson
isy, Linköpings universitet Linköping, 26 March, 2010
Avdelning, Institution
Division, Department
Division of Communication Systems Department of Electrical Engineering Linköpings universitet
SE-581 83 Linköping, Sweden
Datum Date 2010-03-26 Språk Language Svenska/Swedish Engelska/English Rapporttyp Report category Licentiatavhandling Examensarbete C-uppsats D-uppsats Övrig rapport
URL för elektronisk version
http://www.commsys.isy.liu.se/en http://urn.kb.se/resolve?urn=urn:nbn:se:liu:diva-54640 ISBN — ISRN LiTH-ISY-EX–10/4301–SE
Serietitel och serienummer
Title of series, numbering
ISSN
—
Titel
Title
Turboutjämning för HSPA Turbo Equalization for HSPA
Författare
Author
Cagatay Konuskan
Sammanfattning
Abstract
New high quality mobile telecommunication services are offered everyday and the demand for higher data rates is continuously increasing. To maximize the uplink throughput in HSPA when transmission is propagated through a dispersive channel causing self-interference, equalizers are used. One interesting solution, where the equalizer and decoder exchange information in an iterative way, for improving the equalizer performance is Turbo equalization.
In this thesis a literature survey has been performed on Turbo equalization methods and a chosen method has been implemented for the uplink HSPA standard to evaluate the performance in heavily dispersive channels. The selected algorithm has been adapted for multiple receiving antennas, oversampled processing and HARQ retransmissions. The results derived from the computer based link simulations show that the implemented algorithm provide a gain of approximately 0.5 dB when performing up to 7 Turbo equalization iterations. Gains up to 1 dB have been obtained by disabling power control, not using retransmission combining and utilizing a single receiver antenna. The algorithm has also been evaluated considering alternative dispersive channels, Log-MAP decoding, different code rates, number of Turbo equalization iterations and number of Turbo decoding iterations.
The simulation results do not motivate a real implementation of the chosen algorithm considering the increased computational complexity and small gain achieved in a full featured receiver system. Further studies are needed before concluding the HSPA uplink Turbo equalization approach.
Nyckelord
Keywords Turbo equalization, MMSE, HSPA, ISI, decoding, equalization, low complexity, HARQ
Abstract
New high quality mobile telecommunication services are offered everyday and the demand for higher data rates is continuously increasing. To maximize the uplink throughput in HSPA when transmission is propagated through a dispersive chan-nel causing self-interference, equalizers are used. One interesting solution, where the equalizer and decoder exchange information in an iterative way, for improving the equalizer performance is Turbo equalization.
In this thesis a literature survey has been performed on Turbo equalization methods and a chosen method has been implemented for the uplink HSPA stan-dard to evaluate the performance in heavily dispersive channels. The selected algorithm has been adapted for multiple receiving antennas, oversampled process-ing and HARQ retransmissions. The results derived from the computer based link simulations show that the implemented algorithm provide a gain of approximately 0.5 dB when performing up to 7 Turbo equalization iterations. Gains up to 1 dB have been obtained by disabling power control, not using retransmission combining and utilizing a single receiver antenna. The algorithm has also been evaluated con-sidering alternative dispersive channels, Log-MAP decoding, different code rates, number of Turbo equalization iterations and number of Turbo decoding iterations. The simulation results do not motivate a real implementation of the chosen algorithm considering the increased computational complexity and small gain achieved in a full featured receiver system. Further studies are needed before concluding the HSPA uplink Turbo equalization approach.
Acknowledgments
First of all I would like to thank Ning He and Niklas Johansson, my supervisors at Ericsson Radio Access Technologies, for their valuable help and support during my work. This thesis would not have been possible without you. I am also grate-ful to Göran Klang, Yngve Selén, Sorour Falahati and all other members in the department for their encouragement and advices.
I would also like to show my gratitude to Mikael Olofsson, examiner and super-visor at Linköping University, for his help during the thesis work and even more important his great teachings in the telecommunication related courses.
Last but not least, special thanks to my family, friends and girlfriend for their patience and support during the hard time of my thesis work.
Notations
a the scalar a
a the vector a
A the matrix A
A−1 inverse of the matrix A
diag(a) a matrix containing zeros except its diagonal which is set to the values in vector a
0 vector or matrix with all zeros
∀ for all
(·)∗ the elementwise complex conjugate
(·)T the transpose
(·)H the Hermitian or conjugate transpose
|·| the absolute value
δ(t) the Dirac delta function (f ? g) the convolution of f with g
Acronyms
3G 3rd Generation
3GPP 3rd Generation Partnership Project
ACK Acknowledgement
BER Bit Error Rate
BLER Block Error Rate
BPSK Binary Phase-Shift Keying
BS Base Station
CC Chase Combining
CDMA Code Division Multiple Access
CRC Cyclic Redundancy Check
DFE Decision Feedback Equalization
DPCCH Dedicated Physical Control Channel
DS-CDMA Direct Sequence Code Division Multiple Access E-DCH Enhanced Dedicated Channel
E-DPCCH E-DCH Dedicated Physical Control Channel E-DPDCH E-DCH Dedicated Physical Data Channel E-TFC Enhanced Transport Format Combination
EUL Enhanced Uplink
EXIT Extrinsic Information Transfer
FIR Finite Impulse Response
GF Galois Field
GI Guard Interval
GSM Global System for Mobile communications
HARQ Hybrid Automatic Repeat Request
HSDPA High Speed Downlink Packet Access
HSPA High Speed Packet Access
HSUPA High Speed Uplink Packet Access
I In-phase
IBI Inter Block Interference
ICI Inter Chip Interference
IR Incremental Redundancy
ISI Inter-Symbol Interference
ITU International Telecommunication Union
LE Linear Equalization
LLR Log-Likelihood Ratio
LMMSE Linear Minimum Mean Square Error
LMMSE-CE Linear Minimum Mean Square Error Chip Equalization MAI Multiple Access Interference
MAP Maximum A Posteriori
MIMO Multiple Input Multiple Output
ML Maximum Likelihood
MMSE Minimum Mean Square Error
MRC Maximum-Ratio Combining
MSE Mean Square Error
MUD Multi-User Detection
NACK Negative Acknowledgement
OVSF Orthogonal Variable Spreading Factor
PAM Pulse Amplitude Modulation
PCCC Parallel Concatenated Convolutional Code
PDF Probability Density Function
Q Quadrature
QAM Quadrature Amplitude Modulation
QPSK Quadrature Phase-Shift Keying
RC Raised Cosine
RRC Root Raised Cosine
RS Reed-Solomon
RSC Recursive Systematic Convolutional SCCC Serial Concatenated Convolutional Code
SF Spreading Factor
SINR Signal-to-Interference-plus-Noise Ratio
SISO Soft-In Soft-Out
SOVA Soft-Output Viterbi Algorithm
TPC Transmit Power Control
TTI Transmission Time Interval
UE User Equipment
VA Viterbi Algorithm
WCDMA Wideband Code Division Multiple Access
ZF Zero Forcing
List of Figures
Figure 2.1 Structure of rate 1/3 Turbo coder (dotted lines apply for trellis termination only) . Figure source: [6].. . . 7 Figure2.2 4-PAM constellation. . . 8 Figure 2.3 Spreading and despreading of Binary Phase-Shift Keying
(BPSK) symbols. Figure source: [52]. . . 9 Figure2.4 E-DPDCH frame structure. Figure source: [1]. . . 10 Figure2.5 Spreading for E-DPDCH/E-DPCCH. Figure source: [7]. . . 11 Figure2.6 Linear equalizer implemented as a discrete FIR-filter. Figure
source: [21]. . . 13 Figure2.7 An example overview of the High Speed Packet Access (HSPA)
uplink transmission and reception process when two coding blocks and four physical data channels are used. . . 16 Figure3.1 Original concatenated coding system. . . 18 Figure3.2 Concatenated encoder and decoder. . . 18 Figure3.3 Serial concatenated convolutional encoder and decoder. . . 19 Figure3.4 Structure of original Turbo equalizer introduced by Douillard
et al. Figure source: [23]. . . 20 Figure4.1 Simplified overview of chosen Turbo equalization algorithm.
See detailed signal flow in Figure 4.2.. . . 26 Figure 4.2 An example overview of a HSPA uplink reception scheme
utilizing Turbo equalization.. . . 27 Figure4.3 SISO LMMSE Equalizer with two receiver antennas. . . 29 Figure4.4 SISO Max-Log-MAP Turbo decoder. . . 36 Figure4.5 A simplified overview of Hybrid Automatic Repeat Request
(HARQ) transmission combining when two Turbo decoders and three HARQ transmissions are used. . . 39 Figure5.1 Original tap placement of channels used in simulations. For
exact values used in simulations see Appendix B. . . 43 Figure5.2 Turbo equalization iterative gain when simulation setup BPSK6
is used in Pedestrian A channel.. . . 44 Figure5.3 Turbo equalization iterative gain when simulation setup PAM22
is used in Pedestrian A channel.. . . 45 xi
xii LIST OF FIGURES
Figure5.4 Turbo equalization iterative gain when simulation setup PAM22 is used in Pedestrian B channel.. . . 47 Figure5.5 Turbo equalization iterative gain when simulation setup PAM12
is used in Pedestrian B channel.. . . 48 Figure5.6 Turbo equalization iterative gain when simulation setup BPSK6
is used in Pedestrian B channel.. . . 49 Figure5.7 Turbo equalization iterative gain when simulation setup PAM22
is used in TU channel. Except a transport block size of 20000 bits has been used instead of a transport block size of 22000 bits, for achieving a code rate of 0.87. . . 50 Figure5.8 Compare Log-MAP and Max-Log-MAP Turbo decoders when
using Turbo equalization with Pedestrian B channel and simulation setup S-BPSK6. Except a transport block size of 10000 bits has been used instead of a transport block size of 12000 bits, for achiev-ing a code rate of 0.87. . . 51 Figure 5.9 Different combinations of internal Turbo decoder iterations
and Turbo equalization iterations. A different setup with 1 re-ceiver antenna, 4PAM modulation, ETFCS=20000, no retransmis-sions and no TPC has been used in Pedestrian B channel. . . 52 Figure 5.10 Turbo equalization iterative gain when simulation setup
S-BPSK6 used in Pedestrian B(3 kmph) channel. . . 54 Figure 5.11 Turbo equalization iterative gain when simulation setup
S-BPSK6 is used in Pedestrian B(30 kmph) channel. . . 55 FigureC.1 Extrinsic Log-Likelihood Ratio (LLR) distribution after code
List of Tables
Table4.1 Symbol alphabet BPSK . . . 37 Table4.2 Symbol alphabet 4-PAM . . . 37 Table4.3 Conversion from L(˜bk,n,q) to
¯
xk,n, E {xk,n} and ¯vk,n, Cov {xk,n, xk,n} . . . 37 Table5.1 Simulation parameters . . . 42
Table B.1 Propagation Conditions for Multi-Path Fading
Environments . . . 64
Contents
Chapter 1: Introduction 1
1.1 Background . . . 1
1.2 Purpose . . . 2
1.3 Outline . . . 2
Chapter 2: HSPA Uplink 5 2.1 Channel Coding. . . 6
2.2 Interleaving . . . 6
2.3 Modulation . . . 7
2.4 Spreading and Scrambling . . . 8
2.5 Equalization. . . 10
2.6 Retransmission Handling. . . 13
2.7 Detailed Transmission and Reception. . . 14
Chapter 3: Turbo Equalization 17 3.1 The Turbo Principle Development . . . 17
3.2 An Overview of Turbo Equalizers . . . 21
Chapter 4: Algorithm Description 25 4.1 Some Definitions . . . 26
4.1.1 Operators . . . 26
4.1.2 LLR Values . . . 26
4.1.3 Probabilistic Definitions . . . 26
4.2 Chip Level SISO LMMSE Equalizer . . . 28
4.3 Descrambling & Despreading . . . 34
4.4 Soft Demodulation . . . 34
4.5 Deinterleaver, HARQ Combining & Rate Dematching . . . 35
4.6 SISO Turbo Decoder . . . 35
4.7 Interleaver, HARQ & Rate Matching . . . 36
4.8 Soft Modulation . . . 36
4.9 Spreading & Scrambling, Chip Mean & Covariance Calculation . . 38
4.10 Retransmission Handling. . . 38
Chapter 5: Algorithm Evaluation 41 5.1 Simulation Environment . . . 41
5.2 Simulation Setup and Parameters . . . 41 xv
xvi CONTENTS
5.3 Simulation Results and Analysis . . . 44 5.3.1 Performance in Lightly Dispersive Channels . . . 44 5.3.2 Performance in Heavily Dispersive Channels. . . 46 5.3.3 Decoding Algorithm . . . 46 5.3.4 Code Rate. . . 46 5.3.5 Turbo Equalization and Turbo Decoding Iterations . . . 53 5.3.6 UE Velocity . . . 53 5.3.7 Other Discussions . . . 54
Chapter 6: Conclusions and Remarks 57
6.1 Conclusions . . . 57 6.2 Further Work . . . 58
Appendix A: Noise Correlation Matrix 61
Appendix B: Used Propagation Channels 63
Appendix C: LLR Distribution Between Iterations 65
Chapter 1
Introduction
1.1
Background
In the last two decades the cellular networks providing wireless communications have been growing rapidly. The breakthrough came with the second generation systems like Global System for Mobile communications (GSM). These systems were designed considering speech and they have limited data handling capabilities. As the demand for other services than speech increased the 3rd Generation (3G) systems were introduced. Wideband Code Division Multiple Access (WCDMA) is one of the the main 3G air interfaces in the world and it provides higher bit rates than earlier systems, which allow high quality multimedia services. The first WCDMArelease, release 99, is based on dedicated resource allocation per user and is not optimally suited for data traffic, which often utilize the bandwidth resources in a non-continuous asymmetric way. Therefore WCDMA release 5, referred to as High Speed Downlink Packet Access (HSDPA), and release 6, referred to as Enhanced Uplink (EUL) or High Speed Uplink Packet Access (HSUPA), were in-troduced. The specifications for current and previous releases can be found on the official 3rd Generation Partnership Project (3GPP) website [11]. HSDPAand EUL are together referred to as HSPA and considerably improve the data rate compared to previous releases of theWCDMAspecifications.
When the first3G cellular networks were deployed the data rates did not ex-ceed 384 Kbps (kilobit per second). Today there areHSPAnetworks utilizing 2x2 Multiple Input Multiple Output (MIMO) with data rates up to 28 Mbps (megabit per second) downlink and 11.5 Mbps uplink. By combining 2x2 MIMO antenna techniques with 64-Quadrature Amplitude Modulation (QAM), networks with up to 42 Mbps downlink could be ready for deployment in 2010 [10].
Everyday the number of users of mobile telecommunication systems is increas-ing and new high quality services are offered. Today there are over 500 million WCDMA users and more than 200 million of them are utilizing HSPA [26]. To meet the continuously increasing demand for higher data rates and better coverage
2 Introduction
in mobile telecommunication systems several complex data transmission problems need to be solved. Since spectral bandwidth is an expensive and finite resource it is necessary to optimize the spectral efficiency of existing frequency bands, both downlink and uplink. One of the biggest problems to efficiently utilize available bandwidth is Inter-Symbol Interference (ISI) caused by transmission in dispersive propagation channels. As a consequence successful reception of the signal can of-ten be a difficult task.
One receiver technique widely researched the last decade is Turbo equalization [23,48,49, 50], which is the subject of this thesis work. Turbo equalization is an iterative equalization and decoding method for optimizing the reception of signals which suffer fromISIand require equalization. The system is based on the Turbo principle which is used in Turbo codes.
1.2
Purpose
The purpose of this thesis is to understand the potential of Turbo equalization applied to the existingHSPAstandard for improving uplink throughput in heavily dispersive propagation channels. The work is divided into three parts:
• Literature survey. Perform a literature survey on Turbo equalization for evaluation of Turbo equalization studies in terms of functionality, perfor-mance and complexity.
• Implementation. Select a suitable algorithm for implementation in a sim-ulator and include differentHSPAuplink receiver features as oversampling, multiple antenna equalization, Transmit Power Control (TPC) and Hybrid Automatic Repeat Request (HARQ) retransmission handling into the algo-rithm to create a realistic simulation environment.
• Evaluation. Evaluate the performance of the implemented receiver algo-rithm by simulations under lightly dispersive and heavily dispersive propa-gation channels. Finally analyze the simulation results and compare them with results from previous studies on Turbo equalizers.
1.3
Outline
Chapter2 gives a brief overview of the HSPAphysical layer and details on parts critical for Turbo equalization. In Chapter 3 the Turbo principle is explained, an overview of the research performed on Turbo equalization is presented and complexity and performance are discussed. Chapter4describes the implemented algorithm. The simulation environment is described and results are evaluated in Chapter 5. In Chapter 6 the conclusions from the simulation results are sum-marized and suggestions are given for future research. In AppendixA the noise correlation matrix used in the algorithm is explained and in AppendixBthe exact
1.3 Outline 3
channel taps of channels used in the simulations are stated. Appendix C demon-strates how Turbo equalization iteratively improves the processed data.
Chapter 2
HSPA Uplink
In this chapter a brief overview ofHSPAuplink is given. For more detailed infor-mation the reader is referred to [21] or 3GPP specifications [1,2,3,4,5,6,7,8,9]. InHSPAuplink a new transport channel named Enhanced Dedicated Channel (E-DCH) is introduced for handling higher data rates through new features as fast scheduling, HARQ retransmissions and lower spreading factors. HSPA up-link is highly exposed to interference since all active User Equipments (UEs), devices as phones and computers communicating with the mobile network, si-multaneously transmit asynchronously over the same frequency band. There are actually two types of interference, Inter-Symbol Interference (ISI) and Multiple Access Interference (MAI). The interference occurring when multiple UEs non-orthogonal signals add up at despreading is calledMAI. If allUEs are transmitting with the same power level, the signal fromUEs most far away from the receiving Base Station (BS) would be affected by severeMAIor possibly not even detected. The problem is called the Near-Far Effect and is handled by so called Transmit Power Control (TPC) in WCDMA. TPC can shortly be described as the BS is controlling the transmission power of allUEs and trying to make the signal quality of the UEs as equal as possible at reception in the BS. MAI is still present after TPCbut it is possible to reduce it and increase the network capacity by using so called Multi-User Detection (MUD). Implementation examples onMUDare given in [56] and [19].
For low data rates, noise andMAIare the dominating sources of interference. For high data rates of HSPAuplink usually self-interference becomes a more im-portant issue, even if this kind of interference also can be found at low data rates. Self interference arise in the presence of a dispersive multipath channel. In a dis-persive channel the transmitted signal from a singleUE is received in theBSvia multiple different propagation paths due to reflection or refraction in the surround-ings. When these paths have different propagation delays a transmitted symbol is received in the BS at different time instances. This effect makes the symbols re-ceived in theBSinterfere with each other when a stream of symbols is transmitted.
6 HSPA Uplink
This kind of self-interference is often calledISI or Inter Chip Interference (ICI). To combat the errors introduced byISIdifferent equalization techniques are used. In this thesis single user uplink is evaluated and thereby MAI is not treated. Only noise andISIare considered as interference sources and Turbo equalization is incorporated into theHSPAuplink receiver for mitigating the interference. Here follows a description of the basic building modules of the transmit and receive chain inHSPAuplink.
2.1
Channel Coding
Channel coding facilitates more reliable transmission of a modulated signal through a propagation channel. Bit errors caused by interference and noise can be detected and/or corrected by a decoder in the receiver. In HSPAuplink a so called rate 1/3 Parallel Concatenated Convolutional Code (PCCC) Turbo encoder is used. It consists of two 8-state constituent Recursive Systematic Convolutional (RSC) encoders and one Turbo code internal interleaver. The code rate can be increased using puncturing for reaching higher data rates, but this is done in expense of decreased error correction capability and higher transmission power. The internal Turbo code interleaver takes the input bits and outputs them in a totally differ-ent order so that the output bitstream has almost no correlation with the input bitstream. The structure of the used encoder can be seen in Figure2.1. xkin the figure represents input bits while zkand z
0
k represents coded bit streams. Here zk and z0k are assumed to be independent due to the interleaver. Readers interested in further details on the encoder used inHSPAuplink are referred to the 3GPP specification [6].
The corresponding decoder in the receiver is fed with soft bit values represent-ing the probabilities of the received bits. The decoder then utilizes the known internal structure of the encoder for error correction and error detection. Turbo decoding will briefly be described in Chapter3but the interested reader can read more about Turbo decoders in [43] and [51].
2.2
Interleaving
In theHSPAuplink standard an interleaver, called channel interleaver, is applied after the encoder at transmission. Then a deinterleaver, called channel deinter-leaver, is needed before the decoder in the receiver. The purpose of the interleaving, which usually is used in combination with error correcting codes, is to decorrelate possible error bursts in a given time frame caused by ISI or imperfect channel equalization.
The Turbo encoder used inHSPAis based in conventional convolutional codes. Theoretically convolutional codes have infinite duration impulse responses but
usu-2.3 Modulation 7
Figure 2.1: Structure of rate 1/3 Turbo coder (dotted lines apply for trellis termination only) . Figure source: [6].
ally stronger dependency can be found within five times the code constraint length [32]. The channel interleaver distributes these stronger dependent bits within the whole transmitted block which gives the strong error correcting capability for lo-cal error bursts at reception. The channel interleaver also plays a central role in Turbo equalization which will be explained in later chapters.
2.3
Modulation
InHSPAuplink two different real-valued modulation schemes for data transmission are used, BPSKand 4-Pulse Amplitude Modulation (PAM). SimpleBPSKmaps each bit bi to a symbol ai according to [7]
ai=
(
+1, bi= 0
−1, bi= 1,
(2.1)
while 4-PAMmaps two bits bi,1 and bi,2using gray coding to a symbol aiaccording
to [7] ai= +1.3416, bi,1= 0, bi,2= 1 +0.4472, bi,1= 0, bi,2= 0 −0.4472, bi,1= 1, bi,2= 0 −1.3416, bi,1= 1, bi,2= 1 (2.2)
8 HSPA Uplink
or as illustrated in Figure2.2.
Real-valued modulation is applied on each E-DCH Dedicated Physical Data Channel (E-DPDCH), described in Section2.4, and they are later pair-wise added in the form of the orthogonal In-phase (I) and Quadrature (Q) branches, represent-ing real and imaginary parts. This gives the resultrepresent-ing constellation Quadrature Phase-Shift Keying (QPSK) if a pair of BPSKmodulated physical data channels are added and 16-QAMif a pair of 4-PAMmodulated physical data channels are added. The constellation points in 4-PAM are closer to each other compared to BPSK. The benefit with 4-PAM is that each symbol represents two bits while a BPSKsymbol correspond to one bit. 4-PAMresults in higher throughput if the channel and noise conditions, often measured in a Signal-to-Interference-plus-Noise Ratio (SINR) value, are allowing it. The downside is that 4-PAM has a higher SINR requirement while BPSK modulated data still may pass through without errors below this requirement. Given a transport format combination, so called Enhanced Transport Format Combination (E-TFC) in HSPAuplink, there is a fixed mapping to a modulation scheme and other parameters. A proper transport format is selected for a givenUEandBSto ensure that the Bit Error Rate (BER) is kept under an acceptable value allowing reliable transmission between them.
Figure 2.2: 4-PAM constellation.
2.4
Spreading and Scrambling
WCDMA is based on the Direct Sequence Code Division Multiple Access ( DS-CDMA) scheme. All active users in a WCDMA system share the same wide frequency band of 5 MHz simultaneously. Users are separated from each other with user specific codes consisting of a bipolar bit stream. The elements of a bit stream are called chips in a spreading system. A bit stream is thereby called a chip stream. The chip rate used inWCDMAis 3.84 Mcps (Mega chips per second) [2]. Each information symbol which is about to be transmitted is multiplied with a chip stream called spreading code, resulting in a new chip stream. At the receiver the same spreading code is used on the transmitted chip stream to recover the symbols. See Figure2.3where aBPSKmodulated signal is spread and despread. The ratio between the resulting chip rate and the baseband information symbol rate is called spreading factor.
2.4 Spreading and Scrambling 9
Figure 2.3: Spreading and despreading ofBPSKsymbols. Figure source: [52].
Each so called Transmission Time Interval (TTI) is the time interval when data is processed for transmission in the physical layer. In earlier releases of the WCDMAstandard it is set to a minimum of 10 ms for uplink but theTTIhas been reduced to 2 ms to allow faster retransmissions inHSPAuplink. TheE-DPDCHs are the HSPAuplink physical channels for data transmission. The transmission radio frame structure for an E-DPDCH can be seen in Figure 2.4. Each radio frame of 10 ms is as seen divided into 15 parts called slots. A chip rate of 3.84 Mcps will thus result in 2560 chips/slot [2]. The 2 msTTIinHSPAuplink consists of 3 slots which together are defined as a subframe. A subframe is the smallest decodable part of the transmitted signal when a 2 msTTIis used.
Figure2.5illustrates how data is multiplexed to E-DPDCHs using spreading. The information is spread in two steps. First a channelization spreading code is applied separately on each physical data channel and secondly a scrambling code is applied on all of them. The channelization codes used inE-DPDCHs are Orthogonal Variable Spreading Factor (OVSF) codes with Spreading Factors (SFs) down to 2. A property of the OVSF codes is that they in ideal conditions, syn-chronous transmission and reception in non-dispersive channels, preserve orthogo-nality between each other. Multiple physical data channels can be dedicated for a user inHSPAuplink and the same set of channelization codes are used by all users for separating their own physical data channels. In Figure 2.5 ced,k denotes the
applied real-valued channelization codes on modulated symbols in each channel number k. The symbols in each physical data channel k is later weighted by the gain factor βed,kshown in Figure2.5and mapped to theIandQbranches by using
10 HSPA Uplink 3GPP TR 25.808 V6.0.0 (2005-03) 10
Release 6T
7.3
Operation in SHO
8
Physical Channel Structure
8.1
Physical Channel Structure for Data Transmission
8.1.1 E-DPDCH
The E-DPDCH is a new physical channel on which the CCTrCh of E-DCH type shall be mapped. The E-DPDCH definition and attributes are the same as the DPDCH except where noted.
Figure 8.1.1 shows the E-DPDCH frame structure. The E-DPDCH radio frame is divided in 5 subframes, each of length 2 ms; the first subframe starts at the start of each E-DPDCH radio frame and the 5th subframe ends at the end of each
E-DPDCH radio frame.
Data
Slot #0 Slot #1 Slot #i Slot #14
1 radio frame = 10 ms E-DPDCH
Slot #2 1 subframe = 2 ms
2560 chips
Figure 8.1.1: E-DPDCH frame structure E-DPDCH slot formats, corresponding rates and number of bits are specified in table 8.1.1.
Table 8.1.1: E-DPDCH fields
Slot Format #I Channel Bit Rate
(kbps) SF Bits/ Frame Subframe Bits/ Bits/Slot Ndata
0 60 64 600 120 40 1 120 32 1200 240 80 2 240 16 2400 480 160 3 480 8 4800 960 320 4 960 4 9600 1920 640 5 1920 2 19200 3840 1280
8.1.2 Timing
The radio frames of all the E-DPDCHs transmitted by a UE shall be time aligned with the UE's UL DPCCH radio frames.
3GPP
Figure 2.4: E-DPDCH frame structure. Figure source: [1].
the qed,k coefficient also shown in the same figure. Here q = 1 for the real-valued I branch, and q = j for the imaginary-valued Q branch. The E-DCH Dedicated Physical Control Channel (E-DPCCH) seen in the bottom of Figure 2.5 carries control information. E-DPCCHis spread withSF256 using code cec, weighted by
the gain factor βecand mapped using the qeccoefficient. Another control channel
which always is present in an active communication is the Dedicated Physical Con-trol Channel (DPCCH). It is not shown in the figure but information inDPCCHis also multiplied with a gain factor βc and it is mapped to theQbranch. Finally all
chip-streams from the physical channels are summed up to get a complex-valued stream of chips. The complex valued scrambling codes applied to the summed up chip-stream can either be long Gold sequences of length 241, corresponding
to a 10 ms radio frame, or short primitive GF(2) polynomials of length 256. In the uplink the scrambling codes are used to separate users from each other. The fact that they are not perfectly orthogonal makes the overall interference increase with increased number of active users in the same frequency band. Short scram-bling codes can be used to ease the implementation of effective multiuser receiver techniques. Details on spreading code generation and spreading procedure done inHSPAuplink can be found in 3GPP specification [7].
2.5
Equalization
The purpose of performing equalization in the receiver is to reduceISIand restore the transmitted signal back to its original shape. Equalizers can be divided into two groups, namely trellis based equalizers and linear filtering based equalizers. Maximum A Posteriori (MAP) symbol detection and Maximum Likelihood (ML) sequence detection are the commonly used trellis based equalizers. For complexity reasons linear filtering based equalizers, e.g. Linear Equalization (LE) and Decision Feedback Equalization (DFE) are more commonly used instead ofML/MAP equal-izers. DFEalso incorporate previously equalized symbols forISIcancellation. The
2.5 Equalization 11
12 HSPA Uplink
parameters for filtering are selected by a cost criterion like Zero Forcing (ZF) or Minimum Mean Square Error (MMSE). The different equalizers are described in [51] and [40]. Due to the linearity of the spreading operation in WCDMA systems linear receivers can be applied either at chip- or symbol-level. The equalization will then be performed either before or after despreading and the performance might be different depending on chosen method. This is further explained in [24]. A simple receiver can be realized by a time-discrete Finite Impulse Response (FIR) filter w[n] with L filter taps applied to the received signal, as illustrated in Figure 2.6. In the figure, nn represents additive gaussian noise, h[n] represents
the discrete channel impulse response and rn represents the received signal in
the receiver. If the FIR filter w[n] is selected as the complex conjugate of the time-reversed channel impulse response
wM RC[n] = h∗[−n] (2.3)
it becomes a so called Maximum-Ratio Combining (MRC) filter. The well known Rake receiver is aMRCfilter applied on symbol level. MRC filtering cannot ac-tually do any equalization of dispersive channels causing ISI, it only maximizes the filter output signal-to-noise ratio. MRC filtering is an appropriate receiver for channels mainly disturbed by noise. It provides good performance for high spreading factors andBPSKmodulation.
Equalization is needed when transmitting through a dispersive channel. If the filter is selected according to
wZF[n] ? h[n] = 1, (2.4)
where ? denotes linear convolution, it is called Zero Forcing (ZF) equalization. ZF equalization gives full compensation forISIbut it also may increase the noise level after filtering. The noise enhancement can be very severe in the case of highly dispersive channels.
A more realistic solution is to use a filter that provides a trade-off between noise enhancement andISIsuppression. One commonly used method is the Linear Minimum Mean Square Error (LMMSE) equalization which minimizes the mean-square error
ε = E{|ˆsn− sn|2} (2.5)
between the equalized output ˆsn and the transmitted signal sn. For a LMMSE
time-discrete filter, the filter weights can be found by solving the Wiener-Hopf equation [33]
RwM M SE= p ⇐⇒ wM M SE = R−1p. (2.6)
In this expression, R is the channel-output auto-correlation matrix of size L × L, which depends on the channel impulse response and noise level, and p is the channel-output/channel-input cross-correlation vector of size L × 1 which depends on the channel impulse response. For channels with long time dispersion, requir-ing a large L in theFIR filter, the equalizer highly increases in complexity. The
2.6 Retransmission Handling 13
complexity increase due to the filtering itself and the inversion of matrix R.
Figure 2.6: Linear equalizer implemented as a discrete FIR-filter. Figure source: [21].
2.6
Retransmission Handling
When a packet, defined as the smallest decodable part of the transmitted signal, is received and fully decoded in the receiver an Acknowledgement (ACK) or Negative Acknowledgement (NACK) is signaled for telling whether the packet was correctly or incorrectly received respectively. Hybrid Automatic Repeat Request (HARQ) is an error-control method for data transmission which use retransmissions to achieve reliable data transmission over an unreliable channel. If a NACK is signaled the soft bits from the received demodulated symbols are stored in a soft bit buffer and a retransmission of the same information is scheduled. When the transmission arrives the new received soft bits are combined with the one stored in the soft bit buffer for increasing the probability of correct decoding. This continues until the receiver signals an ACK or a predefined number of allowed transmissions is exceeded.
There are two commonly used retransmission schemes inHSPAuplink systems: Chase Combining (CC) and Incremental Redundancy (IR). In CC the retrans-mitted packet contains exactly the same information as the previously received erroneously packet. The packets can be combined coherently before decoding and time diversity gain is then achieved. WhenIRis used the rate matching unit uses a different pattern at retransmission resulting in transmission of different encoded bits in case of puncturing is used. This method provides coding gain since each retransmission can decrease the coding rate as additional parity bits from the
re-14 HSPA Uplink
transmissions are combined with bits of the previous transmissions. There is a fixed number of simultaneousHARQprocesses. When 2 msTTIis applied 8 pro-cesses are used, the corresponding number of propro-cesses for 10 msTTI is 4. The processes are activated synchronously and thereby retransmissions of incorrectly received packets are scheduled 8 or 4 packets later. If the decoding still results in aNACKsignal after the maximum number of allowed transmissions is reached for a process, the soft bit buffer for that process is reset and the data is rescheduled in higher layers.
2.7
Detailed Transmission and Reception
An example of the HSPA uplink transmission and reception chain for a single user is given in Figure 2.7. The figure gives an overview of all components the information goes through. Source coded data, usually meaning some sort of data compression, is input to the transmitter in the form of a HSPAuplink transport block and hopefully restored at the end of the receiver. In the first step of the transmitter a 24 bit Cyclic Redundancy Check (CRC) code is added. The purpose of theCRCcode is error detection at the receiving end. AfterCRCcode is added, the data block is fed into a code block segmentation unit whose purpose is to split the data into several parts since the Turbo encoder used inHSPAuplink handles a maximum of 5114 bits [6]. The data is segmented into 2 Turbo encoders in the example shown in the Figure 2.7. No segmentation is needed if the number of bits are less than 5114 and only a single encoder is used. The encoded blocks are concatenated and sent into the rate matching unit after channel coding. The rate matching unit performs puncturing or repetition for adjusting the amount of data so it is suitable for the current transmission data-rate in the E-DCH Dedicated Physical Data Channels (E-DPDCHs). The settings for rate matching are con-trolled by the transport formatE-TFC signaled by theBS.
In Figure2.74 physical data channels are used, which is the maximum number of available physical data channels. They are separately interleaved and later mod-ulation is applied on each channel for mapping bits to symbols. BPSKor 4-PAMis used depending on current transport format. After modulation each physical data channel is spread using a real-valuedOVSF channelization spreading code. Two codes withSF2 and two codes withSF4 are used inHSPAuplink for reaching the maximum transmission bitrate. There are also other channels present not shown in the figure, E-DCH Dedicated Physical Control Channel (E-DPCCH) and Ded-icated Physical Control Channel (DPCCH), carrying control information. Both control channels use a SF of 256. Each physical channel is also weighted by a gain factor and mapped toI andQbranches as described in Section 2.4. As seen in the overview figure, the resulting chip-streams are summed up to give a single complex-valued stream of chips before scrambling is applied. After the scrambling process power control is applied to the chip-stream. Finally pulse shaping is done using a Root Raised Cosine (RRC) filter with a roll-off factor of 0.22 before the signal is bandpass modulated and sent through the channel.
2.7 Detailed Transmission and Reception 15
The transmitted signal passes through a channel and arrives at the receiver where it is transferred back to baseband. More than one antenna can be used at the receiver, as shown in Figure 2.7, to get diversity gain. Equalization is done after the received signal is filtered with the receiver RRC pulse shaping filter. Usually estimated channel coefficients, extracted from pilot symbols transmitted in the Dedicated Physical Control Channel (DPCCH), is used by the equalizer for estimating filter weights. The despreading, descrambling and I/Q demapping sep-arates the equalized signal back to real-valued chip-streams, each associated to a physical channel. In the soft demodulator symbols are mapped to bit-probabilities, or so called Log-Likelihood Ratios (LLRs), required for optimal decoding. The bit-probabilities are separately deinterleaved and later concatenated to a single stream. The single stream of bit-probabilities is zero-padded or punctured according to the rate-matching in the transmitter. If the block of data is a retransmission andIR is used, the bit-probabilities are also combined with previous transmissions. Re-sulting bit-probabilities are segmented to several code blocks if the HSPAuplink transport block in the transmission process was encoded with multiple encoders. Decoding is done and the decoded bits from all decoders are concatenated if mul-tiple decoders are used. The resulting stream of decoded bits is CRC checked for detecting errors possibly not corrected in the decoding. If an error is detected, the bit probabilities achieved before decoding are stored and a NACK signal is sent to the transmitter. If the maximum number of retransmissions is not reached for the actual data packet, a retransmission is performed 8 subframes later, if 2 ms TTIis used.
16 HSPA Uplink
Despreading & Descrambling
Scrambling Channel estimator Equalizer Channel Channel estimator Power Adjustment
HARQ & Rate Matching Rate Dematching &
HARQ Combining Physical Channel Concatenation Physical Channel Segmentation
Code Block Concatenation
Code Block Concatenation
Code Block Segmentation Code Block Segmentation
Turbo
Encoder EncoderTurbo
CRC Check CRC Attachment Spread-ing Spread-ing Modu -lation Inter-leaver Inter-leaver Modu -lation Modu -lation Soft Demod -ulation Soft Demod -ulation Modu -lation Soft Demod -ulation Soft Demod -ulation Turbo Decoder Turbo Decoder Deinter -leaver Deinter -leaver Deinter -leaver Deinter -leaver Spread-ing Spread-ing Inter-leaver Inter-leaver Square root
Square root Square root
Figure 2.7: An example overview of the HSPAuplink transmission and reception process when two coding blocks and four physical data channels are used.
Chapter 3
Turbo Equalization
A receiver trying to find the most probable transmitted bits by taking all available knowledge between transmission and observation into account would be optimal in terms ofBER. Considering the complex statistical relationship occurring when channel coding, interleaving, modulation, channel knowledge and other factors are involved makes an optimal receiver implementation practically impossible. Most receiver designs are therefore similar to theHSPAreception described in the pre-vious chapter, where the observed signal is processed in multiple steps. One sub-optimal solution is a receiver based on iterative processing of information between the equalization and decoding steps to improveBERin each iteration. The meth-ods based on this iterative processing are usually referred as Turbo equalizers, and they are based on the Turbo principle [18], [29]. This chapter begins with a quick review of the Turbo principle development and its basics since it is a vital part of the theory behind Turbo equalizers. Later in this chapter Turbo equalization studies performed until today are summarized and evaluated.
3.1
The Turbo Principle Development
The Turbo principle was in the beginning introduced for achieving efficient channel coding, so called Turbo coding. This section will briefly present the development until today, where the Turbo principle is utilized in Turbo equalization.
The first concepts of concatenated coding schemes came early as in the 60’s. Forney et al. showed that high error-correction capacity which would require long codes could be reached using a setup with two much shorter codes as seen in Fig-ure3.1[34]. In their proposal the inner encoder is a short block code and the outer encoder is an algebraic Reed-Solomon (RS) code [51]. The encoded signal passes through a channel which disturbs it. The received signal is then processed by the inner decoder and the binary output is passed to the outer decoder for correcting errors which at the inner decoding stage could not be corrected.
18 Turbo Equalization
Figure 3.1: Original concatenated coding system.
The performance in this setup is not optimal due to two reasons. The first prob-lem is the communication between the decoders which is hard-bit based, meaning that the outputs of the component decoders are binary decisions instead of proba-bilities of the received bits. The second issue is that the strong dependence between the encoders is only used one-way, from inner to outer decoder, and not the other way around.
Figure 3.2: Concatenated encoder and decoder.
In the 70’s NASA started to use concatenated codes for space applications [25]. The NASA system had two significant improvements compared to the earlier concatenated codes. NASA started using a convolutional code as an inner code, decoded by the optimal Viterbi Algorithm (VA). Secondly NASA incorporated an interleaver between the coding blocks as can be seen in Figure3.2. The main pur-pose of the interleaver/deinterleaver between the encoders/decoders is to scatter possible error bursts that can, due to the channel or structure of inner decoder, still slip through the inner decoder. This way the outer decoder can work more effectively.
In the late 80’s Battail [15] and Hagenauer et al. [30] introduced a so called Soft-Output Viterbi Algorithm (SOVA) as a solution to the above mentioned "hard-bit"
3.1 The Turbo Principle Development 19
problem. SOVAis a modifiedVAwhich can input soft channel measurements and output soft-bits, probabilities of received bits, from the inner decoder to further increase the effectiveness of the outer decoder.
Figure 3.3: Serial concatenated convolutional encoder and decoder.
The next big progress in the evolution of concatenated codes came when Berrou et al. in the article "Near Shannon limit error-correcting coding and decoding: Turbo codes" [18] proposed a solution to the previous mentioned second problem. Feedback was introduced between decoding modules so that iterative decoding can be done, as shown in Figure 3.3. It is this concept which today is known as Turbo coding. A Turbo code based transmitter has two or more error-correcting code modules pairwise separated with an interleaver and the received signal is later decoded iteratively with the same number of decoding modules. The de-coder modules are so called Soft-In Soft-Out (SISO) modules for best exploiting the information exchange between each other. The Soft-In Soft-Out (SISO) mod-ules that result in very high decoding performance in [18] are based on theMAP algorithm.
TheMAPalgorithm is too complex for Turbo decoding implementation in re-ality. Turbo codes were therefore not used until it was discovered that it is possible to reduce the decoding complexity almost without affecting the performance. The development begun with the suggestions from Koch et al. [35] who proposed the Max-Log-MAP algorithm. TheMAPalgorithm is simplified by doing the calcula-tions in the logarithmic domain and by introducing approximacalcula-tions in some steps. While the complexity is massively cut down the approximations reduce the perfor-mance. The solution to the performance reductions came when Robertson et al. [42] proposed the Log-MAPalgorithm which introduced a correction term to the approximations used in the Max-Log-MAP algorithm. The Log-MAP algorithm show performance close to theMAPalgorithm even though its complexity in
com-20 Turbo Equalization
parison is much lower. Today usually the Log-MAP or Max-Log-MAPalgorithm is used in applications relying on Turbo decoding due to their good complexity versus performance properties.
The original Turbo codes introduced by Berrou et al. are Parallel Concatenated Convolutional Codes (PCCCs). The coding scheme used inWCDMAis based on a PCCC. Knowledge of the PCCCinternal structure details is not necessary for this thesis and will therefore not be discussed. The interested reader is referred to the brief article by B. Sklar [43] for details onPCCCs and [6] for the specific PCCCused inWCDMA.
Figure 3.4: Structure of original Turbo equalizer introduced by Douillard et al. Figure source: [23].
In 1996 Benedetto et al. proposed an alternative iterative coding scheme, Serial Concatenated Convolutional Codes (SCCCs) [17]. Figure 3.3 shows a simplified illustration over SCCC encoding and decoding. Turbo equalization is based on theSCCCstructure. Here the inner encoder in Figure3.3 is replaced by a time-discrete multipath channel and the inner decoder with an equalizer. A multipath time-discrete channel can be seen as a convolutional encoder with code rate 1. Thereby the received signal can be "decoded" using an equalizer as the inner de-coder. A typical Turbo equalizer is a scheme with a SISO-equalizer as the inner decoder and a SISO-decoder as the outer decoder. These modules, which are separated with an interleaver/deinterleaver then process the same set of received data in an iterative way until a certain stop criterion is reached. This is usually a successfulCRCcheck, aLLRmagnitude check or a maximum number of allowed iterations. When the iterations are stopped the decoder finally outputs an estima-tion of the transmitted block of data. Comparing the Turbo equalizaestima-tion scheme in Figure3.4and the decoding part of theSCCCexample in Figure3.3will give a better understanding of the similarities between the concepts.
3.2 An Overview of Turbo Equalizers 21
3.2
An Overview of Turbo Equalizers
The Turbo equalization approach was first proposed by C. Douillard et al. in [23] and later further developed in articles such as [16] and [12]. TheSISO-equalizers and SISO-decoders used in the mentioned papers are based on the suboptimal SOVA and Max-Log-MAP algorithms [42] instead of the MAPalgorithm, due to the very high complexity of theMAP algorithm. Unfortunately these algorithms are still too complex when multiple receiving antennas and/or higher order mod-ulations are used. The complexity of these algorithms also increase exponentially with the length of the channel impulse response. To further reduce the Turbo equalization complexity filter-based equalization solutions can be used instead. One of the early suggestions is a joint equalization-decoding approach based on DFE and convolutional coding proposed by Ariyavistakul and Li [13]. Another is a multiuser filter-based Code Division Multiple Access (CDMA) detector pro-posed in [55] by Wang et al. where aSISO LMMSEequalizer is used instead of a MAP equalizer. In these solutions the linear filter coefficients are updated every time one symbol is processed. A solution with lower complexity than the earlier suggestions was proposed by Glavieux et al. in [28].
The Turbo equalization research took a new phase around the time the articles [50], [49] and [48] were published. In [50] R. Koetter et al. explain the fundamentals of the Turbo equalization principles thoroughly and show that Turbo equalizers based on theMAP-criterion outperforms all other Turbo equalizers in convergence and BER. But they also note that in certain conditions the performance of the LMMSE based Turbo equalizers approach theMAP-based ones. The article [49] introduce some new linear filtering-based Turbo equalization methods. They sug-gest two approximativeLMMSEbased methods and oneDFEbased method. The approximative methods give a significant reduction in complexity by updating fil-ter coefficients more sparsely instead of updating at each detected symbol as in the exactLMMSE equalizer used in [55]. They present a complexity comparison of different receivers and they describe a hybrid approach which utilize the two ap-proximativeLMMSEbased methods. They show that it is possible to achieve the same performance as the exactLMMSEequalizer based Turbo equalizer by using the hybrid method. Another interesting conclusion is that theDFEapproach does not show satisfactory results. In the paper [48] with focus on the LMMSEbased Turbo equalizers, the exactLMMSE algorithm and the suggested approximative LMMSE algorithms are described in detail. The algorithms are also extended to general signal constellations. Tüchler et al. also show that the approximative LMMSE-based Turbo equalizers gives very large complexity reductions while still performing relatively well compared to the exactLMMSEandMAPbased Turbo equalizers. It has also been proven that several previous linear soft interference cancellation based Turbo equalizers, e.g. [28], could be viewed as approximations of the solutions presented by Tüchler et al.
An interesting instrument which can be used for optimizing parameters and algorithms in Turbo equalization is Extrinsic Information Transfer (EXIT)-charts.
22 Turbo Equalization
In the article [31] Hanso et al. conclude, "Future iterative receiver research is ex-pected to rely on EXIT-chart-based design principles for the sake of performing close to the capacity limits, using the principles outlined in [54], [53] and stimu-lating the research community to aspire for achieving near-capacity performance over dispersive, fading wireless channels at the lowest possible complexity and de-lay.". InEXIT-charts theLLRconvergence is plotted on graphs for estimating the behavior of different algorithms or settings. Some examples where EXIT-charts have been used are presented in [49] and [20].
A frequency domainLMMSETurbo equalizer can be used since it can save com-putational complexity. The complexity gain compared with time domain equal-izers depends on block length, ISIchannel impulse response length and LMMSE filter length. According to [47] there is no significant difference in performance compared to time domain based equalizers but a drawback with frequency domain algorithms is that for equalizing properly they require a Guard Interval (GI) or extended filtering for removing Inter Block Interference (IBI) [45].
In [22] dos Santos A. F. et al. propose a Turbo equalizer incorporating a Turbo decoder. Their proposal include a special solution for reducing complexity. Usu-ally one or more Turbo decoding iterations are performed at each Turbo equal-ization iteration. In their paper they suggest that only one single convolutional decoder, of the two convolutional decoders inside a Turbo decoder, is executed in each Turbo equalization iteration. Thereby two Turbo equalization iterations are needed to complete one Turbo decoding iteration since only a half Turbo decoding is performed after equalization is preformed. The advantage with this approach is reduced complexity by allowing more executed equalization steps compared to the number of decoding steps. One requirement for utilizing this advantage is the usage of an equalizer with low complexity, which is possible in the case of a low complexityLMMSEequalizer.
Since retransmission combing will be used in the implementation the subject is a part of the literature study. Except relevant 3GPP documents other papers were studied and one of them is [14] where Assimi et al. describe differentHARQ retransmission methods and show howEXITcharts can be used for interpreting the performance of the different methods. In another paper Takeda et al. describe an implementation of a frequency domainLMMSEbasedDS-CDMATurbo equal-izer where retransmission combing is evaluated [44]. They use Chase Combining and conclude by simulation results that throughput performance is improved when combining retransmissions and Turbo equalization. Another article [46] briefly de-scribes an interesting implementation where Chase Combining is used onHSUPA. They suggest diversity combining of retransmissions in the equalizer instead of bitLLRlevel combining before decoding. Each retransmission is basically seen as an extra receiver antenna. The results show significant reduction in Block Error Rate (BLER), especially in higher order modulations.
trans-3.2 An Overview of Turbo Equalizers 23
mitted together with data symbols or in separate control channels. When Turbo equalization is used the extra knowledge about the transmitted information after decoding can be used for improving the channel estimates at each iteration. This has been widely studied and in [37] several proposals are presented for iteratively improving the channel knowledge.
Chapter 4
Algorithm Description
Different mentioned publications in Chapter 3 are showing that LMMSE-based Turbo equalizers perform well at a reasonable level of computational complexity. Most proposals for time-domainLMMSE-based Turbo equalization receivers have a similar base structure. A time-domain LMMSE-based Turbo equalization al-gorithm with basic features has therefore been chosen for implementation in the simulator. The implemented Turbo equalizer can thereby easily be extended with additional features from different proposals. The main references used in the fol-lowing algorithm derivations are [48] and [58]. The simulation environment is described in Chapter5.
The equalization and decoding modules use soft values as input and soft values as output in the following descriptions. Figure 4.1 gives a brief overview of the information flow and notations used in the described algorithm. In the first section of this chapter some definitions needed for the algorithm descriptions are made. The following sections in this chapter will follow the chain of modules in Figure4.1. In Section 4.2 the equalizer is derived. In Section 4.3 the descrambling and de-spreading process is described while Section4.4derives the soft demodulation. In Section 4.5 deinterleaving, physical channel concatenation, rate dematching and code block segmentation are briefly described. In Section4.6the Turbo decoding process is explained. In Section 4.7 the code block concatenation, rate match-ing, physical channel segmentation and interleaving are described. Section 4.8 describes the soft modulation performed and in Section 4.9the calculations from symbol statistics to chip statistics are presented. TheHARQfunctionality is pre-sented in Section4.10.
Figure4.2where four E-DCH Dedicated Physical Data Channels (E-DPDCHs) and two Turbo decoders are used illustrates a more detailed overview of Figure4.1. The figure can be compared with Figure 2.7 on page 16 where a non-iterative reception is illustrated.
26 Algorithm Description Output of bit decisions yi di di ci xk,n xk,n L(b k,n,q) vk,n L(rb,z) L(rb,z) L(bk,n,q) Input from channel Soft Modulation Soft Demodulation Interleaver , HARQ & Rate Matching Descrambling & Despreading Deinterleaver , HARQ Combining &
Rate Dematching Chip Level SISO
LMMSE Equalizer
Spreading & Scrambling Chip Mean & Covariance Calculation
SISO Turbo Decoder
Figure 4.1: Simplified overview of chosen Turbo equalization algorithm. See detailed signal flow in Figure4.2.
4.1
Some Definitions
4.1.1
Operators
The following operators are defined and used in the algorithm derivations. The operator E{·} is the expectation over the Probability Density Function (PDF), taking the a priori information into consideration. The covariance operator is given by Cov{x, y}, E{xyH} − E{x}E{yH}.
4.1.2
LLR Values
Turbo equalization is based on iteratively exchanging probabilities between the equalizer and decoder. When working with probabilities of binary values it is convenient to work withLLRs. The LLRfor a binary variable b is defined as
L(b) , lnP {b = 0}P {b = 1} (4.1) and is often denoted as the "soft" value of a bit or simply soft-bit. The sign of L(b) is the "hard" value of the bit b, it is often called hard-bit. A positive value of L(b) gives decision +1, or the binary value 0. A negative value gives decision -1, or a binary 1. The magnitude of |L(b)| represents the reliability of the decision. The special case when P(b = 0 ) = P(b = 1 ) = 1 /2 results in theLLRvalue L(b) = 0 .
4.1.3
Probabilistic Definitions
Some probabilistic definitions important for understanding the Turbo principle are presented:
4.1 Some Definitions 27
Channel inputs
Despreading & Descrambling Chip Level SISO LMMSE Equalizer
Channel estimator Channel
estimator
HARQ & Rate Matching Rate Dematching &
HARQ Combining Physical Channel Concatenation
Physical Channel Segmentation SISO Turbo
decoder
Spreading & Scrambling Chip Mean & Covariance Calculation Code Block Concatenation SISO Turbo
decoder Code Block Concatenation
Code Block Segmentation CRC Check Deinter -leaver Deinter -leaver Soft Modu -lation Inter-leaver Inter-leaver Soft Modu -lation Soft Modu -lation Soft Demod -ulation Soft Modu -lation Soft Demod -ulation Soft Demod -ulation Deinter -leaver Soft Demod -ulation Deinter -leaver Inter-leaver Inter-leaver Square root Square root
Figure 4.2: An example overview of aHSPAuplink reception scheme utilizing Turbo equalization.
28 Algorithm Description
A Priori Information is probability information associated with a bit biprior
an equalization or decoding step begins. The a priori information should originate from another source than the received sequence or the code constraints.
Extrinsic Information is probability information associated with a bit bi
provided solely by the processing in an equalizer or a decoder. The extrinsic in-formation is derived by processing all other received bits in a block except the actual bit biand all other a priori probabilities except the one associated with the
actual bit bi. In a decoder the extrinsic information is possible to extract since
the memory in the encoder has introduced dependency among bits. By using this dependency, statistics about the probability of the actual bit bi can be gained
from the surrounding bits. It is also possible to extract extrinsic information from equalization in the same manner since a dispersive discrete channel can be seen as a convolutional encoder with code rate 1.
Observation Information is probability information of the received signal.
A Posteriori Information is probability information associated with a bit bi
when all available sources of information about the bit is taken into account. This can be a priori information, extrinsic information and observation information.
4.2
Chip Level SISO LMMSE Equalizer
The equalizer module is used for equalizing the received signal from the two re-ceiver antennas by also incorporating a priori information, i.e. some statistical knowledge of the chips about to be estimated. The a priori information should come from a source other than the received signal. Inputs to the equalizer are thereby the received signal from the two receiver antennas, the statistics ¯di
rep-resenting the expectation values of the chips based on a priori information, the statistics cirepresenting the covariances of the chips based on a priori information
and the channel estimates for both antennas. In this thesis the channel impulse responses are assumed to be known at the receiver and they are thereby not esti-mated. The equalizer finally outputs estimated chips ˆdi.
The transmitted signal includes the E-DPDCHs, E-DPCCH and DPCCH. Scaling factors, spreading codes, I/Q-mapping and symbols are in the following derivations marked with ed, ec and c respectively for separating the variables for different physical channels. The symbols and spreading codes forE-DPDCHs are written without ed. The transmitted signal from aUEcan then be written as
di= K
X
k=1
βed,kqed,kak,ixk,i div Sk+ βecqecaec,ixec,i div Sec+ βcqcac,ixc,i div Sc (4.2)
where di denote the transmitted chip number i. K is number of utilized
4.2 Chip Level SISO LMMSE Equalizer 29
physical data channel number k. ak,i is the product of channel specific
chan-nelization and user specific scrambling codes for physical data channel number k and chip number i. xk,n represent symbol number n modulated in physical data
channel number k. xk,n ∈ S where S = {α1, α2, ..., α2Q} represents the 2Q-ary signal alphabet. A symbol αrin the signal alphabet corresponds to the bit pattern
sr , {sr,1, sr,q, ..., sr,Q}. The bit vector representing symbol xk,n modulated in
the transmitter is defined as bk,n, {bk,n,1, bk,n,q, ..., bk,n,Q}. i div Skis the integer
division of i by Sk. The Sk used in the integer division denotes spreading factor
for physical data channel number k. βecqecaec,ixec,i div Sec and βcqcac,ixc,i div Sc represents the contribution of the control channels.
yi1 di di ci xk,n vk,n yi2 di noise noise Channel 2 Channel 1
Chip level SISO LMMSE equalizer
Chip mean & covariance calculation Square root
Square root Square root
Figure 4.3: SISO LMMSE Equalizer with two receiver antennas.
Figure 4.3illustrates equalization using two antennas. The received signal is oversampled with λ samples per chip in the receiver. The received signal in the antenna with index α can then be modeled by
yαi =
L
X
l=0
hαldi−l+ nαi (4.3)
where yαi , [yiα(λ − 1), ..., yαi(0)]T is the received signal column vector of size λ number of samples per chip, representing sampled values during chip period i. hαl , [hαl(λ − 1), ..., h
α l(0)]
T is the sampled channel coefficients column vector of
size λ. The total revered discrete channel impulse response of length L + 1 chips can be written as [hαLT, ..., hα0T]T. nα
i , [nαi(λ − 1), ..., nαi(0)]T represents the
column vector of sampled received noise during chip period i. The channel coef-ficients hαl include the convolution of the transmitter and receiver pulse shaping filters and the received noise samples nα
i include the receiver pulse shaping filter.
The following definitions are made to simplify the derivation of the LMMSE equalizer
