Multicarrier Modulation and Coding For Multichannel Systems: ÅL Renger

Multicarrier Modulation and
Coding for Multichannel Systems

PL FRENGER
Department of Information Theory
Chalmers University of Technology
S-412 96 Sweden
Submitted to the School of Electrical and Computer Engineering,
Chalmers University of Technology, in partial fullment of the requirements
for the degree of Licentiate of Engineering
Technical report no. 258L
ISBN 91-7197-445-8
Gteborg, February 1997
G
O
T
EBO
R
G
C
H
A
L
M
E
R
S
T
E
K
NIS
K
A
H
O
G
S
K
O
L
A
i
In this thesis decision directed coherent detectors for single- and multicar-
rier systems employing an MMSE channel estimation technique is pre-
sented. The detectors are designed for Rayleigh fading channels and
DQPSK or -DQPSK modulation and exploit knowledge of the fading
process correlation. Pilot symbols are not used and differential encoding is
needed to resolve phase ambiguities due to decision errors. Analytical and
simulated results are presented. The detectors have signicantly lower bit
error probability than a traditional differential detector and do not suffer
from the problem of an irreducable bit error oor.
Further more, a Parallel Combinatory OFDM (PC-OFDM) system is pre-
sented. The PC-OFDM system uses only a subset of the available sub-carri-
ers in each symbol interval. A bit mapping procedure is proposed using the
Johnson association scheme to choose which sub-carriers to use and -
PSK modulation is used on these sub-carriers. The PC-OFDM scheme can
be designed to have a higher spectral efciency, lower bit error probability,
and lower peak-to-average power ratio than ordinary OFDM systems. Ana-
lytical and simulated results are presented for the AWGN channel as well as
simulated results for the Ricean fading channel. The PC-OFDM system is
however not as robust against fading as an ordinary OFDM system.
4
M
ABSTRACT
ii
The use of Rate Compatible Punctured Convolutional codes (RCPC-codes)
for rate matching in multichannel systems is also presented. A DS-CDMA
system is used for evaluating purposes. The RCPC-codes are shown to pro-
vide a exible and efcient coding scheme which is superior compared to
repetition encoding schemes.
Keywords: Multicarrier modulation, OFDM, Channel estimation, Decision
directed coherent detection, Parallel combinatory signaling, RCPC-codes,
rate matching, multichannel systems, multi-code DS-CDMA.
iii
ABSTRACT i
CONTENTS iii
ACKNOWLEDGMENTS vii
1 INTRODUCTION 1
1.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Short Introduction to OFDM . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Applications of OFDM
OFDM Techniques
1.3 Multi-Code DS-CDMA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.4 Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.5 Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.7 Acronyms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
CONTENTS
iv
2 DECISION DIRECTED COHERENT DETECTION IN
MULTICARRIER SYSTEMS 11
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2 System and Channel Models . . . . . . . . . . . . . . . . . . . . . . . . . 12
Single-Carrier System
Multicarrier System
2.3 Decision Directed Channel Estimation . . . . . . . . . . . . . . . . . 16
Scalar Solution for the Multicarrier System
Vector Solution for the Multicarrier System
2.4 Bit Error Probability and Estimation Error Analysis . . . . . . . 23
Bit Error Probability
Estimation Error Variance
2.5 Numerical Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
Results for the Single-Carrier System
Results for the Multicarrier System
2.6 Chapter Summary. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
Appendix 2A: Channel Correlation Function. . . . . . . . . . . . . 35
3 A PARALLEL COMBINATORY OFDMSYSTEM 37
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.2 System and Channel Models . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.3 Bandwidth Efciency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.4 Bit Mapping Procedure. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
Selecting the Sub-Carriers
Positions of the PSK Symbols
3.5 Bit Error Probability in AWGN . . . . . . . . . . . . . . . . . . . . . . . 44
3.6 Numerical Results on AWGN Channels . . . . . . . . . . . . . . . . 46
v
3.7 BER Performance on Ricean Fading Channels . . . . . . . . . . . 49
3.8 Chapter Summary. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
Appendix 3A: Probability Density and
Distribution Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
4 RATE MATCHING IN MULTICHANNEL SYSTEMS
USING RCPC-CODES 53
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
4.2 RCPC-Codes for Rate Matching . . . . . . . . . . . . . . . . . . . . . . 54
4.3 System and Channel Description . . . . . . . . . . . . . . . . . . . . . . 56
4.4 Numerical Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
Downlink
Uplink
Multipath Channel
4.5 Comparison to Repetition Encoding . . . . . . . . . . . . . . . . . . . 61
4.6 Chapter Summary. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
5 CONCLUSIONS AND FUTURE WORK 65
5.1 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
5.2 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
BIBLIOGRAPHY 69
vii
I have had the enormous privilege to have prof. Arne Svensson as my super-
visor. His support has simply been invaluable. I think he knows how grateful
I am for everything, so Ill just say thanks.
I am also deeply indebted to all my dear colleagues, former and present, at
the department of Information Theory. Special thanks to all of you for mak-
ing this a great place to work in. Thanks in particular to my room neighbor
Tony Ottosson for his never-ending enthusiasm. Thanks also to Thomas
Eriksson for being a lousy badminton player and for all fruitful discussions
in the coffee room. Thanks to Lars Kollberg who has helped my computer in
sickness and in health. Thanks also to Tobias Ringstrm for being the
department hacker.
A huge thank you to all my friends for distracting me and to my family for
encouraging me. At last, but not least, thanks to sa-Karin for always being
around.
I could never have done this without all of you. Once again, thanks.
The work presented in Chapter 4 has been performed in the framework of
the project ACTS AC090 FRAMES, which is partly funded by the European
Community.
Pl Frenger
ACKNOWLEDGMENTS
1
CHAPTER
1.1 BACKGROUND
Digital modulation is the process of mapping digital information (bits) to one
wave form out of a set of possible wave forms that are suitable for transmis-
sion over the available medium. The medium may be a radio channel or a cable
of some kind. The modulated signal may be an electromagnetic wave, or car-
rier, where the bits affect the phase, amplitude, or frequency of the transmitted
signal.
The design of a digital communication system is inuenced by a number of
factors such as the technology available, the type of service aimed for (e.g.
speech, data, facsimile, video, images, etc.), the acceptable cost of the system,
the channel characteristics, new ideas in research, regulations, and politics, to
name but a few. The great complexity and the fast development in this eld are
what makes this area so exiting. One strong trend today towards future digital
communication systems is the demand for higher data rates and systems capa-
ble of supporting many different types of services with different bit error prob-
ability and delay requirements. The aim of this thesis is to contribute to the
development of future communication systems.
INTRODUCTION
1
SHORT INTRODUCTION TO OFDM
2
1.2 SHORT INTRODUCTION TO OFDM
This short introduction to OFDM is not intended to give an extensive over-
view but rather to provide some perspective, to the work presented in this
thesis. Good overviews may be found in e.g. [5], [17], [59], [77], [47]. A
multicarrier modulation system is simply a system using more than one car-
rier for transmission. If the transmission of the individual sub-carriers is
coordinated properly we can design systems with some benets compared
to single-carrier systems.
Multicarrier modulation (MCM) is no longer a novel technique. The
idea of transmitting data in parallel on multiple carrier dates back at least
forty years [5]. The sub-carriers in the earliest multicarrier systems did not
overlap in frequency and were separated using band-pass ltering. Early
papers describing a multicarrier system with orthogonal sub-carriers over-
lapping in frequency were published by Chang in 1966 [9] and by Saltzberg
in 1967 [58]. These systems use innitely long symbols that are strictly
band limited and each sub-carrier overlaps only its two nearest frequency
neighbors. The orthogonality between the overlapping sub-carriers is main-
tained by a time offset, or staggering, of the in-phase and quadrature-phase
components called Staggered Quadrature Amplitude Modulation (SQAM).
These systems are a special form of multicarrier systems called Orthogonal
Frequency Division Multiplexing (OFDM) systems. In 1971 Weinstein and
Ebert [76] described an OFDM system using the Fast Fourier Transform
(FFT) to generate the orthogonal wave forms. The data symbols are pro-
cessed in the transmitter by the inverse FFT and by the FFT in the receiver.
The symbols are time-limited and all sub-carriers overlap each other in the
frequency domain. The use of the FFT makes it possible to design OFDM
systems with a large number of sub-carriers and low implementation com-
plexity. Since OFDM systems transmit data in parallel they have longer
symbol intervals, reducing the sensitivity to impulse noise, and the fre-
quency overlapping gives a system with high spectral efciency. OFDM
systems are however more sensitive to synchronization errors and nonlinear
distortions than single-carrier systems [61].
3
The long symbols also reduce the problem of Inter Symbol Interference
(ISI). When an OFDM signal passes through a time dispersive channel the
received signal will suffer from intersymbol interference. However if we
insert a Cyclic Prex (CP) [5], longer than the length of the channel impulse
response, all ISI will be removed by deleting the CP before the FFT in the
receiver. The CP is also necessary to avoid inter-channel interference (ICI)
caused by the transient effects of the channel. Further more, the CP accounts
for the channel memory and forces the linear convolution performed by the
channel to look like a cyclic convolution. This elegant solution of inserting a
CP enables that channel equalization may be performed by a simple multi-
plication after the FFT in the receiver since multiplication is the dual opera-
tion to cyclic convolution when using the DFT. The cost of using a CP is a
reduction in data rate and a loss of power and in many cases it is better to
endure some ISI and use a shorter CP or to use a short CP together with a
time domain equalizer [72], [16], [35].
Applications of OFDM
Multicarrier systems are being proposed and tested for wireless data trans-
mission in applications as HDTV [73], [13], [22], [78], cellular mobile tele-
phony and PCS [20], [56], [54]. The European DAB-system is yet another
example of a successful use of multicarrier modulation [81], [21]. The pop-
ularity of OFDM for broadcasting applications is due to several factors. One
is the ability to avoid transmitting on certain frequencies, that might be
occupied by existing analogue systems, by e.g. inserting zeros on the associ-
ated sub-carriers before the IFFT in the transmitter. Another reason is the
easy design of Single Frequency Networks (SFN). As long as signals from
different transmitters arrive with time differences less than the length of the
cyclic prex, the intersymbol interference may be removed in the same way
as multipath ISI.
Multicarrier modulation is also widely used in Asymmetric Digital Sub-
scriber Lines (ADSL) [65] and High bit rate Digital Subscriber Lines
(HDSL) [10], [12] systems, often under the name Discrete Multi-Tone
(DMT) modulation. In systems like these, multicarrier solutions provide the
4
possibility of assigning different number of bits and different power to the
carriers. Hence sub-carriers with high Signal to Noise Ratios (SNR) are
assigned more bits and power and poorer sub-carriers are assigned less. The
optimum
1
solution to this bit and power allocation problem is known as the
water-pouring solution [40], [11].
OFDM Techniques
To perform channel equalization by a multiplication in the frequency
domain we need something to multiply with. This is the problem of channel
estimation and several solutions are proposed. If we use differential encod-
ing we may perform differential channel estimation, as in [60], assuming
that the channel is constant over two consecutive symbol intervals. This
technique is simple but not as effective as coherent detection using explicit
channel estimation and will lead to a high bit error probability oor if the
channel is changing too fast. Other more efcient channel estimators based
on the use of pilot symbols are proposed in [59], [34], [71]. The use of a
decision directed Least Mean Square (LMS) algorithm is proposed in [57]
and a decision directed MMSE solution is proposed in [26] and described in
Chapter 2 of this thesis.
The peek to average power ratio of the OFDM signal grows linearly with
the number of sub-carriers and as the number of sub-carriers increase the
amplitude distribution of the transmitted signal tends to be Gaussian. Linear
ampliers must be used to avoid intermodulation distortion and this makes
OFDM signals difcult to amplify. The problem of designing such ampli-
ers is addressed in e.g. [7], [36]. Other solutions are the use of clipping to
reduce the peak power [50], various coding schemes [66], [37], [79], predis-
tortion [6], [4], and selective scrambling [19]. The Parallel Combinatory
OFDM (PC-OFDM) system proposed in [27] and described in Chapter 3
has, as one of its benets, a reduction in the peak to average power ratio.
1. In the sense of maximizing the bit rate for a certain bit error probability under the con-
straint that the total available power is limited.
MULTI-CODE DS-CDMA
5
Synchronization problems in OFDM systems have attained much inter-
est since the synchronization requirements are tighter than for single-carrier
systems. To attain frame synchronization, methods have been proposed
using the inherent redundancy of the cyclic prex of an OFDM signal [59].
Frame synchronization methods using pilot carriers [75], as well as decision
directed [41] approaches, are also proposed. In [14] a carrier frequency syn-
chronization method using the CP is proposed.
The pulse form used in an OFDM system determines its spectral proper-
ties. In [76] a smooth roll-off windowing of the CP is proposed. For nite
length pulses the optimum
1
shapes are derived in [74] for OFDM systems
using offset QAM. Time-limited raised cosine pulses together with a certain
time overlap is employed in [45].
1.3 MULTI-CODE DS-CDMA
Instead of using orthogonal carriers we may use spreading codes for parallel
data transmission. The codes may be Walsh [55] or Gold sequences [69] or
some other suitable sequences. With this change of basis functions we
obtain a system called multi-code direct-sequence code division multiple
access (multi-code DS-CDMA) which is very similar to OFDM. Both
OFDM systems and multi-code DS-CDMA systems may be seen as special
cases of a more general class of multi-pulse Pulse Amplitude Modulation
(multi-pulse PAM) systems [43]. Alternatively we may view both systems
as multichannel systems where a channel corresponds to a carrier in the
OFDM case and a code in the multi-code DS-CDMA case. Multi-code DS-
CDMA systems are strong candidates for future personal communication
systems. Aspects of data rate matching in general multichannel systems
using multi-code DS-CDMA systems as an example are examined in Chap-
ter 4.
1. In the sense of minimizing the out-of-band energy under the constraints of zero ISI
and ICI for a given length.
OUTLINE
6
1.4 OUTLINE
The outline of this thesis is as follows. Chapter 2 deals with decision
directed coherent detection and channel estimation in multicarrier systems.
In Chapter 3 a Parallel Combinatory OFDM system is presented and Chap-
ter 4 deals with data rate matching in multichannel systems using Rate
Compatible Punctured Convolutional codes (RCPC-codes). Some conclu-
sions are drawn in Chapter 5 together with a discussion on future work. A
longer outline of the Chapters 2-4 follows below.
Novel coherent detectors for DQPSK and -DQPSK on Rayleigh fad-
ing channels are presented in Chapter 2. The detectors are based on decision
directed minimum mean square error (MMSE) channel estimation. The pro-
posed detectors may be used in both single-carrier and multicarrier systems.
Analytical results in form of expressions for channel estimation error vari-
ance and bit error probability, both with and without diversity, are derived.
The MMSE based detector has a signicantly lower irreducible bit error
probability oor compared to a conventional differential detector. For most
practical applications the error oor of the MMSE based detector is negligi-
ble. Simulated results are presented to verify the analysis and to justify the
approximations made.
A parallel combinatory OFDM system (PC-OFDM) is described in
Chapter 3. The proposed system is able to obtain a higher bandwidth ef-
ciency than an ordinary OFDM system. The bit error rate performance is
analyzed and simulated. On a Gaussian channel the PC-OFDM system can
obtain lower bit error rates than ordinary OFDM. In some cases we can at
the same time increase the bandwidth efciency compared to an ordinary
OFDM system. Further, the peak to mean power ratio of the PC-OFDM sys-
tem is lower than for OFDM, which reduces the problem of designing linear
power ampliers.
Rate Compatible Punctured Convolutional codes (RCPC-codes) are pro-
posed for rate matching in multichannel multiple access systems in Chapter
4. The bit error rate performance of a multi-code direct sequence code divi-
sion multiple-access (DS-CDMA) based system using RCPC-codes is used
4
CONTRIBUTIONS
7
to evaluate the proposed rate matching scheme. Comparison with repetition
encoding is also made. The RCPC-codes outperform the repetition codes
and can provide a larger span of available channel coding rates. We show
that RCPC-codes provide a exible coding scheme suitable for future
mobile radio communication systems.
1.5 CONTRIBUTIONS
The main contributions in this thesis may be summarized as follows
A decision directed MMSE channel estimator for single- and multi-
carrier systems. This work was presented in parts at the IEEE Vehic-
ular Technology Conference, Atlanta, USA, April 28-May 1, 1996
[25] and at the RVK seminar, Lule and Kiruna, Sweden, June 3-6,
1996, [24]. This work is also submitted to IEEE Transactions on
Vehicular Technology [26].
A parallel combinatory OFDM scheme. This work was presented at
the IEEE Personal, Indoor and Mobile Communications Confer-
ence, Taipei, Taiwan, October 15-18, 1996 [27].
A proposal of the use of RCPC-codes for rate matching in a multi-
code DS-CDMA systems. This work will be presented at the IEEE
Vehicular Technology Conference, Phoenix, USA, May 5-7, 1997
[28].
ACRONYMS
8
1.6 ACRONYMS
The elds of communication- and information theory are overwhelmed by
acronyms. Often, but not always, these acronyms have an obvious interpre-
tation. However, the same letter combination may mean something com-
pletely different to someone familiar to a different eld. Listed below are the
acronyms that appear in this thesis together with the interpretation used
here.
ASDL Asymmetric Digital Subscriber Line.
AWGN Additive White Gaussian Noise.
BER Bit Error Rate.
BPSK Binary Phase Shift Keying.
CDMA Code Division Multiple-Access.
CMFB Cosine Modulated Filter Banks.
CP Cyclic Prex.
DAB Digital Audio Broadcasting.
DFT Discrete Fourier Transform.
DMT Discrete Multi-Tone.
DQPSK Differential Quadri-Phase Shift Keying.
DS-CDMA Direct Sequence Code Division Multiple-Access.
FDMA Frequency Division Multiple-Access.
FFT Fast Fourier Transform.
FIR Finite Impulse Response.
HDSL High bit rate Digital Subscriber Line.
HDTV High Denition Tele-Vision.
IDFT Inverse Discrete Fourier Transform.
IFFT Inverse Fast Fourier Transform.
ICI Inter-Channel Interference.
ISI InterSymbol Interference.
LMMSE Linear Minimum Mean Square Error.
LMS Least Mean Square.
MCM MultiCarrier Modulation.
ML Maximum Likelihood.
ACRONYMS
9
MMSE Minimum Mean Square Error.
NBC Natural Binary Code.
OFDM Orthogonal Frequency Division Multiplexing.
PAM Pulse Amplitude Modulation.
PC-OFDM Parallel Combinatory Orthogonal Frequency Division
Multiplexing.
PCS Personal Communication System.
PSK Phase Shift Keying.
QPSK Quadri-Phase Shift Keying.
RAKE Not really an acronym. Refers to a receiver structure using
a tapped delay line collecting signal energy from many
received signal paths in a fashion somewhat analogous to
an ordinary garden rake.
RC Raised Cosine.
RCPC Rate Compatible Punctured Convolutional.
RS-code Reed Solomon code.
SFN Single Frequency Network.
SNR Signal to Noise Ratio.
SQAM Staggered Quadrature Amplitude Modulation.
TDMA Time Division Multiplexing.
-DQPSK Differential Quadri-Phase Shift Keying with phase incre-
ments 45, 135, -45 or, -135 degrees.
4
11
CHAPTER
2.1 INTRODUCTION
This chapter describes coherent detection in single- and multicarrier systems
on Rayleigh fading channels. Using the correlation of the channel fading pro-
cess we design accurate channel estimators needed for coherent detection.
Differential detection of DQPSK and -DQPSK is widely used today
for radio communication in both single-carrier and multicarrier systems world-
wide. Both the north American and the Japanese digital cellular mobile tele-
phone systems are examples of single-carrier systems using -DQPSK [3].
The Japanese system uses a traditional differential detector combined with
diversity reception to obtain acceptable performance on a fading channel [1],
[2]. The European DAB-system uses DQPSK and applies differential channel
equalization to avoid explicit channel estimation [81]. In systems like the ones
mentioned above there is a need for the performance enhancement given by
coherent detection compared with differential detection. The drawbacks with
differential detection are the irreducible error probability obtained due to fad-
ing and the degraded performance compared to ideal coherent detection.
Coherent detection, though, requires an accurate carrier phase reference. In
general, known pilot symbols need to be inserted periodically in the data
stream to enable channel estimation needed for a coherent detector on a fading
4
4
DECISION DIRECTED
COHERENT DETECTION IN
MULTICARRIER SYSTEMS
2
SYSTEM AND CHANNEL MODELS
12
channel [8], [18], [71]. These symbols increase both the bandwidth and the
required signal to noise ratio for a given bit error probability.
Here we propose to use previously made decisions instead of pilot sym-
bols for DQPSK. Thus we derive a decision directed coherent detector that
can be used in both single-carrier and multicarrier systems. Initially only
one known symbol on each carrier is needed in the transmission. However,
differential encoding is still needed, since carrier phase estimation with
decision directed techniques in QPSK always lead to a 90 degree phase
ambiguity due to decision errors.
The outline of this chapter is as follows. In Section 2.2 the system and
channel models used in this chapter for the single-carrier and multicarrier
systems are presented. Section 2.3 describes the actual decision directed
channel estimation procedure proposed here. Then follows in Section 2.4 a
bit error probability analysis of the proposed detectors on fading channels.
Results with and without diversity are obtained for both the single-carrier
system as well as for the multicarrier system. In Section 2.5 numerical
results are presented and discussed. To justify the approximations made in
the analysis, simulated results are also presented. The chapter is then sum-
marized in Section 2.6.
2.2 SYSTEM AND CHANNEL MODELS
In this chapter we assume Gray encoded DQPSK but the method also works
for -DQPSK. We will use upper-case letters for the multicarrier system
and lower-case letters for the single-carrier system whenever there is a need
to distinguish between them. The transmitted symbol
on sub-carrier at time for the multi-
carrier system and at time for the single-
carrier system is given according to Table 2.1. Here is the symbol
period, is the transmitted amplitude and is the imaginary unit. Note that
for the multicarrier system we may choose to do the differential encoding
both in the time domain, between consecutive symbols on the same sub-car-
rier or in the frequency domain between sub-carriers. We have in this chap-
4
C
n k ,
A jA A jA , , , { } n kT
c
k
A jA A jA , , , { } kT
T
A j
13
ter chosen to make the differential encoding in the time domain for the
multicarrier system to enable easy comparison between the two systems.
Differential encoding in the frequency domain is a topic for future research.
We assume a at fading channel model for the single-carrier system. The
impulse response is
(2.1)
where is the Dirac impulse function, is the channel delay which is
assumed to be known and perfectly compensated for and denotes a sam-
ple from a correlated complex Gaussian random process with zero mean.
The auto correlation of the fading process is
, (2.2)
where is the zeroth order Bessel function of the rst kind, is the
maximum Doppler frequency, is expected value of and the asterix
denotes complex conjugate. This is the same model for a mobile radio Ray-
leigh fading channel as used in e.g. [1] and [15] and is by far the most used
model. It is frequently referred to as Jakes model [38]. At the output of the
receiver lter we obtain at time the received value
, (2.3)
where the noise sequence is a complex uncorrelated Gaussian ran-
dom processes with zero mean and variance .
Table 2.1 Gray encoding used in this paper
Bit Pattern 00 01 10 11
Multi -
carrier
Single-
Carrier
C
n k ,
= C
n k 1 ,
jC
n k 1 ,
jC
n k 1 ,
C
n k 1 ,
c
k
= c
k 1
jc
k 1
jc
k 1
c
k 1
g
k
t ( ) h
k
t
k
( ) =
t ( )
k
h
k
h
k
{ }
k
E h
k
h
k k

[ ] J
0
2k f
d
T ( ) = =
J
0
( ) f
d
E x [ ] x
kT
r
k
c
k
h
k
w
k
+ =
w
k
{ }
w
2
14
Multicarrier System
For the multicarrier system we model the channel as a time varying FIR-l-
ter of length where each tap is assumed constant during at least one mul-
ticarrier symbol. Hence the channel impulse response at time index is
, (2.4)
where is the attenuation and is the delay of the th path. It is
assumed throughout this paper that . Each tap is indepen-
dently modeled as random variable with Rayleigh distributed amplitude and
uniformly distributed phase and a correlation function according to Jakes
model. We will for simplicity assume that the power delay prole is uni-
form, i.e. all channel taps are random variables with the same variance. The
multicarrier system used in this chapter is an Orthogonal Frequency Divi-
sion Multiplexing (OFDM) system but the method will work for other mul-
ticarrier systems providing that the correlation between the received values
on different sub-carriers is known. Assuming rectangular pulse shaping the
th transmitted OFDM symbol can be expressed in complex base-band
notation as
. (2.5)
where is the sum of the symbol time and the cyclic prex length
and is the total number of carriers. The cyclic prex is inserted by the
transmitter in order to remove the intersymbol interference (ISI) and inter-
channel interference (ICI) that would otherwise cause degradation to the
system performance [5]. The received signal is obtained by a convolu-
tion between the transmitted signal and the channel impulse response
as depicted in Fig. 2.1.a. When the number of sub-channels is
L
k
g
k
t ( ) h
i k ,
t
i k ,
( )
i 0 =
L 1
=
h
i k ,

i k ,
i
i k ,
iT
s
N
c
=
k
s
k
t ( )
1
T
s
--------- C
n k ,
e
j2
n
T
s
-----t
n 0 =
N
c
1
; t kT k 1 + ( )T { , }
0 ; otherwise
'
=
T T
s
T
c p
N
c
r
k
t ( )
s
k
t ( )
h
k
t ( ) N
c
15
large each sub-channel will have a narrow bandwidth and we can approxi-
mately say that the fading on each sub-channel is at. The transmitted data
symbols on sub-carrier at time index will then be affected by the chan-
nel by a scaling of amplitude and a phase rotation. It is straightforward to
show [60] that after the DFT in the receiver we have
, (2.6)
where is the received value of the th OFDM symbol at the th sub-
channel (see Fig. 2.1.b). is the channel gain and is complex
additive white Gaussian noise with zero mean and variance . By com-
paring (2.6) and (2.3) we see that the equations are almost equivalent, only
differing in the presence of a frequency subscript index for the multicar-
rier system. For the channel model in Fig. 2.1.b the channel gain is
correlated both in time and in frequency. The correlation function is
dened as
, (2.7)
where is the frequency distance and is the time distance. It is shown
in Appendix 2A that for the channel with a uniform power delay prole
equation (2.7) evaluates to
. (2.8)
n k
R
n k ,
C
n k ,
H
n k ,
W
n k ,
+ =
R
n k ,
k n
H
n k ,
W
n k ,
W
2
n
H
n k ,
n k ,
n k ,
E H
n k ,
H
n n k k + , +
[ ] =
n k
n k ,
J
0
2 f
d
T
tot
k ( )
1
L
---
nL
N
c
--------------
,
_
sin
n
N
c
-----------
,
_
sin
---------------------e
jn L 1 ( ) N
c

=
DECISION DIRECTED CHANNEL ESTIMATION
16
2.3 DECISION DIRECTED CHANNEL
ESTIMATION
In the single-carrier and the multicarrier systems we end up with almost the
same equations at the receiver. In order to make a coherent decision on
which symbols were transmitted we need to know the channel gain or
for the single-carrier and multicarrier system respectively. We may
make use of our knowledge of the channel correlation function given in
(2.2) for the single-carrier system and in (2.8) for the multicarrier case as
well as all previously detected symbols. We will start our development with
the single-carrier system and then generalize the procedure to the multicar-
rier case in two steps.
Assuming the data symbols are all known for , we introduce a
scaled received sample
. (2.9)
Figure 2.1 (a) The multicarrier channel model viewed in the time domain and (b)
the multicarrier channel model viewed in the frequency domain.
.
.
.
.
.
.
Multipath
Frequency Selective
Rayleigh Fading
Channel
Flat Fading
Channel
Flat Fading
Channel
Flat Fading
Channel
(a) (b)
C
N
c
1 k ,
C
1 k ,
C
0 k ,
R
0 k ,
R
1 k ,
R
N
c
1 k ,
s
k
t ( )
r
k
t ( )
h
k
H
n k ,
c
l
{ } l k <
y
l
r
l
c
l
=
17
Previously scaled received samples are strongly correlated with the channel
gain that we want to estimate. We choose an estimate that minimizes the
estimation error in the mean square (MMSE) sense, i.e. we want to mini-
mize
, (2.10)
where denotes the estimate and the absolute value of respectively.
The data to be used in the channel estimation are stored in a data vector con-
taining the last received samples scaled by the transmitted symbol i.e.
, (2.11)
where the apostrophe denotes vector transpose. The problem is now reduced
to estimating the channel gain given that the data vector has been
observed. By noting that all our data are normally distributed as well as the
parameter that we want to estimate we obtain the best estimator in the
MMSE sense as a linear combination of our data as [42], [43]
, (2.12)
where is a column vector with real valued elements. Note that the vec-
tor stays constant as long as the matrixes and can be considered
constant. The correlation matrixes and are dened as
(2.13)
and
h
k
E h
k
h
k
2
[ ]
x x x
K
x
k
y
k 1
y
k 2
y
k K

=
h
k
x
k
h
k
C
hx
C
xx
1
x
k
b
x
k
= =
b K
b C
hx
C
xx
C
hx
C
xx
C
xx
E x
k
x
k
[ ]
w
2
A
2
--------- +
0

K 1
1

0
w
2
A
2
--------- +
K 2

K 1

K 2

0
w
2
A
2
--------- +
= =
18
(2.14)
respectively, where the superscript denotes vector transpose and conju-
gate. When this channel estimator is used together with a coherent detector
the data symbols are not known. In this paper we propose to use past deci-
sions to scale the received samples instead of the unknown transmitted sym-
bols, i.e.
. (2.15)
For PSK constellations we may of course multiply with the conjugate
instead, i.e. if we prefer to implement multiplications instead of
divisions. The decision device is now a traditional coherent DQPSK detec-
tor with input
(2.16)
where the superscript refers to diversity branch number and is the total
number of diversity branches. The channel estimation is done independently
in each branch and the diversity decision is used instead of the unknown
transmitted symbols during channel prediction. The MMSE channel estima-
tor may be described as a -tap nite impulse response lter where the l-
ter coefcients are included in the vector and the input
to the lter are the scaled received samples , see Fig. 2.2.
The MMSE channel estimator for the single-carrier systems is further
described in [68]. We have here used a scaled version of the received signal
to avoid a time varying lter. A time varying solution is given in [67] but it
is more complex to implement. Throughout this paper we assume, for the
single-carrier case, perfectly known maximum Doppler frequency and
noise power . Estimation of these parameters, or and directly,
is a topic for future research.
C
hx
E h
k
x
k
[ ]
1

2

K
= =
y
k
r
k
c
k
=
y
k
r
k
c
k
=
z
k
r
k
d ( )
h
k
d ( )
d 1 =
D
=
d D
K
b b
1
b
2
b
K
=
y
k
f
d
w
2
C
hx
C
xx
19
Scalar Solution for the Multicarrier System
We will now generalize the detector described above for a multicarrier sys-
tem. The rst generalization is to use information from sub-carriers nearby
in frequency as well as in time. By examining (2.8) we see that the channel
gain coefcients are highly correlated both in time and frequency. The
vector containing the data to be used for channel estimation is now dened
as
(2.17)
where the apostrophe denotes vector transpose and is a vector of
scaled previously received samples from nearby sub-chan-
nels at time index given by
(2.18)
where
. (2.19)
Our MMSE channel estimator is now given by
Figure 2.2 The MMSE channel estimator described as a -tap nite impulse
response lter.
. . .
. . .
. . .
r
k
y
k 1
y
k 2
y
k K
b
1
b
2
b
K
h
k
h
c
k
y
k
c
k
z
k
r
k
Other diversity branches

+
K
H
n k ,
X
n k ,
Y
n k 1 ,
Y
n k 2 ,
Y
n k K ,

=
Y
n k ,
N 2M 1 + =
k
Y
n k ,
Y
n M k ,
Y
n M 1 k , +
Y
n M k , +
=
Y
n k ,
R
n k ,
C
n k ,
=
20
(2.20)
where is a column vector with elements. The correlation matrices
are dened as and respec-
tively, and are easily calculated using (2.17) and (2.8). The principle of the
LMMSE channel estimator is illustrated in Fig. 2.3 where the lled circles
represent previously received values and where we use the values in the
shaded box to calculate the estimate . The input to the decision device
is now given by
. (2.21)
We assume perfectly known channel correlation function and noise
power . The scalar channel estimator for the multicarrier system is fur-
ther described in [24].
Figure 2.3 Principle of the scalar MMSE channel estimator for the multicarrier
system. The lled circles represent scaled previously received samples according
to (2.19). The values in the shaded box are used in the vector according to
(2.17) and the values in the dashed box are used in according to (2.18). In
this example sub-channels and previous symbols are used to
calculate the channel estimate .
H
n k ,
C
HX
C
XX
1
X
n k ,
b
X
n k ,
= =
b KN
C
XX
E X
n k ,
X
n k ,
[ ] = C
HX
E H
n k ,
X
n k ,
[ ] =
H
n k ,
Z
n k ,
R
n k ,
d ( )
H
n k ,
d ( )
d 1 =
D
n k ,
w
2
Previously received symbols
F
r
e
q
u
e
n
c
y

t
a
p
s
X
n k ,
Y
n k 1 ,
H
n k ,
Y
n k 4 ,
Y
n 1 k 1 ,
Y
n k 1 ,
Y
n 1 k 1 , +
X
n k ,
Y
n k 1 ,
N 3 = K 4 =
H
n k ,
21
Vector Solution for the Multicarrier System
If we alter the formulation slightly we obtain a vector channel estimator. We
may use the data from all sub-carriers to estimate the channel gain on
every sub-carrier simultaneously. We dene our data vector
(2.22)
where is a vector of scaled previously received samples given by
. (2.23)
The channel vector is
. (2.24)
Our estimate is then given by
(2.25)
where is a matrix with elements. The correlation matrixes
and are dened as and
respectively. The principle of the MMSE vector channel estimator is illus-
trated in Fig. 2.4.a where the lled circles represent scaled previously
received samples according to (2.19). The values in the shaded box are used
in the vector and the values in the dashed boxes are used in according
to the description above. In the example given in Fig. 2.4.a, = 4 previous
symbols are used to calculate the channel estimate . If we write the chan-
nel estimate as
(2.26)
N
c
X
k
Y
k 1
Y
k 2
Y
k K

=
Y
k
N
c
Y
k
Y
0 k ,
Y
1 k ,
Y
N
c
1 k ,

=
H
k
H
0 k ,
H
1 k ,
H
N
c
1 k ,

=
H
k
C
HX
C
XX
1
X
k
b
X
k
= =
b
N
c
KN
c
C
XX
C
HX
C
XX
E X
k
X
k
[ ] = C
HX
E H
k
X
k
[ ] =
X
k
Y
k
K
H
k
H
k
b
X
k
A
1
A
2
A
K
Y
k 1
Y
k 2
Y
k K
A
1
Y
k 1
A
2
Y
k 2
A
K
Y
k K
+ + +
= =
=
22
we see that the channel estimator can be implemented as an dimensional
lter with vectors as inputs and with matrixes of size as l-
ter coefcients, as shown in Fig. 2.4.b. The input to the decision device is
still given by (2.21). Further details about the MMSE vector channel estima-
tor for the multicarrier system are found in [25].
Figure 2.4 (a) Principle of the MMSE vector channel estimator for the
multicarrier system. (b) The MMSE vector channel estimator described as an N
c
dimensional lter.
N
c
Y
k
A
k
N
c
N
c
(b)
Previously received symbols
F
r
e
q
u
e
n
c
y

t
a
p
s
(a)
BIT ERROR PROBABILITY AND ESTIMATION ERROR ANALYSIS
23
2.4 BIT ERROR PROBABILITY AND
ESTIMATION ERROR ANALYSIS
We will evaluate the bit error performance given that we have a channel esti-
mate or obtained by any of the methods previously described in
this paper. For the multicarrier case we may use both the scalar or the vector
solution to obtain a channel estimate on sub-carrier . The input to the deci-
sion device is given by (2.16) for the single-carrier system and by (2.21) for
the multicarrier system. We will also evaluate the estimation error variance
in each diversity branch.
Bit Error Probability
From (C-13) in [55] we nd that the optimum decision boundaries for an
MMSE based detector are equal to the optimum decision boundaries of a
traditional differential detector. The bit error probability of DQPSK is given
by
(2.27)
which is equal to (C-20) in [55] (note however the sign error in (C-20)).
This is in fact the bit error probability for the transmission of a single sym-
bol only (one-shot situation) and it does not take into account the effects of
differential decoding which precedes the detector, and assumes that all pre-
vious decisions are correct.
In order to evaluate the bit error probability a normalized correlation
coefcient, , needs to be found. For the single-carrier system is dened
as
. (2.28)
H
n k ,
h
k
n
P
b
1
2
--- 1

2

2
----------------------
2d
d ,
_
1
2
4 2
2
---------------------
,

_
d
d 0 =
D 1
E r
k
h
k
[ ]
E r
k
2
[ ]E h
k
2
[ ]
------------------------------------------ =
BIT ERROR PROBABILITY AND ESTIMATION ERROR ANALYSIS
24
We have omitted the diversity branch index since we only study equal con-
ditions in the diversity branches. It is straightforward to derive this coef-
cient and it is given by
. (2.29)
It should be noted that the phase of is equal to the phase of and that the
absolute value of is independent of .
The correlation coefcient for the multicarrier system with the MMSE
based scalar channel estimator is dened as
(2.30)
and (2.30) evaluates to
. (2.31)
The bit error probability for the multicarrier system will be larger at the
frequency edges where the number of frequency neighbors is smaller. If we
have many sub-carriers we may neglect this edge effect and say that the
average bit error probability is equal to the bit error probability of the sub-
carrier on the center frequency.
Differential channel estimation as described in for example [60] is often
used in OFDM systems. The bit error performance of a detector using dif-
ferential channel estimation is exactly the same as the scalar MMSE channel
estimation based detector with = 1 and [24].
c
k
C
hx
C
xx
1
C
hx
A
2
w
2
+
---------------------------------- =
c
k
c
k
E R
n k ,
H
n k ,

[ ]
E R
n k ,
2
[ ]E H
n k ,

2
[ ]
---------------------------------------------------------- =
C
n k ,
C
HX
C
XX
1
C
HX
A
2
W
2
+
--------------------------------------- =
N K 1 =
NUMERICAL RESULTS
25
For the multicarrier system with the vector channel estimator we may
evaluate the bit error probability of the sub-carrier on the center frequency
to obtain a lower bound. We end up with an expression of that is
(2.32)
where denotes the th row of .
Estimation Error Variance
For the single-carrier system the variance of the channel estimation error is
dened as . This error variance is given by [42]
(2.33)
where is the variance of . For the multicarrier system the MMSE
scalar channel estimator has an error variance, , that is
(2.34)
where is the variance of . The error variance for the vector chan-
nel estimator is on sub-carrier n is
. (2.35)
2.5 NUMERICAL RESULTS
In this section we present analytical and simulated bit error probability
results. It should be noted that the analytical results are based on the
assumption that the previous decisions used in the channel estimate are cor-
rect. Further more, the analytical results give the bit error probability only
for the transmission of a single symbol and it does not take into account the
effects of differential decoding which precedes the detector. Therefore the
analytical results are lower bounds the true bit error probabilities. The pro-
C
n k ,
C
HX
{ }
n
C
XX
1
C
HX
{ }
n
A
2
W
2
+
----------------------------------------------------------- =
C
HX
{ }
n
n C
HX
2
E h
k
h
2
[ ] =
h
2
C
hx
C
xx
1
C
hx
=
h
2
h
k
{ }
H
2
C
HX
C
XX
1
C
HX
H
2
H
n k ,
{ }
H
2
C
HX
{ }
n
C
XX
1
C
HX
{ }
n
=
NUMERICAL RESULTS
26
posed detectors have been simulated for different maximum Doppler fre-
quencies and these results take into account errors in the channel estimation
due to incorrect decisions and the effect of differential decoding. Compari-
sons are made with a traditional differential detector according to [1] and a
genie based coherent detector where the fading channel is assumed com-
pletely known (ideal coherent detector). The performance of a genie aided
detector is a lower bound to an MMSE based detector. Results are shown
without diversity and with diversity of two independent branches. All simu-
lations are performed with at most 1 million transmitted bits. Bit error prob-
ability is always shown versus mean received energy per bit per double-
sided noise spectral density . We shall begin by presenting results for
the single-carrier system.
Results for the Single-Carrier System
First we will study the bit error rate for various normalized Doppler fre-
quencies. In Fig. 2.5 we show analytical bit error probability for an MMSE
based detector with ve taps ( = 5) for normalized Doppler frequencies
equal to 0.1, 0.05, 0.01 and 0.001. The lower bound obtained by the
genie aided detector is shown as a comparison. We show results without
diversity and with two path diversity. Here we clearly see that the lower
bound is approached for small Doppler frequencies, while the slope of the
curves becomes less attractive with increasing Doppler frequency. The deg-
radation in signal to noise ratio at low bit error probabilities therefore
becomes quite large for large Doppler frequencies. The advantage of this
detector is that it does not show any irreducible error oor as a conventional
differential detector does.
For comparison we show bit error probability for the same parameters
and a differential detector in Fig. 2.6. By comparing these two graphs, we
clearly see the superiority of the MMSE based detector for those SNR val-
ues where the differential detector shows an irreducible error oor. For
smaller SNR values the analytical results show a small gain for the MMSE
based detector but this will disappear when we examine the simulation
results.
E
b
N
0
K
f
d
T
NUMERICAL RESULTS
27
Figure 2.5 Analytical bit error probability for the single-carrier MMSE based
detector with taps for 4 values of normalized Doppler frequency, with and
without diversity. As a reference we also consider a genie based coherent detector
as a lower bound.
Figure 2.6 Analytical bit error probability for a single-carrier differential detector
for 4 values of normalized Doppler frequency, with and without diversity. As a
reference we also consider a genie based coherent detector as a lower bound.
0 5 10 15 20 25 30 35 40
10
4
10
3
10
2
10
1
Coherent without diversity
Coherent with diversity
Lower bound without diversity
Lower bound with diversity
Mean received E
b
/N
0
(dB)
B
i
t

E
r
r
o
r

P
r
o
b
a
b
i
l
i
t
y
f
d
T 0.1 =
f
d
T 0.01 =
f
d
T 0.001 =
f
d
T 0.05 =
K 5 =
0 5 10 15 20 25 30 35 40
10
4
10
3
10
2
10
1
Differential without diversity
Differential with diversity
Mean received E
b
/N
0
(dB)
B
i
t

E
r
r
o
r

P
r
o
b
a
b
i
l
i
t
y
f
d
T 0.1 =
f
d
T 0.01 =
f
d
T 0.001 =
f
d
T 0.05 =
NUMERICAL RESULTS
28
In Fig. 2.7 the simulated results for the single-carrier system are pre-
sented for the same parameters as in the previous two gures. By comparing
the analytical results in Fig. 2.5 and the simulated results in Fig. 2.7 we
clearly see some degradation due to error propagation in the channel estima-
tor and the differential decoding. The difference between the analysis and
the simulated results are small for low bit error probabilities.
In Fig. 2.8 we examine the performance of the MMSE based detector of
various channel predictor orders. The normalized Doppler frequency is
0.05 and results are shown for = 1, 2, 5 and 25. We see that changing the
channel predictor order, , from 1 to 2 lowers the error oor by an order of
magnitude for the case without diversity. For predictor orders higher than 2
the error oor is not visible in the gure.
Figure 2.7 Simulated bit error probability for single-carrier MMSE based detector
with taps for 4 values of normalized Doppler frequency with and without
diversity. As a reference we also consider a genie based coherent detector as a
lower bound.
0 5 10 15 20 25 30 35 40
10
4
10
3
10
2
10
1
Without diversity
With diversity
Mean received E
b
/N
0
(dB)
B
i
t

E
r
r
o
r

P
r
o
b
a
b
i
l
i
t
y
f
d
T 0.1 =
f
d
T 0.01 =
f
d
T 0.001 =
f
d
T 0.05 =
K 5 =
f
d
T
K
K
NUMERICAL RESULTS
29
Results for the Multicarrier System
The multicarrier system can explore the correlation present between the
sub-carriers in order to produce a better channel estimate. For the scalar
MMSE channel estimator we may vary both the number of previous sym-
bols, , used and the number of nearby sub-carriers, , to use. In Fig. 2.9
results for a multicarrier system with sub-carriers are shown.
The channel has independent taps and the normalized Doppler fre-
quency is 0.05. The poorest performance with a high error oor is
obtained when only one previously received symbol is used on the same
sub-carrier ( = 1 and = 1). By using ve previously received samples
the error oor is signicantly reduced and is not seen in the gure, just as in
the single-carrier case. By making use of the correlation between the sub-
carriers the bit error probability is reduced even more. This is clearly seen
by looking at the curves with = 5, = 5 and at the price of higher com-
plexity even better results are obtained with = 7, = 7.
Figure 2.8 Analytical bit error probability for a predictor based detector of various
length, with and without diversity. The normalized Doppler frequency is 0.05.
As a reference we also consider a genie based coherent detector as a lower bound.
0 5 10 15 20 25 30 35 40
10
4
10
3
10
2
10
1
Coherent without diversity
Coherent with diversity
K 1 =
K 5 =
K 25 =
Mean received E
b
/N
0
(dB)
B
i
t

E
r
r
o
r

P
r
o
b
a
b
i
l
i
t
y
K 2 =
f
d
T
K N
N
c
128 =
L 5 =
f
d
T
K N
K N
K N
NUMERICAL RESULTS
30
In Fig. 2.10 we show simulated results for a multicarrier system with the
same parameters as in Fig. 2.9. By comparing the curves marked with plus
signs ( = 5, = 1) and those marked with cross signs ( = 5, = 5) in
Fig. 2.10, we see that the latter performs better at low bit error probabilities
where the error propagation in the channel estimation procedure is not as
severe.
The multicarrier system takes advantage of the correlation between the
sub-carriers. The bit error rate performance will therefore depend on how
correlated the channel is between the different sub-carriers. In Fig. 2.11 we
show the results for a multicarrier system with carriers where we
vary the length of the channel delay prole, , and hence also the correla-
tion between sub-carriers (see (2.8)). Results are shown for = 5, 25, 50
and 75. For large channel lengths the performance will be degraded due to
two factors, one being the reduced correlation between sub-carriers. The
other factor is that we need a longer cyclic prex to remove the intersymbol
interference caused by the channel and this will lead to a loss in signal
Figure 2.9 Analytical bit error probability for the MMSE scalar channel
estimation based multicarrier system with various and . The number of sub-
carriers . The channel has = 5 taps and = 0.05.
0 5 10 15 20 25 30 35 40
10
4
10
3
10
2
10
1
Without diversity
With diversity
K 1 N , 1 = =
K 5 N , 1 = =
K 7 N , 7 = =
Mean received E
b
/N
0
(dB)
B
i
t

E
r
r
o
r

P
r
o
b
a
b
i
l
i
t
y
K 5 N , 5 = =
K N
N
c
128 = L f
d
K N K N
N
c
128 =
L
L
NUMERICAL RESULTS
31
energy per transmitted bit. An interesting property to observe in Fig. 2.11 is
that the irreducible bit error oor is not inuenced by the sub-carrier corre-
lation. In fact the error oor is only inuenced by the normalized Doppler
frequency and the number of previously received samples, , used in
the channel estimation procedure. The use of nearby sub-carriers in the
channel estimation will improve the bit error rate performance for error
probabilities larger than the error oor.
Figure 2.10 Simulated bit error probability for the MMSE scalar channel
estimation based multicarrier system with various and . The number of sub-
carriers = 128. The channel has = 5 independent taps and the normalized
Doppler frequency is = 0.05.
0 5 10 15 20 25 30 35 40
10
4
10
3
10
2
10
1
Without diversity
With diversity
Mean received E
b
/N
0
(dB)
B
i
t

E
r
r
o
r

P
r
o
b
a
b
i
l
i
t
y
K 1 N , 1 = =
K 5 N , 1 = =
K 7 N , 7 = =
K 5 N , 5 = =
K N
N
c
L
f
d
f
d
T K
NUMERICAL RESULTS
32
Finally we show results for the vector channel estimator for the multicar-
rier case. Again the number of sub-carriers is 128 and is set to 0.1.
The channel length is = 5 and results for = 2 and 5 are shown. The ana-
lytical and simulated results, with and without diversity, are shown together
in Fig. 2.12. We see that the difference between the simulated and analytical
results are about 3 dB without diversity and 2 dB with diversity at bit error
probability 10
-3
in the case when = 5. The vector channel estimator
always uses the information between all frequency neighbors and will per-
form better than a scalar channel estimator not using all available informa-
tion, when the channel is highly correlated in frequency.
Figure 2.11 Analytical bit error probability for the MMSE scalar channel
estimation based multicarrier system on channels of various length, = 5, 25, 50
and 75. = 3 previously received samples from = 11 sub-carrier are used for
channel estimation. Normalized Doppler is 0.1 and =128.
0 5 10 15 20 25 30 35 40
10
4
10
3
10
2
10
1
Without diversity
With diversity
Mean received E
b
/N
0
(dB)
B
i
t

E
r
r
o
r

P
r
o
b
a
b
i
l
i
t
y
L 75 =
L 50 =
L 5 =
L 25 =
L
K N
f
d
T N
c
N
c
f
d
T
L K
K
CHAPTER SUMMARY
33
2.6 CHAPTER SUMMARY
Decision directed coherent detectors for single- and multicarrier systems
based on MMSE channel estimation are proposed. We use differential
encoding of the data to avoid the inherent 90 degrees phase ambiguity of
QPSK due to decision errors. The channel model used for the single-carrier
system is a at Rayleigh fading channel with correlation of the fading pro-
cess according to Jakes model. The multicarrier system uses rectangular
pulse shapes and the channel model used is a time varying FIR-lter of
length with independent fading of each tap. We assume perfect know-
ledge of the channel correlation function and the AWGN noise variance in
both cases. The performance of the proposed detectors are evaluated in
terms of channel estimation error variance and bit error probability. The
analysis is performed under the assumption that all previous decisions are
Figure 2.12 Analytical and simulated bit error probability for multicarrier system
with the vector channel estimator. = 2 or 5 previously received samples from all
sub-carrier are used for channel estimation. = 0.1, = 128 and the channel
has = 5 taps.
0 5 10 15 20 25 30 35 40
10
4
10
3
10
2
10
1
Simulated without diversity
Analytical without diversity
Simulated with diversity
Analytical with diversity
Mean received E
b
/N
0
(dB)
B
i
t

E
r
r
o
r

P
r
o
b
a
b
i
l
i
t
y
K 5 =
K 2 =
K
f
d
T N
c
L
L
CHAPTER SUMMARY
34
correct and the differential decoding is not taken into account. Simulations
are performed to study the degradation caused by these effects. The analyti-
cal and simulated results show that the proposed detectors do not suffer
from the problem of an irreducible bit error probability oor and have sig-
nicantly lower bit error probability as compared to differential detection.
CHANNEL CORRELATION FUNCTION
35
APPENDIX 2A
The channel correlation function is dened in (2.7). The channel transfer
function is the discrete Fourier transform of the channel impulse
response. We immediately get
(2A-1)
where the rst equality comes from the fact that different channel taps are
uncorrelated and we get no cross terms. The second equality is obtained by
moving the expectation into the sum. The term is the variance of ,
i.e. the power of the th channel tap. For the uniform power delay prole we
set , independent of , so that the total power of all channel taps
is unity. By doing so we obtain
(2A-2)
If we instead use an exponential power delay prole, i.e.
, (2A-3)
where is the channel delay spread divided with the sampling time
and is a constant chosen so that the total power of all taps
is equal to one we obtain
H
n k ,
{ }
n k ,
E h
i k ,
e
j2ni N
c

i 0 =
L 1
,

_
h
l k k ,
e
j2 n n ( )l N
c

l 0 =
L 1
,

_
E h
i k ,
h
i k k ,
e
j2ni N
c

i 0 =
L 1
J
0
2 f
d
kT ( )
i
2
e
j2ni N
c

i 0 =
L 1
=
= =
i
2
h
i k ,
{ }
i
i
2
1 L = i
n k ,
J
0
2 f
d
kT ( )
1
L
--- e
j2ni N
c

i 0 =
L 1
J
0
2 f
d
kT ( )
1
L
---
1 e
j2nL N
c

1 e
j2n N
c

------------------------------------
J
0
2 f
d
kT ( )
1
L
---
nL
N
c
--------------
,
_
sin
n
N
c
-----------
,
_
sin
-----------------------e
j2n L 1 ( ) N
c

=
=
=
i
2
ce
i m
ds

=
m
ds
T
s amp
T
s
N
c
= c
36
. (2A-4)
By using the approximation we get an expression suitable for
small that is
, (2A-5)
where is the coherent bandwidth of the channel dened as
. Equation (2A-5) is found also in [80] where it is derived
with equality under the assumption that the channel delays are expo-
nentially distributed random variables.
n k ,
J
0
2 f
d
kT ( )
1 e
T
s
N
c
ds

1 e
T
s
N
c
ds
j2n N
c
+ ( )
----------------------------------------------------------- =
e
x
1 x
n
n k ,
J
0
2 f
d
kT ( )
1 j
n
B
c
T
s
----------- +
-----------------------------------
B
c
B
c
1 2
ds
=
i k ,
37
CHAPTER
3.1 INTRODUCTION
One of the drawbacks of OFDM is the large peak to mean power ratio that
causes problems since linear ampliers need to be used to avoid cross talk
interference between sub-channels. In this chapter a proposition of a parallel
combinatory OFDM system (PC-OFDM) that can increase the bandwidth ef-
ciency and at the same time lower the peak to mean power ratio compared to
an ordinary OFDM system, is put forward. In some cases we can decrease the
bit error rate as well. The basic idea is to use not all, but only a subset of the
available sub-carriers in each OFDM symbol. We map some bits to a set of
sub-carriers to use and other bits to the phase of the selected sub-carriers.
This chapter is organized as follows. In Section 3.2 the system and channel
models are presented and in Section 3.3 the bandwidth efciency of the PC-
OFDM system is compared to that of ordinary OFDM. In Section 3.4 a bit
mapping procedure for PC-OFDM is proposed. The bit error probability of the
PC-OFDM system using M-PSK modulation is derived on a Gaussian channel
in Section 3.5. Some analytical and simulated results for the AGWN channel
are presented in Section 3.6 and Section 3.7 deals with the performance of the
PC-OFDM system on Ricean at fading channels. The chapter is then summa-
rized in Section 3.8.
A PARALLEL COMBINATORY
OFDMSYSTEM
3
38
3.2 SYSTEM AND CHANNEL MODELS
We assume for simplicity that we use an OFDM system with rectangular
pulse shaping. The transmitted signal is thus generated in the same fashion
as in (2.5). Parallel combinatory signaling for spread spectrum systems has
previously been studied in [62], [63], [64], and here we use similar princi-
ples for OFDM systems. The PC-OFDM system proposed here does not use
all the available sub-carriers but only a subset of carriers. The
selected carriers are then modulated by -ary PSK. Thus we map
bits to a subset of carriers to use and bits to the phase of the selected
sub-carriers. We dene as the row vector containing the parallel combi-
natory bits, as the row vector containing the PSK bits and nally
. Please note that we have skipped the time subscript
index and only study one single PC-OFDM symbol. The number of bits per
OFDM symbol, , is
(3.1)
where is the largest integer smaller than or equal to and
2
log denotes
the base-2 logaritm. The receiver rst detects the sub-carriers with the
largest amplitude and decodes the parallel combinatory bits accord-
ingly, and then performs traditional -PSK detection on the selected carri-
ers to obtain the . Obviously the ordinary OFDM system is a special
case of a PC-OFDM system where all sub-carriers are used and conse-
quently no bits are used to choose amongst the sub-carriers. The channel
models used in this chapter are the AWGN and the at fading Ricean chan-
nel models.
N
tot
N
pc
N
pc
M m
pc
m
psk
b
pc
b
psk
b
tot
b
pc
b
psk
=
m
tot
m
tot
m
psk
m
pc
+ N
pc
M log
2
N
tot
N
pc
,
_
log
2
+ = =
x x
N
pc
m
pc
M
m
psk
BANDWIDTH EFFICIENCY
39
3.3 BANDWIDTH EFFICIENCY
Assuming that the number of sub-carriers, , is relatively large we dene
the bandwidth of the PC-OFDM system as and the data rate
as . For reasons of simplicity we assume that the length of the
cyclic prex is negligible compared to the useful symbol duration .
This assumption will not affect the comparison between the PC-OFDM sys-
tem and the ordinary OFDM system, since the cyclic prex will be used in
both systems. The bandwidth efciency of the PC-OFDM system may now
be dened as . By plotting the bandwidth efciency, , versus
as in Fig. 3.7, we see that the PC-OFDM system gains in band-
width efciency compared to an ordinary OFDM system for some ranges of
. The ordinary OFDM system uses all available sub-carriers
and hence . We see that a system using BPSK ( = 2)
should use less carriers than a system using 16-PSK ( = 16) to obtain a
maximal bandwidth efciency. This is obvious since when deciding not to
use one sub-carrier for PSK modulation we lose one bit in the BPSK case
and four bits in the 16-PSK case. However, the gain in the number of paral-
lel combinatory bits is in both cases the same. Marked with white and black
circles in Fig. 3.7 are the PC-OFDM systems and the ordinary OFDM sys-
tems that have the same bandwidth efciency. If for example we
can transmit the same number of bits per second per Hz as an ordinary
OFDM system by using = 9, 18, 25, 29 and 31 for the signal constella-
tions BPSK, QPSK, 8-PSK, 16-PSK and 32-PSK respecti vely. Thus the PC-
OFDM system can in some cases obtain a higher bandwidth efciency than
an ordinary OFDM system. By using fewer sub-channels we also decrease
the peak to average power ratio and the employed sub-carriers can transmit
at a higher power which, as we will see in Section 3.5, can decrease the bit
error rate as well.
N
tot
B N
tot
T
s
=
R m
tot
T
s
=
T
cp
T
s
R
b
B =
N
pc
N
tot
N
pc
N
tot
N
tot
N
pc
N
tot
1 = M
M
N
tot
32 =
N
pc
BIT MAPPING PROCEDURE
40
3.4 BIT MAPPING PROCEDURE
The problem of mapping the bits to a PC-OFDM symbol deserves some
attention. First we need to nd out which sub-carriers to use for signaling.
Several solutions to this problem are possible. The obvious one is to gener-
ate a lookup table with entries. The entries may consist of vectors of
length where a zero in one position denotes that the corresponding
sub-carrier should not be used, and a one that it should be used for signaling.
To reduce the storage space needed for this approach we may produce a
smaller look-up table containing only entries that when shifted cyclically
produces all entries. These solutions will however fail when is
large due to the exponential increase in storage needed. We may use a con-
stant weight code instead, preferably with high code rate to avoid losing
data rate. Constant weight codes are however nonlinear to their nature (the
Figure 3.1 Bandwidth efciency of the PC-OFDM system. The black circles
represent the ordinary OFDM systems and the white circles represent PC-OFDM
systems with the same bandwidth efciency. -PSK is used with = 2, 4, 8, 16
and 32.
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
1
2
3
4
5
Npc / Nfft
R
b

/

B

[
b
i
t
s
/
s
/
H
z
]
Bandwidth Efficiency
M = 32
M = 16
M = 8
M = 4
M = 2
[
b
i
t
s
/
s
/
H
z
]
N
pc
N
tot
R
b
B
M M
2
m
pc
N
tot
2
m
pc
m
pc
41
all zero word is not a codeword) and are only available for very specic
lengths and weights. Thus we may use a specic constant weight code if its
weight and lengths happen to coincide with the and we want to
use. To our best knowledge there are no known general family of constant
weight codes for arbitrary lengths and weights with high rates, that are easy
to decode. We have thus chosen a different solution that is described in the
next sub-section.
Selecting the Sub-Carriers
The problem of generating a lookup table consisting of all possible ways to
choose out of may be formulated recursively. The computational
complexity to nd out the contents of a certain row in that lookup table
grows linear with so there is no need to store the table. Let
denote this table. We note that may be written as
(3.2)
where denotes a column vector of length containing only
ones, and a vector of length containing only zeros. The
association scheme expressed in (3.2) is known as the Johnson association
scheme [48]. We now see that if we are interested in nding the contents of
row in we simply compare with to nd out if the rst posi-
tion equals 0 or 1 and the other positions follow by the recursive form of
. We need only to take care of the trivial problems of nding
and to stop the recursion. Obviously this mapping works both
ways, i.e. given the contents of a certain row we can calculate the row index
. We may easily reformulate the problem if we prefer an iterative solution
instead. For mapping of the bits to a row index we have chosen the
natural binary code (NBC) [39].
N
pc
N
tot
N
pc
N
tot
N
tot
A
N
pc
N
tot
,
A
N
pc
N
tot
,
A
N
pc
N
tot
,
1 A
N
pc
1 N
tot
1 ,
0 A
N
pc
N
tot
1 ,
=
1
l
1
N
tot
1
N
pc
1
,
_
=
0 l
0
N
tot
1
N
pc
,
_
=
i A
N
pc
N
tot
,
i l
1
A
N
pc
N
tot
,
A
N N ,
A
0 N ,
i
m
pc
i
42
Positions of the PSK Symbols
If we are not careful in how to map the remaining bits on the selected
sub-carriers we will have error events of high probability where almost all
bits are received incorrectly. This is best explained by an example.
Example 3-1. Suppose we use QPSK with and together with the
bit mapping as given in Fig. 3.2. This system transmits 3 bits in the choice of
sub-carriers and 14 bits in the received phase of the selected sub-carriers. (Note
that we can transmit one extra bit compared to a system using all 8 sub-carriers).
Assuming that we will produce the vector
as input to the OFDM modulator. The rst three
bits select the sub-carriers 1 to 7 and the following bits select the phases of the
carriers. Suppose that due to channel disturbances we receive the vector
where we erroneously have chosen the last sub-
carrier instead of the rst. Following the de-mapping procedure we then receive
the bit vector where every single bit is wrong.
Figure 3.2 (a) mapping of the parallel combinatory bits and (b) mapping of the
PSK bits using Gray encoded QPSK with and .
m
psk
m
tot
N
tot
8 = N
pc
7 =
b
tot
00011001100110011 [ ] =
C 1 1 1 1 1 1 1 0 =
C
0 1 1 1 1 1 1 1 =
b
tot
11100110011001100 [ ] =
pc-bits Carriers to use
0 0 0 1 1 1 1 1 1 1 0
0 0 1 1 1 1 1 1 1 0 1
0 1 0 1 1 1 1 1 0 1 1
0 1 1 1 1 1 1 0 1 1 1
1 0 0 1 1 1 0 1 1 1 1
1 0 1 1 1 0 1 1 1 1 1
1 1 0 1 0 1 1 1 1 1 1
1 1 1 0 1 1 1 1 1 1 1
00
01
11
10
(a)
(b)
1
-1
-j
j
N
tot
8 = N
pc
7 =
43
To avoid this problem we need to place the PSK symbols in certain posi-
tions whenever possible, so that an erroneous decision of sub-carriers results
in as few errors as possible in the PSK modulated bits. Let denote the th
row of . We now assign position numbers, ranging from 1 to ,
to the nonzero coordinates of . This is denoted by a new vector by . Let
denote the th position of the vector and use the following ad hoc
rules for producing .
1. Divide the vector into as many sections of length as possible.
That is let , where the last part may or may not
be of length .
2. For = 0 to do
For each position do
If and the th PSK symbol is not yet assigned
a position in then let and mark that
the th PSK symbol is now assigned a position in .
next
next
3. If there are PSK symbols that are not yet assigned a position of
simply place them at any positions that are not used already.
The algorithm presented above is not known to be optimal. However, an
exhaustive computer search for better mapping procedures has been per-
formed for all cases where and there are several equally good but
no one better. The criterion used was the average number of PSK symbol
errors provided that two carriers are erroneously switched. The bit mapping
procedure is summarized in Fig. 3.3
Figure 3.3 The proposed bit mapping procedure.
I
i
i
A
N
pc
N
tot
,
N
pc
I
i
J
i
I n ( ) n I
J
i
I
i
N
pc
I
i
I'
0
I'
1
I'
l
=
N
pc
k l
n
I'
k
n ( ) 1 = n
J
i
J
i
kN
pc
n + ( ) n =
n J
i
n
k
J
i
N
tot
5
Split
NBC
Decoding
Select
sub-carriers
PSK
Mapping
To
i
Determine
PSK positions
IFFT
BIT ERROR PROBABILITY IN AWGN
44
We now continue with our previous example to see what will happen.
Example 3-1 (continued). The positions of the PSK symbols are now chosen
according to Fig. 3.4. If we receive the same vector as before,
we will now decode it to the bit vector
with ve erroneous bits wrong and twelve cor-
rect bits. We can not expect the parallel combinatory bits to be correct nor the
bits corresponding to the sub-carriers we did not choose, but all the other bits
were in this example received correctly and this is a signicant improvement.
3.5 BIT ERROR PROBABILITY IN AWGN
We will now derive the bit error performance of the PC-OFDM system on
an AWGN channel. Assume that the transmitted word at time index is
(3.3)
where denotes the transmitted value of sub-carrier and the apostrophe
denotes vector transpose. The time subscript index is dropped for conve-
nience. The transmitted amplitude is , where denotes the abso-
lute value of . At the receiver we obtain the vector where
(3.4)
is a vector of i.i.d. Gaussian random variables with variance and
zero mean. Taking the absolute value of the received vector we have a
Figure 3.4 Positions of the PSK symbols in Example 3-1.
C
0 1 1 1 1 1 1 1 =
b
tot
11100001100110011 [ ] =
pc-bits PSK positions
0 0 0 1 2 3 4 5 6 7 0
0 0 1 1 2 3 4 5 6 0 7
0 1 0 1 2 3 4 5 0 7 6
0 1 1 1 2 3 4 0 6 7 5
1 0 0 1 2 3 0 5 6 7 4
1 0 1 1 2 0 4 5 6 7 3
1 1 0 1 0 3 4 5 6 7 2
1 1 1 0 2 3 4 5 6 7 1
k
C
C
0
C
1
C
N
pc
1
0 0

=
C
n
n
k
C
n
A = x
x R C W + =
W
W
0
W
1
W
N
fft
1

=
N
tot

2
R
BIT ERROR PROBABILITY IN AWGN
45
vector of random variables where the values corresponding to the sub-
carriers used for transmission have a Ricean probability density function
and the other values are Rayleigh distributed, that is
(3.5)
where are Ricean and are Rayleigh random variables. There are
variables that have a Rayleigh distribution. The probability
density functions and the distribution functions for Ricean random variables
will be denoted and respectively and for the Rayleigh random
variables we will use and . These functions are found in
Appendix 3A. The smallest of the Ricean variables, which we denote
with has the probability density function [52]
(3.6)
and the distribution function
(3.7)
The largest value of the Rayleigh variables, , has the
probability density function
(3.8)
An error in the parallel combinatory symbol will occur if . This
will occur with the probability
(3.9)
Equation (3.9) may be numerically evaluated. Using (3.9) and assuming that
two PSK symbols are lost every time we choose the wrong subset of sub-
carriers in the detector we get an approximation of the bit error probability
as
N
pc
R
X
0
X
1
X
N
pc
1
Y
0
Y
1
Y
K 1

=
X
n
Y
n
K N
tot
N
pc
=
f
X
x ( ) F
X
x ( )
f
Y
y ( ) F
Y
y ( )
N
pc
X
min
f
X
min
x ( ) N
pc
1 F
X
x ( ) [ ]
N
pc
1
f
X
x ( ) =
F
X
min
x ( ) 1 1 F
X
x ( ) [ ]
N
pc
=
K N
tot
N
pc
= Y
max
f
Y
max
y ( ) K F
Y
y ( ) [ ]
K 1
f
Y
y ( ) =
X
min
Y
max
<
p Prob X
min
Y
max
< [ ] F
X
min
y ( ) f
Y
max
y ( ) y d
0
= =
NUMERICAL RESULTS ON AWGN CHANNELS
46
(3.10)
where
. (3.11)
3.6 NUMERICAL RESULTS ON AWGN
CHANNELS
In Fig. 3.5 to Fig. 3.8 we show the performance of the PC-OFDM system
with for BPSK, QPSK, 8-PSK and 16-PSK respectively and for
= 6, 10, 14, 18, 22, 26, 30 and 32. The solid lines in these gures are
analytical results according to (3.10). Since we have made approximations
in the analysis we have also performed simulations to verify the analytical
results. These simulated bit error curves are the dashed curves marked with
circles. The simulations were performed until at least 2000 bit errors were
obtained or stopped after two million bits were transmitted. By comparing
the analytical and simulated results we see that for high bit error rates the
analysis is slightly pessimistic and that for low bit error rates the approxima-
tions made are tight. We see that the bit error performance decreases when
decreases except for very high error rates. For low bit error rates the
PC-OFDM is superior compared to the conventional OFDM system on an
AWGN channel in terms of the needed to obtain a certain bit error
probability. It is clear that to obtain the lowest bit error rate we should use
, which is very similar to orthogonal signaling [55], but looking at
Fig. 3.1 we see that such a system has very low bandwidth efciency and the
comparison is in that respect unfair. Instead we can trade BER performance
against bandwidth efciency to some extent. The PC-OFDM system using
= 10 sub-carriers and BPSK in Fig. 3.5 has higher bandwidth efciency
P
b
p
m
pc
m
tot
----------
1
2
---
m
psk
m
tot
-----------
N
pc
2
N
pc
------------------ P
psk
2
N
pc
---------
1
2
--- +
,

_
+
1 p ( )
m
psk
m
tot
----------- P
psk
+
P
psk
1
M log
2
---------------erfc
m
tot
N
pc
----------
E
b
N
0
-------

M
-----
,

_
sin
,

_
N
tot
32 =
N
pc
N
pc
E
b
N
0
N
pc
1 =
N
pc
47
than an ordinary OFDM system and also performs better for bit error rates
lower than . For the systems using QPSK in Fig. 3.6 we may com-
pare the PC-OFDM system using 18 sub-carriers with the ordinary OFDM
system. The PC-OFDM system performs better for BER lower than
. Looking at Fig. 3.7 we see that the PC-OFDM system using 8-
PSK with is superior below bit error rates of . In Fig.
3.7 we use 16-PSK and we thus need to use 31 of the available 32 sub-carri-
ers to obtain the same bandwidth efciency as ordinary OFDM. By compar-
ing the results of the ordinary OFDM system with the curve using 30 sub-
carriers we see that the BER performances are almost the same and that only
small gains are feasible by using PC-OFDM. If we use M-PSK modulation
with we will not be able to design a PC-OFDM system having both
lower BER and higher bandwidth efciency compared to an ordinary
OFDM system.
Figure 3.5 Analytical (solid) and simulated (dashed) BER results for the PC-
OFDM system using BPSK ( = 2). = 32 and = 6, 10, 14, 18, 22, 26
and 30.
3 10
5
5 10
5
N
pc
26 = 3 10
2
M 32
1 2 3 4 5 6 7 8 9 10 11
10
6
10
5
10
4
10
3
10
2
10
1
10
0
OFDM
N
pc
= 30
26
22
18
14
10
6
B
E
R
E
b
/N
0
[dB]
M N
tot
N
pc
48
OFDM system using QPSK ( = 4). = 32 and = 6, 10, 14, 18, 22, 26
and 30.
OFDM system using 8-PSK ( = 8). = 32 and = 6, 10, 14, 18, 22, 26
and 30.
1 2 3 4 5 6 7 8 9 10 11
10
6
10
5
10
4
10
3
10
2
10
1
10
0
30
22
26
18
14
10
N
pc
= 6
OFDM
E
b
/N
0
[dB]
B
E
R
M N
tot
N
pc
1 2 3 4 5 6 7 8 9 10 11
10
6
10
5
10
4
10
3
10
2
10
1
10
0
N
pc
= 30
26
22
18
14
10
6
OFDM
B
E
R
E
b
/N
0
[dB]
M N
tot
N
pc
BER PERFORMANCE ON RICEAN FADING CHANNELS
49
3.7 BER PERFORMANCE ON RICEAN
FADING CHANNELS
On a fading channel we will obtain
(3.12)
after the DFT in the receiver where
assuming that the fading is sufciently slow and all ISI components are
removed by the cyclic prex. The notation is used for the DFT of the
channel impulse response affecting sub-carrier at time index . On a at
Ricean channel all sub-channels will have the same attenuation and we can
write the received vector as where is a
Ricean distributed scalar value described by the ratio of the power in the
constant direct path ( ) and the Rayleigh fading indirect path ( ). This
ratio is denoted and in Fig. 3.9 we examine the bit error rate performance
OFDM system using 16-PSK ( = 16). = 32 and = 6, 10, 14, 18, 22, 26
and 30.
1 2 3 4 5 6 7 8 9 10 11
10
3
10
2
10
1
N
pc
= 30
26
22
18
14
10
6
OFDM
B
E
R
E
b
/N
0
[dB]
M N
tot
N
pc
R
k
H
k
C
k
W
k
+ =
H
k
diag H
0 k ,
H
1 k ,
H
N
tot
1 k ,
, , , ( )
=
H
n k ,
n k
R
k
H
k
C
k
W
k
+ = H
k
D I
k
+ =
D I
k
BER PERFORMANCE ON RICEAN FADING CHANNELS

50
by simulating the system with . Note that when
the power of the indirect path is zero and the channel is Gaussian, while the
power of the direct path is zero when and the channel attenuation
is Rayleigh distributed. In Fig. 3.9 we compare the bit error probability of
an ordinary OFDM systems with 32 carriers and a PC-OFDM system using
only 25 of these in each PC-OFDM symbol. The used carriers are modu-
lated using 8-PSK and referring to Section 3.3 we know that the two sys-
tems have the same bandwidth efciency. We see in Fig. 3.9 that the PC-
OFDM is very sensitive to fading and that it performs poorer than an ordi-
nary OFDM system for but when the direct path is 10 times
stronger ( ) the PC-OFDM system performs slightly better. The
conclusion is that the PC-OFDM system is not as robust against fading as
the ordinary OFDM system, which is not a surprise since the bit error rate
for PC-OFDM is high at low in AWGN.
Figure 3.9 Bit error rate performance versus in dB for a at Ricean
channel with Marked with circles are the simulated results
for the PC-OFDM system and with no mark are the simulated results for the
ordinary OFDM system. = 32, = 25 and 8-PSK is used.
0, 5 and 10 dB = =
=
0 and 5 =
10 dB =
E
b
N
0
1 3 5 7 9 11 13 15 17
10
4
10
3
10
2
10
1
0 dB
5 dB
10 dB
PCOFDM
E
b
/N
0
[dB]
B
E
R
E
b
N
0
0, 5 and 10 dB. =
N
tot
N
pc
CHAPTER SUMMARY
51
3.8 CHAPTER SUMMARY
A parallel combinatory OFDM system is proposed. The PC-OFDM system
is capable of obtaining higher bandwidth efciency and lower bit error prob-
ability on AWGN channels than an ordinary OFDM system. The peak to
average power ratio is lower for the PC-OFDM system since fewer sub-car-
riers are used simultaneously. A bit mapping procedure using the Johnson
association scheme together with an algorithm for positioning the PSK sym-
bols is proposed. A good analytical approximation of the bit error probabil-
ity on AWGN channels is derived and simulations are performed. The PC-
OFDM system is however not as robust against fading as an ordinary
OFDM system as shown in Section 3.7.
PROBABILITY DENSITY AND DISTRIBUTION FUNCTIONS
52
APPENDIX 3A
PROBABILITY DENSITY AND DISTRIBUTION FUNCTIONS
Let and be zero mean statistically independent Gaussian random
variables, each having a variance , and let . The random
variable is a Rayleigh distributed random variable and the probability
density function of is given by [55], [52]
(3A-1)
and its distribution function is given by
. (3A-2)
Let and be statistically independent Gaussian random variables
with means and , and let . The random variable is
a Ricean distributed random variable and the probability density function of
is given by
(3A-3)
where and denotes the zeroth order Bessel function of the
rst kind. The distribution function of is given by
. (3A-4)
where
(3A-5)
is the Marcums -function [53].
X
1
X
2
2
X X
1
2
X
2
2
+ =
X
X
f
X
x ( )
x
2
-----e
x 2
2

= , x 0
F
X
x ( ) 1 e
x
2
2
2

= , x 0
Y
1
Y
2
m
1
m
2
Y Y
1
2
Y
2
2
+ = Y
Y
f
Y
y ( )
y
2
------e
y
2
s
2
+ ( ) 2
2

I
0
ys
2
------
,

_
= , y 0
s m
1
2
m
2
2
+ = I
0
Y
F
Y
y ( ) 1 Q
1
s
---
y
--- ,
,

_
= , y 0
Q
1
a b , ( ) e
a
2
b
2
+ ( ) 2 a
b
--
,

_
k
I
k
ab ( )
k 0 =
= , b a 0 > >
Q
53
CHAPTER
4.1 INTRODUCTION
A future mobile communication system must provide services with both high
and low bit rates with various quality requirements. In many applications e.g.
speech and video we might want the system to provide time varying data rates
as well. To allow many users to share a common channel we may divide the
channel into sub-channels in the time, frequency or code domain. One such
sub-channel may be a time slot in a TDMA based system, a certain frequency
range in an FDMA based system or a specic code in a CDMA system. It is
also possible to design a hybrid multiple-access system consisting of compo-
nents from different multiple-access techniques. This multichannel approach
provides a exible and efcient way of solving the multiple rate problem. The
users with the lower data rates are assigned one sub-channel, and users with
higher data rates are assigned several sub-channels. In such a system there is a
need for a exible channel coding scheme to allow different error protection
for different services, and to match the source data rate to a multiple of the
basic channel data rate which is provided by the multiple-access system [51].
RCPC-codes have attained some interest in recent publications [31], [32],
[29] [30], [33], [44], [23]. We show in this paper that RCPC-codes can provide
the large number of different channel coding rates needed for future mobile
communication services. An important feature of RCPC-codes is that the same
RATE MATCHING IN
MULTICHANNEL SYSTEMS
USING RCPC-CODES
4
RCPC-CODES FOR RATE MATCHING
54
decoder can be used for all different code rates, which reduces the receiver
complexity. Whenever multiple data rates are provided by assigning multi-
ple slots, carriers or codes to one user, we will have large steps in the possi-
ble channel data rates provided by the system. RCPC-codes can ll this gap
and provide a exible and efcient method for source data rate matching.
The outline of this Chapter is as follows. In Section 4.2 RCPC-codes and
their use for rate matching purposes are briey described. In Section 4.3 a
multi-code DS-CDMA based system used for evaluating the proposed rate
matching scheme is described together with the channel models used. Sec-
tion 4.4 contains numerical results for a downlink and an uplink scenario. In
Section 4.5 we compare the RCPC rate matching scheme with a repetition
encoding scheme, and the Chapter is summarized in Section 4.6.
4.2 RCPC-CODES FOR RATE MATCHING
RCPC-codes consist of a convolutional mother code with rate and
a puncturing matrix , with rows and columns. The puncturing matrix
consists of zeros and ones, where a zero denotes that the code symbol is not
transmitted. By designing a large set of puncturing matrices we can easily
change the resulting code rate. Generally, from a mother code of rate ,
we obtain a family of codes with the rates
. (4.1)
By introducing a rate compatibility restriction on the puncturing matrices
we can change the channel coding rate during transmission and thus easily
obtain a system with unequal error protection properties [31]. In Fig. 4.1 we
show an example of the rate matching in a communication system based on
multiple channels (time slots, frequency bands or codes) for variable data
rate. Each channel is assumed to transmit 20 kbit/s. We assume a mother
code of rate = 1/4, and that the puncturing matrix contains = 8 col-
umns. By varying the channel coding rate and the number of channels used
we obtain a large number of source data rates supported by the system. All
r 1 n =
P n p
1 n
r
p
np
------,
p
np 1
---------------, ,
p
p 1 +
------------ =
r P p
RCPC-CODES FOR RATE MATCHING
55
possible combinations of coding rates and the number of channels used are
marked with circles in Fig. 4.1. Note that the source data rate 80 kbit/s is
supported by using 5 to 16 channels with channel coder rates 1/4, 4/15, 2/7,
4/13, 1/3, 4/11, 4/10, 4/9, 1/2, 4/7, 2/3, 4/5 and 8/9 respectively, and thus
allowing different error protection for different services.
One advantage of the RCPC codes is that the same Viterbi decoder can
be used for all different puncturing matrixes thus giving a receiver with low
complexity. An alternative way to perform rate matching is to have a higher
rate convolutional encoder concatenated with a repetition encoder that sim-
ply repeats some of the bits before transmission. The repeated bits are com-
bined before Viterbi decoding in the receiver. For a given constraint length
of the convolutional mother code these two approaches have equal computa-
tional complexity.
Figure 4.1 Rate matching with RCPC-codes. All possible combinations of
encoder rates and the number of channels used are marked with circles.
0.3 0.4 0.5 0.6 0.7 0.8 0.9
0
10
20
30
40
50
60
70
80
S
o
u
r
c
e

D
a
t
a

R
a
t
e

[
k
b
i
t

/

s
]
Channel Encoder Rate
1 Channel
2 Channels
3 Channels
4 Channels
5 Channels
SYSTEM AND CHANNEL DESCRIPTION
56
4.3 SYSTEM AND CHANNEL DESCRIPTION
For purposes of evaluating the rate matching properties of the RCPC-codes
a multi-code DS-CDMA system is chosen. The spreading is done using
orthogonal Gold codes [69] of length 128. In this chapter we use QPSK
modulation where the same spreading sequence were used for I- and Q-
spreading. We have simulated the bit error rate for both a situation where all
users are synchronized as well as the case when the users are asynchronous.
This corresponds to a downlink and an uplink scenario, respectively. We
have used both a single path correlated Rayleigh fading channel as well as a
ve path correlated Rayleigh fading channel with equally strong and inde-
pendent paths. The fading processes are generated according to Jakes model
[38] with a normalized Doppler frequency (the Doppler frequency
multiplied with the symbol duration) of 0.1. An interleaver of 50 50 sym-
bols were used and all simulations were performed until at least 2000 errors
occurred. We assume perfect channel estimation and synchronization and
we use single user detectors in the receiver. The RCPC-code used has a con-
straint length of 5 and the generating polynomials were 46, 72, 56 and 66 in
octal notation. The puncturing matrices are chosen according to Table I of
[29]. In the sequel denotes the transmitted energy per information bit.
4.4 NUMERICAL RESULTS
First we show results for a single path Rayleigh fading channel in Fig. 4.2 to
Fig. 4.5 and then follow results for a multipath fading channel with ve
independent and equally strong Rayleigh fading paths in Fig. 4.6. A com-
parison between RCPC-codes and repetition codes is shown in Fig. 4.7.
Downlink
We start with the case where all spreading sequences are transmitted syn-
chronously. In Fig. 4.2 we see the simulated bit error rate results for differ-
ent puncturing of the rate 1/4 convolutional mother code as a function of
f
d
T
s
E
b
NUMERICAL RESULTS
57
. All curves use 64 of the 128 available orthogonal Gold codes. We
thus have to accept lower data rate in order to lower the channel encoder
rate. We see that there is little difference in bit error rate performance when
going from = 1/4 to = 1/3. We also see that we obtain reasonable quality
for coding rates < 2/3. Due to the interference between the spreading
codes caused by the relatively fast fading we obtain an error oor at a bit
error rate of approximately 10
-4
.
When choosing to use a low rate channel code we must allocate more
spreading codes to the users in order to keep the effective data rate constant.
If we assume that the users want to communicate with 80 kbit/s and that one
spreading code can carry a bit rate of 20 kbit/s we can achieve this by
assigning 4 spreading codes to all users and use no channel coding ( = 1).
Other alternatives are of course to use equal to 1/2 or 1/4 and assign 8 or
Figure 4.2 Simulated bit error rate for different channel encoder rates, , as a
function of in dB with 64 orthogonal Gold codes of length 128. The
normalized Doppler frequency is 0.1. Interleaver size is 50 50 symbols.
Downlink scenario.
0 5 10 15 20
10
4
10
3
10
2
10
1
10
0
q
q
q
q
q
q
q
q
q
q
q
v
v
v
v
v
v
v
v
v
v
v
w
w
w
w
w
w
w
w
w
w
w
x
x
x
x
x
x
x
x
x
x
x
s
s
s
s
s
s
s
s
s
s
s
v
w
x
s
r = 8/9
r = 4/5
r = 2/3
r = 4/7
r = 1/2
r = 4/10
r = 1/3
r = 1/4
E
b
/N
0
[dB]
B
E
R
r
E
b
N
0
E
b
N
0
r r
r
r
r
NUMERICAL RESULTS
58
16 spreading codes to the users, respectively. In Fig. 4.3 we show simulated
bit error rate results for a system with 8 active users transmitting at 80 kbit/s
on 20 kbit/s sub-channels where the number of spreading codes used are
determined by the channel encoder rate. We see that although the interfer-
ence level gets higher we gain by using low rate channel codes.
Uplink
We now turn our attention to the case where the users are not synchronized
to each other. Spreading codes transmitted by a single user are however still
synchronized. The delays between users are assumed to be uniformly dis-
tributed over the symbol interval. We assume 8 active users and the total
number of spreading codes used by the system is kept constant to 64. All
users experience an independently Rayleigh fading channel. In Fig. 4.4 we
Figure 4.3 Simulated bit error rates for different as a function of the
channel encoder rate, , with orthogonal Gold codes of length 128. The source
data rate is 80 kbit/s. The normalized Doppler frequency is 0.1. Interleaver size is
50 50 symbols. Downlink scenario.
0.2 0.3 0.4 0.5 0.6 0.7 0.8
10
4
10
3
10
2
10
1
10
0
E
b
/N
0
= 5 dB
10 dB
15 dB
20 dB
B
E
R
Channel encoder rate, r
E
b
N
0
r
NUMERICAL RESULTS
59
see the simulated bit error rate when using 64 of the available 128 spreading
sequences. The bit error rate is degraded compared to the situation where all
users are synchronous due to the increased interference between the users.
We see that the coding is efcient in counteracting the degrading effects of
multiple access interference. The bit error rate performance for the channel
encoders with the highest rate in Fig. 4.4 is far to poor for most applications.
Still the loss in bit error rate performance when going from channel encoder
rate = 1/4 to = 1/3 is reasonably low.
In Fig. 4.5 the 8 asynchronous users all transmit at a data rate of 80 kbit/
s in the same fashion as described previously for Fig. 4.3. Once again we see
that although the puncturing reduces the multiple access interference we
still gain in performance when choosing a lower rate puncturing scheme.
Figure 4.4 Simulated bit error rate as a function of in dB for channel
encoder rates ranging from = 8/9 (top) to 1/4 (bottom) with 64 orthogonal Gold
codes of length 128. The normalized Doppler frequency is 0.1. Interleaver size is
50 50 symbols. Uplink scenario.
0 5 10 15 20 25 30
10
3
10
2
10
1
10
0
4/13
r = 8/9
E
b
/N
0
[dB]
B
E
R
4/5
2/3
4/7
1/2
4/9
4/10
4/11
1/3
2/7
4/15
1/4
E
b
N
0
r
r r
NUMERICAL RESULTS
60
Multipath Channel
In Fig. 4.6 we use a multipath fading channel model with ve equally strong
Rayleigh fading paths. All users are assumed to be synchronous and share
the same channel (downlink scenario). Out of the 128 available spreading
sequences 64 are used for all curves in Fig. 4.6. The tap delays are uni-
formly distributed over one symbol duration. We use a ve nger RAKE in
the receiver and perform maximum ratio combining of the RAKE nger
output values. We assume perfect channel knowledge in terms of path
delays and fading gains. We see that the RAKE receiver obtains a substan-
tial diversity gain giving lower bit error rates for low .
Figure 4.5 Simulated bit error rates for different as a function of the
channel encoder rate, . with orthogonal Gold codes of length 128. The source
data rate is 80 kbit/s. The normalized Doppler frequency is 0.1. Interleaver size is
50 50 symbols. Uplink scenario.
0.2 0.3 0.4 0.5 0.6 0.7 0.8
10
2
10
1
10
0
E
b
/N
0
= 10 dB
15 dB
20 dB
25 dB
30 dB
Channel encoder rate, r
B
E
R
E
b
N
0
r
E
b
N
0
COMPARISON TO REPETITION ENCODING

61
4.5 COMPARISON TO REPETITION ENCODING
An alternative to RCPC-codes is to use a xed convolutional code and con-
catenate it with a simple repetition code where some bits are transmitted
twice for rate matching purposes (c.f. IS-95 [70]). For comparison we show
in Fig. 4.7 the simulated bit error rate results of RCPC-codes, based on a
rate 1/4 convolutional mother code, punctured to the rates 4/9, 1/3 and 4/15
(solid lines). Also shown in Fig. 4.7 are the results of a rate 1/2 convolu-
tional code that is repetition encoded to the same rates (dashed lines). The
channel used is a single path correlated Rayleigh fading channel as
described earlier. Both mother codes have a constraint length of 5 and the
generator polynomials used for the repetition mother code are 46 and 72 in
Figure 4.6 Simulated bit error rate for different channel encoder rates, , as a
function of in dB on a ve path channel with 64 orthogonal Gold codes of
length 128. The normalized Doppler frequency is 0.1. Interleaver size is 50 50
symbols. Downlink scenario.
E
b
/N
0
[dB]
B
E
R
0 2 4 6 8 10 12
10
4
10
3
10
2
10
1
10
0
q
q
q
q
q
q
q
v
v
v
v
v
v
v
w
w
w
w
w
w
w
x x
x
x
x
x
x
s s s
s
s
s
s
v
w
x
s
r = 8/9
r = 4/5
r = 2/3
r = 4/7
r = 1/2
r = 4/10
r = 1/3
r = 1/4
r
E
b
N
0
COMPARISON TO REPETITION ENCODING

62
octal notation [46]. The curves in Fig. 4.7 marked with circles correspond to
the case where the used rate is , that is the used rate is closer to the rate
of the repetition mother code than the rate of the RCPC mother code
. The number of repeated bits is less than the number of punctured
bits for this case. Marked with cross signs are the curves with rate
where we are closest in rate to the RCPC mother code and marked with
lled diamonds are the curves corresponding to the rate , where
the same number of bits are punctured or repeated. We see that the RCPC-
codes outperform the repetition codes for all rates 1/2 > > 1/4. Also we
note that when using repetition codes we can not provide coding rates
higher than the rate of the repetition mother code and that the rate of the rep-
etition mother code must be low enough to provide reasonably good perfor-
mance. Given the facts that the computational complexities of both schemes
are equal, that the repetition scheme is less exible in providing a large span
Figure 4.7 Simulated bit error rate for RCPC-codes (solid) and repetition codes
(dashed) for channel encoder rates = 4/9, 1/3 and 4/15 with 64 orthogonal Gold
codes of length 128. The normalized Doppler frequency is 0.1. Interleaver size is
50 50 symbols. Downlink scenario.
4 9
1 2 ( )
1 4 ( )
4 15
r 1 3 =
r
10 12 14 16 18 20
10
4
10
3
10
2
x
x
x
x
x
x
x
x
x
x
x
x
E
b
/N
0
[dB]
B
E
R
x
r = 1/3
r = 4/15
r = 4/9
r
CHAPTER SUMMARY
63
of coding rates compared to RCPC-codes and that the RCPC-codes show
better performance we conclude that puncturing is superior to repetition for
the purposes of rate matching.
4.6 CHAPTER SUMMARY
The use of RCPC-codes for rate matching in multi-channel systems is pro-
posed and analyzed. RCPC-codes are shown capable of providing a exible
and efcient coding scheme for multi-rate multiple access systems. Simula-
tions are performed using a multi-code DC-CDMA system. RCPC-codes
provide lower bit error probability and a larger span of available coding
rates than a scheme bases on repetition encoding.
65
CHAPTER
5.1 CONCLUSIONS
In Chapter 2 decision directed coherent detectors for single-carrier and multi-
carrier systems based on a minimum mean square error channel estimation and
decision feedback are presented. The MMSE based detector does not suffer
from the problem of an irreducible bit error probability oor. The single-car-
rier system can use the correlation in time between the received values and the
multicarrier system can lower the bit error rate performance further by utiliz-
ing the correlation between the different sub-carriers. The performances of the
proposed detectors are analyzed under the assumption that all previous deci-
sions are correct and the differential decoding is not taken into account. Simu-
lations are performed to study the effects of error propagation and differential
decoding in the detector when decisions are used in channel estimation. We
assume perfect knowledge of the channel correlation function and the additive
white Gaussian noise power. For the single-carrier system this means that the
normalized Doppler frequency is perfectly known and for the multicarrier sys-
tem we need the additional knowledge of the number of independent taps on
the channel. The sensitivity to parameter errors is small, at least for small Dop-
pler and small predictor orders. The ML detector for the single-carrier system
on this channel is derived independently in [15]. Due to some differences in
the implemented channel models, the results are not directly comparable. It
CONCLUSIONS
AND FUTURE WORK
5
CONCLUSIONS
66
seems, however, that our detector is equally good as the ML detector for
small SNR values and moderate Doppler frequencies, while the ML detector
shows a signicant improvement for large SNR values and Doppler frequen-
cies. It should be noted that the ML detector is more complex, since it sam-
ples faster than the symbol rate. A decision directed coherent detector for
16-QAM based on the same principles as those described in Chapter 2 is
found in [49]. Using 16-QAM the channel estimator coefcients will be data
dependent.
In Chapter 3 we show that the PC-OFDM system can be designed to
have higher bandwidth efciency and at the same time lower peak to mean
power ratio than ordinary OFDM. On a Gaussian channel the bit error prob-
ability of PC-OFDM is in some cases less than the BER for ordinary OFDM
for high values. Since OFDM is more robust against impulse noise
than a single-carrier system due to the long symbol interval this would moti-
vate the use of PC-OFDM on channels with AWGN and impulse noise. As
an example of such a channel we can mention communications on high volt-
age power lines that suffer from Gaussian noise due to partial discharges to
the surrounding air and impulse disturbances due to switching on and off
electrical machinery. On fading channels the use of PC-OFDM can only be
motivated by its higher bandwidth efciency and reduced peak to mean
power ratio. The bit error rate will be higher for the PC-OFDM system com-
pared to an ordinary OFDM system.
We show in Chapter 4 that RCPC-codes provide a exible coding
scheme for multi-rate multiple access systems. The same Viterbi decoder is
used for all rates which allow a simple implementation. RCPC-codes can
provide a larger span of channel coding rates compared to repetition encod-
ing. The bit error rate performance of RCPC-codes is lower than that of rep-
etition encoding schemes of the same complexity. Further more, we show
that the additional interference, caused by lowering the code rate and using
more spreading codes to maintain the effective data rate, is compensated for
by the larger coding gain provided.
E
b
N
0
FUTURE WORK
67
5.2 FUTURE WORK
In Chapter 2 we assume that the correlation function of the process is known
as well as the channel noise variance. Further investigations are needed to
show exactly how estimation errors in these parameters will effect the per-
formance of the MMSE based detector. Differential encoding in the fre-
quency domain instead of in the time domain is also a topic for future
research. Another interesting problem is how to combine the good proper-
ties of decision directed schemes and pilot based schemes for channel esti-
mation in the best way. By using very few pilots it is possible to avoid the
differential encoding of the data since the pilots will then resolve the phase
ambiguities. Also the decision directed approach described in Chapter 2 has
poorer bit error probabilities on the sub-carriers at the frequency edges
where the number of frequency neighbors is smaller. Inserting pilots in
these regions might help to reduce the error propagation and thus lower the
error variance of the channel estimator.
Further studies are needed to nd a suitable coding scheme for the PC-
OFDM system described in Chapter 3. One solution using an RS-code to
decode all bits in one PC-OFDM symbol is proposed in [62]. If however the
PSK-modulated bits are much fewer than the parallel combinatory bits the
RS-code will not be able to correct errors due to a wrong detection of sub-
carriers in the receiver. Longer RS-codes with code words spanning over
several PC-OFDM symbols may be preferred instead. An alternative solu-
tion to PC-OFDM might be to design a block code with elements in the eld
GF(M) and codewords of length having as low weight as possible. The
code elements can be mapped on the carriers using ( -1)-PSK modulation
for the nonzero elements and transmitting a zero carrier for the zero ele-
ments of the codeword. This is a topic for further investigations. It is inter-
esting to note that the PC-OFDM system actually performs better than a
system using a water-pouring bit and power allocation scheme on an
AWGN channel where all sub-carriers are equal. Simulation results not pre-
sented in this thesis show that this advantage of the PC-OFDM system is
maintained on channels with non-constant transfer function [27]. The bit
N
tot
M
FUTURE WORK
68
and power allocation from the water-pouring scheme is kept unchanged
and parallel combinatory signaling is used independently on all carriers hav-
ing the same number of bits. This solution is somewhat lumbering but actu-
ally achieves better result than the already excellent results of the water-
pouring scheme. Further studies in this direction are also topics for future
research.
Future work related to the work presented in Chapter 4 is to nd RCPC-
codes for large constraint length and low rates and to nd puncturing
schemes optimized for specic channels. Codes with larger constraint
lengths are needed to enhance the BER performance and codes with lower
rates are needed to provide a larger span of available data rates and to ensure
low enough BER performances for services sensitive to errors.
69
[1] F. Adachi, Postdetection optimal diversity combiner for DPSK differential
detection, IEEE Transactions on Vehicular Technology, Vol. 42, No. 3, pp.
326-337, August 1993.
[2] F. Adachi and K. Ohno, BER performance of DQPSK with postdetection
diversity reception in mobile radio channels, IEEE Transactions on Vehicu-
lar Technology, Vol. 40, No. 1, pp. 237-249, February 1991.
[3] D. M. Balston and R. C. V. Macario, Ed., Cellular Radio Systems. Artech
House: Norwood, 1993.
[4] M.-G. Di Benedetto and P. Mandarini, An application of MMSE predistor-
tion to OFDM signals, IEEE Transactions on Communications, Vol. 44, No.
11, pp. 1417-1420, November 1996.
[5] J. A. C. Bingham, Multicarrier modulation for data transmission: An idea
whose time has come, IEEE Communications Magazine, Vol. 28, No. 5, pp.
5-14, May 1990.
[6] A. Brajal and A. Chouly, Compensation of nonlinear distortions for orthog-
onal multicarrier schemes using predistortion, in Proceedings IEEE Global
Telecommunications Conference, San Francisco, USA, November 28-
December 2, 1994, pp. 1909-1914.
[7] M. Burgos-Garca and F. Prez-Martnez, Simple procedure for optimum
linearization of ampliers in multicarrier applications, Electronics Letters,
Vol. 30, No. 2, pp. 114-115, January 1994.
BIBLIOGRAPHY
70
[8] J. K. Cavers, An analysis of pilot symbol assisted modulation for Rayleigh
fading channels, IEEE Transactions on Vehicular Technology, Vol. 40, No.
4, pp. 686-693, November 1991.
[9] R. W. Chang, Synthesis of band-limited orthogonal signals for multicarrier
data transmission, The Bell Systems Technical Journal, pp. 1775-1796,
December 1966.
[10] J. S. Chow, J. C. Tu, and J. M. Ciof, A discrete multitone transceiver sys-
tem for HDSL applications, IEEE Journal on Selected Aereas in Communi-
cations, Vol. 9, No. 6, pp. 895-908, August 1991.
[11] P. S. Chow, J. M. Ciof, and J. A. C. Bingham, A practical discrete multi-
tone transceiver loading algorithm for data transmission over spectrally
shaped channels, IEEE Transactions on Communications, Vol. 43, No. 2/3/
4, pp. 773-775, February/March/April 1995.
[12] P. S. Chow, N. Al-Dhahir, J. M. Ciof, and J. A. C. Bingham, A multicarrier
E1-HDSL transceiver with coded modulation, European Transactions on
Telecommunications and Related Technologies, Vol. 4, No. 3, pp. 257-266,
May-June 1993.
[13] T. de Couason, R. Monnier, and J. B. Rault, OFDM for digital TV broad-
casting, Elsevier Science Publishers, Signal Processing, Vol. 39, No. 1-2,
pp. 1-32, September 1994.
[14] F. Daffara and O. Adami, A new frequency detector for orthogonal multicar-
rier transmission techniques, in Proceedings IEEE Vehicular Technology
Conference, Chicago, USA, July 25-28, 1995, pp. 804-809.
[15] W. C. Dam and D. P. Taylor, An adaptive maximum likelihood receiver for
correlated Rayleigh-fading channels, IEEE Transactions on Communica-
tions, Vol. 42, No. 9, pp. 2684-2692, September 1994.
[16] N. Al-Dhahir and J. Ciof, Optimum nite-length equalization for multicar-
rier transceivers, in Proceedings IEEE Global Telecommunications Confer-
ence, San Francisco, USA, November 28-December 2, 1994, pp. 1884-1888.
[17] O. Edfors, Low-complexity algorithms in digital receivers, Ph.d. Disserta-
tion, Lule University of Technology, September 1996.
[18] O. Edfors, M. Sandell, J.-J. van de Beek, S. K. Wilson, and P. O. Brjesson,
OFDM channel estimation by singular value decomposition, in Proceed-
ings IEEE Vehicular Technology Conference, Atlanta, USA, April 28-May 1,
1996, pp. 923-927.
71
[19] P. Van Eetvelt, G. Wade, and M. Tomlinson, Peak to average power reduc-
tion for OFDM schemes by selective scrambling, Electronics Letters, Vol.
32, No. 21, pp. 1963-1964, October 1996.
[20] B. Engstrm and C. stberg, A system for test of multiaccess methods
based on OFDM, in Proceedings IEEE Vehicular Technology Conference,
Stockholm, Sweden, June 8-10, 1994, pp. 1843-1845.
[21] ETSI ETS 300 401, Radio broadcasting systems: Digital audio broadcasting
(DAB) to mobile, portable and xed receivers, Sophia Antipolis, France,
1995.
[22] K. Fazel, S. Kaiser, P. Robertson, and M. J. Ruf, Concept of digital terres-
trial television broadcasting, Wireless Personal Communications, Vol. 2,
No. 1-2, pp. 9-27, 1995
[23] Y. Feria and K.-M. Cheung, Seamless data-rate change using punctured
convolutional codes for time-varying signal-to-noise ratio, in Proceedings
IEEE International Conference on Communications, Seattle, USA, June 18-
22, 1995, pp. 342-346.
[24] P. Frenger and A. Svensson, A Decision feedback coherent detector for
OFDM, in Proceedings IEEE Vehicular Technology Conference, Atlanta,
Georgia USA, April 28-May 1, 1996, pp. 1584-1588.
[25] P. Frenger and A. Svensson, A minimum mean square error channel estima-
tor for OFDM, in Proceedings RVK seminar, Lule and Kiruna, Sweden,
June 3-6, 1996, pp. 121-125.
[26] P. Frenger and A. Svensson, Decision directed coherent detection in multi-
carrier systems on Rayleigh fading channels, Submitted to IEEE Transac-
tions on Vehicular Technology, 1996.
[27] P. Frenger and A. Svensson, A parallel combinatory OFDM system, in Pro-
ceedings IEEE Personal, Indoor and Mobile Radio Communications, Taipei,
Taiwan, October 15-18, 1996, pp. 1069-1073.
[28] P. Frenger, P. Orten, T. Ottosson, and A. Svensson, Rate matching in mul-
tichannel systems using RCPC-codes, To appear in Proceedings IEEE
Vehicular Technology Conference, Phoenix, USA, May 5-7, 1997.
[29] J. Hagenauer, Rate compatible punctured convolutional codes, in Proceed-
ings IEEE International Conference on Communications, Seattle, USA, June
7-10, 1987, pp. 1032-1036.
72
[30] J. Hagenauer, Rate compatible punctured convolutional codes (RCPC-
codes) and their applications, IEEE Transactions on Communications, Vol.
36, No. 4, pp. 389-400, April 1988.
[31] J. Hagenauer, N. Seshadri, and C.-E. Sundberg, The performance of rate-
compatible punctured convolutional codes for future digital mobile radio, in
Proceedings IEEE Vehicular Technology Conference, New York, USA, 1988,
pp. 22-29.
[32] J. Hagenauer, N. Seshadri, and C.-E. Sundberg, The performance of rate-
compatible punctured convolutional codes for digital mobile radio, IEEE
Transactions on Communications, Vol. 30, No. 7, pp. 966-980, July 1990.
[33] P. Hoeher, J. Hagenauer, E. Offer, Ch. Papp, and H. Schulze, Performance
of an RCPC-coded OFDM-based digital audio broadcasting (DAB) system,
in Proceedings IEEE Global Telecommunications Conference, Phoenix,
USA, December 2-5, 1991, pp. 44-46.
[34] P. Hoeher, S. Kaiser, and P. Robertson, Pilot-symbol-aided channel estima-
tion in two dimensions, To appear in Proceedings IEEE International Con-
ference on Communications, Montral, Canada, June 8-12, 1997.
[35] M. Itami, H. Takano, H. Ohta, and K. Itoh, A method of equalization of
OFDM signal with inter-symbol and inter-channel interferences, in Pro-
ceedings International Conference on Communication Systems, Singapore,
November 14-18, 1994, pp. 109-113.
[36] M. Johansson, T. Mattsson, L. Sundstrm, and M. Faulkner, Linearization
of multi-carrier ampliers, in Proceedings IEEE Vehicular Technology Con-
ference, Secaucus, New Jersey, USA, 1993, pp. 684-687.
[37] A. E. Jones, T. A. Wilkinson, and S. K. Barton, Block coding scheme for
reduction of peak to mean envelope power ratio of multicarrier transmission
schemes, Electronics Letters, Vol. 30, No. 25, pp. 2098-2099, December
1994.
[38] W. C. Jakes, Ed., Microwave mobile communications. Reissued edition,
IEEE Press: New York, 1993.
[39] N. S. Jayant and P. Noll, Digital coding of waveforms: principles and appli-
cations to speech and video. Prentice-Hall: Englewood Cliffs, N. J., 1984.
[40] I. Kalet, The multitone channel, IEEE Transactions on Communications,
Vol. 37, No. 2, pp. 119-124, February 1989.
[41] K. W. Kang, J. Ann, and H. S. Lee, Decision-directed maximum-likelyhood
estimation of OFDM frame synchronisation offset, Electronics Letters, Vol.
30, No. 25, pp. 2153-2154, December 1994.
73
[42] S. M. Kay, Fundamentals of statistical signal processing: Estimation theory.
Prentice-Hall: Englewood Cliffs, N. J., 1993.
[43] E. A. Lee, D. G. Messerschmitt, Digital communications, 2nd edition. Klu-
wer Academic Publishers: Boston, 1994.
[44] L. H. C. Lee, New rate compatible punctured convolutional codes for
Viterbi decoding, IEEE Transactions on Communications, Vol. 42, No. 12,
pp. 3073-3079, December 1994
[45] R. Li and G. Stette, Time-limited orthogonal multicarrier modulation
schemes, IEEE Transactions on Communications, Vol. 43, No. 2/3/4, pp.
1269-1272, February/March/April 1995.
[46] S. Lin and D. J. Costello, Jr., Error control coding: Fundamentals and appli-
cations. Prentice-Hall: Englewood Cliffs, N. J., 1983.
[47] J.-P. Linnartz and S. Hara, Editorial introduction, Wireless Personal Com-
munications, Vol. 2, No. 1-2, pp. 1-7, 1995.
[48] F. J. MacWilliams and N. J. A. Sloane, The theory of error-correcting codes.
North-Holland Publishing Company: Amsterdam, 1977.
[49] M. F. Nakhjiri and A. Svensson, Decision directed coherent detection of 16-
QAM on fading channels, in Proceedings IEEE Vehicular Technology Con-
ference, Atlanta, Georgia USA, April 28-May 1, 1996, pp. 988-992.
[50] R. ONeill and L. B. Lopes, Performance of amplitude limited multitone
signals, in Proceedings IEEE Vehicular Technology Conference, Stockholm,
Sweden, 1994, pp. 1675-1679.
[51] T. Ottosson and A. Svensson, Multi-rate schemes in DS/CDMA systems,
in Proceedings IEEE Vehicular Technology Conference, Chicago, USA,
1995, July 25-28, pp. 1006-1010.
[52] A. Papoulis, Probability, random variables and stochastic processes, 3rd edi-
tion. McGraw-Hill: New York, 1991.
[53] S. Parl, A new method of calculating the generalized Q function, IEEE
Transactions on Information Theory, Vol. IT-26, No. 1, pp. 121-124, January
1980.
[54] R. Petrovic, W. Roehr, and D. W. Cameron, Multicarrier modulation for
narrowband PCS, IEEE Transactions on Vehicular Technology, Vol. 43, No.
4, pp. 856-862, November 1994.
[55] J. G. Proakis, Digital communications, 3rd edition. McGraw-Hill: New York,
1995.
74
[56] C. Reiners and H. Rohling, Multicarrier transmission technique in cellular
mobile communications system, in Proceedings IEEE Vehicular Technology
Conference, Stockholm, Sweden, June 8-10, 1994, pp.1660-1664.
[57] J. Rinne and M. Renfors, Equalization of orthogonal frequency division
multiplexing signals, in Proceedings IEEE Global Telecommunications
Conference, San Francisco, USA, November 28-December 2, 1994, pp. 415-
419.
[58] B. R. Saltzberg, Performance of an efcient parallel data transmission sys-
tem, IEEE Transactions on Communication Technology, Vol. COM-15, No.
6, pp. 805-811, December 1967.
[59] M. Sandell, Design and analysis of estimators for multicarrier modulation
and ultrasonic imaging, Ph.d. Dissertation, Lule University of Technology,
September 1996.
[60] H. Sari, G. Karam, and I. Jeanclaude, An analysis of orthogonal frequency-
division multiplexing for mobile radio applications, in Proceedings IEEE
Vehicular Technology Conference, Stockholm, Sweden, 1994, pp. 1635-
1639.
[61] H. Sari, G. Karam, and I. Jeanclaude, Transmission techniques for digital
terrestrial TV broadcasting, IEEE Communications Magazine, Vol. 33, No.
2. pp. 100-109, 1995?.
[62] S. Sasaki, J. Zhu, and G. Marubayashi, Performance of the parallel combi-
natory spread spectrum multiple access communication system with the error
control technique, in Proceedings IEEE Second International Symposium
on Spread Spectrum Techniques and Applications, Yokohama, Japan, 1992,
pp. 159-162.
[63] S. Sasaki, H. Kikuchi, H. Watanabe, J. Zhu, and G. Marubayashi, Perform-
ance of parallel combinatory spread spectrum communication systems using
multiphase modulation, in Proceedings of The 17th Symposium on Informa-
tion Theory and Its Applications, Hiroshima, Japan, pp. 421-424.
[64] S. Sasaki, H. Kikuchi, J. Zhu, and G. Marubayashi, Multiple access per-
formance of parallel combinatory spread spectrum communication systems
in nonfading and Rayleigh fading channels, IEICE Transactions on Commu-
nications, Vol. E78-B, No. 8, pp. 1152-1161, August 1995.
[65] K. Sistanizadeh, P. S. Chow, and J. M. Ciof, Multi-tone transmission for
asymetric digital subscriber lines (ADSL), in Proceedings IEEE Interna-
tional Conference on Communications, Geneva, Switzerland, May 23-26,
1993, pp. 756-759.
75
[66] S. J. Shepard, P. W. J. Van Eetvelt, C. W. Wyatt-Millington, and S. K. Barton,
Simple coding scheme to reduce peak factor in QPSK multicarrier modula-
tion, Electronics Letters, Vol. 31, No. 14, pp. 1131-1132, July 1995.
[67] A. Svensson, A coherent detector based on linear prediction and decision
feedback for DQPSK, Electronics Letters, Vol. 30, No. 20, pp. 1642-1643,
September 1994.
[68] A. Svensson, 1 and 2 Stage decision feedback coherent detectors for
DQPSK in fading channels, in Proceedings IEEE Vehicular Technology
Conference, Chicago, USA, July 25-28, 1995, pp. 644-648.
[69] S. Tachikawa, Recent spreading codes for spread spectrum communication
systems, Electronics and Communications in Japan, Part I, Vol. 75, No. 6,
pp. 41-49, June 1992.
[70] TIA/EIA/IS-95, Mobile station-base station standard for dual-mode wide-
band spread spectrum cellular system, Telecommunication Industry Associ-
ation, July 1993.
[71] F. Tufvesson and T. Maseng, Pilot assisted channel estimation for OFDM in
mobile cellular systems, To appear in Proceedings IEEE Vehicular Technol-
ogy Conference, Phoenix, USA, May 4-7, 1997.
[72] A. Vahlin and N. Holte, Use of a guard interval in OFDM on multipath
channels, Electronics Letters, Vol. 30, No. 24, pp. 2015-2016, November
1994.
[73] A. Vahlin and N. Holte, OFDM for broadcasting in presence of analogue
co-channel interference, IEEE Transactions on Broadcasting, Vol. 41, No.
3, September 1995, pp. 89-93.
[74] A. Vahlin and N. Holte, Optimal nite duration pulses for OFDM, IEEE
Transactions on Communications, Vol. 44, No. 1, January 1996, pp. 10-14.
[75] W. D. Warner and C. Leung, OFDM/FM frame synchronization for mobile
radio data communication, IEEE Transactions on Vehicular Technology,
Vol. 42, No. 3, pp. 302-313, August 1993.
[76] S. B. Weinstein and P. M. Ebert, Data transmission by frequency division
multiplexing using the discrete fourier transform, IEEE Transactions on
Communication Technology, Vol. COM-19, No. 5, pp. 628-634, October
1971.
[77] S. K. Wilson, Digital audio broadcasting in a fading and dispersive chan-
nel, Ph.d. Dissertation, Stanford University, August 1994.
76
[78] Y. Wu and B. Caron, Digital television terrestrial broadcasting, IEEE Com-
munications Magazine, Vol. 32, No. 5, pp. 46-52, May 1994.
[79] D. Wulich, Reduction of peak to mean ratio of multicarrier modulation
using cyclic coding, Electronics Letters, Vol. 32, No. 5, pp. 432-433, Febru-
ary 1996.
[80] M. D. Yacoub, Foundations of mobile radio engineering. CRC Press: Boca
Ranton, Florida, 1993.
[81] G. Zimmermann, M. Rosenberger, and S. Dostert, Theoretical bit error rate
for uncoded and coded data transmission in digital audio broadcasting, in
Proceedings IEEE International Conference on Communications, Dallas,
USA, June 23-27, 1996, pp. 297-301.

Multicarrier Modulation and Coding For Multichannel Systems: ÅL Renger

Transféré par

Informations du document

Description originale:

Titre original

Copyright

Formats disponibles

Partager ce document

Partager ou intégrer le document

Options de partage

Avez-vous trouvé ce document utile ?

Ce contenu est-il inapproprié ?

Droits d'auteur :

Formats disponibles

Multicarrier Modulation and Coding For Multichannel Systems: ÅL Renger

Transféré par

Droits d'auteur :

Formats disponibles

Multicarrier Modulation and

Coding for Multichannel Systems

Other diversity branches

BER PERFORMANCE ON RICEAN FADING CHANNELS

COMPARISON TO REPETITION ENCODING

COMPARISON TO REPETITION ENCODING

Vous aimerez peut-être aussi