Professional Documents
Culture Documents
COGNITIVE RADIOS
by
Peiman Amini
Doctor of Philosophy
December 2009
Copyright
c Peiman Amini 2009
of a dissertation submitted by
Peiman Amini
This dissertation has been read by each member of the following supervisory committee
and by majority vote has been found to be satisfactory.
Cynthia Furse
Rong-Rong Chen
Keneth Stevens
Frederic Noo
THE UNIVERSITY OF UTAH GRADUATE SCHOOL
Marc Bodson
Chair/Dean
David S. Chapman
Dean of The Graduate School
ABSTRACT
The demand for wireless services is on the rise and the vast majority of the
spectral resources have already been licensed. Consequently, cognitive radio tech-
nology has been proposed to make secondary use of licensed spectrum. Multi-
carrier communication technology has been suggested to utilize the white spaces
in the spectrum. Orthogonal frequency-division multiplexing (OFDM) was the
first multicarrier technique proposed for cognitive radios. However, OFDM suffers
from significant leakage among the carriers of different users. On the other hand,
filterbank multicarrier (FBMC) communication can overcome the spectral leakage.
Therefore, FBMC has been suggested as an alternative to OFDM for cognitive
radios. In this dissertation, we investigate the implementation issues that need to
be addressed for an actual deployment of FBMC.
Efficient polyphase structures for implementation of FBMC are investigated.
A novel formulation for a family of polyphase structures for staggered modulated
multitone (SMT) is derived. Using our derivation, it is shown that some of the
SMT analysis structures in the literature are not applicable to frequency selective
channels.
A preamble design and related algorithms are proposed for FBMC systems. The
proposed preamble is used to detect the beginning of packet, to adjust an automatic
gain control, to synchronize the carrier frequency and timing phase, and to identify
the channel impulse response. Furthermore, decision directed carrier and timing
tracking algorithms are proposed to track residual timing and carrier offset after
the acquisition. In addition, a decision directed phase lock loop (PLL) is designed
to force any built up phase error to zero. Also, an algorithm is implemented to
track the best timing phase by minimizing a cost function.
This dissertation also reports implementation of a cognitive radio equipped with
filterbank spectrum sensing. The cognitive radio was implemented on a software
defined radio platform. The filterbank spectrum sensor is shown to exhibit superior
performance in terms of the spectral dynamic range when compared to the FFT
based techniques. The radio can detect the presence of interferers on the carrier
that is currently using and move to an unoccupied part of the spectrum.
FBMC is also applied to fault detection on live wires. Optimally designed
synthesis filterbanks are used to confine the test signal to the portion(s) of the
frequency band that are free of live signal. Moreover, optimal analysis filters are
designed which can separate the reflected test tones and minimize leakage from the
live wire signals.
v
I dedicated this work to my family: Masoud, Mahin, Payvand, and Pooyan.
CONTENTS
ABSTRACT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iv
LIST OF FIGURES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . x
ACKNOWLEDGMENTS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiii
CHAPTERS
1. INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Multicarrier Communications
for Cognitive Radios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 OFDM/FFT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.3 Filterbank Spectrum Sensing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.4 Filterbank Communications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.4.1 Staggered Modulated Multitone, SMT . . . . . . . . . . . . . . . . . . . 12
1.4.2 Cosine Modulated Multitone, CMT . . . . . . . . . . . . . . . . . . . . . . 12
1.4.3 Filtered Multitone, FMT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.4.4 A Comparison of FMT, SMT and CMT . . . . . . . . . . . . . . . . . . 13
1.5 Contribution of the Dissertation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.6 Organization of the Dissertation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2. A REVIEW ON FILTERBANK MULTICARRIER
COMMUNICATION SYSTEMS . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.1 Filterbank Multicarrier Communications . . . . . . . . . . . . . . . . . . . . . . 20
2.1.1 Staggered Modulated Multitone, SMT . . . . . . . . . . . . . . . . . . . 20
2.1.2 Cosine Modulated Multitone, CMT . . . . . . . . . . . . . . . . . . . . . . 25
2.1.3 Filtered Multitone, FMT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.2 FBMC Sensitivity to Synchronization Errors . . . . . . . . . . . . . . . . . . . 29
2.2.1 SMT Sensitivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.2.2 CMT Sensitivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
2.3 Sensitivity to Carrier Offset . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.4 Sensitivity to Timing Offset . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
2.4.1 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3. POLYPHASE IMPLEMENTATION OF FILTERBANK MULTICARRIER
COMMUNICATION SYSTEMS . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.1 Polyphase Synthesis Filterbank . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.2 Polyphase Analysis Filterbank . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
3.3 SMT Polyphase Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.3.1 Type-I SMT Polyphase Structure . . . . . . . . . . . . . . . . . . . . . . . 51
3.3.2 Type-II SMT Polyphase Structure . . . . . . . . . . . . . . . . . . . . . . 54
3.3.3 Type-III SMT Polyphase Structure . . . . . . . . . . . . . . . . . . . . . . 57
3.3.4 Equalization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
3.4 CMT Polyphase Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
3.4.1 Equalization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
3.4.2 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
4. PREAMBLE DESIGN FOR FILTERBANK MULTICARRIER
SYSTEMS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
4.1 Preamble Design in OFDM Systems . . . . . . . . . . . . . . . . . . . . . . . . . 75
4.2 FBMC Packet Format . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
4.3 Carrier Acquisition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
4.4 Timing Acquisition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
4.5 Equalization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
4.6 Simulation Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
4.6.1 Sensitivity Discussion for SMT and CMT . . . . . . . . . . . . . . . . . 90
4.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
5. CARRIER AND TIMING OFFSET TRACKING . . . . . . . . . . . 91
5.1 Literature Survey on FBMC Bind
Synchronization Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
5.2 Carrier Tracking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
5.2.1 SMT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
5.2.2 CMT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
5.3 Timing Tracking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
5.4 Simulation Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
5.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
6. COGNITIVE RADIO TESTBED . . . . . . . . . . . . . . . . . . . . . . . . . . 105
6.1 Problem Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
6.2 Technical Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
6.2.1 Channel Sensing and MAC Layer . . . . . . . . . . . . . . . . . . . . . . . 108
6.2.2 Transmitter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
6.2.3 Receiver . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
6.3 System Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
6.4 Implementation Process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
6.5 Spectrum Sensing Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
6.6 Test and Demonstration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
6.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
7. FILTERBANK MULTICARRIER FOR
COGNITIVE LIVE WIRE TESTING . . . . . . . . . . . . . . . . . . . . . . 121
7.1 In-Band and Out-of-Band Interference . . . . . . . . . . . . . . . . . . . . . . . 122
7.2 Filterbank MCR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
7.2.1 Signal Generator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
viii
7.2.1.1 An Analysis of the Raised-Cosine Window
Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
7.2.1.2 An Optimum Choice of the Window Function . . . . . . . . . 130
7.2.1.3 The Relationship Between Time-Domain
Windowing and Filterbank Synthesis . . . . . . . . . . . . . . . . 132
7.2.2 Analysis Filterbanks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
7.3 Cognitive Live Wire Testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
7.3.1 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
8. CONCLUSIONS AND FUTURE RESEARCH . . . . . . . . . . . . . . 137
8.1 Outlook into Future Research . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
8.1.1 Mobile FBMC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
8.1.2 MIMO FBMC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
8.1.3 Pilot Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
8.1.4 Implementation of FBMC Communications . . . . . . . . . . . . . . . 142
8.1.5 Implementation of Filterbank Multicarrier Reflectometry . . . . . 142
APPENDICES
A. ACRONYMS AND ABBREVIATIONS . . . . . . . . . . . . . . . . . . . . 143
B. VARIABLES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
REFERENCES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
ix
LIST OF FIGURES
xii
ACKNOWLEDGMENTS
INTRODUCTION
The demand for ubiquitous wireless services has been on rise in the past and
is expected to remain the same in the future. As a result, the vast majority of
the available spectral resources have already been licensed. In the United States,
government regulatory agencies have allocated different blocks of spectrum from 9
kHz to 300 GHz for various applications. Hence, it appears that the regulated radio
spectrum has been fully occupied and new applications will not have access to the
radio spectrum. However, spectrum bands need to be made available to the public
even if they are already allocated or licensed. It has been noted that the static
frequency allocations have resulted in inefficient usage of the spectrum resources.
For example, measurements have shown that that the actual spectrum utilization
in the 3-4 GHz frequency band is 0.5% and 0.3% in the 4-5 GHz [2]. In fact, most
of the commercial wireless systems transmit intermittently and they need variable
bandwidth over time.
The advancement of digital signal processors (DSPs), general purpose processors
(GPPs) and field-programmable gate arrays (FPGAs) has enabled us to build
reconfigurable radios. These radios can change their configuration to meet the
requirements of communication network they are operating on. This concept is
the foundation of software defined radios where the radio functionality is mostly
accomplished by software programming. In software defined radios (SDR), once
the signal is digitized, the radio functions are implemented using software-driven
components. Hence, the SDR technology has enabled us to build radios that can
transmit over different frequency bands and change their modulation schemes and
pulse shapes based on policies defined in software. Consequently, cognitive radio
2
technology has been proposed where the cognitive radio can share the spectrum
with the incumbent licensed users. In other words, cognitive radio technology
enables us to have secondary (i.e., unlicensed) users that are allowed to transmit
and receive data over portions of the spectra when primary (i.e., licensed) users
are inactive. This is done in a way that the secondary users (SUs) are invisible
to the primary users (PUs). In such a setting, PUs are ordinary terminals within
their base-station centric network or in direct point-to-point communications. PUs
thus do not need to possess much intelligence beyond the ability to communicate
with their peers in their networks. The SUs, on the other hand, should have
the intelligence of sensing the spectrum and using the available resources when
they need them. At the same time, the SUs need to give up the spectrum when
a PU begins transmission. This emerging technology is being investigated and
used by different research organizations and government agencies. The DARPA
next generation program (XG) [3] and wireless network after next (WNAN) [4]
programs have been studying the applicability of dynamic frequency-selective radios
based on cognitive radio concepts. Agile radios dynamically adapt to the channel
environment. These radios assess their environment and the spectrum policies and
regulations and capitalize on the available spectrum in their environment.
The enhancements in communication technology as well as the new requirements
for the public safety agencies indicate that narrow-band real-time voice communi-
cations might not be sufficient for mission critical applications. The International
Telecommunication Union (ITU) and National Emergency Number Association
(NENA) are developing the multimedia location based communication requirements
for inclusion in the next generation of public safety architectures [5]. Furthermore,
in disaster scenarios such as earthquake or hurricane the communication infras-
tructure might be damaged and therefore ad-hoc cognitive wireless communication
becomes more critical [6]. Therefore building radios which can dynamically adapt
to the environment and transmit with high data rates can save lives and reduce the
risks involved in the missions of public safety agencies [5].
The advancements in SDR technology, and the under-utilization of spectrum in
3
sense the spectrum, and this involves some sort of spectral analysis. Since the fast
Fourier transform (FFT) can be used for spectral analysis and at the same time
act as the demodulator of an OFDM signal, OFDM is a suitable candidate for
multicarrier-based cognitive radio systems. However, a number of shortcomings of
OFDM in the application of cognitive radio have been noted in [9] and solutions
to them have been proposed. To elaborate, the source of the problems with the
OFDM solution is the large side-lobes of the frequency response of the filters
that characterize the channels associated with subcarriers in an OFDM system.
Therefore, there is significant interference among the carriers of different SUs as
well as between SUs and PUs in the wireless channel.
If the spectrum sensor lacks a sufficiently high spectral dynamic range, SUs
may not be able to detect the low power PUs and they may interfere with them.
Moreover, if the resolution of the spectrum sensing is low, the radio will not be
able to best harness the wireless resources. It has been shown that FFT as part
of an OFDM data transmission system is not able to provide a sufficiently high
spectral dynamic range for channel sensing. On the other hand, as a channel
sensing tool, filterbank-based spectrum analyzer can be applied to cognitive radios
and its performance found to be close to that of the Thomsons multitaper method
(MTM) [10], which has been proposed as the best candidate for cognitive radios [11].
A multicarrier transceiver is required to feature two major properties in a
cognitive radio system where the SU dynamically fills the spectrum holes. First,
the cognitive radio transmitter must confine the spectral content of the transmitter
within the selected band(s), i.e., spectrum holes. In other words, its out-of-band
interference must be minimized. Second, the receiver should be able to avoid the
in-band interference from the other signals on the channel [12]. In other words, in
order to increase bandwidth efficiency, receivers need to have acceptable out-of-band
rejection capabilities. An OFDM signal has large side lobes and therefore, does
not satisfy the first requirement. Moreover, OFDM/FFT does not satisfy the
FCCs envisioned out-of-band rejection requirements [13]. On the other hand,
filterbank multicarrier can overcome the spectral leakage problems of OFDM at the
5
transmitter side and therefore lead to less interference from SUs to PUs and other
SUs. Filterbank receiver is also capable of providing high out-of-band attenuation.
Therefore, filterbank multicarrier has been suggested as an alternative to OFDM.
Filterbank multicarrier transmitter and receiver are implemented using a synthesis
filterbank at the transmitter and analysis filterbank at the reciver [12]. A block
diagram of a transmitter and a receiver filterbank multicarrier is presented in
Fig. 1.1.
Three classes of filterbank multicarrier (FBMC) have been studied in the lit-
erature. Interestingly, the first multicarrier methods that were developed, prior
to OFDM, were filterbank-based. The first proposal came from Chang in the
1960s, [14], who presented the conditions required for signaling a parallel set of
pulse amplitude modulated (PAM) symbol sequences through a bank of overlapping
vestigial side-band (VSB) modulated filters. A year later, Saltzberg extended the
idea and showed how the Changs method could be modified for transmission of
quadrature amplitude modulated (QAM) symbols [15]. Saltzberg showed that
a perfect reconstruction FBMC system can be implemented using a half-symbol
space delay between the in-phase and the quadrature components of QAM symbols
and by proper transmit and receive pulse-shapes in a multichannel QAM system
while having the maximum spectral efficiency. In 1980s, Hirosaki progressed more
on FBMC and proposed an efficient polyphase implementation for the Saltzberg
method [16][18]. The method proposed by Saltzberg is referred to as OFDM based
on offset QAM or OFDM-OQAM. The offset comes from the half symbol shift
between the in-phase and quadrature of each QAM symbol with respect to each
other. We refer to this method as staggered modulated multitone (SMT), where
the word staggered refers to the fact that the in-phase and quadrature components
in each QAM symbols are time staggered.
The pioneering work of Chang [14] has received much less (direct) attention
than SMT. Nevertheless, the cosine modulated filterbanks that have been widely
studied within the signal processing community [19] are nothing but a reinvention
of Changs filterbank, formulated in discrete time. The use of cosine modulated
6
ej2f0 n
s0 [k]
N hT [n]
j2f1 n
e
s1 [k]
N hT [n] ! Re{} Channel
ej2fN 1 n ej2fc n
sN 1 [k]
N hT [n]
ej2f0 n
s0 [k]
hR [n] N
ej2f1 n
s1 [k]
hR [n] N
Channel
ej2fc n ej2fN 1 n
sN 1 [k]
hR [n] N
filterbanks for data transmission was widely studied in the 1990s. The advance-
ments in digital subscriber line (DSL) technology led to more work on two classes
of FBMC communication systems, namely, filtered multitone (FMT) and discrete
wavelet multitone (DWMT) modulation [20]. More recently, in [21] it has been
shown that DWMT is essentially using cosine-modulated filterbanks. Therefore,
DWMT was renamed to cosine-modulated multitone (CMT). It is also known that
CMT is using vestigial sideband (VSB) modulation to transmit PAM symbols [21].
FMT is another multicarrier communication scheme which has been proposed
for DSL applications [22]. In FMT, the adjacent subcarriers do not overlap. There-
7
fore, FMT is not bandwidth efficient when it is compared with SMT and CMT.
1.2 OFDM/FFT
OFDM has a number of problems for cognitive radio applications [8]. Herein,
we discuss some shortcomings of OFDM in cognitive radio networks. Assuming
that the data symbols are independent, the power spectral density (PSD) of an
OFDM signal can be described as a summation of PSDs of each subcarrier
X
( f ) = i (f ) (1.1)
i
In (1.2), K is the signal level, TS is the period of an OFDM symbol which is the
summation of one FFT block and the guard interval, and fi is the center frequency
of the ith subcarrier.
The sidelobes of the sinc pulse shape is relatively large. Therefore, the out-
of-band energy generated by an OFDM signal is significant. In a cognitive radio
setting, where it is critical to have minimal interference with the PUs, the side-lobes
may cause an unacceptable level of interference to the PUs. The sinc shape of the
subcarrier PSD is a direct result of the abrupt transition among successive OFDM
symbols. The sinc pulse shape can be avoided if we use soft transitions among
successive symbols. Cyclic extension of each OFDM symbol from TS to (1 + 2)TS
and windowing by a raised cosine shape can provide us the soft transition. Fig. 1.2
shows how the successive extended OFDM symbols are overlapped. This scheme
increases the effective duration of each OFDM symbol from TS to (1 + )TS , which
1
results in bandwidth loss of +1
.
Weiss et al. investigated different choices of , and concluded that as large as
1 is needed to obtain a reasonable suppression of the out-of-band energy [8]. Fig.
1.3 depicts an example of the PSD for various choices of . This clearly shows the
large side-lobes of the rectangular window ( = 0) and how the side-lobes decrease
in magnitude as increases.
8
TS TS TS
One point worth noting is that even though the raised cosine window reduces
the side-lobes of the subcarrier spectra, the side-lobes close to the main lobe are
still large (Fig. 1.3). This point is noted in [8] where a subcarrier deactivation
mechanism is used to avoid subcarrier bands near the active PU band. This
mechanism results in a further loss in bandwidth efficiency. Brandes et al. [23], [24]
propose a method for side-lobes reduction by using non-zero value deactivated
subcarriers. It has been reported in [24] that side-lobes at around 60 dB can be
achieved for = 0.2. Unfortunately, the subcarrier cancelation procedure involves
a constraint optimization for each OFDM symbol, which is a computationally
expensive task. Using filterbank solution, on the other hand, lower side-lobes
(90 dB) can be achieved at virtually no additional computational cost [13].
As was mentioned at the beginning of this chapter, the out of band rejection
capability of a cognitive radio receiver is important to minimize the interference
received by a SU from PUs and other SUs. This is a major problem in an OFDM
receiver if proper considerations are not taken into account [8]. The solution to the
potentially weak out-of-band rejection has been studied in the DSL literature [25].
It is shown that this problem can be solved by applying a window to the received
signal prior to passing it to the FFT block for demodulation. Fig. 1.4 presents the
method of receiver windowing. If the window is rectangular, one picks N samples
of a received OFDM symbol, during the time T . These N samples are then passed
to an N -point FFT for demodulation. Using windowing at the receiver, (1 + )N
samples are picked from the OFDM symbol and the window is applied to them.
A Fourier transform is then applied to (1 + )N time-domain samples and the
9
0
=0
10 =0.25
20 =0.5
=1
30
40
i(f), dB
50
60
70
80
90
100
0 2 4 6 8 10
fT
Figure 1.3: The power spectral density of a subcarrier of an OFDM signal when
different choices of the roll-off parameter is used.
T T
2 2
(1 + )TS
sensing mechanism needs to have a high spectral dynamic range. While FFT has
been suggested as one channel sensing method, note that it suffers from a number of
shortcomings that originate from the large side lobes of the frequency response of the
filters that characterize each subcarrier. These sidelobes produce spectrum leakage
from neighboring subcarriers, resulting in significant inaccuracy and low dynamic
range. Thus, with the FFT, SUs are less spectrum agile and cannot detect low
power users. This might not be an important issue in systems where the channel
access is performed in Time Division Duplex (TDD) and Time Division Multiple
Access (TDMA). However, for the systems that incorporate frequency division
multiple access and frequency division duplex (FDMA/FDD), the limitations of
FFT are serious. Hence, in our solution [12], we propose using filterbanks as the
sensing method. By using a filterbank sensing system, the side lobes of the filters
associated with each carrier can be made arbitrarily small by adjusting filter length
and design. As a result, filters are no longer the limiting factor in achieving high
spectral dynamic range. The signal power of the output of the filterbank is then
used to estimate the signal spectrum. A block diagram of a filterbank spectrum
sensor is presented in Fig. 1.5 where h(t) is the prototype filter used for spectrum
sensing, fc + f0 , fc + f1 ,...,fc + fN 1 are the center frequency of the spectrum bands
we are sensing.
11
ej2f0 t
ej2fc n ej2fN 1 t
Haykin showed that the multitapper method (MTM) is a near optimal channel
sensing method [11]. Unfortunately MTM comes at the expense of high computa-
tional complexity. In [12] and [26], it was shown that a filterbank of prolate filters
can be used for sensing with virtually ideal performance. In the filterbank sensing
method, a prototype filter is designed and then modulated to sense each subcarrier.
Implementation can be performed efficiently through a polyphase structure which
is described in Chapter 3.
It is worth noting that filterbank sensing can be applied to other applications
when we need to detect the presence of a signal in a medium. For example, in
wiring applications, test signal should be transmitted on unoccupied portions of the
spectrum, so that it does not interfere with the live signal on the wire. Therefore,
spectrum sensing can be used to detect the presence of the live signal on the wire
and to find the spectrum holes which can be used to transmit a test signal [27].
Moreover, one can use the filterbank to generate test pulse shapes that have low
out-of-band interference to transmit on the spectrum holes on the wire.
12
same data rate, would have a rate of 1/T real symbols on each subcarrier with the
bandwidth of 1/2T . Therefore, the same bandwidth is divided into twice as many
subcarriers in case of CMT to achieve the same data rate.
2
R=
T
(a) FMT
f
2 NR(1+a), a >0
R=
T
(b ) O F DM - OQ A M
f
NR
R/2=
T
( c) C MT
NR/2
recovery in SMT and CMT communication systems which were proposed for other
applications as well. SMT has been a candidate for physical layer of IEEE 802.11
and Third Generation Partnership Project (3GPP) LTE (Long Term Evolution).
Furthermore, a cognitive system is designed for fault detection on the live wires
based on filterbank multicarrier spectrum sensing and test signal generation.
The contribution of this dissertation are as follows:
Much of the recent CFO estimation literature attempts to develop blind CFO
estimators based on correlation properties of SMT signal. These methods
need a large window of signal to detect the carrier and timing offset and
are complex to implement. We focus on design of a preamble for packetized
communication system which may be used in practical systems. We present
a preamble structure which can be used in a packet based communication
system to detect the packet, adjust the AGC gain, and perform carrier acqui-
sition, timing recovery, and channel estimation.
While preamble is used for initial timing and carrier offset estimation, without
any carrier tracking loop, the carrier and timing phase may drift over the
length of the payload. The necessity of carrier and timing offset tracking
arises especially due to the use of long packets in data communication systems
such as 802.11n. We propose decision directed timing and carrier tracking
mechanisms for SMT and CMT. These tracking methods may be used to
track the residual offsets after the initial acquisitions are done.
able to show that our cognitive radios can coexist with other active primary and
cognitive devices.
Multicarrier reflectometry (MCR) for locating faults on live wires has recently
been proposed. Chapter 7 studies the use of filterbanks for generation/synthesis of
MCR test signals and also for signal analysis for fault identification and location.
We note that the test signals have to be confined to the portion(s) of the frequency
band that is (are) free of signals already on the wires in order to avoid interfering
with them. Moreover, for effective analysis of the reflected waves, optimal filters
that separate the test signal tones and also minimize leakage from the existing
signals on the wires should be designed. We discuss the criteria necessary to design
effective MCR systems and develop the relevant filterbank design procedures. In
Chapter 7, the novel idea of cognitive live wire testing is presented. A cognitive
live wire tester measures the live wire signal activities and decides which part of
the spectrum should be used for testing.
Finally, Chapter 8 presents the concluding remarks and future research. Various
areas of research about FBMC are presented in Chapter 8. We discuss that FBMC
can be applied to different communication applications and therefore research work
is required in many areas for FBMC systems to function in different usage scenarios.
CHAPTER 2
for system with high mobility and high doppler effect where orthogonality among
subcarriers might be destroyed and ICI mounts to considerable distortion in OFDM
signals. Furthermore, as discussed in Chapter 1, for asynchronous multicarrier
communications in multiuser systems and cognitive radio networks, big sidelobes
of OFDM subcarriers result in interference and therefore loss of performance. There
would be up to 50% loss in bandwidth efficiency to suppress the sidelobes [9].
It is also interesting to note that the researchers who have studied filterbanks
have invented a class of filterbanks which are called modified DFT (MDFT) fil-
terbank, [30]. Careful study of MDFT reveals that this is in fact a reinvention
of Saltzbergs filterbank (in discrete-time) with emphasis on compression/coding.
The literature on MDFT begins with the pioneering works of Fliege, [31], and later
has been extended by others, e.g., [32] and [33].
Here, we present an overview fn filterbank communication techniques: SMT,
CMT and FMT. We focus more on SMT and CMT where the bandwidth efficiency
is maximum. The polyphase implementation of SMT and CMT are discussed.
Formulation of FBMC sensitivity to timing and carrier offset is presented in Section
3.
Transmitter
sI0 (t) h(t)
!
jsQ
0 (t)
h(t T /2)
ej ( T t+ 2 )
2
sI1 (t)
h(t) !
!
jsQ
1 (t)
h(t T /2) !{} Channel
ej2fc t
j(N 1)( 2
T t+ 2 )
e
sIM 1 (t)
h(t) !
!
jsQ
N 1 (t)
h(t T /2)
Receiver
sI0 (n)
!{} h(t)
sQ
0 (n)
!{} h(t + T /2)
ej ( T t+ 2 )
2
sI1 (n)
!{} h(t)
Channel
!
sQ
h(t + T /2) 1 (n)
!{}
ej2fc t
ej(N1)( T t+ 2 )
2
sIN 1 (n)
! !{} h(t)
sQ
N 1 (n)
!{} h(t + T /2)
where (t) is the Dirac delta function. The complex-valued baseband SMT modu-
lated signal is defined as
22
X
N 1
x(t) = xm (t) (2.4)
m=0
where
X
jm( 2t + )
xm (t) = sIm [l]h(t lT ) + jsQ
m [l]h(t lT T /2) e
T 2 . (2.5)
l=
Analogously, assuming an ideal channel, the output of the receiver, sk [n], consists
of the real and imaginary components sIk [n] and sQ
k [n],
In (2.6), sIk [n] is found as the real part of the signal at the output of the corre-
sponding matched filter with response h(t) and expressed as
2
+ 2 )
yk ( ) = ejk( T x( ) (2.8)
is the demodulated signal before matched filtering. Substituting (2.4) and (2.8) in
(2.8), we obtain
Z X
N 1
jk( 2 + 2 )
sIk [n] = < h( nT ) xm ( )e T d . (2.9)
m=0
X X
N 1 Z
sIk [n] = < h( nT ) sIm [l]h( lT ) (2.10)
l= m=0
j(mk)( 2
)
+ jsQ
m [l]h( lT T /2) e T
+ 2 d.
23
Changing the variable nT to and then taking into account that l varies from
to , (2.11) can be simplified to
X X
N 1 Z
2
sIk [n] = + l]h( lT )h( ) cos (m k)
sIm [n +
l= m=0 T 2
2
sm [n + l]h( lT T /2)h( ) sin (m k)
Q
+ d.
T 2
(2.12)
Starting with (2.13) and following the same line of derivations as above, we obtain
X X
M 1 Z
2
sQ
k [n] = sm [n + l]h( lT )h( + T /2) sin (m k)
I
+
l= m=0 T 2
2
+sQm [l + n]h( lT + T /2)h( + T /2) cos (m k) + d .
T 2
(2.14)
We now proceed to discuss the design of the matched filter h(t). In an ideal
transmission system, the received signal equals the transmitted one and
It follows directly from (5.6) and (5.7), that (2.15) can be met, if h(t) is chosen
such that the following identities are satisfied.
24
Z
2
h( lT )h( ) cos (m k) + d = [m k, l] (2.16)
T 2
Z
2
h( lT T /2)h( ) sin (m k) + d = 0 (2.17)
T 2
Z
2
h( lT )h( + T /2) sin (m k) + d = 0 (2.18)
T 2
Z
2
h( lT + T /2)h( + T /2) cos (m k) + d = [m k, l] (2.19)
T 2
where [m k, l] is the Dirac delta function. [m k, l] is one if m k = 0
and l = 0 and it is zeros otherwise. For convenience of the design, it is common to
constrain h( ) to a real and even (i.e., symmetric around = 0) function of time, .
Under these constraints, one can show that the integrand in (2.17) is anti-symmetric
around = lT /2 + T /4 and the integrand in (2.18) is anti-symmetric around =
lT /2T /4. These, in turn, imply that (2.17) and (2.18) are automatically satisfied.
Hence, in designing h( ), it is sufficient to limit ourselves to the constraints imposed
by (2.16) and (2.19). Also, to further simplify the design of h( ), it is reasonable
to assume that only adjacent subcarrier bands may overlap. When this is the case,
only instances of m and k where m k = 0, +1, 1 need to be considered in (2.16)
and (2.19). Values of m and k where |m k| > 1 are related to nonadjacent
subcarrier bands and thus, their multiplication in (2.16) results in values close to
zero. Also, for m k = 0, and iff h( ) is a root-Nyquist filter, (2.16) equates to
Z
h( lT )h( )d = [0, l] = [l]. (2.20)
Finally, for m k = 1, (2.16) reduces to
Z
2
h( lT )h( ) sin = 0. (2.21)
T
Since, h( ) is a symmetric filter and sin is odd with respect to origin, we can easily
show that the integrand in (2.21) is anti-symmetric around = lT /2 + T /4 and
thus (2.21) is valid. In a similar way, and under the same conditions, one can show
that (2.19) reduces to (2.20). In summary, any realization of h( ) which is even and
real and satisfies the root-Nyquist condition in (2.20), leads to an SMT transceiver
system which satisfies (2.15).
25
X
sm (t) = sm [n](t nT ) (2.22)
n
where sm [n] are real PAM symbols. According to Fig. 2.2, the baseband multicarrier
CMT signal at the transmitter, x(t), is obtained as
X
N 1
x(t) = xm (t) (2.23)
m=0
where
X
+
xm (t) = sm [l]h(t lT )ej 2T (tlT ) ejm( T t+ 2 ) . (2.24)
l=
The received PAM symbol on each subcarrier is the real part of the complex
jt jt
output of the matched filter, h(t)e 2T = h(t)e 2T , viz.,
sck [n] = yk (t) ? h(t)ej 2T t |t=nT (2.26)
and
yk (t) = ejk( T t+ 2 ) x(t). (2.27)
26
Transmitter
s0 (t)
h(t)ej 2T t
ej( T t+ 2 )
s1 (t)
h(t)ej 2T t ! !{} Channel
ej(N 1)( T n+ 2 )
ej2fc t
sM 1 (t)
h(t)ej 2T t
Receiver
s0 [n]
h(t)ej 2T t !{}
ej( T t+ 2 )
s1 [n]
h(t)ej 2T t !{}
Channel
ej(N 1)( T n+ 2 )
ej2fc n
sM 1 [n]
h(t)ej 2T t !{}
f f f
(2.28)
X
N 1 X
+ Z
sck [n] = sm [l]h( )h( (l n)T )ej 2 (nl) ej(mk)( T + 2 ) d. (2.29)
m=0 l=
Furthermore, by changing l n to l, and then taking into account that l varies from
to , we obtain
28
X
N 1 X
+ Z
sck [n] = sm [l + n]h( )h( lT )ejl 2 ej(mk)( T + 2 ) d. (2.30)
m=0 l=
Using (2.59), in order to have an ISI free communications, the following identity
should be satisfied Z
cos(l ) h( )h( lT )d = [l]. (2.32)
2
Noting that, cos(l 2 ) is equal to zero for odd values of l. (2.32) reduces to
Z
h( )h( 2kT )d = [m] (2.33)
These results imply that h(t) should be a square-root Nyquist filter with zero
crossing at the interval of 2T . Hence h(t) is a pulse shape that normally would
1
be used in a system with a data rate of 2T
. This is inline with the fact that we
are transmitting VSB signals with half the bandwidth of a QAM double sideband
signal. Similar to SMT, we assume that the filter h(t) is designed such that only
adjacent subcarriers overlap. Therefore, using (2.59), in order to have ICI free
communication, we need to have
g1 [l] = 0 (2.34)
and
g1 [l] = 0 (2.35)
and Z
g1 [l] = h( )h( lT ) cos + +l d. (2.37)
T 2 2
29
where yk ( ), in the presence of timing, carrier and phase offset, can be written as
jk( 2 +
2 ) x( t ) e j(2f + )
yk ( ) = e T
0
0
+ ( ) (2.43)
X X
N 1 Z
sIk [n] = sIm [l]h( nT )h( t0 lT )
l= m=0
2 2mt0
cos (m k) + + 2f + 0
T 2 T
sQ
m [l]h( nT )h( t0 lT T /2)
2 2mt0
sin (m k) + + 2f + 0 d )
T 2 T
+ (nT ) (2.45)
31
where (t) is
Z
2k
(t) = ( )h(t t0 ) cos k + d. (2.46)
T 2
Applying the variable change nT to , (2.45) can be simplified to
X N X1 Z
I
sk [n] = sIm [n + l]h( t0 lT )h( )
l= m=0
2 2mt0
cos (m k) + + 2f + 2f nT + 0
T 2 T
m [n + l]h( t0 lT T /2)h( )
sQ
2 2mt0
sin (m k) + + 2f + 2f nT + 0 d
T 2 T
+ (nT ). (2.47)
In this chapter, we study the effect of timing and carrier offset with the assump-
tion of no residual phase error. That is assumed that 2f nT 2kt0
T
+ 0 is
being perfectly tracked and compensated, e.g., using a decision directed methods.
Furthermore, we assume that roll off factor of h is smaller than one. Thus at the
receiver, the interference term from the carriers which are more than one carrier
spacing away is negligible. As a result, we only consider the desired subcarrier and
the adjacent ones in our calculations. Hence, we can write sIk [n] as
X X
u=+1
Q
I I I Q
sk [n] = sk+u [l + k]gu [l] + sk+u [l + k]gu [l] + (nT ) (2.48)
l= u=1
where guI [l] and guQ [l] are the impact of the in-phase and quadrature symbols trans-
mitted on the carrier k + u on in-phase element of carrier k, respectively. guI [l] and
guQ [l] can be expressed as
Z
I
gu [l] = h( t0 lT )h( )
2 2ut0
cos u + + 2f d (2.49)
T 2 T
and
Z
guQ [l] = h( T /2 t0 lT )h( )
2 2ut0
sin u + + 2f d. (2.50)
T 2 T
32
In the rest of this chapter, the effect of carrier and timing offset is studied indepen-
dently. We investigate the effect of carrier and timing offset on SINR degradation
as well as the magnitude of ICI and ISI terms. SINR degradation is defined as
SINRA
D(f, t0 ) = (2.51)
SINRS
where SINRA is the signal to noise ratio at the input of the decision block at the
receiver which may have carrier or timing offset. SINRS is the signal to noise ratio
if the transmitter and receiver are completely synchronous. Using (2.48), SINRA
can be calculated as following
2
g0I [0]2 E[sIk [n]]
SINRA = (2.52)
2 P P u=+1 Q I 2 g I [0]2 + 2
E[sIk [n]] l=
2
u=1 gu [l] + gu [l] 0
while SINRS is
2
E[sIk [n]]
SINRS = . (2.53)
2
To the best of our knowledge there is no closed form equation for D(f, t0 ) in case
of SMT. Hence, the distortion can only be studied numerically.
where
jk( T + 2 )
yk ( ) = e [x( t0 ) e2f +0
+ ( ) . (2.55)
X
N 1 X
+ Z
sC
k [n] = sm [l]h( )h( t0 + (l n)T )ej 2 (nl)
m=0 l=
mt
j (mk)( T + 2 ) T 0 +2f +0
e d. (2.56)
33
where (t) is Z
k
+ 2 )
(t) = ( )h(t t0 )ejk( T d (2.58)
Therefore, the received PAM symbol is
X
N + Z
1 X
sk [n] = sm [l + n]h( )h( t0 lT )
m=0 l=
mt0
cos (m k)( + ) l + 2f nT + 2f + 0 d
T 2 2 T
+ 0 (nT ) (2.59)
where 0 (nT ) = <{(nT )}. Using the assumption that the phase offset is perfectly
compensated, one obtains
X X
u=+1
sk [n] = sk+u [l + k]gu [l] + 0 (nT ) (2.60)
l= u=1
where
Z
gu [l] = h( )h( t0 lT )
ut0
cos u + l + 2f d. (2.61)
T 2 2 T
We can express the degradation for CMT is the same way as (2.51). Where SINRA
and SINRS are defined as
g0 [0]2 E[s2k [n]]
SINRA = (2.62)
P Pu=+1
u=1 gu [l] g0 [0]
2 I
E[sk [n]] l=
2 2 + 20
and
E[sk 2 [n]]
SINRS = . (2.63)
20
34
Similar to SMT case, there is no closed form solution in the current literature for
the degradation function of CMT. Hence, numerical results are presented for CMT
as well as SMT.
0
10
20
30 gQ [l]
0
40
Interference power
gI0[l]
50
60 gQ [l]
1
70
gI1[l]
80
90 gQ [l]
1
100
gI1[l]
110
120
130
140
150
6 5 4 3 2 1 0 1 2 3 4 5 6
l
Figure 2.4: The ICI, ISI, and signal energy of SMT when f is 0.02 of carrier
spacing.
0
10
20
30 gQ [l]
0
40
Interference power
gI0[l]
50
60 gQ [l]
1
70
gI1[l]
80
90 gQ [l]
1
100
gI1[l]
110
120
130
140
150
6 5 4 3 2 1 0 1 2 3 4 5 6
l
Figure 2.5: The ICI, ISI, and signal energy of SMT when f is 0.01 of carrier
spacing.
36
0
10 g0[l]
20
g1[l]
30
40 g1[l]
Interference power
50
60
70
80
90
100
110
120
130
140
150
6 5 4 3 2 1 0 1 2 3 4 5 6
l
Figure 2.6: The ICI, ISI, and signal energy of CMT when f is 0.02 of carrier
spacing.
0
10
g0[l]
20
30 g1[l]
40 g1[l]
Interference power
50
60
70
80
90
100
110
120
130
140
150
6 5 4 3 2 1 0 1 2 3 4 5 6
l
Figure 2.7: The ICI, ISI, and signal energy of CMT when f is 0.01 of carrier
spacing.
37
ference is the summation of powers of all the interference terms. In Fig. 2.8, SIR for
SMT and CMT for different values of CFO are presented. It has been shown in [38]
that the performance of CMT and SMT is the same in the presence of carrier
offset. Thus Fig. 2.8 presents one curve for both CMT and CMT systems. As
indicated in Fig. 2.8, for the CFO values less than 0.023 of carrier spacing the SIR
is larger than 30 dB. Therefore, the desired symbols can comfortably be detected
for values of CFO less than 0.023 of carrier spacing. Moreover, one can see that the
CFO requirement for OFDM system which is 0.01 to 0.02 of carrier spacing [39] is
sufficient for SMT and CMT systems as well.
We also study the degradation that is caused by carrier offset both for SMT
and CMT systems and compare it to OFDM. Fig. 2.9 presents the distortion as a
function of carrier offset when the SNR at the receiver input is 10 dB. SMT and
CMT have the same distortion results as predicted in [38]. For a normalized carrier
offset (to the carrier spacing) of 0.1, SMT and CMT have the same results and they
both have 2 dB less distortion than an OFDM system. Fig. 2.10 presents distortion
when the SNR at the receiver input is 30 dB. As we can see, SMT and CMT have
90
CMT and SMT
80
70
Signal To Interference Ratio
60
50
40
30
20
10
0
0 0.023 0.04 0.06 0.08 0.1
Normalized Carrier Offset
Figure 2.8: Signal to interference ratio in the presence of carrier offset for SMT
and CMT.
38
1.8
Distortion in dB
1.2
0.8
0.6
0.4
0.2
0
0 0.02 0.04 0.06 0.08 0.1
Normalized Carrier Offset
Figure 2.9: Distortion of SMT, CMT, and OFDM in the presence of carrier offset
when SNR=10 dB.
16
CMT and SMT
14
OFDM
12
Distortion in dB
10
0
0 0.02 0.04 0.06 0.08 0.1
Normalized Carrier Offset
Figure 2.10: Degradation of SMT, CMT, and OFDM in the presence of carrier
offset when SNR=30 dB.
39
0.8 dB less distortion than OFDM when the normalized carrier offset is 0.1. As a
result, if a system requires high SINR, SMT and CMT outperform OFDM in terms
of CFO immunity. This indeed is an important gain in a system with high data
rate when higher order modulations are required.
0
10
20
30 gQ [l]
0
40
Interference Energy
gI0[l]
50
60 gQ [l]
1
70
gI1[l]
80
90 gQ [l]
1
100 I
g1[l]
110
120
130
140
150
6 5 4 3 2 1 0 1 2 3 4 5 6
l
2
Figure 2.11: The ICI, ISI, and signal energy of SMT when t0 is 64
T.
0
10
20
30 gQ [l]
0
40
Interference Energy
gI0[l]
50
60 gQ [l]
1
70
gI1[l]
80
90 gQ [l]
1
100 I
g1[l]
110
120
130
140
150
6 5 4 3 2 1 0 1 2 3 4 5 6
l
1
Figure 2.12: The ICI, ISI, and signal energy of SMT when t0 is 64
T.
41
0
10
20
30
40
Interference power
g0[n]
50
60
70 g1[n]
80
90 g1[n]
100
110
120
130
140
150
5 4 3 2 1 0 1 2 3 4 5
n
1
Figure 2.13: The ICI, ISI, and signal energy of CMT when t0 is 64
T.
0
10
20
30
40
Interference power
g0[n]
50
60
70 g1[n]
80
90 g1[n]
100
110
120
130
140
150
5 4 3 2 1 0 1 2 3 4 5
n
2
Figure 2.14: The ICI, ISI, and signal energy of CMT when t0 is 64
T.
42
2.4.1 Summary
This chapter investigated the filter design conditions for SMT and CMT. It was
shown that root-Nyquist filter designs can be used for SMT and CMT. Moreover, we
explored the sensitivity of SMT and CMT modulations to carrier and timing offset.
It was shown that assuming perfect phase tracking, it is possible to comfortably
detect the transmitted symbols when there is a CFO as much as 0.02 of carrier
spacing. Furthermore, we saw from the simulation results that SMT and CMT
outperform OFDM in terms of CFO immunity especially when high SNRs are
required. Therefore, the carrier offset requirement for OFDM will be sufficient for
SMT systems. The effect of timing offset was also numerically studied. It was
shown that SMT and CMT are sensitive to timing offset and thus accurate timing
recovery algorithms need to be developed. Furthermore, we saw that for timing
offset of less than 0.016T , the resulting SIR is greater than 30 dB.
43
90
CMT and SMT
80
70
50
40
30
20
10
0
0 0.01601 0.04 0.06 0.08 0.1
Normalized Timing Offset
Figure 2.15: Signal to interference ratio in the presence of timing offset for SMT
and CMT.
1.5
0.5
0
0 0.02 0.04 0.06 0.08 0.1
Normalized Timing Offset
Figure 2.16: Distortion versus normalized timing offset for SNR = 10 dB.
44
16
12
Distortion in dB
10
0
0 0.02 0.04 0.06 0.08 0.1
Normalized Timing Offset
Figure 2.17: Distortion versus normalized timing offset for SNR = 30 dB.
CHAPTER 3
POLYPHASE IMPLEMENTATION OF
FILTERBANK MULTICARRIER
COMMUNICATION SYSTEMS
2k 2k
Using ej T
t
= ej T
(tlT )
, we get
46
X
j 2k
xk (t) = sk [l]h(t lT )e L
(tlT )
l=
X
= sk [l]hk (t lT ) (3.3)
l=
where
2k
hk (t) = h(t)ej L
t
(3.4)
X
xk [n] = sk [l]hk [n lL]. (3.5)
l=
X
N 1
X[z] = Xk (z). (3.6)
k=0
Using multirate signal processing identities [34], one finds from Fig. 3.2 that
2
where WL = ej L . Substituting (3.8) in (3.7), we obtain
X
N 1 X
X(z) = Sk (z L )h[n]z n WLkn . (3.9)
k=0 n=
47
s0 [n] x0 [n]
L h[n] j2
e L n
s1 [n] x1 [n]
L h[n] ! x[n]
2
ej(N 1) L n
sN 1 [n] xN 1 [n]
L h[n]
Figure 3.1: General structure of a synthesis filterbank
X
L1
H(z) = z p El (z L ) (3.11)
l=0
where El (z) is the z-transform of el [n]. Using (3.11), (3.9) can be rearranged
L1
X N
X 1
l L L kl
X(z) = z El (z ) Sk (z )WL . (3.12)
l=0 k=0
PL1
The vector q = k=0 Sk (z L
)WLkl l = 0, , L 1 is the size L IDFT of
the upsampled signal. If the number of symbols, N , is less than L, zeros need to
be appended to form a vector with length of L. The multiplication by El (z L ) is
equivalent to passing the output of IDFT through the polyphase filters.Therefore,
using (3.12), one can design the structure presented in Fig. 3.2.
Using the properties of multirate systems [34], one can simplify the structure
presented in Fig. 3.2. Since, IDFT is simply a constant matrix multiplication, it
48
s0 [n] x[n]
L E0 (z L )
z 1
L-point IDFT
s1 [n]
L E1 (z L )
sN 1 [n]
L
0 z 1
0 EL1 (z L )
Sk (z L ) = Hk (z)X(z). (3.13)
s0 [n] x[n]
E0 (z) L
z 1
s1 [n] L-point IDFT
E1 (z) L
sN 1 [n]
0 z 1
0 EL1 (z) L
s0 [n]
j2 h[n] L
e L n
s1 [n]
h[n] L
x[n]
2
ej(N 1) L n
sN 1 [n]
h[n] L
x[n] s0 [n]
E0 (z ) L L
z 1
s1 [n]
L-point IDFT
E1 (z L ) L
sN 1 [n]
L
z 1
EL1 (z L )
x[n] s0 [n]
L E0 (z)
z 1
s1 [n]
L-point IDFT
L E1 (z)
sN 1 [n]
z 1
L EL1 (z)
2
ej L n
jsI1 (t) !
L h[n]
2
ej(N 1) L n
j N 1 sIN 1 (t) !
L h[n]
jsQ
0 (t) L h[n L/2]
2
ej L n
j 2 sQ !
1 (t) L h[n L/2]
2
ej(N 1) L n
j N sQ !
N 1 (t)
L h[n L/2]
Fig. 3.8 where PSFB is polyphase synthesis filterbanks. The upsampling rate of
L is assumed in the structure and the delay of L/2 which is equivalent to T /2
is performed after the synthesis filterbank on the second polyphase structure. We
call the structure in Fig. 3.8 the synthesis structure for Type-I SMT Polyphase
Structure.
Similar to the changes we made in the synthesis filterbank, we can break the
53
sI0 [n]
jsI1 [n]
j N 1 sIN 1 [n]
P SF B
0
x[n]
jsQ
0 [n]
j 2 sIQ [n]
j N sQ
N 1 [n] P SF B z L/2
0
transmitter structure at Fig. 2.1 into two analysis filterbanks with filters of h(t) and
2
+ 2 )
h(t + T /2). The k 2 from the term ejk( T can be moved to the output of the
filterbanks and the delay of T /2 can be moved to the input signal. Doing these
changes, the receiver part of Fig. 2.1 can be transformed to Fig. 3.9. By replacing
the structures in Fig. 3.9 with polyphase analysis filterbanks (PAFB) presented in
Fig. 3.4, the structure in Fig. 3.10 can be obtained. The structures in Fig. 3.8 and
Fig. 3.10 are used in [34] and [30].
separate filters with the delay of T /2. Alternatively, one may note that we can
define (2.5) as
xm (t) = s0k (t) h(t) (3.16)
where
X X
s0k (t) = sIk [n](t nN ) + j k [n](t nN T /2).
sQ (3.17)
n n
sI0 [n]
h[n] L !{}
2 j
ej L n
! ! sI1 (n)
h[n] L !{}
2 (j)N 1
ej(N1) L n
! ! sIN 1 (n)
h[n] L !{}
y[n]
sQ
0 (n)
h[n + L/2] L !{}
2 j
ej L n
! ! sQ
1 (n)
h[n + L/2] L !{}
2 (j)N 1
ej(N1) L n
! ! sQ
N 1 (n)
h[n + L/2] L !{}
sI0 [n]
!{}
j
sI1 [n]
!{}
P AF B (j)N 1
sIN 1 [n]
!{}
y[n]
sQ
0 [n]
!{}
j
sQ
1 [n]
!{}
z L/2 P AF B (j)N 1
sQ
N 1 [n]
!{}
jsQ
0 [n]
2
z 1
sI0 [n] !
2 L/2 h[n]
j 2 sIQ [n]
2
2
z 1 ej L n
jsI1 [n]
2
!
L/2 h[n]
!
j N sQ
N 1 [n]
2 2
1
ej(N 1) L n
z
j N 1 sIN 1 [n] ! !
2 L/2 h[n]
structure for a combined receiver structure is depicted in Fig. 3.13. The analysis
polyphase structure for Fig. 3.13 can be derived in the same way as the one in
Fig. 3.3. The only difference is that downsampling of L/2 is used instead of L.
Hence, upsampled polyphase components, Ek (z 2 ), are used in the structure. The
fixed structure of the type-II polyphase analysis filterbanks is presented in Fig. 3.14.
At even discrete time indexes, the input of odd carriers are strictly imaginary
and the input of even subcarriers are strictly real.
At odd discrete time indexes, the input of odd subcarriers are strictly real
and the input of even subcarriers are strictly imaginary.
58
jsQ
0 [n]
2
z 1
sI0 [n] s!0 [n] S0! [n]
2 + E0 (z 2 ) L/2
j 2 sIQ [n]
2 z 1
z 1
L point IDFT
z 1
j N sQ
N 1 [n]
2
z 1
j N 1 sIN 1 [n] s!N 1 [n]
2 +
0 z 1
!
SL1 [n]
0 EL1 (z 2 ) L/2 +
sQ
0 [n]
2 !{}
z1
sI0 [n]
h[n] L/2 2 !{}
sQ
1 [n]
2 !{}
j 2
j
e L n
z1
y[n] ! ! sI1 [n]
h[n] L/2 2 !{}
sQ
N 1 [n]
2 !{}
j(N1) 2
L n
(j)N 1
e z1
! ! sIN 1 [n]
h[n] L/2 2 !{}
Therefore, the input of IDFT at even discrete time and odd discrete time indexes
can be written as
(1)p sI2p [n] k = 2p
s0k [2n] = p I (3.18)
j(1) s2p+1 [n] k = 2p + 1
and
j(1)p sQ
2p [n] k = 2p
s0k [2n + 1] = p+1 Q (3.19)
(1) s2p+1 [n] k = 2p + 1,
respectively. If we consider the even time index 2n, the IDFT in Fig. 3.12 can be
written as
X
L1
Sk0 [2n] = s0m [2n]WLkm . (3.20)
m=0
X
L/21
pk
X
L/21
pk
Sk0 [2n] = (1)p sI2p [n]WL/2 + jWLk (1)p sI2p+1 [n]WL/2 . (3.21)
p=0 p=0
In (3.21), the IDFT of size L is broken into two IDFT of size L/2 where all the
inputs are strictly real. Using this, [29] and [28] have proposed using an IDFT of
60
sQ
1 [n]
2 !{}
z1
y[n] s!0 [n] S0! [n] sI0 [n]
L/2 E0 (z 2 ) 2 !{}
sQ
0 [n]
2 !{}
z 1
j z1
s!1 [n] S1! [n] sI1 [n]
L Point IDFT
L/2 E1 (z )2
2 !{}
z 1
sQ
N 1 [n]
2 !{}
(j)N 1
z1
!
SN 1 [n]
sIN 1 [n]
2 !{}
z 1
s!L1 [n]
L/2 EL1 (z 2 )
size L/2, where the input of IDFT is the addition of adjacent subcarriers. To do
this, one may define
X
L/21
pk
Uk [n] = up [n]WL/2 (3.22)
p=0
X
L/21
pk
Uk [2n] = (1)p (sI2p [n] + jsI2p+1 [n])WL/2 . (3.24)
p=0
Next, we define
1
Tk1 [2n] = (Uk [2n] + UL/2k [2n]) (3.25)
2
61
1
Tk2 [2n] = (Uk [2n] UL/2k
[2n]) (3.26)
2
X
L/21
p(kL/2)
+ (1) p
(sI2p [n] jsI2p+1 [n])WL/2
p=0
L/21
X
1 pk
= (1)p (sI2p [n] + jsI2p+1 [n])WL/2
2 p=0
X
L/21
pk
+ (1) p
(sI2p [n] jsI2p+1 [n]) WL/2
p=0
1 X
L/21
pk
= (1)p sI2p [n]WL/2 . (3.27)
2 p=0
X
L/21
pk
Tk2 [2n] =j (1)p sI2p+1 [n]WL/2 . (3.28)
p=0
X
L/21
pk
X
L/21
pk
Sk0 [2n + 1] = j(1)p sQ
2p [n]WL/2 + WLk (1)p+1 sQ
2p+1 [n]WL/2 .
p=0 p=0
where
1
Tk1 [2n + 1] = (Uk [2n + 1] UL/2k
[2n + 1]) (3.31)
2
1
Tk2 [2n + 1] = (Uk [2n + 1] + UL/2k [2n + 1]) (3.32)
2
62
and
X
L/21
pk
2p [n] s2p+1 [n])WL/2 .
(1)p (jsQ Q
Uk [2n + 1] = (3.33)
p=0
Following (3.24), (3.25), (3.26), and (3.29) and also (3.30), (3.31), (3.32) and (3.33),
one arrives at the structure presented in Fig. 3.15. Comparing to Fig. 3.14, the
complexity of the IDFT processing is one half.
Similar methodology can be applied to derive an efficient receiver structure.
Assuming that the transmitted signal can be recovered without any phase rotation
(which might not be correct in frequency selective channels), we have the following
condition at the output of IDFT in Fig. 3.14.
At even discrete time indexes, at the output of IDFT, we need only the real
part of the output of the even subcarriers and only the imaginary part of the
output of odd subcarriers.
At odd discrete time indexes, at the output of IDFT, we need only the
imaginary part of the output of the even subcarriers and only the real part
of the output of odd subcarriers.
0 0
In Fig 3.14, Sm [n] are the outputs of the IDFT. Sm [n] and the desired output
symbols are related as
( I
(1)p S 0 2p [2n] k = 2p
sIk [n] = Q (3.34)
p
(1) S 0 2p+1 [2n] k = 2p + 1
and ( Q
(1)p S 0 2p [2n + 1] k = 2p
sQ
k [n] = I (3.35)
(1)p+1 S 0 2p+1 [2n + 1] k = 2p + 1
I Q 0
where S 0 m [n] and S 0 m [n] are the real and imaginary parts of Sm [n]. At even time
indexes can be written as
X
L1
S 0 2p [2n] = s0m [2n]WL2pm (3.36)
m=0
63
jsQ
0 [n]
2
z 1
sI0 [n] s!0 [n]
2 +
j 2 sIQ [n]
2
z 1
jsI1 [n] s!1 [n] u0 [n] U0 [n]
2 + +
S0! [n]
E0 (z 2 ) L/2
z 1
j N 1 sQ U1 [n]
Subcarrier Separation
L/2 point IFFT
N 2 [n]
2 S1! [n]
E1 (z 2 ) L/2 +
z 1
j N 2 sIN 2 [n] s!N 2 [n] z 1
2 +
j N sQ
N 1 [n]
2
z 1
j N 1 sIN 1 [n] s!N 1 [n] uN/21 [n]
2 + +
UL/21 [n] z 1
0 !
SL1 [n]
EL1 (z 2 ) L/2 +
0
! T !
! ! !
UN/2k [n]/2 2
k [n]
SL/2k [n]
(1)n WLk
pm p(m+L/2)
Using WL/2 = WL/2 , we can write (3.37) as
I X
L/21
pm
2S 0 2p [2n] = (s0 m [2n] + s0 m+L/2 [2n])WL/2
m=0
X
L/21
pm
+ (s0 m [2n] + s0 m+L/2 [2n])WL/2 . (3.38)
m=0
The desired signal on an odd subcarrier at an even time index can be written as
Q
2j S 0 2p+1 [2n] = S 0 2p+1 [2n] S 0 2p+1 [2n]
X
L1
(2p+1)m
X
L1
(2p+1)m
= s0 m [2n]WL s0 m [2n]WL
m=0 m=0
X
L1
pm
X
L1
= WLm s0 m [2n]WL/2 pm
WLm s0 m [2n]WL/2 . (3.40)
m=0 m=0
I pm
Similar to the case of S 0 2p [2n], we can use the periodicity of WL/2 with respect to
m to obtain
Q X
L/21
pm
2j S 0 2p+1 [2n] = WLm (s0 m [2n] + s0 m+L/2 [2n])WL/2
m=0
X
L/21
pm
WLm (s0 m [2n] + s0 m+L/2 [2n])WL/2 . (3.41)
m=0
65
X
L/21
Q
2j S 0 2p+1 [2n] = WLm s0 m [2n] + s0 m+L/2 [2n] (3.42)
m=0
pm
s0 Lm [2n] s0 L/2m [2n] WL/2 . (3.43)
Next we define
1 0
rk1 [2n] = s k [2n] + s0 L/2k [2n] (3.44)
2
+ s0 L/2+k [2n] + s0 Lk [2n]
and
1 0
rk2 [2n] = s k [2n] + s0 L/2k [2n]
2
s0 L/2+k [2n] s0 Lk [2n] . (3.45)
Using (3.44) and (3.45), one can write (3.39) and (3.42) as
I X
L/21
pm
1
S 0 2p [2n] = rm [2n]WL/2 (3.46)
m=0
and
Q X
L/21
pm
j S 0 2p+1 [2n] = WLm rm
2
[2n]WL/2 . (3.47)
m=0
where
Following the same line of thoughts for the odd time indexes, it can be shown
that
Q I
Qk [2n + 1] = S 0 2p [2n + 1] + j S 0 2p+1 [2n + 1]
X
L/21
km
= qm [2n + 1]WL/2 (3.50)
m=0
with
and
1 0
rk1 [2n + 1] = sk [2n + 1] s0 L/2k [2n + 1]
2
+ s0 L/2+k [2n + 1] s0 Lk [2n + 1] (3.52)
and
1 0
rk2 [2n + 1] = s k [2n + 1] s0 L/2k [2n + 1]
2
s0 L/2+k [2n + 1] + s0 Lk [2n + 1] . (3.53)
I
This shows that the desired terms, S 0 Q 0
2p [2n+1] and S 2p+1 [2n+1] can be conveniently
obtained combining rk1 [2n + 1] and rk2 [2n + 1] according to (3.51) and applying an
L/2 point IDFT to the results. The obtained receiver structure is presented in Fig.
3.16.
3.3.4 Equalization
In the case of SMT, if the bandwidth of each subcarrier is sufficiently narrow,
gain of channel on each subcarrier can be approximated by a fixed complex value.
Therefore, once the channel gains are obtained, one may choose to use a single-tap
complex equalizer per subcarrier channel. In that case, the gains of the equalizers
are the inverse of the channel gains at the center frequency of each subcarrier
channel. In order to be able to apply the inverse of the complex channel, we need
67
sQ
1 [n]
2 !{}
z1
sI0 [n]
2 !{}
sQ
0 [n]
2 !{}
s!0 [n] j z1
y[n] q0 [n]
L/2 E0 (z 2 ) Q0 [n] sI1 [n]
2 !{}
z 1
Subcarrier Combination
1
L/2 Point IDFT (j)N 2
z1
z sIN 2 [n]
2 !{}
sQ
N 2 [n]
2 !{}
(j)N 1
z1
QN/21 [n] sIN 1 [n]
z 1 2 !{}
s!L1 [n] qL/21 [n]
L/2 EL1 (z 2 )
s!k [n]/4
s! L/2k [n]/4 ! ! ! rk1 [n]
! qk [n]
(1)n
s! Lk [n]/4 ! !
! rk2 [n] !
n
s! L/2+k [n]/4 (1) WLk
to have access to the complex output of each channel before taking the real value
of the signal. We have access to the complex signal at the polyphase structure of
type-I and type-II. However, in case of the polyphase structure type-III, we do not
have access to the complex output of each subcarrier. This indeed happens because
of our assumption that we do not need the imaginary part of a signal when the real
part is our desired signal and vice versa. Therefore, while the type III transmitter
can be used as an efficient implementation, type III receiver implementation may
not be used in frequency selective channels. Furthermore, same problem exists with
the type III receiver implementation when carrier offset needs to be estimated. We
will have more elaborate description of carrier offset compensation in Chapters 4
and 5.
X
+
x0k [n] = (j)k (j)l (1)lk sk [l]hk [n lL] (3.58)
l=
we have
X 0 (z) is defined as
X
N 1
0
X (z) = Xk0 (z). (3.60)
k=0
and also
X
kn
Hk (z) = h[n]z n W2L . (3.61)
n
X
N 1 X
0
X (z) = j k Sk ((1)k jz L )h[n]z n W2L
kl
. (3.62)
k=0 n
We can use the a 2L polyphase representation for the filter Hk (z) to derive
X
N X
1 2L1
X 0 (z) = j k Sk ((1)k jz L )W2L
kl
El (z 2L )z l (3.63)
k=0 l=0
We have
1/2
X 0 (z) = X(zW2L ). (3.65)
70
1/2
Therefore, in order to derive X(z), we need to replace z by zW2L . X(z) can
be obtained as
X
N X
1 2L1
1/2
X(z) = j k Sk ((1)k z L )W2L
kl
El (z 2L )(zW2L )l . (3.66)
k=0 l=0
Using (3.67), and the properties of multirate systems, one can draw the structure
in Fig. 3.17.
At the receiver, we need to pass the receive signal through the filterbank and
then we need to take the real values of the signal to get transmitted signal from
the VSB signal. After straightforward derivations similar to the derivation of the
transmitter, the complex signal before the real value operation can be written as
1/2
Xk [(1)k z] = (j)k Hk (zW2L )X(z) (3.68)
s0 [n]
E0 (z 2 ) L
j
(1)n j 1/2 1
W2L z
s1 [n]
2L-Point IDFT
E1 (z 2 ) L
1/2 1
W2L z
Nn
(1) j N 1 j N 1
sN 1 [n]
0 1/2 1
W2L z
0 E2L1 (z 2 ) L
where Xk (z) are the signal value before taking the real value in Fig. 3.18. By
replacing Hk (z) by its polyphase elements we get
X
2L1
1/2
k
Xk [(1) z] = (j)k W2L
kl
El (z 2L )(zW2L )l X(z) (3.69)
l=0
Therefore, one can show that Fig. 3.18 presents the polyphase structure for receiver
filterbanks.
3.4.1 Equalization
In the case of CMT, also, if the subcarrier spacing is such that we can use
the flat gain approximation over each subcarrier, one can show that inverse of
channel needs to be applied to the complex value signal before the real value is
obtained. The polyphase CMT implementation described here provides access to
the complex value signal and thus a channel inverse may be applied. More details
on the equalization CMT are investigated in [21] and [41]. Moreover, [21] has
s0 [n]
L E0 (z 2 ) !{}
1/2 1
W2L z
j (1)n
s1 [n]
2L-Point IDFT
L E1 (z 2 ) !{}
1/2 1
W2L z
(j)N 1 (1)N n
sN 1 [n]
!{}
1/2 1
W2L z
L E2L1 (z 2 )
shown that the CMT signal has some features that enable blind adaptation of the
equalizers.
3.4.2 Summary
Polyphase technique for implementation was presented. Polyphase implemen-
tation of synthesis and analysis filterbanks were derived. The general polyphase
structures were used as an basis to come up with a concise derivation of polyphase
structures for SMT. The new formulation was used to show that the third polyphase
implementation is not functional for frequency selective channels. Polyphase struc-
ture for CMT systems was also discussed.
CHAPTER 4
data-aided joint CFO and symbol timing estimator. The training sequence in their
method is similar to the short training in 802.11 OFDM system, and consists of
Nrip identical blocks each of duration of QT where T is duration of one FMT
or SMT symbol. The methodology used in [51] is similar to the synchronization
methods used in preamble of OFDM systems which are described in section 4.1.
This method requires Nrip Q FBMC symbols. Also, more recently and independent
of the work in this dissertation, Fusco et al. [46] have proposed a pilot signal similar
to the long preamble proposed in this dissertation. Fusco et al. [46] use this pilot
signal for timing recovery and carrier phase estimation, based on a cost function
which is different from the one proposed in this dissertation. Simulation results that
compare the accuracy of the timing recovery method of [46] with the one proposed
in this chapter are presented in Section 4.6. Furthermore, the described method
in [51] and [46] does not provide a methodology for AGC and channel estimation.
In this dissertation, we propose a data aided method and a packet format that
address AGC, carrier and timing synchronization and channel estimation [52].
In this chapter, we first present a literature survey on preamble designs used in
OFDM systems. Then we present our proposed preamble design for FBMC and
the corresponding techniques for carrier and timing synchronization. Simulation
results are presented at the end of the chapter.
half a symbol length. To elaborate, in the Schmidl method, if the channel impulse
response is shorter than the cyclic prefix (CP), two identical blocks will be received
at the receiver side when there is no carrier offset. Peaks of the correlation function
among the repeated parts reveal the beginning of packet. The cost function for
packet detection is defined as follows:
X
M/21
y[n + m + M/2]y[n + m]
m=0
M [n] = (4.1)
X
M/21
|y[n + m + M/2]|2
m=0
where y[n] is the received symbol. A packet is detected if |M [n]| passes a given
threshold. One of the main challenges in the Schmidl and Cox method is how
to adjust the threshold so that the packets can be detected reliably while the
probability of misdetection remains at an acceptable level. It has been noted in
the literature that the Schmidl and Cox method suffers from a large plateau on the
peak. This plateau results in inaccuracy in the estimation of timing. As a solution
to this problem, training sequences with sharper metric have been proposed.
After the packet is detected and the initial timing acquisition is performed,
the carrier acquisition must be accomplished. Carrier frequency acquisition is also
normally performed using the preamble structure at the beginning of a packet.
The most popular method is exploiting repetitive structures. These repetitive
structures are phase rotated if carrier offset is present. The induced phase shift
can be used to estimate the carrier offset. Assuming that the y1 [n], and y2 [n] are
the received signals corresponding to the first and the second components of the
repetitive preamble:
y1 [n] = p[n]ej2f n/M + 1 [n] (4.2)
where p[n] is the received signal component of the repetitive preamble; 1 (n) and
2 (n) are the noise components of the received signal. The estimate of carrier offset
can be calculated as:
77
1 X
Np 1
c =
f { y2 [n]y1 [n]} (4.4)
2Np /N n=0
where NP is the length of the preamble. Since returns values in the range of
c | N . Therefore, solutions
[, ), the maximum detectable carrier offset is |f 2Np
have been proposed to increase the lock range of the frequency acquisition method.
The carrier offset can be broken into two parts: the part that is an integer multiple of
the subcarrier spacing, and a part that is a fraction of carrier spacing. Schmidl and
Cox estimate the fractional part using the mentioned repetitive training sequence.
Furthermore, they transmit two PN sequences (P N1 and P N2 ) to calculate the
integer part of the carrier offset. P N1 is transmitted on the even subcarriers, and
P N2 is transmitted on the odd subcarriers. They first detect the fractional part and
compensate it before they process the second part of preamble for estimating the
integer part. Thus when the two PN sequences are processed, only an integer offset
is remaining. As a result, the received PN sequences after DFT are only cyclicly
rotated due to uncompensated integer frequency. The amount of rotation can be
easily calculated by cyclicly rotating the received signal and finding the preamble
that maximizes the correlation with the transmitted PN sequence. The value of
the rotation is the integer part of carrier offset.
The alternative method to extend the lock range of the frequency acquisition
is proposed by Morelli and Mengali [54]. They divide an OFDM symbol to Q > 2
identical parts of N/Q samples. They estimate the CFO
1 X
Q/2
c =
f (q){[q] [q 1]} (4.5)
2/Q q=1
X
o+N 1qN/Q
(q) = y[k + qN/Q]y [k], q = 1, 2, ..., Q/2 (4.6)
k=o
1
We note that since a matched filter h(t) is applied at the input of the receiver, strictly speaking,
the length of the long preamble after filtering at the receiver is at least twice the duration of h(t).
However, since the tails of the response at the beginning and end are small, we found, numerically,
restricting the length of the long preamble to the length h(t) does not incur any significant loss
in performance.
signal field
short training (8 s) long training (8 s) (4 s) data
signal
short training long training field data
symbols; see the design example in [38]. It thus may appear that with the proposed
preamble, FBMC is less bandwidth efficient than OFDM. However, the absence of
guard intervals (cyclic prefix) in FBMC will result in a shorter payload and, thus,
the overall packet length in an FBMC system is expected to be shorter than its
counterpart in OFDM.
where ak are a set of binary numbers with magnitude K, i.e., they take values of
K. We may choose ak s to optimize certain properties of xlong (t), for instance, to
minimize its peak to average power ratio (PAPR). This optimization is of particular
interest as it will allow maximization of signal power during the training phase
which, in turn, improves the accuracy of the carrier frequency and timing phase
estimates. The power spectral density of our proposed preamble is presented in
Fig. 4.3.
80
40
60
80
100
0.5 0 0.5
F
Assuming the channel has an equivalent baseband impulse response c(t), the
long training symbol will be received as
where v(t) is the channel additive noise. Taking the Fourier transform of on both
sides of (4.9) and using (4.8), we obtain
where
N
X
2
1
2k
A(f ) = ak C(f )H f . (4.11)
k=0
T
Squaring both sides of (4.10), we get
Assuming a low noise channel, one may ignore the term |V (f )|2 on the right-hand
side and thus simplify (4.12) to
Assuming that the channel noise v(t) is a complex symmetric white stationary
Gaussian process with an instantaneous variance of v2 , V (f ) also will be a complex-
valued symmetric white stationary Gaussian process with an instantaneous variance
of v2 . Hence, <{A (f )V (f )} will be a real-valued white nonstationary Gaussian
process with an instantaneous variance |A(f )|2 v2 /2 and, accordingly, (4.14) may
be rewritten as
|Ylong (f )|2 = |A(f )|2 + |A(f )|V 0 (f ) (4.14)
or, alternatively,
|Ylong (f )|2
= |A(f f )| + V 0 (f ). (4.16)
|A(f f )|
Since V 0 (f ) is a white noise, a maximum likelihood (ML) estimate of f , say
c , may be obtained by minimizing the following cost function
f
Z !2
c) = |Y long (f )|2
c | df
(f |A(f f
|A(f f c|
Z 2
1 c |2 df.
= |Ylong (f )|2 |A(f f (4.17)
|A(f f c |2
1
If we simply ignore the scaling factor c |2
|A(f f
under the integral (4.17), we
obtain the modified/simplified cost function
Z 2
c
(f ) = c
|Ylong (f )| |A(f f |
2 2
df. (4.18)
1. We ignore the channel effect and simply assume that C(f ) = 1, i.e., an ideal
channel. Noting that the terms H f 2kT
are nonoverlapping, this leads to
1 N
2
X
2
2k
|A(f )| = K
2 H f
2 . (4.20)
T
k=0
2. We note that when channel noise is small, |Ylong (f )|2 resembles the shape of
|A(f f )|2 accurately and, thus, |Ylong (f )|2 may be used to estimate the
magnitude of C(f ) at each of the bands defined by the terms H f 2k T
and,
accordingly, an approximation to |A(f )|2 , may be constructed as
N
1 2
X
2
2k
|A(f )| = K
2 2
|Ck | H f
2 (4.21)
k=0
T
where Ck = Ylong ( 2k
T
) is an estimate of C(f ) at f = 2k
T
.
83
Using the Parsevals theorem for DFT, (4.22) may equivalently be written as
N
1
X
2
where uk [n] are the signal samples at the FFT input in Fig. 4.4. This shows that
the optimum timing phase can be obtained without performing any FFT operation.
In a recent work Fusco et al. [46] have also proposed the use of an isolated
FBMC symbol (similar to the proposed long preamble in this dissertation) for
timing acquisition. They have noted that in the absence of channel distortion, such
a symbol is symmetric with respect to its center and have developed the following
equation for timing acquisition.
84
ylong [n] N
u0 [n] y0 [n]
E0 (z 2 )
z 1
N u1 [n] y1 [n]
ylong [n 1] E1 (z 2 )
z 1 N
2 -point
FFT
.. ..
. .
z 1
N
u N 1 [n] ylong [n]
E N 1 (z 2 ) 2
N 2
ylong [n 2 + 1]
(M 1)/2
X
nopt
= arg max ylong [n i]ylong [n + i] (4.24)
n
i=1
where M is the length of ylong [n]. It is also noted in [46] that the symmetry
property of the isolated FBMC symbol holds approximately in presence of channel
and, thus, argued that the same formula may be used for timing acquisition in
multipath/frequency selective channels.
It is also worth noting that while in the absence of the channels distortion,
both (4.23) and (4.24) provide the optimum timing phase, they only result in a
near optimum timing phase when a channel distortion and/or noise present. We
evaluate the accuracy of the two methods and compare them with each other in
Section 4.6.
4.5 Equalization
Once the preamble is CFO-compensated, and the optimum timing phase is
acquired, assuming a flat gain over each subcarrier channel, the outputs of the
85
signal analyzer of Fig. 4.4 are the training symbols ak s scaled by the channel gains
at the center frequencies 2k
T
, k = 0, 1, , N/2. Moreover, if we assume that these
samples are dense enough, an interpolation may be applied to find the channel gains
at all frequency points where the payload subcarrier channels will be located. Note
that the locations of the center of subcarrier channels depend on the modulation
type, say, being CMT or SMT. Once the channel gains are obtained, one may choose
to use a single-tap complex equalizer per subcarrier channel. In that case, the gains
of the equalizers are the inverse of the channel gains at the center frequency of each
subcarrier channel. It is also possible to use a multitap equalizer per subcarrier.
This has been discussed in detail in [17], for the SMT, where it is argued that
to remove ICI, the equalizers should be fractionally spaced ones. The receiver
structure proposed in [17] is tailored towards implementation of the half-symbol
spaced fractionally spaced equalizers.
In the case of CMT, the equalizers shall be inserted at the points before the <{}
blocks in Fig. 2.2. The efficient CMT implementations proposed in [41] and [21]
provide access to these points and thus equalizers can be easily implemented.
In the case of SMT, if one follows an implementation that mimics the receiver
structure of Fig. 2.1, the equalizers should be inserted at the points right before
where the demodulated signals branch to the <{} and ={} blocks. If that is
the case and decision directed loops are adopted for the equalizers tracking, the
presence of the filters h(t) and h(t + T2 ) within the loops may introduce excessive
delay which may result some undesirable behavior. Fortunately, in the case of SMT
also the efficient polyphase structures that have been proposed in the literature,
say, [17], are such that the <{} and ={} blocks are moved to the output of the
filters h(t) and h(t + T2 ) and thus avoid the problem of loop delay.
where the samples are spaced at the interval T /64, and T , in units of seconds, is
the symbol interval in the case of SMT. We assume a transmission bandwidth of
20 MHz which is divided into N = 64 subcarriers. This results in the subcarrier
spacing (20 MHz)/64 = 312.5 kHz and the symbol interval T = 1/0.3125 = 3.2 s.
Signals are generated at an over-sampled rate of four times faster than their Nyquist
rate, i.e., at a sample interval Ts = T /(4N ) = T /256. This will allow us to adjust
the timing phase with an accuracy of Ts which is four times better than the Nyquist
rate T /N . We also recall that since in the CMT, modulation is VSB, if the same
subcarrier spacing as in the SMT is assumed (because of the reasons mentioned
in [38]), the symbol interval will be T /2.
We use a short preamble similar to that of IEEE 802.11a and g in our packets,
i.e., 10 cycles of a periodic signal with period of 0.8 s. The long preamble is an
isolated SMT symbol in which the even subcarriers are filled up by a set of binary
phase-shift keying (not QAM, OQAM or VSB) symbols, and the odd subcarrier
are filled up with zeros, as in (4.8). The binary symbols ak are selected through
a random search to minimize the peak power of xlong (t). This combined with the
fact ak s are nonzero only at even subcarriers will allow us to reduce the peak
amplitude of xlong (t) to about 9 dB below that of the payload, assuming that the
pilot symbols ak have the same power as the payload symbols sk [n]. We add this
margin of 9 dB to xlong (t) and transmit a high-powered long preamble. Since this
boosts the SNR of the long preamble, it leads to a more accurate carrier estimation
and timing recovery. To allow reproduction of the results presented here by an
interested reader, we note that the samples of xlong (t), at the rate fs = 4N/T , are
generated using the following instructions in MATLAB:
N=64; L=4*N; K=6; alpha=1; gamma=1;
h=sr Nyquist p(K*L,L,alpha,gamma);
a=sign(randn(N/2,1));
xlong=H*a;
where sr Nyquist p(N,M,alpha,gamma) is a square-root Nyquist filter design pro-
gram that has been developed in [37]. The designed filter h[n] has a length of
87
KL + 1 and h[n] ? h[n] has zero crossings at an interval L samples. Also, in the
above MATLAB lines, H is a (KL + 1) (N/2) matrix with the k column of
k k k
hk = [h[0] h[1]ej4 L h[n]ej4 L n h[KL]ej4 L KL ]T , k = 0, 1, , N/2.
(4.26)
Fig. 4.5 presents the mean square error (MSE) of the residual CFO (normalized
to the carrier spacing) after tuning the carrier using the long preamble. The three
methods discussed in Section 4.3 are examined. These methods are: (i) correlation-
based estimation according to (4.19) with the channel included using (4.21); (ii)
correlation-based estimation assuming an ideal channel, i.e., using (4.19) and (4.20);
and (iii) ML-based estimation using the cost function (4.17). For the latter case,
the threshold levels of 10% 25% of the maximum of |A(f )|2 are examined. It is also
assumed that A(f ) is known perfectly. The results presented in Fig. 4.5 have been
averaged over 10,000 randomly generated channels.
From the results presented in Fig. 4.5, the following observations are made.
While at lower SNR values, the correlation-based methods are superior to the ML
estimator, at higher values of SNR the latter performs better. This can be explained
if we recall that the approximation used to derive the ML estimator improves as
SNR increases. In high SNR regime (> 15 dB) all methods result in a relatively
low residual CFO. Hence, in practice, all methods may work satisfactorily and thus
one may choose the one with the lowest complexity. On the other hand, in low
SNR regime (< 15 dB) the correlation-based methods outperform the ML method.
Furthermore, the correlation-based methods have lower computational complexity
than the ML methods; compare the relevant equation in Section 4.3. Noting these,
we conclude that the correlation-based CFO estimation methods are better suited
in any practical FBMC system.
After carrier acquisition, the CFO-compensated long preamble is used for timing
acquisition. In Section 4.4 we developed a formula (equation (4.23)) for timing
acquisition and noted that a different formula (equation (4.24)), applicable to our
packet setup, has been recently proposed by Fusco et al. [46]. To evaluate the
performance of (4.23) and compare it with the results obtained using (4.24), we run
88
0
10
Corr, Ch. included
Corr., Ch. not included
2 ML, Threshold at 10%
10
ML, Threshold at 25%
4
10
f
2
6
10
8
10
10
10
0 5 10 15 20 25 30
SNR, dB
Figure 4.5: Residual CFO of the proposed long preamble-based carrier acquisition
methods. The vertical axis shows the MSE of residual CFO normalized to the
subcarrier spacing of the payload. The horizontal axis indicates the SNR during
the payload part of the packet.
the following experiment. The channel introduced at the beginning of this section
is included and 10,000 SMT packet are examined, each with a randomly selected
channel. No channel noise was added. The short preamble of each packet is used
for course carrier acquisition. The acquired carrier is removed from the preamble
portion and further tuning of carrier is performed using the method discussed in
Section 4.3. Then, (4.23) and (4.24) are used for timing acquisition. Subsequently,
the equalizer coefficients are set using the method presented in Section 4.5. The
payload part of the packet is then processed using the tracking algorithms discussed
in Sections 5.2 and 5.3. As a measure of performance, the mean square error (MSE)
of the recovered symbols compared with the transmitted symbols are evaluated and
averaged across time and all subcarrier symbols. Since there is no channel noise in
this set of simulations, the measured MSE is caused by the residual ISI and ICI.
89
s2
SIR = 10 log10 (4.27)
MSE
where s2 = E[|s[n]|2 ] is the symbol power. The results of this set of tests are
compiled and presented in the form of a histogram in Fig. 4.6. The following
observations are made from the histogram:
For better channels (with smaller multipath effects), Fusco et al. method
performs better. These are cases with SIR of more than 50 dB.
On the other hand, in channels with higher level of distortion, the method
proposed in this dissertation shows superior performance.
Since in practical channels SNR values are often below 30 dB, it is reasonable
to say that both methods have satisfactory performance. Nevertheless, one
3000
Proposed method
Fusco et al method
2500
2000
1500
1000
500
0
20 30 40 50 60 70 80
SIR, dB
Figure 4.6: SIR comparison of (4.23) and (4.24). The histogram are based on
testing over 10000 randomly generated channels.
90
may argue that the method proposed in this dissertation may be preferred
over that of [46], as SIR values in the range of 40 dB or below are more
destructive than those in the range of 50 dB or greater.
4.7 Summary
A packet format for transmission of FBMC signals was proposed. The proposed
packet format follows an structure similar to those of IEEE 802.11a and g, and IEEE
802.11n that are based on OFDM multicarrier signaling. It starts with a short
preamble for AGC adjustment and coarse carrier acquisition. A long preamble
for more accurate tuning of the carrier frequency, timing phase acquisition, and
adjustment of the tap weights of a set of frequency domain equalizer then follows.
Once these synchronization steps are performed, the receiver is ready to detect the
data symbols in the payload part of the packet.
CHAPTER 5
Bolsckei was the first to propose a blind carrier offset estimation method for
SMT [45]. This method is an extension of Gini-Giannakis [57] estimator for single
carrier systems and is applicable to both OFDM and SMT. The correlation function
of the received signal, y[n],
is used to estimate the CFO. It has been shown that cy [n, ] is M periodic with
respect to n if subcarrier weighting is used. The Fourier series of cy [n, ], Cy [k, ]
is then used to derive a blind estimator. It has also been shown that the Fourier
transform of a function of Cy [k, ] with respect to can be used to derive an
estimator for a wider range of carrier offsets. The Bolsckei method needs subcarrier
weighting which is a limiting factor for practical systems. The other issue is
that since the values of subcarrier weights are used in the estimator, the channel
response is needed at the receiver and the channel estimation errors will affect the
performance of Bolsckeis method.
Ciblat and Serpedin have proposed using conjugate correlation which is defined
as
ccy [n, n0 ] = E[y[n]y [n n0 ]] (5.2)
for estimation of carrier offset [47] in SMT systems. This method is an extension of
[58] which is applicable to noncircular modulations. It is shown that the conjugate
correlation of received signal is periodic with period of {0 + Lq }q=0,..,L , where 0 =
(2f mod 1), L is the number of subcarriers, and f is the carrier offset. Cilbat et
al. estimate the frequency of embedded harmonics of the signal by finding the points
that maximize the periodogram of the conjugate correlation function [47]. The
authors have mentioned that since the periodogram has several local maximas, the
coarse estimate might result in outliers. Therefore, they have concluded that their
method is best suited for carrier tracking. The Ciblat method is more complicated
than Bolsckeis method because it requires an FFT based peak search and a steepest
descent fine peak search in addition to the conjugate correlation estimation [59].
93
Fusco et al. have extended a maximum likelihood CFO estimation for OFDM
systems and noncircular modulations [60] to SMT [50]. They have assumed that
there are many subcarriers in the system and thus the received signal is modeled as
a complex Gaussian random variable. They have used the noncircularity of SMT
in a nondispersive channel to derive an ML algorithm for CFO estimation. The
authors have acknowledged that the ML CFO estimator requires computationally
intensive maximization procedure with respect to a continuous parameter and is
not suitable for implementation. They have proposed a bilinear unbiased estimator
for frequency offset. In this method, B[n] is defined as
1 X
L1
B[n] = y[k + nL]y[M k + nL]. (5.3)
L 1 k=1
Then, the v[n] is used in the following equation to calculate the carrier offset:
3 X1
=
f [nv[n]( n)] (5.5)
2L( 2 1) n=1
5.2.1 SMT
In an SMT receiver, the phase and quadrature components of the detected data
symbols, before passing through a decision device (a slicer), are given by
X N X1 Z
I
sk [n] = sIm [l]h( lT )h( nT )
l= m=0
2
cos (m k) + + [n]
T 2
sQ
m [l]h( lT T /2)h( nT )
2
sin (m k) + + [n] d (5.6)
T 2
and
X X
N 1 Z
sQ
k [n] = sIm [l]h( lT )h( + T /2 nT )
l= m=0
2
sin (m k) + + [n] + f T
T 2
+ sQ [l]h( lT + T /2)h( + T /2 nT )
m
2
cos (m k) + + [n] + f T d (5.7)
T 2
95
where [n] is the demodulator carrier phase angle at time nT . Combining (5.6)
and (5.7) and separating the desired and interference terms, we obtain
where k [n] is the interface resulting from ISI and ICI terms. Although, for brevity,
the channel noise is not included in (5.6) and (5.7), one can argue that k [n] may
include the channel noise as well.
Assuming that f is small enough such that f T 1, hence we have
The goal of the carrier tracking loop is to force [n] to zero. We assume a
receiver structure as in Fig. 5.1. We obtain an averaged estimate of the phase error
[n] as !
X
N 1
[n] = sk [n]sk [n] (5.11)
n=0
where sk [n] is the detected data symbol after passing sk [n] through a slicer and (x)
denotes the angle associated with the complex variable x. The loop filter output
is an estimate of the phase error in y[n] arising from the CFO. Fig. 5.2 depicts the
structure of carrier tracking for SMT .
5.2.2 CMT
Following the CMT transmitter structure that was discussed in Chapter 2 and
was presented in Fig. 2.2 and assuming a phase error [n] at the analysis filterbank
96
sk [n]s
Analysis sk [n]s
y[n] Slicer
Filter Bank
Phase
ej Estimator
Loop [n]
Filter
Figure 5.1: A PLL equipped FBMC receiver. The input y[n] is the demodulated
received signal.
2n
ejm( M + 2 )
sIm [n]
Re{} h[n] M
sQm [n]
decision
Im{} h[n + M/2] M
Phase Error
Channel z1 kp Subcarriers
Combination
Phase Compensation
ki z1
Integrator Loop Filter
input, if we switch the <[] blocks and the sampler, one finds that the input to the
<[] block at the kth subcarrier channel is given by
X
M 1 X
+ Z
sC
k [n]
j[n]
=e sm [l]h( lT )h( nT )ej 2T (nT lT ) ej(mk)( T + 2 ) d
m=0 l=
(5.12)
where the superscript C on sC
k [n] is to emphasize that it is complex-valued.
Separating the terms associated with the desired symbol, sk [n], and the inter-
R
ference terms in (5.12) and noting that h2 ( )d = 1, we obtain
k [n] sk [n]e
sC for k = 0, 1, , N 1.
j[n]
+ k [n], (5.13)
where, as in the case of SMT, k [n] is the interference resulting from ISI and ICI
terms as well as channel noise. Also, following the same line of thoughts as in the
case of SMT, one finds that the PLL structure presented in Fig. 5.1 is applicable
to CMT as well, with (5.11) replaced by
!
X
N 1
[n] = sk [n]sC
k [n] (5.14)
n=0
of ISI and ICI, its imaginary part contains a significant level of ISI and ICI. When
the carrier phase is known, the imaginary part of sC
k [n] is simply ignored and thus
has no impact on the decision value sk [n]. However, the relatively large variance
of the imaginary part of sC
k [n] results in a noisy estimate of [n]. Nevertheless, in
systems with the packet format, we have numerically found that since the preamble
98
s0 [n]
Re{ } decision
N
j 2N n
h[n]e
Im{}
Phase Compensation ej( N n+ 2 )
s1 [n]
Re{ } decision
Channel
!
h[n]ej 2N n N
Im{} sin1 {}
sN1 [n] Ns2
ej(N1)( N n+ 2 )
Re{ } decision
h[n]ej 2N n N
Im{}
z 1 kp
ki z 1
Integrator Loop Filter
allows a very good estimate of CFO, to track the residual CFO, in the PLL, one
may use a loop filter with a sufficiently small gain for suppression of the noisy
component of [n].
where sm [n, ] is the detected symbol at the output of the mth subcarrier channel,
at time n, when the timing offset value is and sm [n, ] is obtained after passing
sm [n, ] through a slicer. The optimum timing offset is thus tracked by searching
for a value of that minimize [n, ]. A typical early-late gate method, [62], may
be adopted for this purpose.
0.3
0.2
0.1
[ n ]
0.1
0.2
0 100 200 300 400 500 600 700 800 900 1000
S y mb ol i n d e x , n
0.1
0.05
[ n ]
0.05
0.1
0 100 200 300 400 500 600 700 800 900 1000
S y mb ol i n d e x , n
Figure 5.4: Performance of the PLL for carrier tracking in a CMT receiver. The
top figure shows the phase error, [n], at the loop filter input. The lower figure
shows the phase jitter, [n], of the input signal to the analysis filterbank.
101
0.3
0.2
0.1
[ n ]
0.1
0.2
100 200 300 400 500 600 700 800 900 1000
S y mb ol i n d e x , n
0.1
0.05
[ n ]
0.05
0.1
100 200 300 400 500 600 700 800 900 1000
S y mb ol i n d e x , n
Figure 5.5: Performance of the PLL for carrier tracking in an SMT receiver. The
top figure shows the phase error, [n], at the loop filter input. The lower figure
shows the phase jitter, [n], of the input signal to the analysis filterbank.
102
0.035
Without tracking
With tracking
0.03
0.025
0.02
MSE
0.015
0.01
0.005
0
0 2000 4000 6000 8000 10000
ti m e i n d e x , n
Figure 5.6: Mean square error at the output of an SMT receiver, averaged over
all subcarriers, with and without a timing tracking loop.
Although the results presented here are for the SMT, the same results are obtained
for the CMT. For the results presented in Fig. 5.6, it is assumed that there is a
difference of 10 ppm (part per milion) between the transmitter symbol clock and
its counterpart at the receiver. As seen, without timing tracking, the MSE at the
receiver output increases with time. The timing tracking loop fixes the problem
and results in an MSE that remains constant, at a level slightly above the noise
level. For this simulation, the SNR was set equal to 30 dB. This has an associated
noise level of 0.001.
Fig. 5.7 compares the performance of CMT and SMT when both carrier and
timing tracking loops are active. At SNR values of 15 dB or less both methods
perform virtually the same. However, at higher values of SNR, the CMT degrades.
This difference is believed to be mostly due to the higher phase error/jitter at the
carrier recovery loop filter output in the CMT.
If the residual carrier offset is small, one can decrease Kp and/or KI and have
103
1
10
SMT
CMT
0
MSE 10
1
10
2
10
3
10
0 5 10 15 20 25 30
SNR, dB
Figure 5.7: Comparison of the MSE of CMT and SMT in tracking mode.
5.5 Summary
A preamble is used to perform the initial timing and carrier offset. To resolve
any residual CFO and/or timing offset, tracking algorithms were developed. A
decision directed PLL was designed that forces any built up phase error to zero.
An algorithm was designed to track the best timing phase by minimizing a cost
function. SMT and CMT FBMC communication systems were studied. Through
104
1
10
SMT
CMT1
CMT2
0
10 CMT3
MSE
1
10
2
10
3
10
0 5 10 15 20 25 30
SNR, dB
Figure 5.8: Comparison of the MSE of CMT for three cases: CMT1 (KP = 0.1208,
KI = 0.0068), CMT2 (KP = 0.0837 and KI = 0.001), CMT3 (KP = 0.0193 and
KI = 0.0001), and SMT in tracking mode.
computer simulations it was found that for most parts both systems perform about
the same. Only the carrier tracking look in the CMT found to be more jittery than
its counterpart in the SMT.
CHAPTER 6
power users when users with high and low power are present [12], [74], [26].
Spectrum sensing is performed on each node before transmission and the results
are passed to a basestation which combines all the sensing information to compile a
channel state information (CSI). CSI is also used by the basestation for channel
allocation. Control channels are used for exchanging sensing information and
control messages such as channel assignment for the leaf nodes.
Design decisions such as dividing the tasks between FPGA and DSP, and
choosing the appropriate methods to implement each block are made in order to
optimize the usage of the resources on hardware. First the modem is simulated
in Matlab and Simulink. Then we use System Generator for DSP to implement
the FPGA blocks. Simulink and Real Time Workshop (RTW) are used to make
individual modules for the DSP. TI Code Composer Studio compiles the DSP
subsystems and combines different DSP modules.
The modeling and simulation environment that supports this work is based on
Discrete Event System Specification (DEVS) [75]. DEVS is a formalism derived
from generic dynamic systems theory. It has well-defined concepts of coupling
of components, hierarchical, modular model construction, and an object-oriented
substrate supporting repository reuse. DEVS has been implemented and used
as a practical simulation tool in a variety of implementations. The DEVSJava
environment is used in this work along with Progressive Simulation Based Design
and development (PSBD) technique [76] to simplify the development process.
the system should be decomposed. A single real modem itself comprises many
individual modules each a candidate for co-simulation along with other simulated
objects and the environment. Fig. 6.1 shows the PSBD approach of designing a
single transceiver including packet generation, coding, decoding, channel sensing,
channel model and emulated PUs.
As we proceed with PSBD, we start using one SDR board. Most of the func-
tionalities of cognitive radio depend on sensing the medium. Channel sensing is
implemented in the earliest stage as shown in Fig. 6.1. To test sensing module we
emulate PU traffic on different frequency bands using a signal generator, while ad-
justing some design parameters such as analogue to digital converter (ADC) gains,
frequency axis margins, and power threshold of PU detection. The transmitted
traffic of PU by the signal generator is a multi-band waveform which is generated
using a Matlab script and uploaded to the device with Agilent Waveform Download
Assistant via Ethernet connection with simulator on a PC. All simulated objects are
running on the same PC, while the SDR board (running implemented components)
is connected to that PC via an Ethernet cable.
SU
Wireless Channel
Simulated
PU1
Simulated
PU2
Simulated
PU3 Emulator (Vector Signal Generator)
Figure 6.1: Progressive simulation based design (PSBD) of a single cognitive modem. The implementation starts from the
sensing component and progressively more of the simulated models (left dotted box) are implemented (right dotted box).
The rectangles are DEVS models simulated, and parallelograms are implemented components on SDR
109
110
CSI vector with the time that an activity is last seen on each carrier to specify
the availability of each channel. The basestation tries to allocate a carrier which is
less used and has neighboring carriers that are less used as well. When assigning
channels, basestation weighs each carrier by the average usage in 3 neighboring
channels. In this way idle areas of the spectrum are chosen more frequently.
6.2.2 Transmitter
The software defined modem provides two modes of operation to process two
different types of services. One service is a 19.2 kbps computer-to-computer data
stream while the other service is a 16 kbps Continuously Variable Slope Delta
Modulation (CVSD) vocoded voice. The data stream is encoded using a rate 1/2
convolutional encoder. 8 Phase-Shift Key (8PSK) symbol mapping is used for data.
For the voice stream, we use a Reed-Solomon encoder and Quaternary Phase-Shift
Key (QPSK) signalling.
To use only one digital upconverter from baseband to Intermediate Frequency
(IF) for both services, the packet assembly is done such that the symbol rate at
the input of the pulse shaping filter is 20 kbps for both data and voice streams.
The transmitted packet consists of two major parts: a 192-sample cyclic preamble,
generated using three identical Binary Phase Shift Key (BPSK) modulated pseudo
noise (PN) sequences of length 64, and a payload constructed using the data or
voice streams output of the symbol mappers.
Upconversion is done using Cascaded-Integrator-Comb (CIC) filters and a pulse-
shaping filter (PSF) whose coefficients are chosen to achieve the Nyquist-M prop-
erty.
6.2.3 Receiver
The received signal is first down-converted to baseband. The baseband signal
is then passed to the synchronization and channel equalization modules, both of
which are implemented in the fractional space. The fractional spacing between the
samples is chosen to be Ts /2, where Ts is the symbol interval.
Synchronization is performed using a cyclic preamble. Cyclic preamble is chosen
111
in our model because it can serve the dual purpose of estimating timing and carrier
offsets while at the same time equalizing the channel effects when coupled with a
cyclic equalizer. [78]. The repetition structure of the cyclic preamble allows us to
detect the start of the packet as well as the carrier offset. This method exhibits good
performance and is easy to implement. Packet detection is performed by computing
the autocorrelation of the received signal. We correlate the signal with a shifted
version of itself and find the position of the preamble by identifying the interval
over which the autocorrelation is significantly large [34]. After compensating for
the carrier offset, we make use of a half symbol spaced fractionally-spaced adaptive
equalizer to compensate for the channel distortion, any residual carrier offset and
to obtain the correct timing phase [34]. The equalizer coefficients obtained using
this algorithm are further fine-tuned using a decision-directed adaptive scheme.
availability of resources, the inherent characteristics of these cores, and the extra
functionalities offered by TI and Xilinx. We make use of the already available Xilinx
Logicore Blocksets for FPGA and the optimized DSP libraries written for vectors
of complex numbers for C64x+ core.
Interfacing between the DSP and FPGA is done using the Video Processing Sub-
system (VPSS) data port. The DSP VPSS is a DM6446 DSP 16-bit synchronous
video transfer port modified to support transfer of nonvideo data to and from the
DSP. The VPSS consists of a Video Processing Front End (VPFE) and a Video
Processing Back End (VPBE). The VPFE is used as an input interface to the DSP
and the VPBE as an output interface from the DSP to FPGA. In the FPGA, a
VPSS data port module, also consisting of a VPBE and a VPFE, is implemented
to interface with the DSP VPSS. The data bus inside the FPGA is a 32-bit and
the VPSS of DM6446 DSP bus is a 16-bit. On the other hand, custom registers,
a shared memory of eight 32-bit words between the DSP and the FPGA On-Chip
Peripheral Bus (OPB), are used as configuration registers. As a result, the fast
VPSS 32-bit bus is our gateway between DSP and FPGA.
The tasks developed for the FPGA are implemented using the System Generator
for DSP. System Generator, an add-on to Simulink provided by Xilinx, produces
a highly optimized FPGA realization, since each module used in the architecture
maps to an FPGA library component that has been carefully constructed and
optimized for the FPGA target device. Moreover, the System Generator provides
us with a visual representation of the system that not only serves as the design
specification, but as the behavioral simulation model and the source definition for
the hardware. The system Generator implementation also facilitates the rapid
investigation of various design options in the system [79].
To develop the DSP subsystem, the algorithms targeting the DSP processor are
first implemented in Simulink blocks. RTW is next used to produce the first version
of the code for the individual Simulink blocks. Each block is then individually tested
in Simulink external simulation. Although RTW is able to generate stand-alone
C code for the Simulink blocks, it can only be used for rapid prototyping and
113
testing since the code it generates is not optimized for a specific DSP or GPP
target. The RTW generated code often needs extra memory and processing power
and the optimization burden is put on the compiler only. The RTW thus cannot
be used to implement the complete DSP subsystem whose requirements include
realtime performance in terms of memory and speed and special data alignment.
To overcome this problem, a Target Language Compiler (TLC) file is developed to
customize the code generation. In writing the TLC, it is feasible to use optimized TI
DSP libraries DSPLiband compiler optimization techniques such as giving feedback
to the compiler. Furthermore, the wrapper TLC (unlike inline TLC) saves a single
version of each algorithm and therefore simplifies code maintenance. The wrapper
TLC code (written for the individual blocks) can be reused in the independent
compilation of the complete DSP subsystem project, without involving the RTW.
Finally, the C code of the complete DSP subsystem (either generated by RTW or
written as wrapper TLC) is compiled by TI Code Composer Studio (CCS). CCS
makes use of the high performance VelociTI architecture of DM6446 to optimize
the code down to the programming level optimization.
1c
2008 IEEE. Quoted, with permission, from P. Amini, E. Azarnasab, P. Amini, S. Akoum
and B. Farhang-Boroujency, An experimental cognitive radio for first responders, Proceeding
of IEEE DySpan, IEEE Dynamic Spectrum Access Conference, October 14-17, 2008, Chicago.
Transmitter
DSP (c64x+) FPGA (Vitex IV, XC4VX35)
Rf = 0
RS Enoder
Binary Source
Conv Encoder
Rf = 1
Symbol Mapping Rd = 0
Rf = 2 To RF Frontend
Framing M1 CPSCC M2 CIC Modulation
V P F E V P BE
Receiver
114
Rd = 0 From RF Frontend
Packet Detection Carrier Recovery M1 CPSCC M2 CIC Demodulation
Framing M1 CPSCC M2 CIC Modulation
V P F E V P BE
Receiver
DSP (c64x+) FPGA (Vitex IV, XC4VX35)
Rd = 0 From RF Frontend
Packet Detection Carrier Recovery M1 CPSCC M2 CIC Demodulation
Rf = 2
RS Decoding
FS Equalizer Demapping
Viterbi
Rf = 3
Rd = 1
Binary Data
Rd = 2 From RF Frontend
Filterbank Sensing AGC M3 Lowpass Filter Demodulation
Sensing Timer Rd = 2
V PFE V P BE
an endless loop, uses the content of the Rd register to select the appropriate DSP
function to be applied on the incoming data. The transmission is initiated, as shown
in Fig. 6.2, in the DSP core. The binary source is an arbitrary bitstream of voice
or data incoming to the DSP. The voice data are input from the pcm3008 stereo
audio codec at 48 KHz and encoded by the CVSD having a data rate of 16kbps.
The binary voice stream is zero padded to achieve a data rate of 20kbps. The data
traffic, on the other hand, is a fixed computer-to-computer data stream. This data
stream has a lower bit error rate than the voice stream and is retransmitted in the
MAC layer (if missed or corrupted) to ensure data integrity of critical information.
The transmit packet is sent to the FPGA through the VPSS. The custom register
Rf is set to zero if the source is audio and one if nonaudio. In the FPGA, depending
on the data content, either Reed Solomon (RS) coding or convolutional coding
combined with interleaving is performed. Following the source coding in the FPGA,
the binary vectors are retransferred to the DSP (Rd = 0) to perform modulation
and framing. Data stream transmitted using 8PSK and QPSK is implemented for
transmitting voice. The binary vector is finally sent back to the FPGA (Rf = 2)
to upconvert the signal to IF and eventually transmit it over the air. Digital up
conversion in the FPGA consists of three blocks. Combined CIC and pulse shaper
(CPSCIC) that follow the Nyquist-M criterion, a CIC integrator, and a Direct
Digital Synthesizer (DDS) to modulate the signal to the IF. The CPSCIC filter we
used is an 80-tap FIR filter generated using the Xilinx FIRCOMPILER provided
by the System Generator for DSP. The baseband signal was modulated to an IF of
30 MHz at a sampling rate of 80 MHz. Note that in the SFF SDR platform, the
FPGA has access to IO, Data Conversion and RF modules.
At the receiver side (Fig. 6.3), two separate functionalities are first performed in
the FPGA, digital down conversion (DDC) and sensing. The resulting signal is then
passed to the DSP (Rd = 1) where synchronization tasks and symbol demapping
are performed. After demapping, the signal is sent back to the FPGA for decoding
(Rf = 2 for voice and Rf = 3 for data). The decoded signal is finally passed to
the DSP (Rd = 2). This timer activates the sensing module 10 times a second for
117
almost 4s. The timer control circuit disables the transmitter functionality while
the sensing is performed. The sensing data are first demodulated to baseband
by means of a DDS whose frequency is centered at the IF. A lowpass polyphase
decimator is used to filter the required signal band and bring the sampling rate
down to 5 MHz. Note that in order to make use of the maximum dynamic range
of the ADC, AGC is developed in the DSP to control the gain of the analogue
amplifiers available on the data conversion module before the signal is digitized.
The resulting signal is then sent to the DSP (Rd = 3) for further processing. In
the DSP, the DSPLib library for C64x+ is used for efficient implementation of the
filterbank sensing. A filterbank is implemented in polyphase structure using 256
8-tap polyphase elements which are the decimated coefficients of a prolate filter.
The output of the polyphase elements are then passed through FFT. The output
energy of the filterbanks is then averaged over three decimated samples. The sensing
information is compared with a tunable threshold to locate possible active primary
users and create a 32-byte channel state information. This information is then
transmitted to the basestation. The basestation compiles the sensing information
from all of the users and creates a common CSI to be used for channel assignment.
Real SU1
Wireless Channel
Real SU2
Si mulated
SU1
Si mulated
SU2
Si mulated
PU1
Figure 6.4: The testbed setup for examining the performance of filterbank sensing.
samples and the prototype filter length are 2048. The results of FFT, FFT with
a Hanning window and filterbank are averaged over three decimated samples. The
calculated PSD from 462MHz to 467MHz for these three methods are depicted
in Fig. 6.5. As we can see in Fig. 6.5, filterbank sensing is able to show all the
transmitted sinewaves clearly. FFT, on the other hand, has a considerable spectrum
leakage which results in missing the three sinewaves at 463.367MHz, 463.545, and
463.985MHz. FFT with hanning while having better dynamic range than FFT,
also misses two of the sinewaves.
30
40
50
Magnitude (dB)
60
70
80
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Normalized Frequency
Figure 6.5: Power spectral density (PSD) measurements by FFT, FFT with hanning window, and filterbank
119
120
available in the band. The testbed was first presented in the 2007 Software Defined
Radio and Technical Exhibition as part of Smart Radio Challenge [64]. The radio
was demonstrated and tested for two days in the exhibitions in the presence of
interfering cognitive and software defined radios from other companies, universities
and research institutions. It was shown that an audio signal can be transmitted
from one cognitive radio to other one while interferes are present in the band
and the radios can move to an unused part of the band when interference from
primary or secondary devices become available in their band. The testbed was
also presented to a panel of judges from major companies in field of cognitive
and software defined radio and the paper describing the implementation [26] was
awarded the best paper award [80]. Furthermore, as a result of successful demon-
stration at the SDR Technical conference, we were invited to the present our testbed
in the Demonstration Track of the 2008 IEEE conference on Dynamic Spectrum
Access Network (DySpan) [68]. We were able to show in the three days of our
demonstration at DySpan that our testbed can coexist with cognitive radios from
various companies and universities.
6.7 Summary
A cognitive radio network for first responders was developed. The system
performs channel sensing to identify the presence of PU and transmits over unused
portions of the spectrum. It was shown that filter spectrum sensing provides a
reliable mechanism for detecting the presence of primary users. The system was
presented at two demonstration environments and it was shown that the cognitive
radio can keep transmitting voice in the presence of various types of interferers.
CHAPTER 7
Aging wiring is one of the forefront issues in maintaining older aircrafts. Inci-
dents such as Swiss Air Flight 111 and TWA Flight 800 have led to a heightened
public awareness of these critical wiring issues [81].
Conventional wiring inspection is typically limited to visual inspection during
routine maintenance or highly invasive testing during major modifications. But
visual inspection does not provide enough information to detect flaws, especially
within wire bundles. Therefore, methods to locate small faults before they create
system level problems are desired. Some faults may cause impedance discontinuities
that are too small to detect. On the other hand, there are methods that can locate
the fault when its impedance discontinuity is larger, such as when the wire vibrates
against a metal structure creating an intermittent short circuit have more promise
of locating these anomalies. This requires being able to test the wires continuously
while they are live and in flight, which requires a new class of reflectometry methods
[1], [82-85].
Recently, a new method of live wire testing called multicarrier reflectometry
(MCR) has been proposed [1]. Reflectometry is essentially a system identification
problem. A test signal is sent across the wire under test, and the reflected signal
is used to study the condition of the wire [1], [82-85] . The wire may be treated
as a plant whose input is the test signal, and its output is the reflected signal.
Identification of the plant, or equivalently, the condition of the wire, is the goal.
1 c
2009 IEEE. Quoted, with permission, from P. Amini, C. Furse, and B. Farhang-Boroujency,
Filterbank multicarrier reflectometry for cognitive live wire testing, IEEE Sensor Journal, vol.
9, no. 12, December 2009.
122
The plant response can be obtained by comparing its input and output signals.
In time domain reflectometry (TDR) [84], for example, the input is chosen to be
a step function or impulse, and accordingly the plant output/reflected signal is
the step or impulse response of the plant/wire. In MCR, the test signal is chosen
such that the plant response can be measured in the frequency domain. For this
purpose, the test signal is chosen to be the summation of a number of tones, and the
plant/wire frequency response is measured by comparing the amplitude and phase of
the transmitted tones with their reflected counterparts, [86]. This method of system
identification was applied to wire testing in [1]. An MCR test equipment selects one
or more portions of the spectrum that is/are free of live wire signals for transmission
of the tones/test signal. This leads to frequency agility and adaptability ideal for
live wire testing.
The work presented in [1], although pioneering in the sense that it for the
first time introduces the concept of MCR, has not given any due attention to the
optimality of the test signals and the filters that are used for signal analysis. We
concentrate on these aspects of MCR and its goal is to refine the MCR method
that was proposed in [1]. We intend to minimize the signal to interference and
noise ratio (SINR) for both the live signal and the test signal used for diagnosis.
This chapter is organized as follows. In Section 7.1, we describe the problems
of out-of-band interference in signal synthesis (i.e., interference with the live wire
signals) and in-band interference while performing signal analysis (i.e., the interfer-
ence that tester receives from live wire signals). The proposed filterbank solution
and the filter design methodologies are presented in Section 7.2. The concept of
cognitive live wire testing is described in Section 7.3. The concluding remarks are
drawn in Section 7.3.1.
i.e., a rectangular window with duration of T . Moreover, it was suggested that the
phase angles i should be chosen to minimize PAPR of x(t).
Defining
X
v(t) = ai sin (i t + i ) (7.3)
iS
X(f ) = V (f ) W (f ) (7.4)
10
!10
|U(f)|2
!20
!30
!40
!50
0 0.1 0.2 0.3 0.4 0.5
normalized frequency8 f
FFT processor. Here again, the application of the rectangular window is equivalent
to applying a set of narrowband filters whose frequency responses are sinc pulses.
Such responses suffer from poor stopband rejection which in turn means the live
wire signal energy is picked up by the analysis filters, reducing the accuracy of the
measurements. We refer to this type of interference as the in-band interference.
The problems associated with the rectangular window are well known to the
signal processing community, and measures to alleviate them have been studied in
different contexts. In spectrum analysis the problem of spectral leakage is solved
by replacing the rectangular window with other window types, e.g., a Hamming
or Hanning window, [87]. For digital subscriber lines (DSL) some of the problems
are very similar to those encountered in live wire testing. There, a communication
signal is transmitted over an unshielded telephone line which is exposed to radio
signals over some portions of the frequency band. Therefore, similar to our case
here, portions of the band should not be used for communication, [88]. Moreover,
the synthesized signals should be designed to avoid these radio bands. In the
125
analysis part of the system (the receiver), proper filtering operation should be
performed in order to avoid any significant portion of the radio signal - indeed, a
very similar problem to the live wire testing. In DSL, the problem of leakage and
thus filtering is solved by replacing the rectangular window with a window that
tapers at the sides, [89]. This avoids the sharp edges that are the main source of
spectral leakage/poor filtering.
We use the windowing method proposed in [89] to reduce the out-of-band
interference. However, for signal analysis, we resort to a direct filterbank solution.
For live wire signals, the filterbank analysis method is a good choice since we can
design (as shown in Section 7.2.2) optimal filterbanks that minimize the in-band
interference and assure near-perfect separation of the tones in a test signal. This
application of filterbanks is unique to live wire testing, because the test signals are
a sum of pure tones. We also study some details of the windowing solution of [89]
and develop new insights that lead to a method of improving the window function
that has been commonly used in the past. We also show that the proposed signal
synthesis with windowing may be formulated in the framework of filterbanks.
interference
system
identification
....
analysis
filterbank
for more details), one can estimate the samples of the frequency response of the
system (the wire). The measured frequency response is then used to study the
time-domain characteristics of the wire. These operations are performed in the
system identification block. We request the reader to refer to [1] for the details of
how this conversion can be performed. In the following, we present the details of
the signal generator and ABF blocks of Fig. 7.2 and discuss the criteria that should
be used to design these blocks optimally.
2. Its spectral content should be confined within the selected band. In other
words, its out-of-band interference should be minimized.
The first property is required as any realistic test has to be completed within a
finite time. This finite time can be easily set by truncating x(t) (or, equivalently,
the tones) outside of a preselected interval.
Once the tones are truncated, their spectrum will spread across the frequency
axis, e.g., see (7.4), and as a result the spectral spread over the frequencies that
are not part of the intended band. We minimize the out-of-band interference by
first adding the tones and then applying a well-designed window function to the
summation.
The PAPR is minimized through a three-step signal synthesis. We first choose
a discrete-time periodic signal [n] with a small PAPR and with a period of, say, N
samples. Such a periodic signal has a Fourier series with N harmonics at frequencies
2k
N
, for k = 0, 1, , N 1. It is also desirable to choose [n] such that its harmonics
have equal power. This results in equal excitation at all frequencies where we wish
to estimate the frequency response of the system. Equal excitation is desirable and
commonly used in system identification, as it results in equal accuracy at all the test
points [90]. Fortunately, sequences with the properties of small PAPR and equal
power harmonics exist. One class of such sequences are the so called polyphase
codes devised by Chu [91]. They are defined as
(
Kn2
ei N , N even
[n] = Kn(n+1) (7.5)
i
e N , N odd
where K is a number relatively prime to N . Note that [n], for any value of n
has the instantaneous power of unity. Hence, its average power is also unity, and
accordingly it has a PAPR of one or, equivalently, 0 dB. This is indeed the minimum
PAPR that any signal may have.
In the second step, we find the Fourier series coefficients of [n] by taking the
DFT of one cycle of it. The result which we call [k] is used to generate the
continuous-time version of [n] as
X
N 1
2k
(t) = [k]ej N
t
. (7.6)
k=0
128
where fc is the carrier frequency (the center of the band) and <{} denotes the real
part of.
We note that after interpolation, the PAPR is no longer at the minimum level
0 dB. The interpolated signal always has a PAPR which is greater than 0 dB.
However, we argue that the use of polyphase codes, which start with a minimum
PAPR, leads to a continuous-time test signal with a small PAPR. We also note
that different polyphase codes result in different PAPR values and thus examine
all choices of [n] for each code length, N , to find the one that results in minimum
PAPR. The test signal generated in this way for all values of N (we examined
values of N in the range of 10 to 100) have PAPR values of around 2.5 dB. This
is relatively small PAPR. To compare, the phase randomization method that used
in [1] results in signals with PAPRs that are 1 or 2 dB greater.
Recalling that the test signal should have a finite duration, in the third step of
test signal generator, a time window should be applied to v(t), as in (7.1). The
common window suggested in the literature is the raised-cosine function defined as
1 cos( Tt0 )
, 0 t T0
1, 2 T0 t T
w(t) = (7.8)
1 + cos( (tT )
)
T0
, T t T + T0
2
0, otherwise
where T0 is the width of role-off time at each side of the window, and T + T0 is
the total duration of the window. Although the raised-cosine window is commonly
used in the practice of signal processing, as discussed below, it is not necessarily
the best window function. Next, we present an analysis of w(t) that leads to some
insight that we use to propose an improved window function.
129
cos(f T0 )
W (f ) = T sinc(f T ) ejf (T +T0 ) . (7.9)
1 4f 2 T02
where
t 1, |t| T /2
= (7.11)
T 0, otherwise
and
t t T0 /2
h(t) = sin . (7.12)
2T0 T0 T0
Clearly, the first term on the right-hand side of (7.9) is the Fourier transform of
sin(x)
(t/T ) and we have defined sinc(x) = x
. The second term is obtained by
taking the Fourier transform of h(t + T0 /2). The last term (a linear phase term)
arises because the two terms on the right-hand side of (7.10) are the time shifted
versions of (t/T ) and h(t + T0 /2), respectively.
It is instructive to note that the Fourier transform of h(t) appears as a multi-
plicative factor to the sinc pulse on the right-hand side of (7.9) and thus has the role
of attenuating the side-lobes of sinc(f T ). A good choice of h(t) can significantly
improve the out-of-band interference of the spectral content of x(t).
To develop an in-depth understanding of the impact of h(t) on the magnitude
response of W (f ), in Fig. 7.3 we have presented |W (f )|, |T sinc(f T )| and |H(f )|.
The relationship between the parameters T and T0 and some features of these plots
are also presented. In particular, we may note that the width of the main lobe of
|H(f )| is equal to 1.5/T0 and, thus, decreases as T0 increases. On the other hand,
narrowing the latter width is desirable, as it results on attenuating more of the side
lobes of |T sinc(f T )|, thus, improving the out-of-band interference of the spectrum
of x(t).
130
20
T | si n c(f T )|
10 | H (f )|
| W (f )|
M agn i tu d e re sp on se s, d B
!10
2/T
!20
!30
1.5/T0
!40
!50
!60
!1 !0.5 0 0.5 1
Fre q u e n c y, f
Figure 7.3: An example of the magnitude response of the window function w(t)
and its factors.
10
rai se d - c osi n e
M agn i tu d e re sp on se s, d B 0 p rolate
!10
!20
!30
!40
!50
!60
!1 !0.5 0 0.5 1
Fre q u e n c y, f
In contrary, the choice of h(t) in the raised-cosine window, for a given time span
T0 , results in fixed response in the frequency domain. In particular, the width of
its main lobe is fixed at 1.5/T0 . Fig. 7.4 compares the magnitude responses of h(t)
when it is selected to be a prolate function and when it is selected according to
(7.12), both with the same time duration T0 . However, the prolate design that we
have selected has a slightly wider main lobe, in order to achieve higher attenuation
of its side lobes. Here, at the cost of slightly wider main lobe, we achieve 10 to
20 dB improvement in suppression of the side lobes.
To conclude our discussion in this section, we present in Fig. 7.5 a repeat of the
spectrum that we previously presented in Fig. 7.1 after applying a window function
based on the proposed prolate design. As seen, and expected, the out-of-band
spectra are significantly attenuated from 20 dB to 40 dB to below 100 dB. As
a result the interference from test signal to the live signal on the wire is orders of
132
|Xtest(f)|2 !50
!100
!150
0 0.1 0.2 0.3 0.4 0.5
normalized frequency, f
Figure 7.5: An example of the test signal after applying w(t). The window
function w(t) is based on a prolate design.
magnitude smaller than noise level. Therefore, the effect of test signal on the live
signal SINR is negligible.
This result clearly shows that the ith tone in x(t) is obtained by first applying
a rectangular window to the continuous-time tone eji t and then passing the result
through the modulated filter h(t)eji t , centered at i . Hence, the time domain
windowing has an equivalent filterbank synthesis where each tone passes through a
filter in the filterbank that is centered at its respective frequency.
2. Set the gain of G(f ) at f = 0 (this is at the middle of its passband) equal to
one.
3. The filter frequency response should have zero crossings at integer multiples
of the frequency spacing of the tones to avoid interference among different
tones.
Es = hT h (7.16)
2. To set the gain of G(f ) at f = 0 equal to one, the following constraint should
be imposed
cT
0h = 1 (7.18)
3. The constraints that impose zero gain at integer multiples of the frequency
spacing between the tones are obtained as
cT
i h = 1, for i = 1, 2, , N 1, (7.19)
The above results suggest that to design the prototype filter g[n], one should
minimize Es subject to the constraints (7.18) and (7.19). Hence, using the method
of Lagrange multipliers, we define
X
N 1
0 h 1) +
= hT h + 0 (cT i cT
i h (7.20)
i=1
!10
!20
magnitude, dB
!(0
!40
!50
!60
!%0
!80
!0.5 0 0.5
normalized frequency, f
8N + 1 = 129. By increasing the filter length (L), one can arbitrarily control
the stopband attenuation of this filter. Therefore, using the filterbank analysis
structure, the interference from the live signal on the wire to the test signal is not
a limiting factor in decreasing the effective SINR in our diagnosis system.
begin a communication session over the portion of the spectrum that is not used by
other radios. Clearly, extension of this concept to live wire testing is straightforward
and does not need any further elaboration. We only need to equip the live wire
tester with spectrum sensor, i.e., an spectrum analyzer.
Many spectrum analyzers essentially use a filterbank to extract the spectral
energy of the analyzed signal at different portions of the frequency band. Moreover,
recent studies have identified filterbanks as the natural choice for spectrum sensing
in cognitive radios, [94]. We also propose using filterbank method for spectrum
sensing in live wire testing, and note that since in our system setup an analysis
filterbank is connected to the wire this addition comes at virtually no additional
cost.
7.3.1 Summary
Multicarrier reflectometry (MCR) has recently been proposed as an effective
method of testing live wires [1]. This chapter presented a more detailed study of
the signal processing tools necessary in the implementation of MCR and discussed
how these tools may be perfected. We noted that to minimize the interference
between the test equipment and live wire signals, filterbanks are the best signal
processing tools. We thus developed novel methods for this application using
filterbanks. We showed that using a filterbank based on the prolate window design,
the interference level from the test signal to the live wire signal can be controlled
and decreased significantly. Hence the effective SINR of the live signal will be
minimally affected by the test signal. Furthermore, it was shown that the optimum
filters can be designed for analysis filterbanks. The stop band attenuation of the
analysis filterbanks can also be made arbitrary small by increasing the length of the
filters. We also borrowed an idea from the field of cognitive radios and suggested
the concept of cognitive live wire testing.
CHAPTER 8
There is high demand for wireless services and the vast majority of the spectrum
has already been licensed. It appears that the regulated radio spectrum has been
fully occupied and new applications will not have access to the radio spectrum.
It has been noted that the static frequency allocations have resulted in inefficient
usage of the spectrum resources. On the other hand, the advancement of SDRs has
enabled us to build radios that can change their pulseshape, modulation scheme and
carrier frequency. Consequently, cognitive radio technology has been proposed that
uses the SDR technology to make secondary use of licensed spectrum. Multicarrier
communication technology has been suggested as a suitable candidate to utilize the
white spaces in the spectrum. The subcarriers that appear to be unattended are
added to the spectrum pool while busy parts of the spectrum are not aggregated to
avoid collision with PUs. OFDM/FFT was the first multicarrier technique which
was proposed for data communication and spectrum sensing in cognitive radios.
However, OFDM suffers from large side-lobes in the frequency response of the filters
that characterize the subcarriers in an OFDM system. Therefore, there is significant
interference among the carriers of different SUs as well as between SUs and PUs. On
the other hand, filterbank multicarrier can overcome the spectral leakage problems
of OFDM at the transmitter side and therefore lead to less interference from SUs to
PUs and other SUs. Filterbank receiver is also capable of providing high out-of-band
attenuation. Moreover, filterbanks can provide us with a robust spectral analysis
tool with high spectral dynamic range. Therefore, filterbank multicarrier has been
suggested as an alternative to OFDM for cognitive radio applications. In this
138
aforementioned tasks. The algorithms for carrier frequency and timing recovery
as well as channel identification/equalizer adjustment and methods for carrier and
timing tracking loops were proposed. The proposed ideas were evaluated and their
satisfactory performance was presented through computer simulations.
In order to have complete packet based design which is suitable for implemen-
tation in data networks, we designed timing and carrier tracking algorithms. While
our proposed preamble may be used for initial timing and carrier offset estimation,
without any tracking loop, the carrier and timing phase may drift over the length
of the payload. Furthermore, carrier and timing offset tracking are more necessary
for data communication system with large payload sizes such as IEEE 802.11n.
We proposed decision directed timing and carrier tracking mechanisms for SMT
and CMT. A decision directed PLL was used to force any built up phase error to
zero. Timing was tracked by minimizing a cost function. Simulation results were
presented to show the satisfactory performance of the proposed methods.
As a case study for implementation of a filterbank based cognitive radio network,
cognitive radios equipped with filterbank spectrum sensing were implemented. This
work was done to evaluate the performance of our proposed filterbank sensing in
an actual implementation. The cognitive radio was implemented on the Small
Form Factor (SFF) Software Defined Radio (SDR) platform, provided by Lyrtech
and Texas Instruments (TI). Our filterbanks spectrum sensing was shown to exhibit
superior performance in terms of the spectral dynamic range when compared to the
conventional FFT based techniques, i.e., periodogram method. Different processing
tasks were divided between a TI c64x+ DSP and a Xilinx Virtex IV FPGA while an
ARM9 core was used to host Greenhills operating system. To tackle the complexity
of development, we used progressive simulation based design. To develop the
network of the cognitive radio modems, we deployed the available real modems with
simulated primary and secondary users in a central simulation. Three real nodes
were used in our testbed which perform channel sensing and data transmission.
If the presence of a primary user is detected on a carrier that is being used, the
cognitive radio moves to an unoccupied part of the spectrum. Our implementation
140
was tested at a major exhibition and a cognitive radio demonstration track and it
was shown that our radio can coexist not only with primary devices but also with
other cognitive radios.
Filterbank multicarrier techniques were also applied to fault detection of live
wires. Multicarrier refelctometry (MCR) as an effective method for testing live
wires had been recently proposed. We studied the use of filterbanks for genera-
tion/synthesis of MCR test signals and also for analysis of the reflected signal. We
noted that the test signals have to be confined to the portion(s) of the frequency
band that is (are) free of the live wire signals in order to avoid interference with the
critically important live ware signals. Moreover, for effective analysis of the reflected
waves, optimal filters that separate the test signal tones and also avoid minimal
leakage from the live wire signals should be designed. We discussed the criteria
necessary to design effective MCR systems and developed the relevant filterbank
design procedures. We also borrowed an idea from cognitive radio research and
introduced the novel idea of cognitive live wire testing, where the tester first
measures the live wire signal activities and then decides on which part of the
spectrum may be used for testing.
other hand, if the orthogonality is lost, there would be interference among adjacent
and nonadjacent subcarriers that results in considerable distortion. Hence, FBMC
systems are more suitable than OFDM systems for environments with high mobility
and high doppler effect. Designing all the components of an FBMC systems that
can reliably perform in frequency selective fast fading channels presents us many
new areas of research. Using adaptive filters and decision feedback equalizers for
channel estimation, joint data and channel estimation, performance of different
filter designs in different channel models are few examples of the research topics in
mobile FBMC systems.
VARIABLES
This appendix provides definitions used in the text. The following variables are
used in Chapters 1 to 6.