Fulltext01 PDF

Linköping Studies in Science and Technology
Dissertation No. 1344
Contributions to Reconfigurable Filter Banks

and Transmultiplexers
Amir Eghbali
Division of Electronics Systems

Department of Electrical Engineering
Linköping University, SE–581 83 Linköping, Sweden
WWW: http://www.es.isy.liu.se
E-mail: amire@isy.liu.se
Linköping 2010
Contributions to Reconfigurable Filter Banks and Transmultiplexers
c 2010 Amir Eghbali

Department of Electrical Engineering,

Linköping University,
SE–581 83 Linköping,
Sweden.
ISBN 978-91-7393-296-7
ISSN 0345-7524
Printed by LiU-Tryck, Linköping, Sweden 2010

to my family...
Abstract
A current focus among communication engineers is to design flexible radio sys-
tems to handle services among different telecommunication standards. Thus, low-
cost multimode terminals will be crucial building blocks for future generations of
multimode communications. Here, different bandwidths, from different telecom-
munication standards, must be supported. This can be done using multimode
transmultiplexers (TMUXs) which allow different users to share a common chan-
nel in a time-varying manner. These TMUXs allow bandwidth-on-demand. Each
user occupies a specific portion of the channel whose location and width may vary
with time.
Another focus among communication engineers is to provide various wideband
services accessible to everybody everywhere. Here, satellites with high-gain spot
beam antennas, on-board signal processing, and switching will be a major comple-
mentary part of future digital communication systems. Satellites provide a global
coverage and customers only need to install a satellite terminal and subscribe to the
service. Efficient utilization of the available limited frequency spectrum, calls for
on-board signal processing to perform flexible frequency-band reallocation (FFBR).
In an integrated communication system, TMUXs can operate on-ground whereas
FFBR networks can operate on-board. Thus, successful design of dynamic commu-
nication systems requires flexible digital signal processing structures. This flexibil-
ity (or reconfigurability) must not impose restrictions on the hardware and, ideally,
it must come at the expense of simple software modifications. In other words, the
system is based on a hardware platform whose parameters can be modified without
a need for hardware changes.
This thesis outlines the design and realization of reconfigurable TMUX and
FFBR structures which allow dynamic communication scenarios with simple soft-
ware reconfigurations. In both structures, the system parameters are determined
in advance. For these parameters, the required filter design problems are solved
only once. Dynamic communications, with users having different time-varying
bandwidths, are then supported by adjusting some multipliers, commutators, or a
channel switch. These adjustments do not require hardware changes and can be
performed online. However, the filter design problem is solved offline. The thesis
provides various illustrative examples and it also discusses possible applications
of the proposed structures in the context of other communication scenarios, e.g.,
cognitive radios.
i
Acknowledgments
I would like to thank my supervisor Professor Håkan Johansson for giving me the
opportunity to work as a Ph.D student. However, I should not forget to sincerely
thank him for his patience, inspiration, and guidance in helping me deal with my
problems.
I would also like to thank my co-supervisor Docent Per Löwenborg for discussions
and feedback.
Special thanks have to go to all members of my family for their support. Not all
problems can be solved by computers, books, and discussions, etc. One mostly re-
quires emotional support and encouragement from beloved ones. God has blessed
me with the best of these! I just do not know how to be thankful... I will never be
able to do this...
The former and present colleagues at the Division of Electronics Systems, Depart-
ment of Electrical Engineering, Linköping University have created a very friendly
environment. They always kindly do their best to help you. You never feel alone
even if you come from another country and do not speak fluent Swedish. Actually,
you feel it like being at home!
Last but not least, I should thank all my friends whom have made my stay in
Sweden pleasant.
Amir Eghbali
Linköping, September 2010
iii
Contents
1 Introduction 1
1.1 Motivation and Problem Formulation . . . . . . . . . . . . . . . . . . 1
1.2 Thesis Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2 Basics of Digital Filters 9

2.1 FIR Filters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.2 IIR Filters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.2.1 Note on Stability . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.3 Polyphase Decomposition . . . . . . . . . . . . . . . . . . . . . . . . 11
2.4 Special Classes of Filters . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.4.1 Complementary Filters . . . . . . . . . . . . . . . . . . . . . . 12
2.4.2 Linear-Phase FIR Filters . . . . . . . . . . . . . . . . . . . . 13
2.4.3 Nyquist (M th-band) Filters . . . . . . . . . . . . . . . . . . . 15
2.4.4 Hilbert Transformers . . . . . . . . . . . . . . . . . . . . . . . 16
2.5 FIR Filter Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3 Basics of Multirate Signal Processing 21

3.1 Sampling Rate Conversion: Conventional . . . . . . . . . . . . . . . 21
3.1.1 Noble Identity . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.2 Sampling Rate Conversion: Farrow Structure . . . . . . . . . . . . . 25
v
Contents Contents
3.2.1 Design of the Farrow Structure . . . . . . . . . . . . . . . . . 28

3.3 General M -Channel FBs . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.3.1 Filter Design for Modulated FBs . . . . . . . . . . . . . . . . 31
3.4 General M -Channel TMUXs . . . . . . . . . . . . . . . . . . . . . . 31
3.4.1 Mathematical Representation of TMUXs . . . . . . . . . . . 31
3.4.2 Duality of FBs and TMUXs . . . . . . . . . . . . . . . . . . . 33
3.4.3 Approximation of PR in Redundant TMUXs . . . . . . . . . 34
4 Flexible Frequency-Band Reallocation For Real Signals 37

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
4.1.1 Contribution and Relation to Previous Work . . . . . . . . . 38
4.1.2 Choice of the FFBR Network . . . . . . . . . . . . . . . . . . 38
4.1.3 MIMO FFBR Network Configuration . . . . . . . . . . . . . 40
4.2 FFBR Network Based on Variable Oversampled Complex Modu-
lated FBs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
4.2.1 Efficient Realization of the FFBR Network . . . . . . . . . . 41
4.3 Alternative I . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
4.3.1 Complex Versus Real Sampling . . . . . . . . . . . . . . . . . 43
4.3.2 Arithmetic Complexity: Hilbert Transformer . . . . . . . . . 43
4.3.3 Arithmetic Complexity: DFT with Complex Inputs . . . . . 44
4.3.4 Arithmetic Complexity: Complex FFBR Network . . . . . . 45
4.4 Alternative II . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
4.4.1 Arithmetic Complexity: Real FFBR Network . . . . . . . . . 46
4.5 Comparison . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
4.5.1 Arithmetic Complexity: Complex Versus Real FFBR . . . . . 49
4.5.2 Arithmetic Complexity: Alternative I Versus Alternative II . 50
4.5.3 Performance: Alternative I Versus Alternative II . . . . . . . 52
4.6 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
4.6.1 Measure of Complexity . . . . . . . . . . . . . . . . . . . . . 54
4.6.2 Applicability of Alternatives I and II . . . . . . . . . . . . . . 56
4.6.3 Filter Bank Design . . . . . . . . . . . . . . . . . . . . . . . . 56
5 A Multimode Transmultiplexer Structure 57

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
5.2 Problem Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
5.3 Multimode TMUX Structure . . . . . . . . . . . . . . . . . . . . . . 59
5.3.1 Channel Sampling Rates . . . . . . . . . . . . . . . . . . . . . 59
5.3.2 Sampling Rate Conversion . . . . . . . . . . . . . . . . . . . . 60
5.3.3 Subcarrier Frequencies . . . . . . . . . . . . . . . . . . . . . . 61
5.4 Filter Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
5.4.1 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
5.5 Implementation and Design Complexity Issues . . . . . . . . . . . . 64
5.6 TMUX Application . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
5.7 Analysis Using Multirate Building Blocks . . . . . . . . . . . . . . . 68
5.8 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
vi
Contents Contents
6 A Class of Multimode Transmultiplexers Based on the Farrow

Structure 73
6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
6.2 Prerequisites . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
6.2.1 Problem Formulation . . . . . . . . . . . . . . . . . . . . . . 74
6.2.2 Some General Issues . . . . . . . . . . . . . . . . . . . . . . . 75
6.3 Proposed Integer SRC Multimode TMUX . . . . . . . . . . . . . . . 76
6.3.1 Variable Integer SRC Using the Farrow Structure . . . . . . . 76
6.3.2 Approximation of Perfect Reconstruction (PR) . . . . . . . . 79
6.3.3 Filter Design . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
6.3.4 Filter Design Parameters . . . . . . . . . . . . . . . . . . . . 80
6.3.5 Filter Design Criteria . . . . . . . . . . . . . . . . . . . . . . 80
6.4 Proposed Rational SRC Multimode TMUX . . . . . . . . . . . . . . 83
6.4.1 TMUX Illustration . . . . . . . . . . . . . . . . . . . . . . . . 83
6.4.2 Efficient Variable Rational SRC . . . . . . . . . . . . . . . . . 84
6.4.3 Approximation of PR . . . . . . . . . . . . . . . . . . . . . . 87
6.5 TMUX Performance . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
6.5.1 Effects of Bp on the SRC Error . . . . . . . . . . . . . . . . . 90
6.6 Direct Filter Design . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
6.6.1 Design Example . . . . . . . . . . . . . . . . . . . . . . . . . 95
6.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
7 Reconfigurable Nonuniform Transmultiplexers Using Uniform

Modulated Filter Banks 99
7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
7.2 Problem Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
7.3 Nonuniform TMUXs Using Modulated FBs . . . . . . . . . . . . . . 102
7.4 System Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
7.4.1 Channel Sampling Periods . . . . . . . . . . . . . . . . . . . . 105
7.4.2 TMUX Illustration . . . . . . . . . . . . . . . . . . . . . . . . 105
7.4.3 Choice of GB . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
7.4.4 Choice of Center Frequency . . . . . . . . . . . . . . . . . . . 105
7.5 Implementation Cost . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
7.5.1 Choice of M and ρ . . . . . . . . . . . . . . . . . . . . . . . . 110
7.5.2 Filter Design Restrictions . . . . . . . . . . . . . . . . . . . . 112
7.6 Comparison with Existing Multimode TMUXs . . . . . . . . . . . . 112
7.6.1 Flexibility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
7.6.2 Spectrum Efficiency . . . . . . . . . . . . . . . . . . . . . . . 113
7.6.3 Direct or Indirect Design . . . . . . . . . . . . . . . . . . . . 113
7.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
vii
Contents Contents
8 Applications to Cognitive Radios 115

8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
8.2 Approach I: Use of DFBR Networks . . . . . . . . . . . . . . . . . . 117
8.2.1 Structure of the DFBR Network . . . . . . . . . . . . . . . . 118
8.2.2 User Bandwidth Versus Multiplexing Bandwidth . . . . . . . 119
8.2.3 Reconfigurability . . . . . . . . . . . . . . . . . . . . . . . . . 119
8.2.4 Modifications . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
8.3 Approach II: Use of TMUXs . . . . . . . . . . . . . . . . . . . . . . . 122
8.3.1 Structure of the TMUX . . . . . . . . . . . . . . . . . . . . . 123
8.3.2 Reconfigurability . . . . . . . . . . . . . . . . . . . . . . . . . 123
8.3.3 Modifications . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
8.4 Choice of Frequency Shifters . . . . . . . . . . . . . . . . . . . . . . . 125
8.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
9 Conclusion and Future Work 129
A Derivation of (6.23) 131
B Derivation of (6.35) 135
viii
Acronyms and Abbreviations

ADC Analog to Digital Converter
AFB Analysis Filter Bank
CLS Constrained Least-Squares
CMFB Cosine Modulated Filter Bank
DAC Digital to Analog Converter
DFBA Dynamic Frequency-Band Allocation
DFBR Dynamic Frequency-Band Reallocation
DFT Discrete Fourier Transform
ESA European Space Agency
EVM Error Vector Magnitude
FB Filter Bank
FDM Frequency Division Multiplexed
FIR Finite-length Impulse Response
FFBR Flexible Frequency-Band Reallocation
FBR Frequency-Band Reallocation
GB GuardBand
GRB Granularity Band
GSM Global System for Mobile communications
ICI Inter-Carrier Interference
IDFT Inverse Discrete Fourier Transform
IIR Infinite-length Impulse Response
ISI Inter-Symbol Interference
IS-54 Interim Standard-54
IS-136 Interim Standard-136
LPTV Linear Periodic Time-Varying
LS Least-Squares
LTI Linear Time-Invariant
MF/TDMA Multiple Frequency/Time Division Multiple Access
MIMO Multi-Input Multi-Output
MDFT Modified Discrete Fourier Transform
MSE Mean Square Error
NPR Near Perfect Reconstruction
PFBR Perfect Frequency-Band Reallocation
PR Perfect Reconstruction
RF Radio Frequency
QAM Quadrature Amplitude Modulation
SFB Synthesis Filter Bank
SISO Single-Input Single-Output
SRC Sampling Rate Conversion
TMUX Transmultiplexer
WLAN Wireless Local Area Network
ix
x
1
Introduction
1.1 Motivation and Problem Formulation

Communication engineers aim to design flexible radio systems to handle services
among different telecommunication standards [1–10]. Along with the increase in
(i) the number of communication standards (modes), and (ii) the range of services,
the requirements on flexibility and cost-efficiency of these radio systems increase
as well. Hence, low-cost multimode1 terminals will be crucial building blocks for
future generations of communication systems. Multistandard communications re-
quire to support different bandwidths from different telecommunication standards.
Table 1.1 shows the bit rate, number of users sharing one channel, and the chan-
nel spacing of some popular cellular telecommunication standards, e.g., interim
standard-54/136 (IS-54/136), global system for mobile communications (GSM),
and IS-95 [11]. To include such standards in a general telecommunication system,
one should handle a number of different bandwidths. Consequently, any user can
use any standard which suits its requirements on bandwidth, transmission quality,
etc. Assume, for example, that a communication channel is shared by three users
A, B, and C which respectively transmit video, text, and audio. With bandwidth-
on-demand, any user can, at any time, decide to send either of video, text, and
audio. Furthermore, at any time, any user can decide to use any center frequency.
1 This is also referred to as multiband, multistandard, universal [7].
1
1. INTRODUCTION
Table 1.1: Bit rate, number of users sharing one channel, and channel spacing in
different telecommunication standards.
Standard Bit Rate No. of Users Channel Spacing

IS-54/136 48.6 Kbps 3 30 KHz
GSM 271 Kbps 8 200 KHz
IS-95 1.2288 Kbps 798 1250 KHz
To support multimode communications, we thus need a system which allows

different numbers of users, having different bit rates, to share a common channel.
Transmultiplexers (TMUXs) allow different users to share a common channel [12].
Consequently, multimode TMUXs constitute one of the main building blocks in
multistandard communications. Multiple access schemes such as code division
multiple access, time division multiple access, frequency division multiple access,
and orthogonal frequency division multiple access are special cases of a general
TMUX structure [13–15]. To support bandwidth-on-demand, the characteristics of
the TMUXs must vary with time. Such a communication system has a dynamic
allocation of bandwidth. Each user occupies a specific portion of the channel whose
location and width may vary with time.
The principle of such a communication system is shown in Fig. 1.1. Here,
the whole frequency spectrum is shared by P users. Each user Xp has a band-
width of π(1+ρ)
Rp , p = 0, 1, . . . , P − 1, and Rp can be an integer or a rational value.
Furthermore, ρ is the roll-off factor and a guardband (GB) of ∆ separates the
user signals2 . To support such a scenario, we can, in principle, use conventional3
nonuniform TMUXs or FBs, e.g., [16–31]. In a dynamic communication system,
these conventional TMUXs and FBs would require either predesign of different
filters or online filter design. This becomes inefficient when simultaneously consid-
ering the increased number of communication scenarios and the desire to support
dynamic communications. Therefore, it is vital to develop low-complexity TMUXs
which dynamically support different communication scenarios with reasonable im-
plementation complexity and design effort. One aim of this thesis is to introduce
TMUXs which allow different numbers of users, having different bandwidths, to
share the whole frequency spectrum in a time-varying manner.
As a promise of future digital communication systems, communication engi-
neers also aim to support various wideband services accessible to everybody ev-
erywhere [32–39]. Here, satellites with high-gain spot beam antennas, on-board
signal processing, and switching will be a major complementary part of future dig-
ital communication systems [32–37]. Because of the global coverage of satellites,
customers only need to install a satellite terminal and subscribe to the service.
The European space agency has proposed three major network structures for
2 The choice of ∆ does not restrict the analysis and design of the TMUX and, hence, throughout
this thesis we will mostly assume ∆ = 0.

3 This is due to the duality of filter banks (FBs) and TMUXs [12].
2
1. INTRODUCTION
Case I: D<0
X0 X1 X2 XP-1 X0
wT
0 2p
X0 X1 X2 XP-1 X0
wT
0 2p
Case II: D=0
X0 X1 X2 XP-1 X0
wT
0 2p
X0 X1 X2 XP-1 X0
wT
0 2p
Case III: D>0

D D
X0 X1 X2 XP-1 X0
wT
0 2p
D D
X0 X1 X2 XP-1 X0
wT
0 2p
Figure 1.1: Problem formulation where P users share the frequency spectrum.
broadband satellite-based systems in which satellites communicate with the users

through multiple spot beams [37]. Therefore, we need efficient reuse of the limited
available frequency spectrum by satellite on-board signal processing [32–57]. This
calls for flexible frequency-band reallocation (FFBR) networks [40–50] also referred
to as frequency multiplexing and demultiplexing [40, 50–56].
The digital part of the satellite on-board signal processor is a multi-input multi-
output system. The number of input signals can differ from that of the output sig-
nals. Furthermore, the input/output signals can have different bandwidths. Such
a communication system must support different communication and connectiv-
ity scenarios. One such main scenario is based on multiple frequency/time division
multiple access (MF/TDMA). Here, the bandwidth of each incoming signal is com-
posed of a number of adjacent smaller frequency bands (subbands). Each subband
is occupied by one (a few) user (users). This MF/TDMA scheme slices the chan-
nel both in time and frequency [58]. At any time, any portion of the channel can
be used by any user. The on-board signal processor reallocates all subbands to
different output signals and center frequencies.
The principle of this operation is illustrated in Fig. 1.2. Here, different users
3
1. INTRODUCTION
Output signal 1
3 1
FFBR Network
Input signal 1 wTout [rad]
In 1 Out 1 p
1 2 3
wTin [rad] Output signal 2
p
Out 2 5 4
In 2 p
4 5 6 Out 3
p
2 6
wTout [rad]
p
Figure 1.2: Frequency-band reallocation (FBR) for an FFBR network where any
signal in any of the two input signals can be reallocated to any position in any of
the three output signals.
are present at the input of the FFBR networks and each of them must be real-
located to different center frequencies. In a dynamic communication system, the
bandwidth and center frequency of the users may change in a time-varying man-
ner. This necessitates FFBR networks which can dynamically perform reallocation
of users with different bandwidths. Consequently, some requirements are imposed
on FFBR networks such as flexibility, low complexity, near perfect frequency-band
reallocation, simplicity, etc. [37]. In practice, one may need GBs between the sub-
bands so that the network is realizable. It is one aim of this thesis to outline flexible
and low complexity solutions for such FFBR networks. Although the idea of FFBR
networks stems from satellite-based communications, they are generally applicable
to systems which require frequency multiplexing and demultiplexing. This thesis
will also outline some of these applications in the context of cognitive radios.
To successfully design dynamic communication systems, communication engi-
neers require high levels of flexibility in digital signal processing structures. This
flexibility must not restrict the hardware and, ideally, it must come at the expense
of simple software modifications. This is frequently referred to as reconfigurabil-
ity [4, 6, 59–62] meaning that the system is based on a hardware platform whose
parameters can be modified without hardware changes.
This thesis outlines solutions for the reconfigurable communication scenarios
discussed above. It is a result of the research performed at the Division of Electron-
ics Systems, Department of Electrical Engineering, Linköping University between
October 2006 and August 2010. The research during this period has resulted in
the following publications [43–46, 63–68]:
1. A. Eghbali, H. Johansson, and P. Löwenborg, “Flexible frequency-band

reallocation MIMO networks for real signals,” in Proc. Int. Symp. Image
Signal Processing Analysis, Istanbul, Turkey, Sept. 2007.
4
1. INTRODUCTION
2. A. Eghbali, H. Johansson, and P. Löwenborg, “Flexible frequency-band

reallocation: complex versus real,” Circuits Syst. Signal Processing, DOI
10.1007/s00034-008-9090-3, Jan. 2009.
3. A. Eghbali, H. Johansson, and P. Löwenborg, “An arbitrary bandwidth

transmultiplexer and its application to flexible frequency-band reallocation
networks,” in Proc. Eur. Conf. Circuit Theory Design, Seville, Spain, Aug.
2007.
4. A. Eghbali, H. Johansson, and P. Löwenborg, “A multimode transmulti-

plexer structure,” IEEE Trans. Circuits Syst. II, vol. 55, no. 3, pp. 279–283,
Mar. 2008.
5. A. Eghbali, H. Johansson, and P. Löwenborg, “A Farrow-structure-based

multi-mode transmultiplexer,” in Proc. IEEE Int. Symp. Circuits Syst.,
Seattle, Washington, USA, May 2008.
6. A. Eghbali, H. Johansson, and P. Löwenborg, “A class of multimode trans-

multiplexers based on the Farrow structure,” Circuits Syst. Signal Processing,
2010, submitted.
7. A. Eghbali, H. Johansson, and P. Löwenborg, “On the filter design for a

class of multimode transmultiplexers,” in Proc. IEEE Int. Symp. Circuits
Syst., Taipei, Taiwan, May. 24-27, 2009.
8. A. Eghbali, H. Johansson, and P. Löwenborg, “Reconfigurable nonuniform

transmultiplexers based on uniform filter banks,” in Proc. IEEE Int. Symp.
Circuits Syst., Paris, France, May 30-June 2, 2010.
9. A. Eghbali, H. Johansson, and P. Löwenborg, “Reconfigurable nonuniform

transmultiplexers based on uniform filter banks,” IEEE Trans. Circuits Syst.
I - Regular Papers, accepted for publication.
10. A. Eghbali, H. Johansson, and P. Löwenborg, and H. G. Göckler, “Dy-

namic frequency-band reallocation and allocation: From satellite-based com-
munication systems to cognitive radios,” J. Signal Processing Syst., DOI
10.1007/s11265-009-0348-1, Feb. 2009.
These papers are covered in Chapters 4–8. The following papers were also published
during this period but they are not included in this thesis:
1. A. Eghbali, O. Gustafsson, H. Johansson, and P. Löwenborg, “On the com-

plexity of multiplierless direct and polyphase FIR filter structures,” in Proc.
Int. Symp. Image Signal Process. Analysis, Istanbul, Turkey, Sept. 2007.
2. G. Mehdi, N. Ahsan, A. Altaf, and A. Eghbali, “A 403-MHz fully differential

class-E amplifier in 0.35 um CMOS for ISM band applications,” in Proc.
IEEE EWDTS 2008, Lviv, Ukraine, Oct. 9-13, 2008.
5
1. INTRODUCTION
3. A. Eghbali, H. Johansson, T. Saramäki, and P. Löwenborg, “On the design

of adjustable fractional delay FIR filters using digital differentiators,” in Proc.
IEEE Int. Conf. Green Circuits Syst., Shanghai, China, June 21-23, 2010.
1.2 Thesis Outline

The thesis consists of nine chapters where Chapters 2 and 3 deal with the back-
ground material. The main contributions of the thesis appear in Chapters 4–8.
Chapter 2 reviews the basics of digital filters. It includes the definition of finite-
length impulse response and infinite-length impulse response filters; polyphase de-
composition; and some special classes of filters. The minimax, least-squares (LS),
and the constrained LS filter design problems are also treated.
Chapter 3 discusses sampling rate conversion (SRC) using conventional struc-
tures and the Farrow structure. Furthermore, the noble multirate identities and
efficient SRC structures are considered. In addition, FBs and TMUXs are studied.
The perfect reconstruction is treated and its approximation by redundant TMUXs
is considered. Finally, the filter design problem for redundant TMUXs is outlined.
Chapter 4 is based on [43, 45] and it discusses approaches for realizing FFBR
networks. The chapter introduces two alternatives for processing real signals using
real input/output and complex input/output FFBR networks. It is shown that the
real case has less overall number of processing units. In addition, the real system
eliminates the need for two Hilbert transformers and is suitable for systems with
a large number of users. Finally, issues related to performance and the trend in
arithmetic complexity with respect to (i) the prototype filter order, (ii) the number
of FB channels, (iii) the order of the Hilbert transformer, and (iv) the efficiency in
FBR are also considered.
Chapter 5 covers [46, 63] and it introduces a multimode TMUX capable of gen-
erating a large set of bandwidths and center frequencies. The TMUX utilizes fixed
integer SRC, Farrow-based variable rational SRC, and variable frequency shifters.
The building blocks, their operation, and the filter design problem along with some
design examples are considered. It is shown that, by designing the filters only once
offline, all possible combinations of bandwidths and center frequencies are obtained
online. This requires simple adjustments of the variable delay parameter of the
Farrow-based filters and the variable parameters of the frequency shifters. Using
the rational SRC equivalent of the Farrow-based filters, the TMUX is described in
terms of conventional multirate building blocks. The performance and functionality
tests of the FFBR network, discussed in Chapter 4, are also illustrated.
Chapter 6 considers a class of multimode TMUXs proposed by [64–66]. The
TMUXs use the Farrow structure to realize polyphase components of general in-
terpolation/decimation filters. This allows integer SRC with different ratios to
be realized using fixed filters and a few variable multipliers. In conjunction with
variable frequency shifters, an integer SRC multimode TMUX is presented and its
filter design problem, using the minimax and LS methods, is treated. A model
of general rational SRC is then constructed where the same fixed subfilters are
6
1. INTRODUCTION
used to perform rational SRC. Efficient realizations of this rational SRC scheme
are presented. Similarly, variable frequency shifters are utilized to derive a general
rational SRC multimode TMUX. By processing quadrature amplitude modulation
signals, the performance of the TMUX is also discussed.
Chapter 7 is based on [67, 68] and it introduces reconfigurable nonuniform
TMUXs based on fixed uniform modulated FBs. The proposed TMUXs use cosine
modulated FBs and modified discrete Fourier transform FBs. Users can occupy
different bandwidths and center frequencies in a time-varying manner. The filter
design, realization, and the reconstruction error are discussed. Further, the system
parameters and the implementation cost are treated. The chapter also compares
the proposed TMUXs to those in Chapters 5 and 6.
Chapter 8 is based on [44] and it deals with two approaches for frequency
allocation and reallocation used in the baseband processing of cognitive radios.
These approaches can be used depending on the availability of a composite signal
comprising several user signals or the individual user signals. With composite
signals, the FFBR network in Chapter 4 is used. To process individual users, the
TMUXs in Chapters 5–7 can be used. Discussions on reconfigurability with respect
to cognitive radios are also provided.
Chapter 9 gives some concluding remarks and open issues for future research.
7
1. INTRODUCTION
8
2
Basics of Digital Filters
This chapter reviews some basics of digital filters. First, finite-length impulse
response (FIR) and infinite-length impulse response (IIR) filters are discussed.
Section 2.3 treats the polyphase decomposition. Some classes of filters, viz., power
complementary, Nyquist, linear-phase FIR, and Hilbert transformers are discussed
in Section 2.4. Finally, Section 2.5 outlines the minimax, least-squares (LS), and
the constrained LS (CLS) filter design problems.
2.1 FIR Filters

A causal1 FIR filter of order N has an impulse response with N + 1 coefficients
h(0), h(1), . . . , h(N ). The transfer function of an N th-order FIR filter is [69]
N
X
H(z) = h(n)z −n . (2.1)
n=0
In the time domain and with an input sequence x(n), the output sequence is
N
X
y(n) = h(k)x(n − k) ⇔ Y (z) = H(z)X(z). (2.2)
k=0
1 A filter is causal if h(n) = 0, n < 0. A non-causal FIR filter can be made causal by insertion
of a proper delay.
9
2. BASICS OF DIGITAL FILTERS
x(n) T T T T
h0 h1 h2 hN-1 hN
y(n)
Figure 2.1: Direct form realization of an N th-order FIR filter.
x(n)
h0 h1 h2 hN-1 hN
y(n) T T T T
Figure 2.2: Transposed direct form realization of an N th-order FIR filter.
There are different ways to realize (2.2) and two are shown in Figs. 2.1 and 2.2
where the impulse response values are h0 , h1 , . . . , hN . The FIR filters allow one
to use non-recursive algorithms for their realization thereby eliminating problems
with instability. This thesis always deals with non-recursive stable FIR filters.
Figures 2.1 and 2.2 need N + 1 multiplications, N two-input additions, and N
delay elements.
2.2 IIR Filters

If the length of h(n) is infinite, the filter is called IIR where
PN −n
n=0 a(n)z
H(z) = PN . (2.3)
1 − n=1 b(n)z −n
With b(n) = 0, n = 0, 1, . . . , N, an IIR filter reduces to an FIR filter. Realization

of IIR filters requires recursive algorithms which may give rise to problems of
instability. As the poles of IIR filters are not in the origin (as opposed to FIR
filters), their design has extra degrees of freedom. However, care must be taken to
place the poles inside the unit circle to ensure stability.
2.2.1 Note on Stability

The z-transform of h(n) is defined by the Laurent series [69–73]
+∞
X
H(z) = h(n)z −n . (2.4)
n=−∞
10
This transform exists if h(n) decays to zero as n approaches −∞ and +∞. If [72]
|h(n)|≤M1 K1n , n≥0, (2.5)

|h(n)|≤M2 K2n , n≤0, (2.6)
then (2.4) converges for

K1 < |z| < K2 . (2.7)
jθ
As z can have a radius r and an angle θ of the form z = re , (2.4) will converge
on every concentric circle with K1 < r < K2 . For right-hand (left-hand) sided
sequences, (2.4) will converge on concentric circles exterior (interior) to some ra-
dius, say Kc , determined by the radius of the largest (smallest) pole [72]. If (2.4)
converges for r = 1, the Fourier transform of h(n) exists and it is defined as [69, 71]
+∞
X
H(ejωT ) = h(n)e−jnωT . (2.8)
n=−∞
2.3 Polyphase Decomposition

The transfer function in (2.1) can be decomposed as
∞
X
H(z) = h(nL)z −nL
n=−∞
X∞
+z −1 h(nL + 1)z −nL (2.9)
n=−∞
...
∞
X
+z −(L−1) h(nL + L − 1)z −nL ,
n=−∞
which can be rewritten as [12, 69, 70]

L−1
X
H(z) = z −i Hi (z L ). (2.10)
i=0
Here, Hi (z) are the polyphase components and
hi (n) = h(nL + i), i = 0, 1, . . . , L − 1. (2.11)
This decomposition is frequently referred to as the Type I polyphase decomposition.

The Type II polyphase decomposition of (2.1) is [12]
L−1
X
H(z) = z −(L−1−i) Ri (z L ), (2.12)
i=0
11
where Ri (z) = HL−1−i (z) [12]. The Type I and II polyphase decompositions allow
one to efficiently realize the analysis and synthesis filter banks (FBs) of general
FBs, respectively [12].
With polyphase realization, the filters operate at the lowest possible sampling
frequency. Although polyphase decomposition reduces the implementation cost,
the total number of multiplications and additions does not change. This cost
reduction is achieved by operating the adders and multipliers at a lower sampling
frequency. To realize an N th-order FIR filter using the L-polyphase decomposition,
we need L subfilters of length roughly NL+1 . To do so, (2.2) is rewritten as [70]
L−1
X L−1
X L−1
X
Y (z) = Yl (z L )z −l = Xi (z L )z −i Hj (z L )z −j , (2.13)
l=0 i=0 j=0
where Yl (z), Xi (z), and Hj (z) are the polyphase components of Y (z), X(z), and
H(z), respectively. In a matrix form, (2.13) becomes
Y0 (z L ) X0 (z L )
   
 Y (z L )   X (z L ) 
 1   1 
 = H(zL )  (2.14)
.. ..
 
   
 .   . 
YL−1 (z L ) XL−1 (z L )
where
H0 (z L ) z −L HL−1 (z L ) ... z −L H1 (z L )
 

 H1 (z L ) H0 (z L ) ... z −L H2 (z L ) 

L
H(z ) =  .. .. .. ..
. (2.15)
.
 
 . . . 
HL−1 (z L ) HL−2 (z L ) ... H0 (z L )
2.4 Special Classes of Filters

Some classes of digital filters are more suitable for multirate systems. The sequel
introduces some of these classes.
2.4.1 Complementary Filters

The filters Hk (z), k = 0, 1, . . . , K, are power complementary if [12]
K
X
|Hk (ejωT )|2 = c, c > 0. (2.16)
k=0
In general, Hk (z) are complementary of order p if [74]

K
X
|Hk (ejωT )|p = c, p ∈ N, c > 0. (2.17)
k=0
12
In special cases, the magnitude and power complementary filters satisfy (2.17) for
p = 1 and p = 2, respectively. Higher order complementary filters, e.g., p > 2, can
generate ordinary magnitude and power complementary filters while maintaining
superior cut-off characteristics [74]. Strictly (or delay) complementary filters are
those who add up to a delay as [12, 69]
K
X
Hk (ejωT ) = cz −D0 , c6=0. (2.18)
k=0
2.4.2 Linear-Phase FIR Filters

The FIR filters can have a linear phase so as to preserve the shape of the signals.
This requires h(n) to be either symmetric or antisymmetric as [69]
Symmetric : h(n) = h(N − n), n = 0, 1, . . . , N (2.19)

Antisymmetric : h(n) = −h(N − n), n = 0, 1, . . . , N. (2.20)
Then, we have about N2 distinct coefficients thereby reducing the number of mul-
tipliers. However, this does not change the number2 of adders. The frequency
response of a linear-phase FIR filter can be expressed as
N ωT
H(ejωT ) = e−j( 2 +c)
HR (ωT ) = ejΘ(ωT ) HR (ωT ), (2.21)
where HR (ωT ) is the real zero-phase frequency response with c = 0 and c = π2

for symmetric and antisymmetric h(n), respectively. The magnitude response, i.e.,
|HR (ωT )|, always assumes real positive values whereas HR (ωT ) could be negative.
The phase response is [69, 75]
(
Θ(ωT ), HR (ωT )≥0
Φ(ωT ) = (2.22)
Θ(ωT )−π, HR (ωT ) < 0.
In general, the linear-phase response can be of the form [75]
Φ(ωT ) = −αωT + β. (2.23)
Depending on h(n) being symmetric or antisymmetric and N being odd or even,

four types of linear-phase FIR filters are defined as [69, 75]
Type I : h(n) = h(N − n), N even

Type II : h(n) = h(N − n), N odd
Type III : h(n) = −h(N − n), N even
Type IV : h(n) = −h(N − n), N odd. (2.24)
2 For Type III linear-phase FIR filters, the number of adders is also reduced.
13
Table 2.1: Typical locations of zeros for linear-phase FIR filters.
Type Location
I Arbitrary
II ωT = π
III ωT = 0, π
IV ωT = 0
These four types have different expressions for HR (ωT ) as [75]

 P N2


 h( N2 ) + 2 n=1 h( N2 − n) cos(nωT ) Type I

 N −1
 2 P 2 h( N −1 − n) cos( n+1 ωT )

Type II
n=0 2 2
HR (ωT ) = (2.25)
 2 P N2 −1 h( N − 1 − n) sin((n + 1)ωT ) Type III
n=0

 2

 P N2−1 N −1

2 n=0 h( 2 − n) sin( n+1

2 ωT ) Type IV.
Further, [75]
(
− N ωT
2 Type I,II
Φ(ωT ) = (2.26)
− N ωT
2 + π
2 Type III,IV.
The group delay τg (ωT ) and the phase delay τp (ωT ) are defined as [69, 75]
dΦ(ωT )
τg (ωT ) = − , (2.27)
d(ωT )
and
Φ(ωT )
τp (ωT ) = − . (2.28)
ωT
The shape of a periodic signal is preserved3 if τp (ωT ) is almost constant in the

passband. This makes the delay of all signal components approximately equal. For
nonperiodic signals, τg (ωT ) may be used. For a constant phase delay, β in (2.23) is
forced to be zero whereas for a constant group delay, β in (2.23) can be arbitrary.
Linear-phase FIR filters have a constant group delay of τg (ωT ) = N2 .
The zeros of a real-valued linear-phase FIR filter are either real or as complex
conjugate pairs. If the zeros appear off the unit circle, they are mirrored with
respect to the unit circle. This thesis focuses on Types I or II as we deal with
lowpass filters. Table 2.1 shows typical locations of the zeros for different linear-
phase FIR filters.
3 The shape of a periodic bandpass or highpass signal is preserved if β in (2.23) is a multiple
of 2π and α is constant [75].
14
2.4.3 Nyquist (M th-band) Filters

A lowpass non-causal filter h(n) of order N is said to be M th-band if any of its
polyphase components, i.e., Hk (z), satisfies [12, 69, 76]
1
Hk (z M ) = . (2.29)
M
Here, N = KM − m with K and m being integers. Then,
k = M − m mod M (2.30)
m
where m mod M represents the remainder of M . In general and for a non-causal
h(n), this gives (
1
M n=0
h(n) = (2.31)
0 n = ±M, ±2M, . . .
meaning that every M th sample, except the center tap, is zero. This reduces
the number of multipliers and adders required to realize the filter. If h(n) is an
M th-band filter, its delayed version is also an M th-band filter [12]. In the causal
case, H(z) is an M th-band filter if the kth polyphase component has the form
1 −nk
Hk (z) = M z . In the time domain, this becomes
(
1
M n = nk
h(nM + k) = (2.32)
0 otherwise.
For an M th-band filter, the passband and stopband edges are, respectively, [77]
π(1 − ρ)
ωc T =
M
π(1 + ρ)
ωs T = , (2.33)
M
where ρ is the roll-off factor (excess bandwidth [75]) and 0 < ρ < 1 so that the
π
transition band contains ωT = M . In the context of FBs, ρ can assume any value
such that ρ > 0 [78]. In brief, H(z) has a real zero-phase frequency response where
1 π
HR (ωT ) = , ωT = . (2.34)
2 M
Furthermore, the passband and stopband ripples are related to each other as
δs ≤(M − 1)δc . (2.35)
If H(z) is an M th-band filter, the sum of M shifted copies of H(z) results in a

constant. In other words,
M
2π
X
k
H(zWM ) = c, WM = e−j M , c > 0. (2.36)
k=0
15
An alternative to (2.36) is obtained from (2.18) with D0 = 0 [12]. Generally, the

impulse response of a Nyquist filter could be causal or non-causal; FIR or IIR;
linear-phase or nonlinear-phase; and real or complex. This thesis always designs
real causal linear-phase FIR Nyquist filters. Nyquist filters find applications in, e.g.,
transmultiplexers [79], spectrum sensing for cognitive radios [61, 80, 81], sampling
rate conversion [12, 69], and pulse shaping in communications [82, 83].
2.4.4 Hilbert Transformers

The spectrum of a real-valued signal is Hermitian symmetric around ωT = 0 and
H(ejωT ) = H ∗ (e−jωT ). This results in some redundancy between the portions of
the spectrum at negative and positive values of ωT [84]. Thus, the information of a
real-valued signal can be obtained from its spectrum for ωT ∈[0, π]. It is also desir-
able for, e.g., single sideband communications, to discard the negative frequencies
and only process the positive part [85]. To preserve the positive frequencies, the
real signal x(n) is passed through a complex linear-phase filter [84]
(
jωT 2 0 < ωT < π
H(e )= (2.37)
0 −π < ωT < 0.
From (2.37), we see that there is some ambiguity at ωT = 0 [84]. The corresponding
IIR non-causal impulse response is

 1
 n=0
2j
h(n) = nπ odd n (2.38)

0 otherwise.

The complex output sequence is then

y(n) = x(n) ∗ h(n) = x(n) + jx(n) ∗ hi (n), (2.39)
where ∗ represents convolution and [84]
(
2
nπ odd n
hi (n) = (2.40)
0 even n.
Further, (
−j 0 < ωT < π
Hi (ejωT ) = (2.41)
j −π < ωT < 0.
In the literature, (2.41) is also referred to as the Hilbert transformer [12, 86, 87].
This thesis uses the term Hilbert transformer for (2.37). From (2.39), we can see
that the real and imaginary parts of y(n) are related by a Hilbert transform, i.e.,
a phase shift of π2 at all frequencies as in (2.41). One way to design a Hilbert
transformer is to shift a real lowpass half-band filter G(z) of length 2N as [69, 84]
N −1
H(z) = j2G(−jz) = (−1) 2 z −N + jE(−z 2 ), (2.42)
16
where
z −N − E(z 2 )
G(z) = . (2.43)
2
In the FIR case, E(z 2 ) has a linear phase with a group delay of N samples. Further,
E(z) is a wideband lowpass filter. This thesis shifts a real lowpass half-band filter
to obtain a Hilbert transformer. Thus, we have causal linear-phase FIR filters.
2.5 FIR Filter Design

The frequency response of an ideal digital filter is equal to unity in the passband(s)
and zero in the stopband(s). In other words,
(
1 in passband(s)
H(ejωT ) = (2.44)
0 in stopband(s).
Furthermore, there are no transition band(s) resulting in a brick-wall characteristic.

Such a filter has an infinite length, e.g., an ideal lowpass sinc function, as
(
1 n=0
h(n) = sin(n) (2.45)
n n 6= 0
and is not realizable. To get a realizable filter, one approximates this ideal transfer
function in the passband(s) and stopband(s) by allowing transition band(s) as well
as some ripples. Thus, the practical specification for a digital filter is
1 − δc ≤ |H(ejωT )| ≤ 1 + δc , ωT ∈ Ωc
jωT
|H(e )| ≤ δs , ωT ∈ Ωs . (2.46)
Here, δc and δs are, respectively, the passband and stopband ripples with Ωc and
Ωs being the passband and stopband regions. One can generally have filters with
multiple passband and stopband regions. Then, the specifications must be satisfied
for all of these regions. Further, one can allow different ripples in these regions.
As an example, in a lowpass filter, Ωc covers [0, ωc T ] whereas Ωs covers [ωs T, π].
Here, ωc T and ωs T are the passband and stopband edges, respectively.
After estimating the filter order, h(n) must be determined such that (2.46) is
satisfied for desired values of Ωc , Ωs , δc , and δs . A commonly used formula to
estimate the order of a linear-phase FIR filter is the Bellanger’s formula [88]
2 2π
NB ≈ − log10 (10δs δc ) . (2.47)
3 ωs T − ωc T
For reasonable orders, (2.47) gives a good approximation. For general nonlinear-
phase FIR filters, such formulae do not exist. Then, a manual search is the only
17
6
100*(NB−NK)/NK
0.8
−50
0.6
0.4 −100
0.2 −150
(ωsT−ωcT)/π −200
δs=δc [dB]
Figure 2.3: Relative comparison of the orders estimated by (2.47) and (2.48).
way to find the filter order. Note that there exist other formulae to estimate the
order, e.g., Kaiser [89], as
√
−20 log10 ( δs δc ) − 13
NK ≈ . (2.48)
14.6(ωs T − ωc T )/2π
This thesis uses the Bellanger’s formula. For large values of δc and δs , (2.47) and
(2.48) may result in negative orders but such large ripples may not be practical
also. As an example, with δc = δs = 0.5, ωs T = 0.3π, and ωc T = 0.2π, we get
NB = −5.3059 and NK = −9.5608. Throughout this thesis, the ripples are chosen
so that they (i) are practical, and (ii) ensure positive orders. This is achieved if
• δs δc < 0.1 in (2.47).

26
• δs δc < 10− 20 in (2.48).
Figure 2.3 shows a relative comparison of these positive orders for some typical
values of ωs T − ωc T and δs = δc . As can be seen, there is a maximum of 10%
difference between NB and NK . With the values of δs , δc , ωs T , and ωc T used in
this thesis, this difference is about 5%. Consequently, the conclusions of the thesis
are valid even if (2.48) is used. However, (2.48) slightly changes the fomulations of
complexity, etc. Generally and for very small or large ωc T , these formulae suffer
from estimation inaccuracies. However, there are other methods to estimate the
18
filter order as in, e.g., [90]. As [90] complicates the derivations of the arithmetic
complexity provided in this thesis, we do not use it.
The filter design problem finds h(n) so as to satisfy a specific criterion. This
criterion could be the energy, maximum ripple, or combinations of them leading to
LS, minimax, or CLS approaches. The general minimax design problem is
min δ, subject to (2.49)
|H(ejωT ) − 1| ≤ δ, ωT ∈ Ωc
|H(ejωT )|≤W (ωT )δ, ωT ∈ Ωs .
On the other hand, the LS design problem is
|H(ejωT )|2
Z Z
jωT 2
min |H(e ) − 1| d(ωT ) + d(ωT ). (2.50)
ωT ∈Ωc ωT ∈Ωs W (ωT )
Regarding CLS, one could minimize the stopband (passband) energy with con-
straints on the passband (stopband) ripples. This thesis formulates the CLS design
problem as
Z
|H(ejωT ) − 1|2 d(ωT ) ≤ δ, ωT ∈ Ωc
ωT ∈Ωc
|H(ejωT )|≤δdes , ωT ∈ Ωs .
Here, δdes is the desired maximum stopband ripple. Further, W (ωT ) is a weighting
function. A large W (ωT ) results in small (large) stopband approximation errors
for minimax (LS) designs. This thesis assumes frequency independent weighting
functions and, thus, W (ωT ) is constant in the frequency range of interest.
19
20
3
Basics of Multirate Signal
Processing
This chapter treats some basics of multirate systems. Sections 3.1 and 3.2 dis-
cuss the sampling rate conversion (SRC) based on the conventional structures and
the Farrow structure. Then, filter banks (FBs) are defined in Section 3.3 where
their input-output relation and the perfect reconstruction (PR) conditions are con-
sidered. As duals of FBs, transmultiplexers (TMUXs) are outlined in Section 3.4.
Finally, redundant TMUXs with non-overlapping filters and their filter design prob-
lem are treated.
3.1 Sampling Rate Conversion: Conventional

Different parts of a multirate system operate at different sampling frequencies.
Consequently, there is a need for SRC between these parts. This can be performed
by interpolation (decimation) which increases (decreases) the sampling frequency
of digital signals [12, 69]. An alternative, to perform SRC on digital signals, is
to first construct the corresponding analog signal and, then, resample it with the
desired sampling frequency. However, it is more efficient to perform SRC directly
in the digital domain. By changing the sampling frequency, the implementation
cost for a given task can be reduced as the adders and multipliers can operate
at a lower rate. Interpolation and decimation are two-stage processes comprising
lowpass filters as well as downsamplers and upsamplers. The block diagrams of
21
3. BASICS OF MULTIRATE SIGNAL PROCESSING
(a) (b)
x(m) M y(n) x(n) L y(m)
Figure 3.1: (a) M -fold downsampler. (b) L-fold upsampler.
x(m) H(z) M y(n) x(n) L H(z) y(m)
Figure 3.2: Decimation by M . Figure 3.3: Interpolation by L.
upsamplers and downsamplers are shown in Fig. 3.1. A downsampler retains every
M th sample of the input signal as [12, 69]
y(n) = x(nM ). (3.1)
In the frequency domain, (3.1) becomes [12, 69]

M −1
1 X 1
k
Y (z) = X(z M WM ), (3.2)
M
k=0
where WM is defined as in (2.36). The output signal is the sum of M stretched (by
1
k
converting z to z M ) and shifted (through the terms WM ) versions of X(z). Note
1
that X(z M ) is not periodic by 2π. Adding the shifted versions gives a signal with
a period of 2π so that the Fourier transform can be defined.
An upsampler adds L−1 zeros between consecutive samples of x(n) and [12, 69]
(
n
x( L ) if n = 0, ±L, ±2L, . . .
y(n) = (3.3)
0 otherwise.
In the frequency domain, (3.3) becomes [12, 69]
Y (z) = X(z L ), (3.4)
and the whole frequency spectrum is compressed by L giving rise to images. The
upsampler and downsampler are linear time-varying systems [12].
Unless x(n) is lowpass and bandlimited1 , downsampling results in aliasing. Con-
sequently, decimation requires an extra filter as in Fig. 3.2. This anti-aliasing filter
H(z) limits the bandwidth of x(n) as the original signal can only be preserved if
π
it is bandlimited to M . In Fig. 3.2,
+∞
X
y(n) = x(k)h(nM − k). (3.5)
k=−∞
1 This is not necessary to avoid aliasing. For example, if X(ejωT ) is nonzero only at
ωT ∈[ω1 T, ω1 T + 2π
M
] for some ω 1 T , there is no aliasing [12].
22
As upsampling causes imaging, interpolation requires a filter as in Fig. 3.3. This

lowpass anti-imaging filter H(z) removes the images and [12]
+∞
X
y(n) = x(k)h(n − kL). (3.6)
k=−∞
For SRC2 by a rational ratio M L , interpolation by L in Fig. 3.3 must be followed by

decimation by M in Fig. 3.2. Consequently, the cascade of the anti-imaging and
anti-aliasing filters results in one filter, say G(z). Thus, the output is [12]
+∞
X
y(n) = x(k)g(nM − kL). (3.7)
k=−∞
This thesis will frequently use this cascade and its dual, i.e., interpolation by M
followed by decimation by L. Generally, G(z) is a lowpass filter with a stopband
edge at [12, 69]
π π π
ωs T = min( , ) = . (3.8)
M L max(M, L)
In practice, there is a roll-off factor as in (2.33). If M and L are mutually co-
prime numbers, a decimator can be obtained by transposing the interpolator. For
mutually coprime M and L, the following three systems
1. Upsampling by M followed by downsampling by L
2. Downsampling by L followed by upsampling by M
3. Upsampling by kM followed by downsampling by kL followed by multiplier
1
k where k > 1
are equal [91]. Note that (3.7) generally fits into the frame work of a linear dual-rate
system [92] which can always be represented via a kernel function as
+∞
X
y(n) = p(k, n)x(k). (3.9)
k=−∞
3.1.1 Noble Identity

The noble identity allows one to move the filtering operations inside a multirate
structure. If H(z) is a rational function, i.e., a ratio of polynomials in z or z −1 ,
the noble identities can be defined as in Fig. 3.4. Combination of these noble
identities and the polyphase decomposition enables efficient realizations of multi-
rate structures. Efficient structures for integer decimation and interpolation are,
respectively, shown in Figs. 3.5 and 3.6.
L
2 If L > M (L < M ), we have interpolation (decimation) by a rational ratio
M
> 1 (M
L
> 1).
This thesis frequently refers to SRC by a rational ratio Rp > 1.
23
x(m) H(zM) M y(n) <=> x(m) M H(z) y(n)
x(n) M H(zM) y(m) <=> x(n) H(z) M y(m)
Figure 3.4: Noble identities which allow us to move the arithmetic operations to
the lower sampling frequency.
x(m) H(z) M y(n)

Mfs fs
x(m) M H0(z) y(n)

H0(z)
Mfs fs
z-1
M H1(z) x(m) H1(z) y(n)

Mfs fs
z-1
M HM-1(z) HM-1(z)
Figure 3.5: Decimation with polyphase decomposition and noble identities.
x(n) M H(z) y(m)

fs Mfs
x(n) H0(z) M y(m)

H0(z)
fs Mfs
z-1
H1(z) M x(n) H1(z) y(m)

fs Mfs
z-1
HM-1(z) M HM-1(z)
Figure 3.6: Interpolation with polyphase decomposition and noble identities.
24
Table 3.1: Types of the linear-phase FIR filters Sk (z).
Nk k Type
even even I
even odd III
odd even II
odd odd IV
x(n)
SL(z) S2(z) S1(z) S0(z)
m m m
y(n)
Figure 3.7: Farrow structure with fixed subfilters Sk (z) and variable fractional
delay µ.
3.2 Sampling Rate Conversion: Farrow Structure
In conventional SRC and if the SRC ratio changes, new filters are needed. This
reduces the flexibility in covering different SRC ratios. By utilizing the Farrow
structure [93], shown in Fig. 3.7, this can be solved in an elegant way. The Farrow
structure is composed of linear-phase finite-length impulse response (FIR)3 subfil-
ters Sk (z), k = 0, 1, . . . , L, with either a symmetric (for k even) or antisymmetric
(for k odd) impulse response. According to Table 3.1, these subfilters could have
any of the four types of the linear-phase FIR filters discussed in Section 2.4.2.
When Sk (z) are linear-phase FIR filters, the Farrow structure is often referred
to as the modified Farrow structure [94]. Throughout this thesis, we simply refer
to it as the Farrow structure. The Farrow structure is efficient for interpolation
whereas, for decimation, it is better to use the transposed Farrow structure [3, 95]
so as to avoid aliasing. This chapter only considers integer and rational SRC ratios.
Then, the decimators are obtained by transposing the corresponding interpolators
[12]. This is in contrast to the irrational case which is more subtle [3, 95]. The
subfilters can also have even or odd orders Nk . With odd Nk , all Sk (z) are general
filters whereas for even Nk , the filter S0 (z) reduces to a pure delay. The transfer
function of the Farrow structure is
3 With infinite-length impulse response (IIR) filters, care must be taken to avoid transients as
µ may change for every sample.
25
L
X
H(z, µ) = Sk (z)µk (3.10)
k=0
Nk
L X
X
= sk (n)z −n µk
k=0 n=0
X L
N X N
X
= sk (n)µk z −n = h(n, µ)z −n .
n=0 k=0 n=0
Here, |µ| ≤ 0.5 and N is the order of the overall impulse response
L
X
h(n, µ) = sk (n)µk . (3.11)
k=0
Further, µ is the fractional delay value4 which defines the time difference between
each input sample and its corresponding output sample. In the rest of the thesis,
we use h(n) and H(z) instead of h(n, µ) and H(z, µ), respectively. Assuming Tin
and Tout to be the sampling period of x(n) and y(n), respectively, µ is5 [63, 65, 96]
Even Nk : [nin + µ(nin )]Tin = nout Tout

Odd Nk : [nin + 0.5 + µ(nin )]Tin = nout Tout (3.12)
where nin (nout ) is the input (output) sample index. If µ is constant for all input
samples, the Farrow structure delays a bandlimited signal by a fixed µ. Figure 3.8
shows two delayed versions of a bandlimited signal x(n) = sin( nπ
12 ) where µ = 0.25
and µ = 0.45. In both cases, one set of Sk (z) has been used and only µ is modified.
In general, SRC can be seen as delaying every input sample with a different
µ. This delay depends on whether one performs decimation or interpolation. For
interpolation, one can obtain new samples between any two consecutive samples of
x(n). With decimation, one can shift the original samples (or delay them in the
time domain) to the positions which would belong to the decimated signal. Hence,
some signal samples will be removed but some new samples will be produced.
Thus, by controlling µ for every input sample, the Farrow structure performs SRC.
For decimation, Tout > Tin whereas interpolation results in Tout < Tin . As an
example, Fig. 3.9 illustrates two versions of a bandlimited signal x(n) = sin( nπ
12 )
where a rational SRC by Rp = 1.75 is performed. In both cases, the same Sk (z)
as those in Fig. 3.8 have been used and only µ(nin ) is modified for every input
sample.
4 In the modified Farrow structure, 0 < µ < 1.

5 In the implementation, a group of input samples are present in the delay elements of Sk (z).
For every µ, its corresponding x(n) must be aligned with the point of symmetry of Sk (z).
26
0.8
x(n)
0.6 x(n−0.25)
x(n−0.45)
0.4
0.2
Amplitude
−0.2
−0.4
−0.6
−0.8
−1
5 10 15 20
n
Figure 3.8: Application of the Farrow structure to delay x(n) = sin( nπ

12 ).
0.8 sin(n1ωT1)
sin(n2ωT1/1.75)
0.6
sin(n3ωT1*1.75)
0.4
0.2
Amplitude
−0.2
−0.4
−0.6
−0.8
−1
5 10 15 20 25 30 35
Samples
Figure 3.9: Application of the Farrow structure to perform SRC on x(n) = sin( nπ
12 ).
27
3.2.1 Design of the Farrow Structure

Generally, Sk (z) are designed so that H(z) approximates an allpass transfer func-
tion with a fractional delay µ over the frequency range6 of interest [94, 95, 97–104].
The desired causal magnitude and unwrapped phase responses are
Hdes (ejωT ) = e−j(∆+µ)ωT , (3.13)

Φdes (ωT ) = −(∆ + µ)ωT, (3.14)
where
maxk (Nk )
∆= . (3.15)
2
The main advantage of the Farrow structure is its ability to perform rational SRC
using only one set of Sk (z) and by simple adjustments of µ. In the non-causal case
and with L subfilters, the Taylor series expansion of (3.13) is [105]
L L
−jµωT
X (−jµωT )k X (−jωT )k
e ≈ = µk . (3.16)
k! k!
k=0 k=0
Comparing (3.10) and (3.16), one way to obtain a fractional delay filter is to deter-
mine Sk (z) so that they approximate kth-order differentiators [102]. Other methods
to design the Farrow structure can be found in, e.g., [94, 95, 97–104].
3.3 General M -Channel FBs

An M -Channel FB splits the input signal into the M subbands Xm (z), m =
0, 1, . . . , M − 1, using the analysis FB (AFB) filters Hm (z). To reconstruct the
original input signal, we need the synthesis FB (SFB) filters Fm (z). Furthermore,
upsamplers and downsamplers by P are also required as in Fig. 3.10. The output
of a general M -channel FB is
P −1 M −1
1 X n
X
Y (z) = X(zWP ) Hm (zWPn )Fm (z) (3.17)
P n=0 m=0
where WP is defined as in (2.36). Ideally, the output signal is scaled (by α) and
delayed (by β) version of the input signal, i.e., y(n) = αx(n − β). Such a system
is referred to as PR. If a FB is near PR (NPR), some aliasing and distortion exist.
Therefore, α is frequency dependent and the distortion transfer function is
M −1
1 X
V0 (z) = Hm (z)Fm (z), (3.18)
P m=0
6 The input of the Farrow structure must be bandlimited to this frequency range.
28
Analysis FB x0(m) Synthesis FB

H0(z) P P F0(z)
x1(m)
x(n) H1(z) P P F1(z) y(n)
xM-1(m)
HM-1(z) P P FM-1(z)
Figure 3.10: General M -channel FB.
whereas the aliasing transfer functions are

M −1
1 X
Vl (z) = Hm (zWPl )Fm (z), l = 1, 2, . . . , P − 1. (3.19)
P m=0
These FBs are generally linear periodic time-varying (LPTV) systems with a period
M . If there is no aliasing, we have a linear time-invariant (LTI) system [12]. In a
PR FB,
V0 (ejωT ) = c, c > 0 (3.20)

Vl (ejωT ) = 0, l = 1, 2, . . . , P − 1. (3.21)
If P = M , the FB is maximally decimated and the number of samples in the set

Xm (z) equals that of X(z). The choice P < M leads to oversampled FBs [12]. If
V0 (z) is allpass (has linear-phase), we have no amplitude (phase) distortion.
Figure 3.11 shows an M -channel maximally decimated FB with the AFB filters
Hm (z) and the SFB filters Fm (z). Here,
M
X −1
l
Y (z) = T0 (z)X(z) + Tl (z)X(zWM ). (3.22)
l=1
The term
M −1
1 X
T0 (z) = Fm (z)Hm (z) (3.23)
M m=0
is the distortion transfer function and
M −1
1 X l
Tl (z) = Fm (z)Hm (zWM ), l = 1, 2, . . . , M − 1 (3.24)
M m=0
are the aliasing transfer functions.
29
Analysis FB Synthesis FB
x(n) H0(z) M M F0(z)
H1(z) M M F1(z)
å y(n)
HM-1(z) M M FM-1(z)
Figure 3.11: M -channel maximally decimated FB.
To obtain the AFB and SFB filters,

PNone can modulate a single N th-order linear-
phase FIR prototype filter G(z) = n=0 g(n)z −n . With cosine modulation [106–
108],
π M +1
hm (n) = 2g(n) cos[(m + 0.5) (N − n + )], (3.25)
M 2
π M +1
fm (n) = 2g(n) cos[(m + 0.5) (n + )] = hm (N − n). (3.26)
M 2
In a PR cosine modulated FB (CMFB), N = 2KM − 1 and K (the overlapping
factor [109]) is an integer. For complex modulated FBs,
−mn
hm (n) = g(n)WM , (3.27)
fm (n) = hm (n). (3.28)
In the maximally decimated case, we can use modified discrete Fourier transform
FBs (MDFT FBs) [110–115]. An M -channel MDFT FB can equivalently be realized
as (see Figs. 7.11 and 7.12 of [76])
M
• Two SRC stages with ratios 2 and 2 while adding some phase offset between
these stages.
• Two separate FBs where the phase offset is applied outside the AFBs and
SFBs.
If an MDFT FB is PR, N is an integer as KM + s where 0≤s < M . The choice of
AFB and SFB filters, having uniform or nonuniform passbands, leads to uniform
or nonuniform FBs [12] which can also be obtained by modulation as [23]
jπαm
(n− Lm2−1 ) jπαm
(n− Lm2−1 )
hm (n) = am gm (n)e− Mm + a∗m gm
∗
(n)e Mm , (3.29)
− jπα m
Mm (n−
Lm −1
) jπαm
(n− Lm2−1 )
fm (n) = bm gm (n)e 2 + a∗m gm
∗
(n)e Mm . (3.30)
Here, αm = (Km + 0.5) and gm (n) is the (possibly complex) prototype filter of
length Lm with Mm being the decimation factor in each branch. Each branch has
30
x0(n)
s0(n) P F0(z) H0(z) P s^0(n)
Channel
x1(n) y(n) ^
y(n)
s1(n) P F1(z) D(z) H1(z) P s^1(n)
e(n)
xM-1(n)
sM-1(n) P FM-1(z) HM-1(z) P ^s (n)
M-1
Figure 3.12: General M -channel TMUX.
a center frequency as ± πα
Mm with Km being an integer where am and bm define the
m
modulation phase. As opposed to uniform FBs, nonuniform FBs achieve a more

general time and frequency tiling [92]. Note that sine modulated FBs (SMFBs)
can be obtained similar to (3.25) and (3.26). The exponentially modulated FBs
(with complex filters) are a combination of SMFBs and CMFBs [107, 108].
For any FB, the AFB and SFB filters can be FIR or IIR. Further discussion on
these issues is not the focus of this thesis and the interested reader is referred to,
e.g., [12, 69, 76].
3.3.1 Filter Design for Modulated FBs

To design the prototype filter G(z), we can use any standard filter design technique,
e.g., [12, 69, 76, 78, 106, 113–119]. The MDFT FB has a typical lowpass G(z) with
a stopband edge as ωs T = 2π M [76]. The CMFB has a typical lowpass G(z) with a
stopband edge as ωs T = π(1+ρ)2M and a 3-dB cutoff frequency at ωT = 2Mπ
[117, 120].
If 0 < ρ≤1, only the adjacent branches overlap. With 1 < ρ≤2 (or ρ > 2), two
(or at least three) adjacent branches overlap [78]. In both FBs, G(z) satisfies the
power complementary property.
3.4 General M -Channel TMUXs

A TMUX converts the time multiplexed components of a signal into a frequency
multiplexed version and back [121]. It allows several users to transmit and receive
over a common channel. A TMUX, e.g., [12, 13, 15–20, 24, 30, 46, 63–69, 76, 79,
116, 117, 119–124], is also referred to as a FB transceiver, e.g., [21, 125–128].
3.4.1 Mathematical Representation of TMUXs

Assume a series of symbol streams sk (n), k = 0, 1, . . . , M − 1, either generated
by different users or parts of a signal generated by one user. Assume also that we
want to transmit these signals through a channel. As in Fig. 3.12, we can pass
sk (n) through the transmitter (pulse shaping) filters Fk (z). Then, (3.6) gives
31
(a)
F0(z) F1(z) FM-1(z)
wT
p
(b)
F0(z) F1(z) FM-1(z)
wT
p
(c)
F0(z) F1(z) FM-1(z)
wT
p
Figure 3.13: M -channel TMUX filters. (a) Overlapping. (b) Marginally overlap-
ping. (c) Non-overlapping.
∞
X
xk (n) = sk (m)fk (n − mP ). (3.31)
m=−∞
The filters Fk (z) take symbols of sk (n) and put pulses fk (n) around them. Here,
M users transmitP through one common channel described by a possibly complex
LD
LTI filter D(z) = n=0 d(n)z −n followed by an additive noise e(n). At the receiver
side, the receiver filters Hk (z) separate the signals and only a downsampling by P
is needed to get the original symbol streams. Ignoring the channel,
M
X −1
Ŝi (z) = Sk (z)Tki (z P ), i = 0, 1, . . . , M − 1 (3.32)
k=0
where
P −1
P 1 X
Tki (z ) = Fk (zWPl )Hi (zWPl ), (3.33)
P
l=0
and WP is defined as in (2.36). Typical characteristics of Fk (z) and Hk (z) are

shown in Fig. 3.13. Similar to FBs, TMUXs can be redundant (P > M ) or critically
sampled (P = M ). To avoid inter-symbol interference (ISI), a level of redundancy
may be needed such that P − M ≥LD [129]. The output of the TMUX in (3.32) is
P
X −1
Ŝi (z) = Tii (z)Si (z) + Tki (z)Sk (z) (3.34)
k=0,k6=i
where Tii (z) and Tki (z) represent the ISI and the inter-carrier interference (ICI),
respectively [79]. Note that the ISI (ICI) is sometimes also referred to as interband
32
Synthesis FB Analysis FB
X0(z) M F0(z) H0(z) M X^0(z)
^
X1(z) M F1(z) y(n) y(n) H1(z) M X^1(z)
å
XM-1(z) M FM-1(z) HM-1(z) M X^M-1(z)
Figure 3.14: M -channel critically sampled TMUX.
(cross-band) ISI [21]. It is desired to have
|Tii (z) − z −ηi | ≤ δISI

|Tki (z)| ≤ δICI (3.35)
with δISI and δICI being the desired ISI and ICI where ηi is the delay in each
branch i of the TMUX.
If an LTI filter is placed between an upsampler and a downsampler of ratio M ,
the overall system is equivalent to the decimated (by M ) version of its impulse
response [12]. In this case, designing Fk (z) and Hk (z) so that the decimated (by
M ) version of Fk (z)Hm (z) becomes a pure delay if k = m and zero otherwise, the
TMUX becomes PR. In terms of (3.34), this means
P −1
1 X 1 1
Tii (z) = Fi (z P WPl )Hi (z P WPl ) = αz −β , (3.36)
P
l=0
P −1
1 X 1 1
Tki (z) = Fk (z P WPl )Hi (z P WPl ) = 0. (3.37)
P
l=0
In a PR system, ŝk (n) = αsk (n − β). The PR properties are independent of the
length and causality of filters, etc., and can be satisfied for both critically sampled
and redundant TMUXs. However, for the critically sampled case, there may not
always exist FIR or stable IIR solutions. Therefore, some redundancy makes the
solutions feasible [20, 22, 24, 26, 27] and it also simplifies the PR conditions.
3.4.2 Duality of FBs and TMUXs

Duality [12] of TMUXs and FBs allows one to obtain the TMUX of Fig. 3.14 from
the FB of Fig. 3.11 where
M
X −1
Y (z) = Xm (z M )Fm (z). (3.38)
m=0
33
This signal is then transmitted over a common channel. With ŷ(n) = y(n) and
ignoring the scaling factors,
M
X −1 M
X −1
X̂d (z) = Xm (z)Fm (z 1/M WM
k
)Hd (z 1/M WM
k
), d = 0, 1, . . . , M − 1. (3.39)
k=0 m=0
Like aliasing and distortion in FBs, we can define two error sources for TMUXs,
i.e., ISI and ICI. In terms of (3.35)–(3.37) and for a critically sampled TMUX, we
have
MX−1
ISI = Fd (z 1/M WM
k
)Hd (z 1/M WMk
), (3.40)
k=0
and
M
X −1 M
X −1
ICI = Fm (z 1/M WM
k
)Hd (z 1/M WM
k
). (3.41)
m=0,m6=d k=0
The duality of FBs and TMUXs applies to both critically sampled and redundant
systems. It has been shown that if a FB is free from aliasing, the corresponding
TMUX is free from ICI [76].
3.4.3 Approximation of PR in Redundant TMUXs

In a PR TMUX and for any two branches k and m, the decimated version of the
cascade of the SFB and AFB filters is a pure delay if k = m and zero otherwise
[12]. In this regard, we use [Fk (z)Hm (z)]zeroth to represent the zeroth polyphase
component of Fk (z)Hm (z). To approximate PR, these ideal conditions must be
approximated as close as desired. Thus, the minimax optimization problem for an
NPR TMUX is
|[Fk (ejωT )Hm (ejωT )]zeroth − 1|≤δ, ωT ∈ [0, π], k = m

|[Fk (ejωT )Hm (ejωT )]zeroth |≤W (ωT )δ, ωT ∈ [0, π], k6=m
where W (ωT ) is the weighting function. Note that (3.42) considers non-causal
filters. It is well known that increasing the order of the SFB and AFB filters allows
one to decrease δ and, hence, improve the approximation of PR.
To simplify (3.42), this thesis uses redundant TMUXs with non-overlapping
filters as shown in Fig. 3.15. The TMUXs are also nonuniform. Consequently, the
ISI in (3.34) would result from the filters in one branch of the TMUX. In general
nonuniform TMUXs, the ICI in (3.34) becomes time-varying7 . However, due to the
redundancy, the stopband attenuation of the filters still controls the ICI. Therefore,
the ICI can be made as small as desired by increasing this stopband attenuation.
To meet NPR conditions, Fk (z)Hk (z) should approximate a Nyquist filter as close
as desired. Then, the SFB and AFB filters should be designed such that
7 In LTI systems, the output at any frequency only depends on the input at the same frequency.
These nonuniform TMUXs are LPTV systems. Then, the output at any given frequency is
dependent on the input at a finite set of frequencies [30].
34
F0(z) F1(z) F2(z) FM-1(z)
wT
2p
Figure 3.15: Filters of a nonuniform non-overlapping TMUX.
• They have sufficiently small ripples in their stopbands.
• The zeroth polyphase component of Fk (z)Hk (z) approximates an allpass

transfer function.
Thus, the simplified minimax design problem is
|[Fk (ejωT )Hk (ejωT )]zeroth − 1|≤δ, ωT ∈ [0, π]

|Fk (ejωT )|≤W1 (ωT )δ, ωT ∈ Ωs
|Hk (ejωT )|≤W2 (ωT )δ, ωT ∈ Ωs
where k = 0, 1, . . . , M − 1. Furthermore, W1 (ωT ) and W2 (ωT ) are the weighting

functions with Ωs being the stopband region as in (2.46). In the least-squares (LS)
sense, (3.43) becomes
min (3.44)
Z
|[Fk (ejωT )Hk (ejωT )]zeroth − 1|2 d(ωT )
ωT ∈[0,π]
|Fk (ejωT )|2

Z
+ d(ωT )
ωT ∈Ωs W1 (ωT )
|Hk (ejωT )|2
Z
+ d(ωT ).
ωT ∈Ωs W2 (ωT )
This thesis frequently uses Fk (z) = Hk (z) and W1 (ωT ) = W2 (ωT ) = W (ωT ).
Then, Fk (z) and Hk (z) are the spectral factors of a Nyquist filter [130–138]. Specif-
ically, with linear-phase FIR filters and Fk (z) = Hk (z), the resulting Nyquist filter,
i.e., Fk (z)Hk (z), has double zeros in the z-plane. Further, we will always design
real lowpass filters and variable frequency shifters will modulate the users into in-
termediate frequencies. Similar to (2.51), we will also use the constrained LS design
method. This thesis does not consider the effects of the channel when designing
the TMUXs but some methods can be found in, e.g., [21, 108, 121, 128].
35
36
4
Flexible Frequency-Band
Reallocation For Real Signals
This chapter discusses a new approach for implementing flexible frequency-band

reallocation (FFBR) networks for bentpipe [38], or transponder [36], satellite pay-
loads. We consider two alternatives to process real signals using real and com-
plex input/output FFBR networks. After some general and historical introduc-
tion in Section 4.1, Section 4.2 briefly reviews the FFBR network. Alternative I,
shown in Fig. 4.7, and its arithmetic complexity are treated in Section 4.3. Sec-
tion 4.4 introduces Alternative II, shown in Fig. 4.9, and it covers the formulation of
frequency-band reallocation (FBR) for real signals, functionality illustration, and
the arithmetic complexity. The comparison of the two alternatives is discussed in
Section 4.5. Finally, Section 4.6 gives some concluding remarks.
4.1 Introduction
As discussed in Section 1.1, the European space agency has proposed three major
network structures for broadband satellite-based communications in which satellites
communicate with users through multiple spot beams. This necessitates to reuse
the limited available frequency spectrum by satellite on-board signal processing [32–
57]. The digital part of the satellite on-board signal processor, or the payload, is a
multi-input multi-output (MIMO) system with input/output signals composed of
different users with different bandwidths. The on-board signal processor reallocates
37
4. FLEXIBLE FREQUENCY-BAND REALLOCATION FOR REAL SIGNALS
all users to different output signals and positions in the frequency spectrum. With
bandwidth-on-demand, the bandwidths of different users may vary with time. This
can be handled by dividing the input beam into a number of granularity bands
(GRBs). At any time, any user can occupy any rational number of GRBs. The
FFBR network operates on multiplexing bandwidths which are integer multiples of
the GRBs. Each multiplexing bandwidth consists of (i) a user bandwidth which is
a rational multiple of the GRB, and (ii) some guardband (GB). There are several
requirements on FFBR networks as [42]
• Flexibility to handle all FBR scenarios on users with different bandwidths.
• Low complexity to reduce the implementation cost. The amount of improve-
ments in system capacity and implementation cost is foreseen to be about
one or two orders of magnitude [37].
• Near perfect frequency-band reallocation (PFBR) to satisfy any communica-
tion performance metric, e.g., error vector magnitude (EVM) [139, 140].
• Simplicity resulting in simple system analysis and design.
4.1.1 Contribution and Relation to Previous Work

There are four types of payloads, viz., bentpipe, full processing, partial process-
ing, and hybrid [38]. This chapter focuses on bentpipe payloads whose principle
is shown in Fig. 1.2. Here, users with different bandwidths are reallocated to dif-
ferent positions in the frequency spectrum. In dynamic communications, these
bandwidths and positions may change in a time-varying manner.
As shown in Fig. 4.1, the FBR based on filter banks (FBs) uses decimation and
interpolation to generate frequency shifts of users [40–42, 48–56]. These approaches
can be classified as maximally decimated FBs [40, 55], tree-structured FBs [40,
48–54], overlap-save discrete Fourier transform (DFT) and inverse DFT (IDFT)
techniques [56], and oversampled complex modulated FBs [41, 42].
In [42], a new class of FFBR networks based on finite-length impulse response
(FIR) variable oversampled complex modulated FBs was introduced. The system
in [42] processes complex signals. Thus, the analytic representation of the real
uplink satellite signals must be processed by the FFBR network. The frequency
multiplexed results should then be converted to real signals for retransmission. This
requires one complex FFBR network and two Hilbert transformers. This chapter
refers to this solution, shown in Fig. 4.7, as Alternative I.
This chapter introduces another alternative to process real signals through a real
input/output FFBR network which has less arithmetic complexity. This solution
is referred to as Alternative II and is shown in Fig. 4.9.
4.1.2 Choice of the FFBR Network

All the FFBR networks in [40–42, 48–56] process complex signals. This chapter
shows that if both of the structures in Figs. 4.7 and 4.9 are applicable, the approach
38
Decimation Interpolation
(a) x v1 v2 v3 y
H(z) M M G(z)
T1 T1 T2 T1 T1
(b) X(ejwT1)
X0 X1 X2 XM1
0 2p/M 4p/M 6p/M 2p wT1
(c) H(ejwT1)
0 2p/M 4p/M 2p wT1
(d) V1(ejwT1)
X1
0 2p/M 4p/M 2p wT1
(e) V2(ejwT2)
X1 X1 X1 X1
0 2p 4p 2Mp wT2
(e) V3 (ejwT1)
X1 X1 X1 X1
0 2p/M 4p/M 2p wT1
(f) G(ejwT1)
0 4p/M 6p/M 2p wT1
(g) Y(ejwT1)
X1
0 4p/M 6p/M 2p wT1
Figure 4.1: FBR using decimation and interpolation. Here, only one FB channel
is shown but channel combiners can produce outputs from several FB channels.
39
Fixed Analysis FB
y0(n)
H0(z) M M G0(z)
Channel Combiner
y1(n)
x(n) H1(z) M M G1(z) y(n)
yq-1(n)
HN-1(z) M M GM-1(z)
Adjustable Synthesis FB
Figure 4.2: N -channel FFBR network utilizing fixed (adjustable) AFB (SFB).
in Fig. 4.9 is superior to that of Fig. 4.7. By using other FFBR networks, only
the exact number of operations changes but the superiority of Fig. 4.9 will still
be preserved. This chapter focuses on the FFBR network in [42] because of the
reasons outlined in [43].
4.1.3 MIMO FFBR Network Configuration

The FFBR network, considered here, is an m-input n-output system where m≤n.
However, it suffices to only discuss the single-input single-output (SISO) case as
the MIMO case is a duplication of fixed SISO structures (refer to Figs. 17 and 23
of [42]) along with some modifications1 . Further discussions on the MIMO system
for both m < n and m = n can be found in [42, 141].
4.2 FFBR Network Based on Variable Oversam-

pled Complex Modulated FBs
As shown in Fig. 4.2, the FFBR network uses fixed analysis FB (AFB) filters
Hk (z), k = 0, 1, . . . , N − 1, to split x(n) into N subbands. Then, downsam-
pling/upsampling by M and the adjustable synthesis FB (SFB) filters Gk (z) per-
form FBR. As adjustable SFB filters result in a high implementation cost, the
FFBR network is realized using fixed SFB filters and an adjustable channel switch.
This requires appropriate choices of system parameters and filter characteristics
but it reduces the implementation cost.
1 For the MIMO case, the channel switch operates between several SISO structures. Further-
more, if m < n, some branches at the output of the DFT block in Fig. 4.3 are set to zero.
40
a0 b0 g0 aN1
aN aN
x(n) M P0(zLWN ) PN1(zLWN ) M
z1 z1
Channel Switch
a1 b1 g1 aN2
aN aN
M P1(zLWN ) IDFT DFT PN2(zLWN ) M
z1 z1
aN1 bN1 gN1 a0
aN aN
M PN1(zLWN ) P0(zLWN ) M y(n)
FFBR Network
Figure 4.3: Efficient DFT- and IDFT-based implementation of Fig. 4.2.
4.2.1 Efficient Realization of the FFBR Network

Figure 4.3 shows the architecture of the N -channel FFBR network with complex
multipliers, DFT, IDFT, polyphase components, and input/output commutators.
The system assumes that the complex input signal is divided into Q GRBs where
N
Q= , A > 1, A ∈ N (4.1)
A
and the GRBs are separated by a GB of 2∆ = 2πǫ Q with 0≤ǫ≤1. The choice of
ǫ defines the transition band of the filters. A large transition band would reduce
the amount of the frequency spectrum covered by the GRBs and, hence, there is a
trade-off. Any user can occupy any rational number of GRBs. To suppress aliasing
and to shift the GRBs by all values of 2πqQ , q = 0, 1, . . . , Q − 1, we should choose
M to be a multiple of Q as2
M = BQ, B ∈ N. (4.2)
To attenuate aliasing by the stopband of the filters, the passbands and transition
bands of the shifted terms should not overlap. This is achieved if
N
M≤ < N. (4.3)
1 + Nπ∆
Assuming the length-S linear-phase FIR prototype filter
S−1
X N
X −1
P (z) = z −n p(n) = z −i Pi (z N ), (4.4)
n=0 i=0
the AFB filters are

Hk (z) = βk P (zWNk+α )
N
X −1
= βk [z −i αi Pi (z N WNαN )]WN−ki . (4.5)
i=0
2 Note A
that N = M B = M L where L is the number of FB channels per GRB.
41
|PR(wT)|
2D
wT
p p/N 0 p/N p
Figure 4.4: Characteristics of P (z).
Here, k = 0, 1, . . . , N − 1, and
2π
WN = e−j N (4.6)
αi = WN−αi (4.7)
(k+α)(S−1)
βk = WN 2
. (4.8)
Further,
ωT (S−1)
P (ejωT ) = e−j 2 PR (ωT ), (4.9)
where PR (ωT ) is the real zero-phase frequency response which should approximate
the magnitude response shown in Fig. 4.4. In addition, α is a real-valued constant to
place the filters at the desired center frequencies. The multipliers βk compensate for
the phase rotations because of the substitution of P (z) with P (zWNk+α ). Therefore,
all the AFB filters become linear-phase FIR filters with the same delay as P (z).
The multipliers γk are
γk = βk WN−k , (4.10)
whereas the SFB filters are
Gk (z) = µkr Hckr (z), (4.11)
with
ckr = k + Asr (4.12)

mr N (S−1)
µkr = WN 2M
(4.13)
(
Bsr sr ≥ 0
mr = (4.14)
M + Bsr sr < 0.
The parameter sr is the number of GRBs by which the subband r is shifted. It is

positive (negative) if that subband is shifted to the right (left). This information
is required to program the channel switch. Specifically, programming the channel
switch requires knowledge of L, sr , and the number of GRBs that each user oc-
cupies. According to (4.13), the constants µkr can be made equal to unity by a
proper choice of S and at the cost of some additional delay. This chapter assumes
µkr = 1.
42
cos(wt)
Anti-Image ADC I I
xa(t) Digital
sin(wt) xa(t) ADC Processor
Anti-Image ADC Q Q
Figure 4.5: Complex sampling. Figure 4.6: Real Sampling.
x(n) fs/2 y(n)

fs fs
xa(t) ADC Hilbert 2 Complex 2 Hilbert DAC ya(t)
FFBR
Figure 4.7: Alternative I: complex FFBR network with Hilbert transformer.
4.3 Alternative I
This section discusses Alternative I which consists of a complex FFBR network
and two Hilbert transformers.
4.3.1 Complex Versus Real Sampling

To process the real uplink satellite signal xa (t) by the FFBR network of Fig. 4.3,
it must be converted into its analytic representation. This can be done either by
complex sampling using two 90 degree out-of-phase analog to digital converters
(ADCs) or by real sampling which consists of an ADC followed by a digital pro-
cessor. The digital processor is composed of a Hilbert transformer followed by a
downsampler [57].
In Fig. 4.5, the ADCs operate at f2s whereas the ADC of Fig. 4.6 operates at
fs . Besides the requirements to attenuate the images arising from the mixers of
Fig. 4.5, the high sensitivity to match the two ADCs using analog components
makes the real sampling approach more attractive [57, 73]. Hence, the structure in
Fig. 4.7 can be used to process real signals and its arithmetic complexity will be
discussed in the sequel.
4.3.2 Arithmetic Complexity: Hilbert Transformer

The analytic representation of a real signal x(n) has a zero-valued spectrum for all
negative frequencies [69]. To generate the analytic representation, x(n) is applied
to a complex half-band filter HHB,c (z). As in (2.42), this filter is derived by shifting
the frequency response of a real lowpass half-band filter HHB,r (z) by π2 radians3
as [69, 84]
HHB,c (z) = jHHB,r (−jz). (4.15)
3 A scaling by two may also be required depending on whether H
HB,c (z) is used for interpo-
lation or decimation (refer to Fig. 4.7). However, this does not affect the filter order.
43
From (2.35) and (2.47), the order of a lowpass real half-band linear-phase FIR4
filter can be approximated as
−2 2π
NH ≈ log10 (10δs 2 ) (4.16)
3 ωs T − ωc T
where δs is the stopband ripple with ωs T and ωc T being the stopband and passband
edges, respectively. A real half-band linear-phase FIR filter has an even order of
the form 4m + 2 where m is an integer [12]. Thus, it is of Type I. To process a
specific GRB by the FFBR network, that GRB should be covered by the passband
of HHB,c (z) which, according to Fig. 4.8, requires
2π
π 2∆ + k Q
ωs T = + ,
2 2
2π
π 2∆ + k Q
ωc T = − . (4.17)
2 2
h i
Here, k∈ 0, 1, . . . , ⌊ Q
2 − ǫ⌋ with ⌊x⌋ being the floor of x. In this way, ωs T < π
and ωc T > 0 and (4.16) becomes
−2 Q
NH ≈ log10 (10δs 2 ) . (4.18)
3 ǫ+k
Having chosen δs , Q, and ǫ, the factor k relates NH to spectrum efficiency η(k) as
kπ
2π − 2(∆ + Q) k+ǫ
η(k) = =1− . (4.19)
2π Q
The efficiency is the ratio between the part of the frequency spectrum used by the
FFBR network and the whole frequency spectrum. In other words, it is the ratio
between the passband of HHB,c (z) and 2π. Ideally, k = 0 but in systems with a
large N , a small k will result in a large NH . The order of HHB,c (z) equals that of
HHB,r (z) as (4.15) does not alter the filter order.
4.3.3 Arithmetic Complexity: DFT with Complex Inputs

The arithmetic complexity for an N -point complex-input DFT is O(N 2 ) and it
requires N 2 complex multiplications and N (N − 1) complex additions [69]. Using
specialized fast Fourier transform algorithms, the arithmetic complexity for the
DFT of a radix-2 length sequence can be reduced to O(N log2 N ). Any sequence
length can be made radix-2 by zero padding [69].
This chapter focuses on the overall number of real operations. We select the
algorithms in [142–145] which require N (log2 N − 3) + 4 real multiplications and
3N (log2 N −1)+4 real additions. Other DFT realization techniques, e.g., [146–149],
may change the exact number of real operations but not the main conclusions.
4 At typical stopband attenuations of 60−80dB, the arithmetic complexity of an infinite-length
impulse response (IIR) half-band filter is around 60 − 70% of that of an half-band FIR filter [99].
Hence, IIR half-band filters may result in less arithmetic complexity. The main conclusion of this
chapter is independent of the choice of IIR and FIR filters.
44
2D+4p/Q
Hilbert Transformer Characteristics 2D+2p/Q
2D
wTin
2pa/Q 2p
Figure 4.8: Illustration of the GRBs and the characteristics of HHB,c (z).
4.3.4 Arithmetic Complexity: Complex FFBR Network

As shown in Fig. 4.3, the building blocks of the FFBR network, viz., polyphase
components, DFT, IDFT, and complex multipliers, are sandwiched between in-
put/output commutators [69]. These blocks produce samples in N parallel branches
but the output commutator retains one sample out of these branches. Conse-
quently, to produce one output sample, the number of operations required by these
building blocks must be divided by N . Both the Alternatives I and II use the
architecture in Fig. 4.3. Hence, we can ignore this division by N for comparison of
the arithmetic complexity. The computational workload to produce one complex
sample by the N -channel complex FFBR network of Fig. 4.3 is
S
• 2N real filters of length N for the AFB and SFB filters operating on complex
data requiring 4S real multiplications and 4(S − N ) real additions5 .
• Complex multipliers αi , βk , γk resulting in 4N complex multiplications6 .
• N -point complex-input IDFT and DFT with 2N (log2 N − 3) + 8 real multi-

plications and 6N (log2 N − 1) + 8 real additions.
To convert complex operations to real operations, [144] assumes that a complex

multiplication requires 3 real multiplications and 5 real additions (3/5) whereas
[146] and [69] use (3/3) and (4/2) assumptions, respectively. One complex addition
requires 2 real additions. Thus, using the (4/2) assumption, the overall arithmetic
complexity becomes 2N log2 N +2(4+2S +5N ) real multiplications and 6N log2 N +
2(4+2S −N ) real additions whereas (3/3) results in 2N log2 N +2(4+2S +3N ) real
multiplications and 6N log2 N + 2(4 + 2S + N ) real additions. These operations run
fs
at 2M where M < N for an oversampled system. The two Hilbert transformers in
Fig. 4.7, add the arithmetic complexity of two complex half-band filters running at
fs
2 if implemented using the polyphase decomposition.
5 Depending on S and N , the polyphase components may not have the same lengths. The
arithmetic complexity, derived here, is thus the worst-case scenario. However, if S is chosen such
that µkr = 1, all the polyphase components will have similar lengths.
6 According to (4.7), choosing α = 0, 0.5 further reduces the arithmetic complexity. This
reduction applies to both the alternatives and is therefore not considered here.
45
x(n) fs y(n)
fs fs
xa(t) ADC Real DAC ya(t)
FFBR
Figure 4.9: Alternative II: real FFBR network without Hilbert transformer.
4.4 Alternative II
To implement the FFBR network for real signals, Fig. 4.9 suggests to sample the
real signal at fs and feed the sampled data into the real FFBR network at the
same sample rate. Basically, the real FFBR network uses the structure of Fig. 4.3
but the polyphase filters operate on real data. Specifically, the polyphase filters
Pk (z), k = 0, 1, . . . , N − 1, process real signals. Here, x(n) is a real signal but
the output of the DFT, in the SFB, is a complex signal. As the multipliers αi
are complex, we need to only compute the real part of their output. Then, the
polyphase filters, in the SFB, can process real samples which would reduce the
implementation cost. Similar to Section 4.3 and as in Fig. 4.10, the N -channel real
FFBR network assumes that the input signal is divided into Q GRBs separated
by a GB of 2∆. Each GRB is divided into a number of uniform FB channels
constructed by complex modulating a real linear-phase FIR prototype filter P (z)
as in (4.5). Like the complex-input case, FBR is performed by dividing the real
input signal into subbands using the AFB filters Hk ; shifting the subbands; and
recombining the shifted subbands using the SFB filters Gk . However, only half of
Hk and Gk are involved in the FBR since the spectrum of a real signal spreads in
[−π, π] whereas the AFB and SFB filters are defined in [0, 2π].
Figure 4.11(a) shows the frequency spectrum of an input signal consisting of
three users. This pattern has been generated by the nonuniform TMUX in [46, 63]
which we shall discuss in Chapter 5. As discussed in Section 4.1.1, FBR needs
fixed AFB and SFB filters and a time-varying channel switch. Figures 4.12 and
4.13 show two different channel switches. The corresponding multiplexed output
signals are shown in Figs. 4.11(b) and (c), respectively.
4.4.1 Arithmetic Complexity: Real FFBR Network

Using earlier discussions, the arithmetic complexity of the real FFBR network is
• 2S real multiplications and 2(S − N ) real additions due to real-input AFB

and SFB filters Pk (z), k = 0, 1, . . . , N − 1.
• Complex multipliers βk and γk requiring 2N complex multiplications.
• N -point complex-input IDFT and DFT resulting in 2N (log2 N − 3) + 8 real

multiplications and 6N (log2 N − 1) + 8 real additions.
46
(a) Granularity Band

2p/Q
wTin
-p 0 2pa/Q 2p/Q+2pa/Q p
(b) Guardband
X2 X1 X0 X0 X1 X2
wT
-p 0 p
(c)
H0 H1 H2 H3 H4 H5 H6 H7
wT
-p 0 p
(d)
G0 G1 G2 G3 G4 G5 G6 G7
wT
-p 0 p
(e)
H0G0+H1G1 H2G2+H3G3+H4G4+H5G5 H6G6+H7G7
Y2 Y1 Y0 Y0 Y1 Y2
wT
-p 0 p
(f)
G2 G3 G4 G5 G6 G7 G0 G1
wT
-p 0 p
(g)
6 6 6 3 3
H2G2+H3G3+H G +H6 G 6 6
H6G6+H7G7 H0G0+H1G1
4 4 5 5
Y0 Y2 Y1 Y1 Y2 Y0
wT
-p 0 p
Figure 4.10: FBR by real FFBR network with Q = 8 = N2 . (a) GRBs. (b)–(e)
Recombination of channels. (b), (c), (f), and (g) Combination of channels and
reallocation of subbands; Hkm stands for Hk shifted m GRBs to the right.
47
(a)
0
Mag. [dB]
−20
−40 X0 X1 X2
−60
−π −0.75π −0.5π −0.25π 0 0.25π 0.5π 0.75π π
ωT [rad]
(a)
0
Mag. [dB]
−20
−40 X1 X2 X0
−60
−π −0.75π −0.5π −0.25π 0 0.25π 0.5π 0.75π π
ωT [rad]
(a)
0
Mag. [dB]
−20
−40 X2 X0 X1
−60
−π −0.75π −0.5π −0.25π 0 0.25π 0.5π 0.75π π
ωT [rad]
Figure 4.11: Spectrum of (a) Real input to the FFBR network. (b) and (c)
Multiplexed output signals based on the FBR scenarios in Figs. 4.12 and 4.13.
Q = M = 10, N = 20, ǫ = 0.125, α = 0.5.
Analysis Bank Output

Synthesis Bank Input
Figure 4.12: Scenario I resulting in the Figure 4.13: Scenario II resulting in the
values of sr in (4.12) to be {3, −2, −2}. values of sr in (4.12) to be {2, 2, −3}.
48
• Complex multiplication of αi on the real outputs of the AFB filters resulting

in 2N real multiplications.
• Complex multiplication of the DFT outputs by αi only to compute the real

part requiring 2N real multiplications and N real additions.
With a (4/2) assumption, the overall arithmetic complexity is 2N log2 N + 2(4 +

S + 3N ) real multiplications and 6N log2 N + 2(4 + S − 32 N ) real additions whereas
a (3/3) assumption results in 2N log2 N + 2(4 + S + 4N ) real multiplications and
fs
6N log2 N + 2(4 + S − 12 N ) real additions operating at M .
According to Fig. 4.10, the FB channels between [π, 2π] are not involved in
the FBR. Thus, more savings in the arithmetic complexity of the IDFT, DFT,
multipliers βk , and γk can be achieved. The amount of these savings depends on
the channel switch position and is thus not considered here7 . The main advantage
of Alternative II is the elimination of the Hilbert transformers which is independent
of the channel switch position.
4.5 Comparison
This section compares the arithmetic complexity of the alternatives in Sections 4.3
and 4.4. We consider equal performances and, for this performance, a comparison
of arithmetic complexity is done. First, we will discuss the arithmetic complexi-
ties of the real and complex FFBR networks. Second, the structures in Figs. 4.7
and 4.9 will be considered. Finally, Alternatives I and II will be compared for a
16-quadrature amplitude modulation (QAM) signal and in terms of EVM. This
comparison shows that both structures can be designed to have similar perfor-
mances but Alternative I would then require additional filters due to the Hilbert
transformers. In the sequel and unless otherwise mentioned, the system parameters
are those in the caption for Fig. 4.11.
4.5.1 Arithmetic Complexity: Complex Versus Real FFBR

According to Sections 4.3.4 and 4.4, the arithmetic complexity depends on S, the
length of the prototype filter, and N , the number of FB channels. Figure 4.14
compares the number of real operations for a 20-channel FFBR network in real
and complex cases. The complex FFBR network has more computational workload.
Further, it has a larger rate of increase in the number of real operations.
The operating frequency of the real FFBR network is twice that of the complex
FFBR network. Thus, the ratio between the number of operations in one time unit,
as shown in Fig. 4.15, compares the arithmetic complexity at the same operating
frequency. For large S, the real FFBR network performs roughly 10 − 15% more
operations in one time unit. According to Sections 4.3.4 and 4.4, for large S, the
7 Using quick Fourier transform [150] and splitting the DFT/IDFT kernel into discrete sine
and cosine transforms, more arithmetic complexity savings inside the DFT/IDFT can also be
achieved.
49
Real FFBR Total

6
Complex FFBR Total
Real FFBR Add.
Real FFBR Mult.
Real operations (*1000)
5 Complex FFBR Add.
Complex FFBR Mult.
200 300 400 500 600 700 800

Prototype filter length
Figure 4.14: Number of real operations in the real and complex FFBR networks
without considering their frequency of operation.
total arithmetic complexity is dominated by 8S. This can be verified by multiplying

the total number of real operations in the real FFBR network by two as its operating
frequency is twice that of its complex counterpart. Consequently, for both real and
complex FFBR networks, the numbers of real operations in one time unit are
asymptotically equal.
4.5.2 Arithmetic Complexity: Alternative I Versus Alterna-

tive II
In addition to the individual FFBR networks, the scheme in Fig. 4.7 needs two
Hilbert transformers which add the arithmetic complexity of two complex half-
band filters running at f2s . For large N , the transition band of HHB,c (z) becomes
smaller thereby increasing NH . To get around this, one can choose not to use some
of the GRBs to widen the transition band (see Fig. 4.8). This reduces both NH and
the spectrum efficiency. Figure 4.16(a) shows the order of HHB,c (z) for different
percentages of efficiency in a 50-channel complex FFBR network8 .
A high efficiency comes at the expense of a large NH . For example, to achieve
an efficiency of 95% with δc = δs = 0.001, the estimated order of HHB,c (z) is
NH = 74. According to Fig. 4.9, this filter operates on a real sequence at the input
side of the FFBR network and on a complex sequence at the output side of the
FFBR network. As discussed in Section 2.4.3, the impulse response of a half-band
filter is symmetric and every other coefficient is zero. Consequently, to process the
8 The superiority of Alternative II is more pronounced for systems with large N .
50
1.3
1.28
1.26
1.24
Complexity ratio
1.22
1.2
1.18
1.16
1.14
1.12
1.1
200 300 400 500 600 700 800
Prototype filter length
Figure 4.15: Number of real operations in the real FFBR over that of the complex
FFBR. Here, the number of operations in one time unit are considered.
(a)
600
Order of HHB,c(z)
500
400
300
200
100
0 10 20 30 40 50 60 70 80 90 100
Efficiency (%)
(b)
1.4
Complexity ratio
1.2
1 250th order
0.8 300th order
0.6 350th order
400th order
0.4
0.2
0 10 20 30 40 50 60 70 80 90 100
Efficiency (%)
Figure 4.16: Order of HHB,c (z) and the ratio between the total number of real
operations in one time unit for the alternatives of Figs. 4.7 and 4.9 with δs = 0.001.
Here, the minimum (maximum) value for the order of HHB,c (z) is 3 (667).
51
real sequence at the input side of Fig. 4.9, we need NH4+2 real multiplications and
NH 9
2 real additions. Similarly, the complex sequence at the output side of Fig. 4.9
NH +2
can be processed at the cost of 2 real multiplications and NH real additions.
Example: Assuming NH = 74, we need 43 (3NH + 2) = 168 real operations
which run at M = 50 2 times higher frequency than the complex FFBR network.
As an example, using a 350th order P (z) and converting all the operations to the
same operating frequency, the arithmetic complexity of Alternative II, in one time
unit, is about 85% of that of Alternative I. Figure 4.16(b) shows the ratio of the
number of real operations in Alternatives I and II for four different prototype filter
orders. With increase in spectrum efficiency, Alternative II results in more savings
in arithmetic complexity10 .
Although the real FFBR network performs around 10 − 15% more operations
in one time unit, it eliminates two complex half-band filters. Thus, Alternative II
has no limitation on the efficiency which means that it has an efficiency of 100%.
At this efficiency11 and for practical values of δs and ǫ in systems with a large
N , the arithmetic complexity of Alternative II, in one time unit, is always less
than 50% of that of Alternative I. Figures 4.17–4.19 show the trend of arithmetic
complexity savings for some values of N , 10−2 ≤δs ≤10−5 , and 0.05≤ǫ≤0.95. These
figures show the ratio between the number of real operations in one time unit. The
values of S are chosen as S = N m + 1, m = 1, 2, . . . , Q. Then, the value of NH is
estimated with k = 0 in (4.18). For these values of S and NH , the total number
of real operations, in one time unit, is computed and the average of their ratios is
plotted.
4.5.3 Performance: Alternative I Versus Alternative II

To compare the performance of Alternatives I and II in a 20-channel system, a
16-QAM signal is processed and the output constellation diagrams are shown in
Figs. 4.20 and 4.21. Here, S = 421 and NH = 154. For comparison, the EVM is
used which is a metric of transmitter quality in modern communication systems
[139, 140]. The EVM is defined as
v
u PNs −1 
u |e(k)| 2
EV M dB = 20 log t PNsk=0
−1
. (4.20)
2
k=0 |sref (k)|
Here, e(k) = s(k) − sref (k) is the complex error sequence with s(k) and sref (k)
being the length-Ns measured and reference complex sequences, respectively.
9 Ifthe application allows, one can further save the arithmetic complexity by only computing
the real part of the result at the output side of Fig. 4.9.
10 For small efficiencies, the complexity of H
HB,c (z) is not significant and both the alternatives
have roughly similar complexities. If the complexity of the FFBR network reduces, the superiority
of Alternative II will further be pronounced.
11 To get realizable filters, a transition band is needed and, hence, the efficiency of 100%, refers
ǫ
to that achieved by k = 0 in (4.19) which becomes 1 − Q .
52
N=20
1
Complexity Ratio
0.8
0.6
0.4
0.2
0.8
10
0.6 8
0.4 6
4 −3
x 10
0.2 2
ε δs
Figure 4.17: The ratio between the number of real operations in one time unit for
the alternatives of Figs. 4.7 and 4.9 with different values of δs and ǫ.
Besides the characteristics of P (z), Alternative I uses two complex half-band

filters whose characteristics further affect the EVM. In this example, the filters were
designed so that both the alternatives result in roughly equal EVMs. The resulting
EVMs for the user X0 in Fig. 4.11, after being processed by Alternatives I and
II are −29.60dB and −30.44dB, respectively. The characteristics of the designed
P (z), Hk (z) and HHB,c (z) are shown in Fig. 4.22.
If another EVM is desired for Alternative I, both P (z) and HHB,c (z) should be
redesigned. On the other hand, Alternative II will only require redesign of P (z).
The performance of Alternative I is determined by both HHB,c (z) and P (z). If
P (z) is designed so as to approximate PFBR as close as desired, the ripples of
HHB,c (z) must be roughly of the same12 size as those of P (z). Otherwise, the
overall EVM will be degraded. Assuming S = 421 and NH = 154, the arithmetic
complexity of Alternative II is about 63% of that of Alternative I. It is desirable
to design HHB,c (z) and P (z) jointly so as to achieve the same performance with
smaller orders. Irrespective of the design method, Alternative I will always require
extra arithmetic complexity due to HHB,c (z).
Due to the use of the same filter design method, i.e., minimax, in both alter-
natives, it is appropriate to compare these structures in terms of the arithmetic
12 This may even require a larger NH than that estimated by (4.18).
53
N=50
0.8
Complexity Ratio
0.6
0.4
0.2
0.8
10
0.6 8
0.4 6
4 −3
x 10
0.2 2
ε δs
complexity with roughly equal requirements on performance, i.e., EVM. The other
comparison which focuses on performance rather than the arithmetic complexity,
is more appropriate if different filter design methods are used.
4.6 Concluding Remarks

The next subsections will discuss some issues that were not treated earlier.
4.6.1 Measure of Complexity

This chapter uses the number of real arithmetic operations and real multiplications
rather than complex multiplications are used. Multiplications by specific complex
numbers may require less real multiplications than a general complex multiplica-
tion. Consequently, the real operations give a more accurate measure of arithmetic
complexity [151]. When converting complex operations to real operations, either
of (3/3) or (4/2) assumptions does not alter the total number of real operations
but the former (latter) results in less multipliers (adders).
As in Section 4.3.4, this chapter does not consider the division by N in the
arithmetic complexity of the FFBR networks. Doing so, Alternative II becomes
54
N=100
0.8
0.6
Complexity Ratio
0.4
0.2
0.8
10
0.6 8
0.4 6
4 x 10
−3
0.2 2
ε δs
1 1
0.5 0.5
Quadrature
Quadrature
0 0
−0.5 −0.5
−1 −1
−1 −0.5 0 0.5 1 −1 −0.5 0 0.5 1
In−Phase In−Phase
Figure 4.20: Alternative I. Figure 4.21: Alternative II.
55
(a)
0
Mag. [dB]
−20
−40
0 0.2π 0.4π 0.6π 0.8π π
ωT [rad]
(b)
0
Mag. [dB]
−20
−40
0 0.2π 0.4π 0.6π 0.8π π 1.2π 1.4π 1.6π 1.8π 2π
ωT [rad]
(c)
0
Mag. [dB]
−20
−40
−π −0.8π −0.6π −0.4π −0.2π 0 0.2π 0.4π 0.6π 0.8π π
ωT [rad]
Figure 4.22: Characteristics of (a) P (z). (b) Hk (z). (c) HHB,c (z).
more superior and the upper bound in Fig. 4.16(b) hovers around 0.5 instead of
1.4.
4.6.2 Applicability of Alternatives I and II

The hybrid, full processing, and partial processing payloads need complex data in
the subsequent stages of the on-board signal processing. Then, the only choice will
be Alternative I so that the intermediate complex signals can be used for necessary
processing, e.g., modulation/demodulation, coding/decoding, etc. Alternative II
can thus be used if the need for intermediate complex signals can be eliminated.
4.6.3 Filter Bank Design

To design the FB for the real FFBR network, the method of [42] can be used.
In contrast to Alternative I, Alternative II uses half of the spectrum of a complex
signal. Due to this, a worst-case degradation of 6dB in the EVM may be introduced
which can be overcome by halving the approximation error.
56
5
A Multimode Transmultiplexer
Structure
This chapter introduces a multimode transmultiplexer (TMUX) which can generate

a large set of bandwidths and center frequencies. It utilizes fixed integer sampling
rate conversion (SRC), Farrow-based variable rational SRC, and variable frequency
shifters. Following an introduction in Section 5.1, Section 5.2 outlines the problem
for the TMUX. Then, Section 5.3 discusses the building blocks and the TMUX
operation whereas Section 5.4 deals with the filter design. After a discussion on
implementation complexity in Section 5.5, two applications of the proposed TMUX
are covered in Section 5.6. In Section 5.7, the TMUX is analyzed using conventional
multirate building blocks. Finally, Section 5.8 concludes the chapter.
5.1 Introduction
As discussed in Section 1.1, a current focus in the communications area is to de-
velop flexible radio systems which seamlessly support services across several radio
standards [1–10]. A major research topic in this area is to cost-efficiently imple-
ment multimode transceivers. The simplest approach for multimode problems is to
use a custom device for each communication mode [8]. With the growing number
of communication standards, this approach is becoming inefficient. Thus, it is vital
to develop new low-cost multimode terminals.
A TMUX allows several users to share a common channel and multimode
57
5. A MULTIMODE TRANSMULTIPLEXER STRUCTURE
Granularity Band (BGRB) Guardband (GB)

(a) 2p/Q
1BGRB
0 2pa/Q 2p/Q+2pa/Q 2p-2p/Q 2p wT
(b) +2pa/Q
1.5BGRB 1.75BGRB 2.2BGRB 1BGRB

0 2p wT
(c)
3.2BGRB 2.7BGRB
0 2p wT
(d)
5.9BGRB 1BGRB
0 2p wT
Figure 5.1: Formulation of the GRBs and examples of different user signals having
arbitrary bandwidths.
communications require multimode TMUXs that support different (time-varying)

bandwidths. In other words, the users can request any bandwidth at any time. As
discussed in Section 3.4, a TMUX is the dual of a filter bank (FB) whose synthesis
FB (SFB) and analysis FB (AFB) consist of a parallel connection of a number of
branches. Each branch is realized by digital bandpass interpolators and decima-
tors. Multimode TMUXs thus require interpolators and decimators with variable
parameters. These blocks can be constructed using variable upsamplers (downsam-
plers) and bandpass filters with variable center frequencies and bandwidths. With
conventional interpolators and decimators discussed in Section 3.1, the implemen-
tation cost of such multimode TMUXs grows proportional to the level of flexibility
in center frequencies and bandwidths.
5.2 Problem Formulation
Similar to Fig. 1.1, we assume that the whole frequency spectrum is occupied by
a number of users having different bandwidths and center frequencies. According
to Fig. 5.1, one can divide the whole frequency band into Q granularity bands
(GRBs) separated by a guardband (GB) of 2πǫ Q where α is as in (4.5). Any user p
can occupy any rational number of GRBs. This chapter models the input patterns
on which the flexible frequency-band reallocation (FFBR) network in Chapter 4 is
operating. For this purpose and in accordance with [42], the values of the GB and
the GRB are chosen as 2πǫ 2π 2πǫ
Q and Q − Q where 0≤ǫ≤1. One can generally select
any value for the GBs and the GRBs according to the system requirements.
58
jw0n ^ n
-jw
e e 0
x0(n0) L F(z) H0(z) H0(z) ^

F(z) L x^ (n0)
0
jw1n ^ n
-jw
e e 1
^
y(n) y(n)
x1(n1) L F(z) H1(z) H1(z) ^
F(z) L ^x (n1)
1
^
ejwP-1n e-jwP-1n
xP-1(nP-1) L F(z) HP-1(z) HP-1(z) ^
F(z) L ^x (nP-1)
P-1
Figure 5.2: Proposed multimode TMUX consisting of fixed integer SRC, variable
rational SRC, and variable frequency shifters. Here, Hp↓ (z) and Hp↑ (z) represent
Farrow-based filters for decimation and interpolation, respectively.
5.3 Multimode TMUX Structure

This section introduces a multimode TMUX, shown in Fig. 5.2, which can generate
arbitrary1 bandwidths and center frequencies. A GRB is the minimum bandwidth
a user can occupy. The users are separated by GBs of ∆≥0 so that the TMUX
is slightly redundant. We mostly consider ∆ = 0. As discussed in Section 3.4.3,
the proposed multimode TMUX requires a small redundancy so as to generate all
possible modes without channel interference and filter redesign. Redundancy may
be of particular interest if there exists severe channel distortion in some frequency
bands [20]. We also assume that any user p can occupy a rational Rp (t) number of
GRBs2 where 1 ≤ Rp (t) ≤ Q.
The TMUX generates a GRB through interpolation by an integer L. To get
bandwidths which are rational multiples of the GRB, the Farrow-based filter per-
forms decimation by rational ratios Rp . As discussed in Section 3.2, for interpola-
tion (decimation) one can use the Farrow (transposed Farrow) structure. To place
the users in appropriate positions in the frequency spectrum, variable frequency
shifters are utilized. In the AFB, the received signal is first frequency shifted such
that the desired signal can be processed in the baseband. Then, a Farrow-based in-
terpolator (by ratio Rp ) followed by decimation by an integer L obtains the desired
signal. Figure 5.3 plots the frequency spectrum at the output of different blocks
with a uniformly distributed random input.
5.3.1 Channel Sampling Rates

In multimode systems, the users X0 , X1 , . . . , XP −1 , can generally have different
data rates. This means that in one time frame, the number of processed samples
1 The bandwidths and center frequencies are in practice limited to rational numbers due to
finite precision. At the expense of additional implementation cost, any precision can be achieved.
2 The value of R (t) is constant during the time frame in which the user signal p is transmitted.
p
In the sequel, the time index t will be omitted.
59
(a)
20
Mag.
0
−20
−40
−60
0 0.25π 0.5π 0.75π π 1.25π 1.5π 1.75π 2π
ωT [rad]
1
(b)
20
Mag.
0
−20
−40
−60
0 0.25π 0.5π 0.75π π 1.25π 1.5π 1.75π 2π
ωT2 [rad]
(c)
20
Mag.
0
−20
−40
−60
0 0.25π 0.5π 0.75π π 1.25π 1.5π 1.75π 2π
ωT2 [rad]
(d)
20
Mag.
0
−20
−40
−60
0 0.25π 0.5π 0.75π π 1.25π 1.5π 1.75π 2π
ωT2 [rad]
(e)
20
Mag.
0
−20
−40
−60
0 0.25π 0.5π 0.75π π 1.25π 1.5π 1.75π 2π
ωT [rad]
1
Figure 5.3: (a) Output of F (z). (b) Output of Hp↓ (z). (c) and (d) Outputs of
frequency shifters. (e) Output of Hp↑ (z). L = 12, Rp = 2.5, ωp = 0.3π, ω̂p = 0.3π,
and T2 = Rp T1 .
in each branch of Fig. 5.2 can be different from the others. Mathematically, the
sampling periods of the TMUX inputs, i.e., T0 , T1 , . . . , TP −1 , must satisfy
T0 R0 = T1 R1 = . . . = TP −1 RP −1 = LTy , (5.1)
where T is the sampling period of y(n). In the proposed redundant TMUX,

PP −1 Rpy PP −1 Rp
p=0 L < 1 which is in contrast to a critically sampled TMUX with p=0 L =
1.
5.3.2 Sampling Rate Conversion

As shown in Fig. 5.2, integer interpolation and decimation by L require lowpass
filters F (z) and F̂ (z), respectively. The stopband and passband edges of these filters
are defined as in (2.33). This also sets the value of the GRB as 2π(1+ρ) L . Further,
SRC by a rational value Rp is performed by the Farrow-based filters resulting in
the set of bandwidths 2π(1+ρ) L Rp , p = 0, 1, . . . , P − 1. Each of the systems Hp↓ (z)
60
and Hp↑ (z) employs a filter with a transfer function given by (3.10) and it performs
SRC by a variable rational ratio Rp . The values of µ are given by (3.12). Further
details can be found in Section 3.2.
5.3.3 Subcarrier Frequencies

According to Section 5.2, the input signal is divided into a number of GRBs sepa-
rated by a GB of 2πǫQ . Having generated the user signals, they must be modulated
into specific locations in the frequency spectrum. To avoid the inter-carrier inter-
ference (ICI), the user signals should not overlap. Consequently, for user k
χ0
(
2 if p = 0
ωk = Pk−1 χk (5.2)
p=0 χp + 2 if p6=0.
Here, χp = ⌈Rp ⌉ 2π
Q , p = 0, 1, . . . , k, and ⌈x⌉ is the ceiling of x. The ceiling
operation ensures that users do not share3 a GRB. In general, the bandwidths
are time-varying and only the GRB is fixed. With no frequency-band reallocation
(FBR), we have ω̂k = ωk . In case of FBR, ω̂k becomes
ω̂k = ωk + Γ(t, ωk ) (5.3)
where Γ(t, ωk ) is a time-varying function expressing the FBR or frequency multi-

plexing or any frequency offset.
5.4 Filter Design

Due to redundancy and according to Section 3.4.3, the ICI is determined by the
stopband attenuation of F (z) and F̂ (z). In each branch and ignoring Hp↓ (z) and
Hp↑ (z), the transfer function from xp (np ) to x̂p (np ) is the zeroth polyphase com-
ponent of F (z)F̂ (z) [12]. This polyphase component controls the inter-symbol
interference (ISI). To make this polyphase component unity4 , F (z)F̂ (z) must be
an Lth-band filter. Thus, F (z) and F̂ (z) are the spectral factors of an Lth-band
filter (refer to Section 2.4.3). This also holds when Hp↓ (z) and Hp↑ (z) are present
provided that they approximate a fractional delay filter with a delay µ throughout
their respective frequency bands. For Hp↓ (z), only the GRB needs to be covered
whereas in the AFB, the whole band except for a small band near π must be cov-
ered. The reason is that the output of F (z) is bandlimited to the GRB. However,
in the AFB, y(n) is processed by Hp↑ (z). Consequently, the complexity of Hp↓ (z)
will be less than that of Hp↑ (z). We thus need to determine F (z) and F̂ (z) such
3 This is specific for the FFBR network in Chapter 4.

4 For causal filters, one of the polyphase components should be a pure delay.
61
1
Magnitude 0.8
0.6
0.4
0.2
0
0 0.1π 0.2π 0.3π 0.4π 0.5π 0.6π 0.7π 0.8π 0.9π π
ωT [rad]
0.01
0.005
Error
−0.005
−0.01
0 0.1π 0.2π 0.3π 0.4π 0.5π 0.6π 0.7π 0.8π 0.9π
ωT [rad]
Figure 5.4: Magnitude response and approximation error of the Farrow structure
designed by (5.5) with δ4 = 0.01 and ω2 T = 0.9π. Here, 6 subfilters with orders
{24, 24, 8, 20, 6, 8} have been used.
that
|[F (ejωT )F̂ (ejωT )]zeroth − 1| ≤ δ1 , ωT ∈ [0, π],

|F (ejωT )| ≤ δ2 , ωT ∈ [ω1 T, π],
|F̂ (ejωT )| ≤ δ3 , ωT ∈ [ω1 T, π], (5.4)
where ω1 T = π(1+ρ)
L , and [F (ejωT )F̂ (ejωT )]zeroth denotes the zeroth polyphase
component of F (z)F̂ (z). The Farrow-based filter H(z, µ) should be designed such
that
|H(ejωT , µ) − e−jωT µ | ≤ δ4 , ωT ∈ [0, ω2 T ], µ∈[−0.5, 0.5]. (5.5)
Additionally, ω2 T = ω1 T for the SFB. In the AFB, the spectral width of y(n), in
Fig. 5.2, determines ω2 T . For example, at a typical spectrum utilization percentage
of 90%, we have ω2 T = 0.9π.
All δi , i = 1, 2, 3, 4, in (5.4) and (5.5) can be reduced to any desired level by
simply increasing the filter order. To design the TMUX, one should
• solve (5.4) to get F (z) and F̂ (z).
• solve (5.5) to get the Farrow subfilters in the SFB and AFB.
62
1
Magnitude 0.8
0.6
0.4
0.2
0
0 0.1π 0.2π 0.3π 0.4π 0.5π 0.6π 0.7π 0.8π 0.9π π
ωT [rad]
−3
x 10
1
0.5
Error
−0.5
−1
0 0.1π 0.2π 0.3π 0.4π 0.5π 0.6π 0.7π 0.8π 0.9π
ωT [rad]
Figure 5.5: Magnitude response and approximation error of the Farrow structure
designed by (5.5) with δ4 = 0.001 and ω2 T = 0.9π. Here, 7 subfilters with orders
{38, 38, 18, 32, 14, 20, 6} have been used.
Having solved these problems once offline, only µ and the variable frequency shifters
change online. The filter pair F (z) and F̂ (z) can be designed as outlined in, e.g.,
[77, 130–138], whereas the Farrow-based filters may be designed as described in,
e.g., [94, 95, 97–104]. The proposed multimode TMUX can thus be designed to
approximate perfect reconstruction (PR) as close as desired for all possible modes
by separately solving three conventional filter design problems offline.
5.4.1 Example
A series of filters with δ1 = δ4 = {0.01, 0.001}, ω1 T = 0.0875π, ω2 T = 0.9π, and
L = 12 are assumed. As the stopband attenuation of F (z) and F̂ (z) suppresses the
ICI, they have been designed with different values of δ2 = δ3 . With fixed δ1 , there
are similar constraints on [F (ejωT )F̂ (ejωT )]zeroth and H(ejωT , µ). The stopband
attenuation of F (z) and F̂ (z) is then the only parameter which changes. Figures 5.4
and 5.5 show the magnitude response and the approximation error of the Farrow
structure for µ = {0, 0.05, 0.1, 0.15, . . . , 0.5}. These structures have been designed
to approximate allpass transfer functions in the frequency band ωT ∈[0, 0.9π] and
the resulting values for δ4 are, respectively, 0.01 and 0.001.
The magnitude responses at the stopband of F (z) = F̂ (z) and the passband of
[F (ejωT )F̂ (ejωT )]zeroth for some of the designed filters are also shown in Figs. 5.6
63
−30
Magnitude [dB] −40
−50
−60
−70
0.09π 0.3π 0.5π 0.7π 0.9π π
ωT [rad]
0.01
0.005
Magnitude
−0.005
−0.01
0.2π 0.4π 0.6π 0.8π π
ωT [rad]
Figure 5.6: Magnitude response in the stopband (top) of F (z) = F̂ (z) and the
passband (bottom) of [F (ejωT )F̂ (ejωT )]zeroth for δ1 = 0.01 and L = 12.
and 5.7. For all of these filters, the values of δ1 are, respectively, 0.01 and 0.001
but they have different stopband attenuations. Figure 5.8 shows the average error
vector magnitude (EVM), discussed in Section 4.5.3, for three multimode setups
in a 16-quadrature amplitude modulation (QAM) signal. The values of δ1 = δ4 set
a lower bound on the EVM. However, the EVM can be decreased to any level and
for all possible modes by decreasing δi , i = 1, 2, 3, 4. Figure 5.9 plots the frequency
spectrum of y(n) for these multimode setups. As can be seen, ∆ = 0.
5.5 Implementation and Design Complexity Issues

The previous section showed that the proposed TMUX can be designed to have as
small errors as desired for all possible modes through three separate offline filter
designs. This is attractive compared to solutions that require either one set of
filters for each mode or online filter design. There is still room for complexity
reductions by modifying the proposed structure. This section points out some of
these.
A motivation to using integer interpolation to generate a GRB is that regular
integer interpolation structures are more efficient than Farrow-based structures
when implementing an interpolator with a relatively large conversion ratio L [152].
64
−30
Magnitude [dB] −40
−50
−60
−70
0.09π 0.3π 0.5π 0.7π 0.9π π
ωT [rad]
−3
x 10
1
0.5
Magnitude
−0.5
−1
0.2π 0.4π 0.6π 0.8π π
ωT [rad]
Figure 5.7: Magnitude response in the stopband (top) of F (z) = F̂ (z) and the
passband (bottom) of [F (ejωT )F̂ (ejωT )]zeroth for δ1 = 0.001 and L = 12.
This is true if multi-stage structures [77] are utilized which should indeed be applied
for large L. If the bandwidth of the users often matches the GRB, this option
(and the dual in the AFB) appears as the most natural choice. If the users often
occupy wider bandwidths than the GRB, it may then be worth to use a smaller
L. The Farrow-based filters in the SFB and AFB can then both work either as an
interpolator or as a decimator. This allows us to find the best trade-off between
the complexities of the integer and rational SRC parts. Some results are available
for interpolators and decimators [152] but the problem is more complex here as we
deal with TMUXs. The overall optimum will depend on how often the users take
on narrow or wide bandwidths.
Another issue is the filter design. The previous section outlined the separate
filter design which is attractive as known techniques can be adopted. Although this
gives a good suboptimum overall solution, it is slightly overdesigned. To reduce
the complexity, one can design all filters simultaneously which can, in principle,
be done using standard nonlinear optimization techniques. This has successfully
been used for fixed FBs and TMUXs [79] but the problem is much more complex
here as we deal with multimode TMUXs. This implies that the requirements must
be satisfied for all possible modes. Consequently, the number of constraints grows
with the number of modes. Simultaneous optimization may therefore be practically
feasible only for problems that have a few modes.
65
(a) δ1=δ4=0.01
Setup 1: R ={1.3,2.5,3.7}
−30 p
EVM [dB]
Setup 2: R ={1.1,1.9,4.1}
p
−35 Setup 3: R ={1.75,2.7,5.1}

p
−40
40 50 60 70 80 90
Stopband attenuation [dB]
(a) δ1=δ4=0.001
−30
EVM [dB]
−40
−50
40 45 50 55 60 65 70 75 80 85
Stopband attenuation [dB]
Figure 5.8: The resulting EVM for some multimode setups at different stopband at-
tenuations of F (z)F̂ (z) with fixed errors in [F (ejωT )F̂ (ejωT )]zeroth and H(ejωT , µ).
5.6 TMUX Application
This section considers two applications of the proposed TMUX. Having designed
the TMUX for an EVM of −100dB, achieved with δi = 10−5 , i = 1, 2, 3, 4, the
structure in Fig. 5.10 can be used for functionality/performance test of the FFBR
network in Chapter 4. The values for the GRB and GB are chosen as 2π Q − Q
2πǫ
2πǫ
and Q where 0≤ǫ≤1. To verify the functionality, four different user signals
{X0 , X1 , X2 , X3 } with Rp = {1.75, 1.25, 2, 3.5} and ωp = {0.2π, 0.6π, π, 1.6π} are
assumed. The frequency spectrum
P of the input and the multiplexed output of the
FFBR network with Q = p ⌈R p = 10 GRBs are shown in Figs. 5.12(a) and
⌉
5.12(b). The scenario of FBR in Fig. 5.12(b), results in ω̂p = {π, 1.8π, 1.4π, 0.4π}.
To illustrate the performance, Fig. 5.12(c) shows the values of EVM for different
stopband attenuations of P (z) in (4.4) and with a 16-QAM signal. The stopband
attenuation of P (z) in (4.4) mainly determines the error of the FFBR network [42].
66
(a)
0
Mag. [dB]
−20
−40
−60
0 0.25π 0.5π 0.75π π 1.25π
ωT [rad]
(b)
0
Mag. [dB]
−20
−40
−60
0 0.25π 0.5π 0.75π π 1.25π
ωT [rad]
(c)
0
Mag. [dB]
−20
−40
−60
0 0.25π 0.5π 0.75π π 1.25π 1.5π 1.75π
ωT [rad]
Figure 5.9: Spectrum of y(n) for multimode setups of Fig. 5.8. (a) Setup 1. (b)
Setup 2. (c) Setup 3.
x0(n0) ^x (n0)
0
x1(n1) Synthesis FFBR Analysis ^x (n1)
1
FB Network FB
xP-1(nP-1) x^P-1(nP-1)
Figure 5.10: Setup for functionality/performance test.
L-th Band Filter Granularity Band (BGRB)=2p(1+r)/L Guardband

2p/Q
1BGRB
0 2pa/Q 2p/Q+2pa/Q 2p-2p/Q 2p wT

+2pa/Q
2π 2πǫ 2π(1+ρ)
Figure 5.11: The GB, the GRB, and the L-th band filter. As Q − Q = L ,
we can have ρ≤ L(1−ǫ)
Q − 1.
67
(a)
0
Mag.
−20
−40 X X X X
0 1 2 3
−60
−80
0 0.25π 0.5π 0.75π π 1.25π 1.5π 1.75π 2π
ωT [rad]
(b)
0
Mag.
−20
−40 X3 X0 X2 X1
−60
−80
0 0.25π 0.5π 0.75π π 1.25π 1.5π 1.75π 2π
ωT [rad]
(c)
EVM [dB]
−20
−40
40 45 50 55 60 65 70
Stopband attenutation of P(z) [dB]
Figure 5.12: (a) and (b) Functionality test of an FFBR network. (c) Performance
test of an FFBR network. L = 12, Q = 10, ǫ = 0.125.
^ n
-jw
e jwmn e p
xm(nm) L Fm(z) Bm Cm(z) Am Ap D (z)

p
Bp F^p(z) L x^p(np)
Farrow Decimator Farrow Interpolator
Figure 5.13: Equivalent path between xm (nm ) and x̂p (np ).
5.7 Analysis Using Multirate Building Blocks

This section describes the proposed TMUX in terms of conventional multirate
building blocks [18]. This is done by utilizing the rational SRC equivalent of the
Farrow-based filter [153]. In each branch, the Farrow-based filter for interpolation
A
by Rp = Bpp > 1 can be modeled as the cascade of upsampling by Ap , the finite-
length impulse response (FIR) filter Dp (z), and downsampling by Bp . Similarly,
a cascade of upsampling by Bp , the FIR filter Cp (z), and downsampling by Ap
A
can be used to model decimation by Rp = Bpp > 1. Consequently, each branch
can be modeled as in Fig. 5.13. Considering the frequency shifters, Fig. 5.13 can
be redrawn as in Fig. 5.14 where zm = ejωm T and ẑp = ej ω̂p T . This structure is
similar to [19] with some differences as:
1. The nonuniform TMUX in [19] does not utilize frequency shifters and the
68
xm(nm) L Fm(z) Bm Cm(z) Am z (z^p/zm)z Ap D (z)

p
Bp F^p(z) L x^p(np)
Farrow Decimator Farrow Interpolator
Figure 5.14: Equivalent path between xm (np ) and x̂m (np ) considering the effects
of frequency shifters in the frequency domain.
^ Bm
xm(nm) LBm F(zBm)Cm(z) Am Am F(z )Dm(z) LBm x^m(nm)
Figure 5.15: Simplified equivalent path between xm (nm ) and x̂m (nm ).
ẑ
term zmp z does not appear in the formulations. As the filters of Fig. 5.2 are
lowpass and the frequency shifts of a filter do not alter the characteristics
ẑ
of its baseband equivalent, the term zmp z does not affect the analysis of the
TMUX. Therefore, similar analysis as that in [19] can be used for the present
TMUX.
2. Instead of single SFB and AFB filters in [19], the present TMUX uses the
cascade of a periodic filter, i.e., Fm (z Bm ) or F̂p (z Bp ), and the FIR equivalent
of the Farrow structure, i.e, Cm (z) or Dp (z).
Note also that instead of designing bandpass filters, the proposed TMUX designs
lowpass filters. Then, frequency shifters modulate the user signals. This not only
gives bandpass characteristics but also increases the reconfigurability regarding the
center frequencies. To clarify the second difference, Fig. 5.14 is redrawn in Fig. 5.15
where p = m and Bp = Bm = B. Figure 5.16 shows the transfer function of the
two cascaded filters, i.e., Fm (z B )Cm (z) or Dm (z)F̂m (z B ). This cascaded filter is
lowpass with passband and stopband edges as
π(1 − ρ)
ωc T = , (5.6)
LB
2π π(1 + ρ)
ωs T = − . (5.7)
B LB
From [19], the blocked transfer function of Fig. 5.2 is T (z) = Φ(z)Ψ(z) where
T
Φ(z) = φ0 (z) φ1 (z) ... φP −1 (z) ,

Ψ(z) = ψ0 (z) ψ1 (z) ... ψP −1 (z) . (5.8)
For the proposed TMUX,
φp (z) = Dp (z)F̂p (z Bp ),
ẑp ẑp
ψp (z) = Fm (( z)Bm )Cm ( z). (5.9)
zm zm
69
(1+r)p/(LB) 2p-(1-r)p/(LB)
2p/B-(1-r)p/(LB) 2p/B+(1-r)p/(LB) 4p/B-(1-r)p/(LB)
wT
(1-r)p/(LB) 2p/B-(1+r)p/(LB)
2p/B
2p/B+(1+r)p/(LB)
4p/B 2p-(1+r)p/(LB) 2p
4p/B-(1+r)p/(LB)
Figure 5.16: The cascade of a periodic filter and the FIR equivalent of the Farrow
structure. The dashed line is the FIR equivalent of the Farrow structure, i.e., Cm (z)
or Dm (z), whereas the solid line is the periodic filter, i.e., Fm (z B ) or F̂m (z B ).
To further simplify this, assume Fm (z) = F (z) and F̂m (z) = F̂ (z) for m =
0, 1, . . . , P − 1. If p = m, then ẑp = zm and (5.9) becomes
φp (z) = Dp (z)F̂ (z Bp ) (5.10)

ψp (z) = F (z Bp )Cp (z). (5.11)
The values of the ICI and ISI are mainly determined by the expressions in (5.8)
ẑ
which themselves depend on the ratio zmp in (5.9). For the desired signal X̂d (z),
the relation m = p = d holds. Then, ωd = ω̂d = ωd also holds and the system
can approximate PR as close as desired via a proper design of the SFB and AFB
filters. If m 6= p (or equivalently, ωm 6= ω̂p ), the signals are considered as undesired
and will be attenuated by the AFB filters. The reason is that ωk in (5.2) ensures
no overlap of the user signals. Therefore, the undesired signals which have passed
ẑ ẑ
through the SFB filter Fm (( zmp z)Bm )Cm ( zmp z), will fall in the stopband of the
AFB filter Dp (z)F̂p (z Bp ) and be attenuated. The amount of the ICI and ISI is
determined by the cascade of these filters.
In conclusion, if the TMUX is designed for a worst-case error δw , it can produce
arbitrary bandwidths while approximating PR with smaller errors than δw . Then,
we need a fixed set of filters and the only parameter to support arbitrary band-
widths is the fractional delay of the Farrow structure. Although the same analysis
methods as for the existing TMUXs can be used here, the implementation is differ-
ent. The conventional rational SRC blocks (upsamplers, downsamplers, and filters)
are only used for modeling whereas the TMUX implements these blocks implicitly
using integer and rational SRC.
5.8 Conclusion
This chapter introduced a multimode TMUX capable of generating a large set of
bandwidths and center frequencies. The TMUX utilizes fixed integer SRC, Farrow-
based variable rational SRC, and variable frequency shifters. It needs only one
offline filter design beforehand. Then, all possible combinations of bandwidths and
center frequencies are easily obtained by online adjustment of (i) the variable delay
70
parameters of the Farrow-based filters, and (ii) the variable parameters of the fre-
quency shifters. Design examples were provided for illustration. Furthermore, the
TMUX was described in terms of conventional multirate building blocks allowing
one to use the design techniques based on the blocked transfer function.
71
72
6
A Class of Multimode
Transmultiplexers Based on the
Farrow Structure
This chapter introduces a class of multimode transmultiplexers (TMUXs) in which

the Farrow structure realizes the polyphase components of general lowpass filters.
After a brief introduction, Section 6.2 gives some prerequisites whereas Section 6.3
considers the design of approximately Nyquist filters. Further, a TMUX capable
of performing integer sampling rate conversion (SRC) is proposed. Extending the
integer SRC to rational SRC, Section 6.4 introduces a multimode TMUX which
can generate arbitrary bandwidths. The TMUX performance is investigated in
Section 6.5 and some concluding remarks are given in Section 6.7.
6.1 Introduction
As discussed in Section 1.1, multimode communications require to support different
bandwidths, resulting from various telecommunication standards, e.g., global sys-
tem for mobile communications (GSM), interim standard-54/136 (IS-54/36), and
IS-95 [11]. To include these standards in a general telecommunication system, we
should handle a number of different bandwidths. Thus, multimode communica-
tions require that several users, with different bit rates, share a common channel.
A TMUX allows different users to share a common channel and multimode TMUXs
thus constitute one of the main building blocks in multistandard communications
[121]. With bandwidth-on-demand, the TMUX characteristics must vary with
73
6. A CLASS OF MULTIMODE TRANSMULTIPLEXERS BASED ON THE
FARROW STRUCTURE
time. This supports a dynamic allocation of bandwidth so that each user occupies
a time-varying portion of the channel.

TMUXs, as duals of filter banks (FBs), are composed of a synthesis FB (SFB)
followed by an analysis FB (AFB). The SFB and AFB consist of a parallel con-
nection of a number of branches [12]. Each branch is realized by digital inter-
polators/decimators. In uniform TMUXs, the bandwidths and center frequencies
of these interpolators/decimators are fixed. Multimode TMUXs require interpo-
lators/decimators with variable parameters. In other words, we need nonuniform
TMUXs and FBs which can modify their characteristics in a time-varying manner.
With fixed traditional TMUXs and FBs, e.g., [16–31], a set of filters, designed for a
FB, can be adopted for a TMUX [12]. This does not apply to multimode TMUXs
though. The use of conventional TMUXs, in multimode communications, requires
either predesign of different filters or online filter design. This becomes complicated
when supporting dynamic communications.
Chapter 5 introduced a multimode TMUX consisting of fixed integer SRC and
Farrow-based variable rational SRC. It needs three sets of filters and the Farrow
structure in the AFB has a higher arithmetic complexity. This chapter introduces
an alternative method to design approximately Nyquist filters where the Farrow
structure realizes the polyphase components of general lowpass filters. The zeroth
polyphase component is a Type I linear-phase finite-length impulse response (FIR)
filter whereas the remaining polyphase components are realized by the Farrow
structure. Specific constraints are imposed on these lowpass filters such that they
are approximately power complementary. The power complementary property al-
lows us to obtain integer SRC multimode TMUXs. Incorporating the integer SRC,
in a general rational SRC structure, a rational SRC multimode TMUX is then
constructed. Different filter design techniques are considered and compared. In
contrast to the TMUX in Chapter 5, the Farrow structures in the SFB and AFB
of the present chapter have equal complexities. The main advantage of the pro-
posed TMUXs is to support dynamic communications with reasonable design effort
and an offline filter design. The proposed TMUXs are near perfect reconstruction
(NPR). We do not cancel the inter-carrier interference (ICI) and inter-symbol in-
terference (ISI). Instead, we suppress them by a proper filter design.
6.2 Prerequisites
This section discusses some prerequisites and general issues.
6.2.1 Problem Formulation

As in Fig. 1.1, we assume the whole frequency spectrum to be shared by P users.
Ap
Each user has a bandwidth of π(1+ρ)
Rp , p = 0, 1, . . . , P −1, and Rp = Bp with integers
74
FARROW STRUCTURE
Ap and Bp . Furthermore, ρ is the roll-off factor and a guardband (GB) of ∆≥0

separates the users. This chapter assumes ∆ = 0 to allow a minimum GB between
the users. The choice of ∆ does not affect the filter design problem. Generally, P
can vary with time. For a discussion independent of time, the sequel assumes that
in each time frame, P is time-invariant. Therefore, in each time frame, the TMUX
operates as a fixed nonuniform TMUX.
Assume α time frames (modes) and the delay of the TMUX being Nd . When
changing P , a new packet of data, containing Ns samples, is sent. Due to transients,
the receiver first receives Nd unreliable samples. The ratio between the number of
reliable samples and the total number of samples gives the efficiency as
αNs 1
η= = α−1 Nd
. (6.1)
αNs + (α − 1)Nd 1 + α Ns
Here, the term α−1α is asymptotically equal to unity. Then, choosing Ns > Nd
increases η and the effect of Ns is more pronounced than that of α. Thus, for every
Nd , we can choose a proper Ns so as not to degrade η. Note that the proposed
TMUXs have linear-phase FIR filters and Nd is known [152].
6.2.2 Some General Issues

Some general issues need to be outlined. First, to avoid ICI, the users do not
overlap and the TMUXs are slightly redundant. Redundancy simplifies the filter
design as discussed in Section 3.4.3. It also enables one to avoid infeasible cases
[20, 22, 24, 26, 27].
Second, we consider packets of data separated by time slots. This is similar
to time division multiple access systems like, e.g., GSM. The TMUX parameters
change only when switching from one mode to another. For each mode, the TMUX
is fixed and it works as any regular fixed TMUX. In other words, the proposed
TMUX realizes a (large) number of fixed TMUXs. The switching between them is
done by changing a few multipliers (in the Farrow structure) and the SRC ratios.
Thus, for each and every mode, we should ensure that the ISI and ICI are small,
as in any regular fixed TMUX. Transients are only present when switching from
one mode to another. An alternative to using the proposed TMUXs is to use a set
of different and separately designed fixed TMUXs. When changing from one mode
to another, we then change one fixed TMUX to another, and we are facing exactly
the same problem regarding transients. Note that [67, 68] solves the problem with
transients but it does not provide a full flexibility as compared to the proposed
TMUXs. A thorough discussion on the differences and similarities of [67, 68] and
the proposed TMUXs can be found in [68] and Chapter 7.
Third, this chapter uses the Farrow structure as outlined in Section 3.2. Here,
Sk (z) are designed to obtain a fractional delay filter, i.e., H(z) = z −µ . Although we
have linear-phase FIR filters Sk (z), we use the term the Farrow structure instead
of the modified Farrow structure [94]. The Farrow structure and the transposed
Farrow structure [3, 95] are used, respectively, for interpolation and decimation, as
seen later in Figs. 6.4, 6.5, 6.14, and 6.15.
75
FARROW STRUCTURE
jw0n ^ n
-jw
e e 0
x0(n0) R0 G0(z) ^ (z)

G R0 x^0(n0)
0
jw1n ^ n
-jw
e e 1
^
y(n) y(n)
x1(n1) R1 G1(z) ^ (z)
G R1 ^x (n1)
1 1
jwP-1n ^ n
-jw
e e P-1
xP-1(nP-1) RP-1 G (z) ^ (z)

G RP-1 ^xP-1(nP-1)
P-1 P-1
Figure 6.1: Integer SRC multimode TMUX composed of variable integer SRCs and
adjustable frequency shifters. The actual realization of SRC with an integer ratio
Rp is performed by the structures in Figs. 6.4 and 6.5. The SRC structures in this
figure are only used for analysis purposes.
6.3 Proposed Integer SRC Multimode TMUX

With general lowpass filters, we can construct an integer SRC multimode TMUX
as in Fig. 6.1. It consists of upsamplers/downsamplers Rp , p = 0, 1, . . . , P − 1;
lowpass filters Gp (z) and Ĝp (z); and adjustable frequency shifters ωp and ω̂p . If
Tp is the sampling period in branch p, then
T0 T1
= = . . . = Ty , (6.2)
R0 R1
PP −1
where Ty is the sampling period of y(n). If p=0 R1p < 1, the TMUX is redundant
PP −1
whereas p=0 R1p = 1 gives a critically sampled TMUX.
6.3.1 Variable Integer SRC Using the Farrow Structure

The Type I polyphase decomposition of a filter Gp (z) is [12, 69, 70]
Rp −1
X
Gp (z) = z −m Gp,m (z Rp ). (6.3)
m=0
If Gp (z) is a general causal ideal lowpass filter of order N , we have

( N
z − 2 in passband
Gp (z) = (6.4)
0 in stopband.
Comparing (6.3) and (6.4) gives
 N −m
 − 2
Gp,m (z) = z Rp
in passband (6.5)
 0 in stopband.
76
FARROW STRUCTURE
yp,m(z) Fp,m(z)
yyp,m(n) yFp,m(n)
mQ 5 3 1 P
mp,m 4 2
p,m mp,m mp,m mp,m mp,m mp,m
SQ(z) S5(z) S3(z) S1(z) SP(z) S4(z) S2(z) S0(z)
x(n) x(n)
Figure 6.2: Realization of Ψp,m (z). Figure 6.3: Realization of Φp,m (z).
Therefore, a general interpolation/decimation filter can be designed by choosing

Gp,0 (z) to be an N0 -th order Type I linear-phase FIR filter and utilizing the Farrow
structure to realize Gp,m (z), m = 1, 2, . . . , Rp − 1, of odd1 order N1 as
N
N0 = = N1 + 1, (6.6)
Rp
where
L
X
Gp,m (z) = Sk (z)µkp,m , (6.7)
k=0
with
−m 1
µp,m = + ⇒ µp,m = −µp,Rp −m . (6.8)
Rp 2
Note that (6.8) relates the SRC ratio Rp to the values of µ in Fig. 3.7. For every
new Rp , we should only compute Rp new values of µp,m whereas Sk (z) need not
change. Specifically, (6.8) can be computed recursively as
−1 1
µp,m+1 = + µp,m , µp,0 = . (6.9)
Rp 2
Efficient Variable Integer SRC

Using (6.7) and (6.8), Gp,m (z) and Gp,Rp −m (z) can be written as
Gp,m (z) = Φp,m (z) + Ψp,m (z), (6.10)
Gp,Rp −m (z) = Φp,m (z) − Ψp,m (z). (6.11)
According to Figs. 6.2 and 6.3 with Q = 2⌊ L+1 L
2 ⌋ − 1 and P = 2⌊ 2 ⌋, we have
⌊L
2⌋
X
Φp,m (z) = Gp,2k (z)µ2k
p,m , (6.12)
k=0
⌊ L+1
2 ⌋
X
Ψp,m (z) = Gp,2k−1 (z)µ2k−1
p,m . (6.13)
k=1
1 With proper modifications, even-order filters can also be designed [152].
77
FARROW STRUCTURE
Fixed Variable
x(n) Gp,0(z) 0
k 1
fs mp,m y(m)
Sk(z) Rp fs
Rp-1
k = 0, 1, ..., L
Figure 6.4: Efficient interpolation by a variable integer ratio Rp using fixed subfil-
ters, variable multipliers, and commutators.
Variable Fixed
0 Gp,0(z) y(m)
1
k fs/Rp
x(n) mp,m
fs Sk(z)
Rp-1
k = 0, 1, ..., L
Figure 6.5: Efficient decimation by a variable integer ratio Rp using fixed subfilters,
variable multipliers, and commutators.
In this way, two polyphase components can be realized at the cost of one [152].
Incorporating commutators, as in Figs. 6.4 and 6.5, SRC by a variable integer ratio
Rp requires fixed filters Sk (z) and Gp,0 (z); variable multipliers µkp,m ; and commu-
tators. Due to (6.8), only the distinct values of µkp,m must be considered. Thus,
variable integer SRC needs either a set of precomputed values or some variable
multipliers.
Arithmetic Complexity
The arithmetic complexity of the fixed parts, in Figs. 6.4 and 6.5, results from the
N0 -th order filter Gp,0 (z); the Farrow structure composed of L+1 subfilters of N1 -th
order; and L additional structural adders2 . Then, we roughly require 3N1 (L+2) 2 +
3L+5
2 fixed arithmetic operations. For each extra coefficient, in S k (z), we need
3(L+2)
2 additional fixed arithmetic operations. The structures in Figs. 6.4 and 6.5
require some variable multipliers as well. The Farrow-based SRC is generally more
efficient than regular SRC, except for an Rp which is either small or factorable into
integer parts [154]. This is true even if one does not consider the reconfigurability
in SRC which is obtained by using the Farrow structure.
2 The decimator of Fig. 6.5 has an additional multi-input structural adder.
78
FARROW STRUCTURE
6.3.2 Approximation of Perfect Reconstruction (PR)

Generally, TMUXs have two sources of interference as in (3.35). The filters between
Xp (z) and X̂p (z) cause ISI, whereas the filters between Xi (z) and X̂p (z), give rise
to ICI. In Fig. 6.1, a desired signal X̂d (z), can be expressed as
RX
d −1 P −1
X Rp
mR zd R p 1 zd 1
X̂d (z) = Xp (z Rd WRd p ( ) )Gp (z Rd WRmd )Ĝd (z Rd WRmd ) (6.14)
m=0 p=0
zp zp
2π
where zp = ejωp T , zd = ej ω̂d T , and Wα = e−j α . If p6=d (p = d), then (6.14)
represents the effect of ICI (ISI) on X̂d (z). Generally, the ICI has P − 1 terms and
the complete ICI is represented by their summation. In the proposed redundant
TMUX, the users do not overlap in the transition bands and passbands. Thus, all
ICI terms fall in the stopband of Ĝd (z). With brickwall lowpass filters,
(
1 in passband
Gp (z) = (6.15)
0 in stopband.
Then, δICI = 0 and δISI = 0 in (3.35). Thus, a redundant PR TMUX requires

brickwall lowpass filters. Such a PR TMUX is unrealizable. This chapter deals with
realizable NPR TMUXs. Therefore, the ICI (ISI) is controlled by the stopband
(passband) ripples of Gp (z) and Ĝp (z). By decreasing these ripples, arbitrarily
good NPR TMUXs can be obtained.
In maximally decimated nonuniform FBs, the AFB and SFB filters should have
specific center frequencies. Otherwise, infeasible cases occur [20, 22, 24, 26, 27] and
NPR FBs may not even be achievable. To avoid this, we can shift some AFB and
SFB filters to specific center frequencies [27]. Duality of FBs and TMUXs gives
rise to similar problems for critically sampled nonuniform TMUXs. The proposed
redundant TMUX does not have infeasible cases as the users do not overlap. The
frequency shifters are chosen so as to ensure feasibility. Feasibility is achieved if
(with ∆ = 0) ( π(1+ρ)
R0 if p = 0
ωp = 1
Pp−1 2 (6.16)
π(1 + ρ)[ Rp + k=0 Rk ] if p6=0.
Note that these frequency shifters are similar to those in [27].
6.3.3 Filter Design

With p = d, zp = zd , and Rp = Rd in (6.14), we have X̂p (z) = Xp (z)Fp (z) where
Rp −1
X 1 1
Fp (z) = Gp (z Rp WRmp )Ĝp (z Rp WRmp ). (6.17)
m=0
This is the zeroth polyphase component of Gp (z)Ĝp (z). Therefore, Sk (z) and
Gp,0 (z) must be determined such that
79
FARROW STRUCTURE
• Gp (z) and Ĝp (z) have small stopband ripples.

• Fp (z) approximates an allpass function.
In other words, Gp (z)Ĝp (z) approximates an Rp -th band filter as defined in Sec-
tion 2.4.3.
6.3.4 Filter Design Parameters

The free design parameters are the coefficients of Sk (z) and Gp,0 (z). The values
of N1 , N0 , L, ρ, µkp,m , and Rp are fixed during the optimization. Note that the
filter design problem is solved only once and offline. After determining Sk (z) and
Gp,0 (z) only once, a large set of SRC ratios can be supported.
6.3.5 Filter Design Criteria

To design the filters3 , we can use both least-squares (LS) or minimax methods
or their combinations. However, the application mainly determines the design
method. For example, one could be interested in minimizing (i) the ripples, or (ii)
the energy of the filter frequency response in specific bands. In some applications,
combinations of energy and ripples may also be useful. The sequel considers the
minimax and LS approaches which are the most commonly used methods [75].
Minimax Design
The minimax filter design problem is
min δ subject to
∀Rp
|Fp (ejωT ) − 1| ≤ δ, ωT ∈ [0, π]

|Gp (ejωT )|≤W (ωT )δ, ωT ∈ [ωs T, π], (6.18)
where δ and W (ωT ) are, respectively, the ripples and the weighting function with
π(1 + ρ)
ωs T = . (6.19)
Rp
This chapter uses flat weighting functions reducing W (ωT ) to a constant W .
The filter specifications are satisfied for every mode. Compared to conventional
TMUXs, this filter design is more complicated but solved only once and offline.
Figure 6.6 shows the filters resulting from (6.18) and Rp = {2, 3, . . . , 30}. The
values of ρ, W , L, N1 , and N0 are 0.2, 1, 5, 17, and 18, respectively. Furthermore,
δ = 2.56×10−3 . The characteristics of the filters with W = 0.2 and W = 5 are,
respectively, shown in Figs. 6.7 and 6.8. Here, δ = 1.025×10−2 and δ = 9.79×10−4 ,
respectively.
3 To avoid having two sets of S (z), we assume that G (z) = Ĝ (z) but the TMUX discussion
k p p
is treated in a general case where Gp (z) may differ from Ĝp (z).
80
FARROW STRUCTURE
−3
x 10
3
2
Fp(ejωT)−1
1
0
−1
−2
−3
0 0.2π 0.4π 0.6π 0.8π π
ωT [rad]
0
Gp(ejωT) [dB]
−20
−40
−60
0 0.2π 0.4π 0.6π 0.8π π
ωT [rad]
Figure 6.6: Approximate Rp -th band filters designed with W = 1 in (6.18).
0.02
Fp(ejωT)−1
0.01
0
−0.01
−0.02
−0.03
0 0.2π 0.4π 0.6π 0.8π π
ωT [rad]
0
Gp(ejωT) [dB]
−20
−40
−60
0 0.2π 0.4π 0.6π 0.8π π
ωT [rad]
Figure 6.7: Approximate Rp -th band filters designed with W = 0.2 in (6.18).
81
FARROW STRUCTURE
−3
x 10
3
2
Fp(ejωT)−1
1
0
−1
−2
−3
0 0.2π 0.4π 0.6π 0.8π π
ωT [rad]
0
Gp(ejωT) [dB]
−20
−40
−60
0 0.2π 0.4π 0.6π 0.8π π
ωT [rad]
LS Design
One way to formulate the LS filter design problem is
Z π
1 π
Z
jωT 2
min |Fp (e ) − 1| d(ωT ) + |Gp (ejωT )|2 d(ωT ). (6.20)
∀Rp 0 W ωs T
The designed LS filters are shown in Figs. 6.9–6.11. Here, the same parameters as
those in Figs. 6.6–6.8 have been used.
A Note on Filter Order

For the same ρ and passband/stopband ripples, we can compare N in (6.6) with
that estimated by the well-known formulae, e.g., Bellanger [88], Kaiser [89], as
discussed in (2.47) and (2.48). This comparison shows that N in (6.6) is about
20% − 30% larger than those estimated by [88, 89]. However, [88, 89] do not
account for the power complementary property. This property requires an excess
order of about 20% − 30% [68] to those estimated by [88, 89]. Consequently, the
proposed design method does not necessarily increase the arithmetic complexity
as compared to regular filter design methods. Further, the filter order (and the
number of delay elements) of the Farrow-based and the regular SRC structures
differ by a maximum of 10% [154].
82
FARROW STRUCTURE
−3
x 10
3
2
Fp(ejωT)−1
1
0
−1
−2
−3
0 0.2π 0.4π 0.6π 0.8π π
ωT [rad]
0
Gp(ejωT) [dB]
−20
−40
−60
0 0.2π 0.4π 0.6π 0.8π π
ωT [rad]
6.4 Proposed Rational SRC Multimode TMUX

Section 6.3.1 revealed that the Farrow structure can implement variable integer
SRC. Additional downsamplers/upsamplers allow one to construct a rational SRC
multimode TMUX. The TMUX, in Fig. 6.12, consists of upsamplers/downsamplers
Ap , Bp , p = 0, 1, . . . , P − 1; lowpass filters Gp (z) and Ĝp (z); and adjustable fre-
quency shifters ωp and ω̂p . If Tp is the sampling period in branch p, then
B0 B1
T0 = T1 = . . . = Ty , (6.21)
A0 A1
where Ty is the sampling period of y(n). The proposed redundant TMUX assumes
PP −1 Bp PP −1 Bp
p=0 Ap < 1 whereas p=0 Ap = 1 gives a critically sampled TMUX.
6.4.1 TMUX Illustration

To illustrate the TMUX, we can study one of its branches. Redundancy allows us to
add signals from other branches and to construct a composite signal. Figure 6.13
plots the outputs of different blocks with a uniformly distributed random input
where Rp = 19 9 . In Fig. 6.13(e), a lowpass filter can remove the images allowing
one to retrieve the desired symbols by a downsampler.
83
FARROW STRUCTURE
−3
x 10
3
2
Fp(ejωT)−1
1
0
−1
−2
−3
0 0.2π 0.4π 0.6π 0.8π π
ωT [rad]
0
Gp(ejωT) [dB]
−20
−40
−60
0 0.2π 0.4π 0.6π 0.8π π
ωT [rad]
Figure 6.10: Approximate Rp -th band filters designed with W = 0.2 in (6.20).
6.4.2 Efficient Variable Rational SRC

Ap
With conventional upsamplers, downsamplers, and filters, SRC by a ratio Bp re-
quires [12, 69]
• Downsampling by Bp after interpolation by Ap .
• Upsampling by Bp before decimation by Ap .
For a fixed Ap , changing the values of Bp gives a number of rational SRC ra-
tios. According to Section 6.3.1, integer SRC by Ap needs a set of fixed filters.
A
Thus, any rational ratio Bpp can be handled using fixed filters and different upsam-
plers/downsamplers Bp . To construct structures for variable rational SRC, we can
replace SRC by Ap in Fig. 6.12, with its equivalent structure from Figs. 6.4 and
6.5.
In Fig. 6.14, some outputs of the Farrow-based interpolation are not needed.
The commutator retains every Bp -th sample. In Fig. 6.15, some inputs of the
Farrow-based decimation are zero. This saves the arithmetic complexity as these
samples need not be processed. This is illustrated in Fig. 6.16 for a group of input
samples {x0 , x1 , . . . , xAp −1 }. For other groups, e.g., {xAp , xAp +1 , . . . , x2Ap −1 }, the
location of these zero-valued samples changes but the savings in the arithmetic
84
FARROW STRUCTURE
−3
x 10
3
2
Fp(ejωT)−1
1
0
−1
−2
−3
0 0.2π 0.4π 0.6π 0.8π π
ωT [rad]
0
Gp(ejωT) [dB]
−20
−40
−60
0 0.2π 0.4π 0.6π 0.8π π
ωT [rad]
jw0n ^ n
-jw
e e 0
x0(n0) A0 G0(z) B0 B0 ^ (z)

G A0 x^0(n0)
0
jw1n ^ n
-jw
e e 1
^
y(n) y(n)
x1(n1) A1 G (z) B1 B1 ^ (z)
G A1 ^x (n1)
1 1 1
^ n
-jw
ejwP-1n e P-1
xP-1(nP-1) AP-1 BP-1 BP-1 ^ (z)

G AP-1 ^x (nP-1)
GP-1(z) P-1 P-1
Figure 6.12: Rational SRC multimode TMUX composed of variable rational SRCs
and adjustable frequency shifters. The actual realization of rational SRC with a
ratio Rp is performed by the structures in Figs. 6.14 and 6.15. The SRC structures
in this figure are only used for analysis purposes.
85
FARROW STRUCTURE
(a)
Mag. [dB]
20
0
−20
−40
0 0.25π 0.5π 0.75π π 1.25π 1.5π 1.75π 2π
ωT1 [rad]
(b)
Mag. [dB]
20
0
−20
−40
0 0.25π 0.5π 0.75π π 1.25π 1.5π 1.75π 2π
ωT2 [rad]
(c)
Mag. [dB]
20
0
−20
−40
0 0.25π 0.5π 0.75π π 1.25π 1.5π 1.75π 2π
ωT2 [rad]
(d)
Mag. [dB]
20
0
−20
−40
0 0.25π 0.5π 0.75π π 1.25π 1.5π 1.75π 2π
ωT2 [rad]
(e)
Mag. [dB]
20
0
−20
−40
0 0.25π 0.5π 0.75π π 1.25π 1.5π 1.75π 2π
ωT3 [rad]
Figure 6.13: Output spectrum. (a) Interpolation by Ap = 19. (b) Downsampling

by Bp = 9. (c) and (d) Frequency shifters ωp = ω̂p = 0.5684π. (e) Upsampling by
Bp = 9. This illustration uses the filters in Fig. 6.6 where T1 , T2 , and T3 represent
different sampling periods at the input/output of the SRC blocks.
Fixed Variable
x(n) Gp,0(z) 0 Bp:1

1
fs k
mp,m y(m)
Sk(z) R p fs
Ap-1
k = 0, 1, ..., L
Ap
Figure 6.14: Efficient interpolation by a variable rational ratio Rp = Bp using the
Farrow structure.
86
FARROW STRUCTURE
Variable Fixed
0 Gp,0(z) y(m)
1
k Rp fs
x(n) Bp mp,m
fs Sk(z)
Ap-1
k = 0, 1, ..., L
Ap
Figure 6.15: Efficient decimation by a variable rational ratio Rp = Bp using the
Farrow structure.
0
Bp-1 zeros Gp,0(z) y(m)
Bp+1 k R p fs
mp,m
Bp-1 zeros
Sk(z)
2Bp+1
Ap-1 k = 0, 1, ..., L
A
Figure 6.16: Efficient decimation by Rp = Bpp using the Farrow structure and by
incorporating the effect of the upsampling by Bp into Fig. 6.15.
complexity are preserved. Using (6.3), (6.7), and (6.8) with Rp = Ap , we have
Ap −1 L
X X −m 1 k
Gp (z) = Gp,0 (z Ap ) + z −m Sk (z Ap )( + ) . (6.22)
m=1
Ap 2
k=0
If some inputs of the Ap branches are zero, we can discard their correspond-
ing polyphase components in (6.22). Thus, only a subset of the values in m =
1, 2, . . . , Ap − 1, will be used.
6.4.3 Approximation of PR
Similar to (6.14), a desired X̂d (z) in Fig. 6.12 can be written as (see Appendix A)
P d −1 B
−1 AX p −1 mAp Bd
X X Ap Bd
Bp zd BAp
kA
X̂d (z) = Xp (z Ad Bp WAd ( ) p WBp p )×
p=0 m=0 k=0
zp
mBd
Bd
zd B1p k 1
) WBp )Ĝd (z Ad WAmd ).
B
Gp (z Ad Bp WAd p ( (6.23)
zp
Note that (6.14) is a special case of (6.23) in which Ap = Rp and Bp = 1. Similar
to (6.17) and with p = d, (6.23) gives X̂p (z) = Xp (z)Fp (z) where
Ap −1 Bp −1
X X 1 1
Fp (z) = Gp (z Ap WAmp WBk p )Ĝp (z Ap WAmp ). (6.24)
m=0 k=0
87
FARROW STRUCTURE
zd
Like Section 6.3.2, the choice of zp allows one to control the ISI and ICI. For
Ap
SRC by Bp ,
we can design the integer SRC by Ap and, then, perform the integer
upsampling/downsampling by Bp . In this case, (4.17) becomes [69]
π(1 + ρ)
ωs T = . (6.25)
max(Ap , Bp )
With fixed Ap in Fig. 6.12, aliasing is avoided if
Ap
≥Bp ≥2. (6.26)
1+ρ
Equations (6.25) and (6.26) give
π(1 + ρ)
ωs T = . (6.27)
Ap
Figure 6.17 shows the magnitude of X̂d (z) for 12 multimode setups consisting of
P = {2, 3, 4, 5, 6} users. Here, xp (n) = δ(n) and each multimode setup has a
number of SRC ratios as in Table 6.1. Although the filters are designed for sets of
Ap , the choice of zzdp enables one to control the ISI and ICI. We can directly use
(6.23) in the filter design but this approach is not favorable as it complicates the
design. Instead, we can control ωp by (6.16) and use the simpler design problems
of (6.18) and (6.20).
The proposed TMUX can also be analyzed using the method in [19]. This
was earlier performed in Chapter 5. The blocked transfer function of Fig. 6.12 is
T (z) = Φ(z)Ψ(z) where
T
Φ(z) = φ0 (z) φ1 (z) ... φP −1 (z) , (6.28)

Ψ(z) = ψ0 (z) ψ1 (z) ... ψP −1 (z) . (6.29)
and
φp (z) = Ĝp (z), (6.30)

zd
ψp (z) = Gp (z ). (6.31)
zp
6.5 TMUX Performance

To illustrate the TMUX performance, the error vector magnitude (EVM) [139, 140]
and the multimode setups of Table 6.1 with 16-quadrature amplitude modulation
(QAM) data are considered. Each multimode setup corresponds to one specific time
frame as in Section 6.2.1. According to Fig. 6.18, the TMUX provides bandwidth-
on-demand and the whole frequency spectrum is shared by any number of users4 .
Figure 6.19 shows the average EVM for these multimode setups.
88
FARROW STRUCTURE
−3
(a) δ
x 10 ISI
5
Mag.
−5
0 0.25π 0.5π 0.75π π 1.25π 1.5π 1.75π 2π
ωT [rad]
(b) δ
ICI
0.01
Mag.
0.005
0
0 0.25π 0.5π 0.75π π 1.25π 1.5π 1.75π 2π
ωT [rad]
Figure 6.17: The ISI and ICI in Fig. 6.12 with xp (n) = δ(n) for the multimode
setups of Table 6.1 and the filters used in Fig. 6.6.
Table 6.1: SRC ratios for the multimode setups of Fig. 6.19. As an example, for
the first setup, R0 = 29 23
10 and R1 = 11 .
Setup A B P
1 [29, 23] [10, 11] 2
2 [15, 27, 23, 6] [4, 5, 4, 1] 4
3 [29, 17, 27, 19] [4, 4, 8, 3] 4
4 [20, 19, 17, 9, 19, 17] [3, 2, 3, 1, 2, 3] 6
5 [25, 27, 29] [7, 8, 7] 3
6 [13, 24, 13] [2, 7, 5] 3
7 [10, 9, 7] [1, 2, 3] 3
8 [28, 23] [11, 10] 2
9 [9, 4, 3] [2, 1, 1] 3
10 [11, 7, 14] [2, 3, 3] 3
11 [29, 23, 9] [5, 10, 2] 3
12 [30, 21, 17, 28, 18] [1, 2, 3, 5, 5] 5
89
FARROW STRUCTURE
20
Mag. [dB] 0
−20
−40
0 0.25π 0.5π 0.75π π 1.25π 1.5π 1.75π 2π
ωT [rad]
20
Mag. [dB]
0
−20
−40
0 0.25π 0.5π 0.75π π 1.25π 1.5π 1.75π 2π
ωT [rad]
20
Mag. [dB]
0
−20
−40
0 0.25π 0.5π 0.75π π 1.25π 1.5π 1.75π 2π
ωT [rad]
20
Mag. [dB]
0
−20
−40
0 0.25π 0.5π 0.75π π 1.25π 1.5π 1.75π 2π
ωT [rad]
Figure 6.18: Transmitted signal for the first four multimode setups of Table 6.1.
The LS filters have a smaller EVM and the filters designed with W = 1 are
superior to those with W = 5 and W = 0.2. Thus, the EVM is determined by the
(i) stopband of Ĝp (z), and (ii) passband of Fp (z). Although the LS approach is
superior according to EVM, some systems may constrain the ripples. Then, the
more appropriate option would be to use the minimax or the constrained LS (CLS)
approaches.
Irrespective of the design technique, this TMUX has an indirect filter design.
In other words, the filters are designed only for sets of Ap . Consequently, the
constraints in the filter design are only satisfied for the values of Ap . Then, rational
A
SRC by a ratio Bpp is realized by choosing the sets of Bp as in (6.26). However,
as Fig. 6.19 shows, the TMUX can satisfy any desired ICI and ISI by a proper
filter design. Without additional constraints, due to Bp , the designed filters are
suboptimal and they have some overdesign as we shall see in the next subsection.
6.5.1 Effects of Bp on the SRC Error

A comparison between (6.17) and (6.24) reveals that downsampling by Bp results
in a sum of Bp terms. Thus, a larger Bp would result in a larger error. However,
ensuring (6.26) allows one to attenuate these terms by the stopband of Ĝp (z).
4 For illustration purposes, the values of A and B are chosen so that y(n), in Fig. 6.12,
p p
occupies between 90 − 99% of the frequency range [0, 2π].
90
FARROW STRUCTURE
−40
−42
Average EVM [dB]
−44
−46
−48
LS, W=1
LS, W=5
LS, W=0.2
−50
Minimax, W=1
Minimax, W=0.2
Minimax, W=5
1 2 3 4 5 6 7 8 9 10 11 12
Multimode setup
Figure 6.19: Average EVM of 16-QAM signals in multimode setups of Table 6.1
for the TMUX in Fig. 6.12 and with the filters in Figs. 6.6–6.11.
To illustrate this, four values of Ap are chosen randomly and the values of Bp
A
are determined using (6.26). This gives a set of Rp = Bpp . Further, a cascade of
interpolation by Rp and decimation by Rp , with the filters in Fig. 6.6, is performed.
Figure 6.20(a) shows the resulting EVM. As can be seen, a larger Bp increases the
EVM. Figure 6.20(b) shows the EVM for all 232 possible unique values of Rp
achieved by Ap = {2, 3, . . . , 30} and (6.26) with ρ = 0.2. There is an upper bound
on the EVM mainly determined by the stopband of Ĝp (z). Thus, the stopband
attenuation can be increased to compensate for the additional error due to Bp .
As Fig. 6.20(b) shows, the EVM varies in a range of about 10 dB. Hence,
increasing the stopband attenuation, by 10 dB, would decrease the highest EVM
to a desired level. According to [88], increasing the stopband attenuation by 10
dB, would increase the order of linear-phase FIR filters by about 10%. On the
other hand, direct optimization would require an increase of 23229 = 800% in the
optimization complexity. This shows that it may indeed be preferable to have a
slight overdesign and avoid direct optimization. However, direct optimization can
be performed for limited sets of Ap and Bp as we shall see in the sequel.
91
FARROW STRUCTURE
(a)
−48
−50
EVM [dB]
−52 Ap = 6
Ap = 13
−54 Ap = 18
−56 Ap = 27
2 4 6 8 10 12 14 16 18 20 22
Bp
(b)
−46
−48
EVM [dB]
−50
−52
−54
−56
5 10 15 20 25 30
Rp
Ap
Figure 6.20: Effects of Bp on the error for the cascade of interpolation by Rp = Bp
A
and decimation by Rp = Bpp . (a) Increase of EVM with the increase in Bp for
a fixed Ap . (b) Upper bound of EVM for all possible values of Rp achieved by
Ap = {2, 3, . . . , 30} and (6.26) with ρ = 0.2.
x(n) hn(k) y(n)
Figure 6.21: Equivalent model for each branch in the TMUX of Fig. 6.12.
6.6 Direct Filter Design
As discussed before, one can design the TMUX filters for the sets of Ap . Specifically,
the specifications are satisfied only for the sets of Ap . Then, one can choose proper
A
values of Bp so as to perform rational SRC by a ratio Rp = Bpp . This indirect
design method results in (i) suboptimal filters, and (ii) overdesign. To overcome
these, we can include additional filter design constraints arising from the sets of
Bp . Then, the specifications are satisfied for all sets of Ap and Bp . This makes the
design direct.
Ad
In one branch of the TMUX, i.e., a cascade of interpolation by Bd and decima-
92
FARROW STRUCTURE
Ad
tion by Bd , the input-output relation is
d −1 B
AX d −1
X 1 1
Y (z) = Ĝ(z Ad WAmd )G(z Ad WBl d WAmd )X(zWBlAdd )
m=0 l=0
BX
d −1
= Tl (z)X(zWBlAdd ), (6.32)
l=0
where
AX
d −1
1 1
Tl (z) = Ĝ(z Ad WAmd )G(z Ad WBl d WAmd ), l = 0, 1, . . . , Bd − 1. (6.33)
m=0
Here, (6.32) is obtained by setting Ap = Ad , Bp = Bd , and zp = zd in (6.23).

Consequently, T0 (z) reduces to (6.17). In the time-variant dual rate system [92]
represented by (6.32), the output signal is the result of operating (see Footnote 7
in Chapter 3)
• T0 (z) on X(z).
2lπAd
• Tl (z), l = 1, . . . , Bd − 1, on the frequency shifted (by Bd ) versions of the
input, i.e., X(zWBlAdd ).
Ideally,
T0 (z) = 1
Tl (z) = 0, l = 1, . . . , Bd − 1. (6.34)
PBd −1
Then, l=0 Tl (z) = 1. As shown in Fig. 6.21, the system can also be modeled as
the operation of a time-varying periodic filter hn (k) on the input signal as [155]
X X
y(n) = x(k) g(mBd − Ad k)ĝ(nAd − mBd )
k m
X X
= x(n − k) g(mBd − nAd + kAd )ĝ(nAd − mBd ),
k m
X
= x(n − k)hn (k). (6.35)
k
where X
Hn (z) = hn (k)z −k . (6.36)
k
Thus, we have Bd impulse responses (see Appendix B)

X
hn (k) = g(mBd − nAd + kAd )ĝ(nAd − mBd ), n = 0, 1, . . . , Bd − 1. (6.37)
m
93
FARROW STRUCTURE
One can determine hn (k) by feeding δ(n − m), m = 0, 1, . . . , to the system and
computing the corresponding outputs ym (n). The impulse responses hn (k) are
then given as hn (n − m) = ym (n).
In a PR system, the output signal is a delayed version of the input signal.
Ignoring the delay and defining the error e(n) = y(n) − x(n), we have
Z π
1
e(n) = (Hn (ejωT ) − 1)X(ejωT )ejnωT d(ωT ). (6.38)
2π −π
Ideally, e(n) = 0 and, then, Hn (ejωT ) = 1. In practice, Hn (ejωT ) can generally

only approximate unity in the frequency range of interest. Consequently, the filter
design problem should determine hn (k) so that e(n) is minimized according to
some criterion. As e(n) depends on both Hn (ejωT ) and X(ejωT ), one generally
requires knowledge about the spectrum of the input signal to determine optimum
filters. In practice, there is no complete knowledge about X(ejωT ) and one accepts
a suboptimum solution instead. In this regard, it can be convenient to make use
of the Lp -norm of a general signal S(ejωT ) given by [69]
s Z π
jωT 1 p
kS(e )kp = p |S(ejωT )| d(ωT ). (6.39)
2π −π
Using the triangle inequality of integrals in (6.38), we have

Z π
1
|e(n)|≤ |Hn (ejωT ) − 1||X(ejωT )|d(ωT ) (6.40)
2π −π
which in terms of L∞ -norm gives
|e(n)|≤kHn (ejωT ) − 1k∞ kX(ejωT )k∞ . (6.41)
As
kS(ejωT ) − 1k∞ = max{|S(ejωT ) − 1|}, (6.42)
ωT
minimizing the maximum of |Hn (ejωT ) − 1| corresponds to minimizing the max-

imum of e(n) resulting from that particular hn (k). From earlier discussions, the
indirect minimax filter design problem can be rewritten as
Case I: min δ subject to

∀Ad
|Fd (ejωT ) − 1| ≤ δ, ωT ∈ [0, π]

|Gd (ejωT )|≤δs , ωT ∈ [ωs T, π]. (6.43)
Here, δs is the desired stopband attenuation and ωs T is given by (6.27). Further,

Fd (z) is given by (6.17). The indirect CLS filter design problem is
Case II : min δ subject to

∀Ad
94
FARROW STRUCTURE
Table 6.2: The SRC ratios for the multimode setups used in Fig. 6.22.
Setup Ap Bp P
1 [15, 18, 13] [4, 7, 2] 3
2 [17, 19, 9] [4, 7, 2] 3
3 [20, 19, 13, 17] [3, 3, 4, 3] 4
4 [10, 3, 5] [1, 1, 2] 3
5 [9, 4, 3] [2, 1, 1] 3
6 [11, 7, 14] [2, 3, 3] 3
Z π
|Fd (ejωT ) − 1|2 d(ωT ) ≤ δ, ωT ∈ [0, π]
0
|Gd (ejωT )|≤δs , ωT ∈ [ωs T, π]. (6.44)
One can express the direct minimax filter design problem as
Case III : min δ subject to

∀Ad ,Bd
Bd −1
1 X
|Hn (ejωT ) − 1| ≤ δ, ωT ∈ [0, π]
Bd n=0
|Gd (ejωT )|≤δs , ωT ∈ [ωs T, π] (6.45)
where ωs T is given by (6.25). Similarly, (6.40) can also be written in terms of the
L2 -norm leading to the direct CLS filter design problem as
Case IV : min δ subject to

∀Ad ,Bd
Bd −1 Z π
1 X
|Hn (ejωT ) − 1|2 d(ωT ) ≤ δ, ωT ∈ [0, π]
Bd n=0 0
|Gd (ejωT )|≤δs , ωT ∈ [ωs T, π]. (6.46)
6.6.1 Design Example

This section considers some design examples for the direct and indirect filter de-
signs. Here, the values of L, N1 , and δs are 5, 17, and 0.01 respectively. Further-
more, (6.43), (6.44), (6.45), and (6.46) result, respectively, in the values of δ to
be 6.21×10−4 , 1.28×10−7 , 2.70×10−4 , and 2.49×10−8 . Note that in all of these
filter design problems, (6.26) holds and their difference lies in the fact whether they
include Bd in the optimization or not.
Figure 6.22 shows the average EVM for 16-QAM signals according to the mul-
timode setups of Table 6.2. Generally, the CLS approach results in a smaller EVM
95
FARROW STRUCTURE
−38
−40
Average EVM [dB]
Case I
−42
Case II
Case III
Case IV
−44
−46
−48
1 2 3 4 5 6
Multimode setup
Figure 6.22: The EVM values for the multimode setups in Table 6.2.
than the minimax method. Furthermore, the direct filter design in Case III (IV)
reduces the EVM compared to the indirect filter design of Case I (II). Compared
to the direct minimax filter design, the direct CLS method brings a larger improve-
ment in system performance.
With δs = 0.01, the EVM values hover around −40dB. In other words, the
ICI is controlled by δs and the four filter design problems decrease the ISI either
directly or indirectly. The overall EVM is determined by both ISI and ICI. The
values of δs and δ are correlated. For the same filter orders, decreasing δs would
increase δ. For example, with the filter orders used in Fig. 6.22, having δs = 0.001
results in about 4dB larger EVM than that depicted in Fig. 6.22. In this case,
the EVM is mainly determined by δ and the difference between these filter design
problems is less pronounced. A desired EVM can be achieved by choosing proper
values of L and N1 .
Although the direct filter design has a smaller EVM, it requires to solve a
more complex design problem. This increased design complexity is a result of
the increased number of constraints and is proportional to the number of sets
{Ap , Bp }. Consequently, the convergence time will increase and issues regarding
memory problems may also arise. However, the memory problems can partially be
alleviated by careful choice of the number of grid points for ωT , etc. As this TMUX
requires offline filter design, the memory problems are generally more important
than the convergence time.
96
FARROW STRUCTURE
6.7 Conclusion
A class of multimode TMUXs was introduced in which the Farrow structure real-
izes general interpolation/decimation filters. These TMUXs support variable SRC
ratios using fixed filters and variable multipliers. Efficient realization structures
are derived and different filter design techniques are compared.
This TMUX does not need online filter design. This comes at the expense of
a more complicated filter design problem but it suffices to solve it only once and
offline. Then, we need to simply adjust some multipliers online. In terms of EVM,
the LS approach is better than the minimax method but some applications may
necessitate the use of the minimax method.
Different filter design formulations result in different performances. However,
the direct filter design has a better control over the TMUX noise but it has a larger
design complexity.
97
FARROW STRUCTURE
98
7
Reconfigurable Nonuniform
Transmultiplexers Using
Uniform Modulated Filter Banks
This chapter introduces reconfigurable nonuniform transmultiplexers (TMUXs)

based on uniform modulated filter banks (FBs). After a brief introduction in
Section 7.1, Section 7.2 outlines the problem formulation. The proposed TMUX
is introduced in Section 7.3 where its filter design, realization, and reconstruction
error are discussed. Section 7.4 discusses the system parameters and topics such as
channel sampling periods, guardbands (GBs), and center frequencies. Section 7.5
treats the implementation cost and the parameter selection whereas Section 7.6
compares the proposed TMUX to the existing multimode TMUXs. Some conclud-
ing remarks are given in Section 7.7.
7.1 Introduction
A TMUX enables one to transmit different data streams through a single chan-
nel [12]. As discussed in Section 3.4, TMUXs consist of a synthesis FB (SFB)
followed by an analysis FB (AFB) which are a parallel connection of a number of
branches. Each branch is realized by digital interpolators/decimators. The SFB fil-
ters cover different regions of the frequency spectrum thereby packing independent
data streams into adjacent frequency bands. This leads to uniform or nonuniform
TMUXs where the passbands of the SFB filters determine the frequency bands.
Nonuniform TMUXs need different interpolators and decimators necessitating
99
7. RECONFIGURABLE NONUNIFORM TRANSMULTIPLEXERS USING
UNIFORM MODULATED FILTER BANKS
different bandpass filters. To freely determine the bandwidths of each data stream,
we should either (i) design a large set of filters, or (ii) design the filters online.
In other words, if the frequency division multiplexed (FDM) scenario (multimode
setup) changes, new sets of filters are required. This becomes involved if the FDM
scenarios change in a time-varying manner. We consequently need TMUXs which,
in a time-varying manner, cover different FDM scenarios.
The duality of FBs and TMUXs allows us to use the conventional filter design
methods in the case of fixed TMUXs. As the characteristics of multimode TMUXs
change in a time-varying manner, they do not generally allow one to use these
conventional filter design techniques. Consequently, multimode TMUXs require
special filter design techniques [46, 63–66]. However, it is desired to obtain mul-
timode TMUXs based on conventional FBs so as to use the existing design and
realization techniques.

Chapters 5 and 6 used the Farrow structure [93] to obtain redundant near perfect
reconstruction (NPR) multimode TMUXs. The filters are designed only once and
offline. Then, the TMUXs are reconfigured by adjusting (i) the fractional delay of
the Farrow structure, and (ii) some frequency shifters.
Nonuniform FBs and TMUXs can also be obtained by combining some of the
AFB and SFB filters and assigning1 them to each user. Although this gives a
nonuniform system, it may result in infeasible cases [20, 22, 24, 26, 27]. Other
alternatives are, to name a few, nonuniform modulated FBs [23]; tree structured
FBs [25]; semi-infinite programming [16, 31]; dividing nonuniform FBs into uniform
FBs [28]; FBs with complex filters [29]; and the recombination structures [17, 24,
26]. These may have less problems associated with infeasible cases but they have a
low degree of reconfigurability. They obtain nonuniform filters through dedicated
adders, modulators, or structures. To change the AFB and SFB filters, these
adders, modulators, or structures must change.
The cosine modulated FBs (CMFBs) and modified discrete Fourier transform
FBs (MDFT FBs) are well-studied. They are generally uniform and we can obtain
uniform TMUXs using the duality of TMUXs and FBs [116]. This chapter outlines
reconfigurable nonuniform TMUXs based on fixed uniform modulated FBs. The
TMUXs use parallel (polyphase) processing and any user is processed by a number
of TMUX branches. One branch represents one granularity band (GRB) and any
user occupies integer multiples of the GRB. This only requires some adjustable
commutators with no additional arithmetic complexity. Consequently, the band-
width of each user can change by simple adjustments of these commutators. The
proposed TMUXs can be either critically sampled or redundant.
With the proposed TMUXs, it suffices to design the filters only once offline. As
the TMUXs use standard modulated FBs, e.g., CMFBs and MDFT FBs, we can
obtain perfect reconstruction (PR) or NPR TMUXs. This chapter shows how
1 This is also referred to as the split-and-add method [20].
100
• To use fixed uniform modulated FBs to obtain reconfigurable nonuniform

TMUXs.
• To determine the system parameters so as to minimize the arithmetic com-

plexity.
In the proposed TMUXs, each polyphase component is transmitted by a feasi-

ble uniform TMUX. Thus, there are no infeasible cases. The data rates for all
polyphase components are equal. As a number of these components belong to one
specific user, different users will have different data rates.
Note that [21] uses the concept in [22] to convert TMUXs with a rational sam-
pling rate conversion (SRC) into TMUXs with an integer SRC. Then, similar to
the proposed TMUX, polyphase components of input signals are processed in dif-
ferent branches. However, each branch of [21] has a specific SRC factor thereby
decreasing the realization regularity.
7.2 Problem Formulation

As in Fig. 1.1, we assume that P users share the whole frequency spectrum. Each
user occupies a time-varying portion of the frequency spectrum. By adjusting this
portion at any point, any user can have any center frequency. Here, ∆ determines
the amount of overlap or GB. Any scenario can be obtained by an appropriate ∆.
Generally, we can have
• Case I with ∆ < 0 in which case different user spectra overlap and we do not
have a GB.
• Case II with minimal GB where ∆ = 0 and we have neither overlap nor extra
GB.
• Case III with ∆ > 0 resulting in an extra GB.
Depending on ∆, we can have critically sampled (Case I) or redundant (Case II and

III) TMUXs. For example, the European space agency has outlined three network
topologies for broadband communication systems using satellites [37]. In these
satellites, users are multiplexed between (and within) different satellite beams [39]
which necessitates ∆ > 0 [43, 44]. Without intermediate processing, e.g., multi-
plexing as discussed above, one may choose Case I so as to increase the spectrum
efficiency. For practical reasons, one may anyhow allow ∆ > 0 relaxing the sub-
sequent signal processing, e.g., band limitation or digital to analog conversion.
Redundant TMUXs may be necessary when the channel distortion is significant in
some bands [20].
101
X (z)
0,0
M F (z)
0
H (z) M X^ (z)
0 0,0
X (z)
0,1
M F (z)
1
H (z) M X^ (z)
1 0,1
X
0,M0-1
(z) M F
M0-1
(z) H (z) M X^ (z)
M0-1 0,M0-1
X (z)
1,0
M F
M0+1
(z) H (z) M X^ (z)
M0+1 1,0
X (z)
1,1
M F (z) H (z) M X^ (z)
M0+2
å y(n) ^
y(n) M0+2 1,1
X (z) M F (z) H (z) ^

X (z)
1,M1-1 M0+M1 M0+M1
M 1,M1-1
X
P-1,0
(z) M F
M-MP
(z) H (z) M X^ (z)
M-MP P-1,0
X
P-1,1
(z) M F
M-MP+1
(z) H
M-MP+1
(z) M X^ (z)
P-1,1
X (z) M F (z) H (z) M

^
X (z)
P-1,MP-1 M-1 M-1 P-1,MP-1
Figure 7.1: Proposed nonuniform TMUX with M channels and P ≤M users.
7.3 Nonuniform TMUXs Using Modulated FBs

We can treat polyphase components of different users as independent data streams
and transmit them in any TMUX branch2 . Consequently, different users have
different bandwidths and a nonuniform TMUX is obtained.
Assume the structure in Fig. 7.1 and
Mp −1
X
Xp (z) = z −l Xp,l (z Mp ), p = 0, 1, . . . , P − 1. (7.1)
l=0
2 This is similar to multicarrier modulation applied to multiple users.
102
Then,
P
X −1 MXp −1
Y (z) = Xp,l (z M )Fαp,l (z). (7.2)

p=0 l=0
The integer Mp is the number of branches that the user Xp (z) occupies. It corre-
sponds to the bandwidth assigned to Xp (z).
To incorporate ∆, we should use the SFB and AFB filters with proper center
frequencies. One can allow a GB by not using some TMUX branches. To exemplify,
assume M0 = 1 and M1 = 2. Then, X0 (z) uses one branch corresponding to F0 (z)
while X1 (z) uses two branches corresponding to F2 (z) and F3 (z). In this way, one
branch, corresponding to F1 (z), acts as a GB. Thus, the index to the proper filter
is ( P
p−1
m=0 Mm + l + pDg if p 6= 0
αp,l = (7.3)
l if p = 0,
where l = 0, 1, . . . , Mp −1. Furthermore, Dg is the (generally time-varying) number
of branches acting as GBs. In Fig. 7.1, we assume Dg = 1. In the AFB,
M
X −1
X̂p,l (z) = Ŷ (z 1/M WM
m
)Hαp,l (z 1/M WM
m
). (7.4)
m=0
Setting ŷ(n) = y(n) and after some manipulations, (7.4) gives

P
X −1 MXp −1 M
X −1
X̂p,l (z) = Xp,l (z) Fαp,l (z 1/M WM
m
)Hαp,l (z 1/M WM
m
). (7.5)
p=0 l=0 m=0
With p = l = d in (7.5), the inter-symbol interference (ISI) can be represented as

M
X −1
X̂d,d (z) = Xd,d (z) Fαd,d (z 1/M WM
m
)Hαd,d (z 1/M WM
m
). (7.6)
m=0
As 0≤αp,l ≤M − 1, (7.6) can be simplified and some manipulations allow one to

express the ISI similar to (3.40). To compute the inter-carrier interference (ICI),
we have
P −1 Mp −1 M −1
X X X
X̂pd ,ld (z) = Xp,l (z) Fαp,l (z 1/M WM
m
)Hαpd ,ld (z 1/M WM
m
) (7.7)
p=0,p6=pd l=0,l6=ld m=0
which can be simplified so that the ICI is computed similar to (3.41). If an NPR
TMUX has a maximum mean square error (MSE) of ǫmax in each branch, the MSE
for Xp,l (z) is
Np,l −1
1 X Np
ǫp,l = [x̂p,l (n) − xp,l (n)]2 , Np,l = (7.8)
Np,l n=0 Mp
103
Xp,0(z)
Xp,1(z)
Xp(z) Xp(z)
Xp,M -1(z)
p
Figure 7.2: Time-varying commutators in the proposed TMUX where Mp can vary
with time.
where ǫp,l ≤ǫmax and Np is the length of Xp (z). Here, the total MSE for Xp (z)
depends on that of its polyphase components. For Xp (z), the MSE is (ǫmax = 0
for a PR TMUX)
Np −1
1 X
ǫp = [x̂p (n) − xp (n)]2 (7.9)
Np n=0
Mp −1 Np,l −1
1 X X
= [x̂p,l (n) − xp,l (n)]2
Mp Np,l n=0
l=0
Mp −1
1 X 1
= ǫp,l ≤ Mp ǫmax = ǫmax .
Mp Mp
l=0
Thus, independent of Mp , the user Xp (z) has a maximum MSE of ǫmax . In other
words, even if the polyphase components of each user are processed in Mp branches,
the overall reconstruction error for that specific user is only determined by the ap-
proximation error of the corresponding FB and for the case with Mp = 1. Therefore,
we need to design the prototype filter for the corresponding FB. Then, a recon-
figurable nonuniform TMUX is obtained by (i) interchanging the AFB and SFB,
and (ii) adding some time-varying commutators. Figure 7.2 shows the architecture
of these time-varying commutators. These adjustable commutators do not require
extra arithmetic complexity but they do give some control hardware overhead, i.e.,
multiplexing of the polyphase components. Specifically, (7.3) describes this con-
trol mechanism but it does not actually require any arithmetic complexity. Such a
control mechanism is generally common for any polyphase realization.
7.4 System Parameters

The TMUX requires to design the prototype filter, say G(z), of the correspond-
ing FB as discussed in Section 3.3.1. This design requires the passband and
stopband ripples δc and δs ; roll-off ρ; and the number of TMUX channels M .
These are the basic system parameters but the application can also constrain
them. One such constraint arises from the multimode setups represented by Mp =
{M0 , M1 , . . . , MP −1 }. With these parameters, some characteristics such as the
channel sampling periods Tp , center frequencies Ωp , and the GB ∆ are determined.
104
7.4.1 Channel Sampling Periods

The proposed TMUX assigns Mp branches to the user Xp (z) which has a sampling
M
period of Tp . Then, Tp Mp = Ty where Ty is the sampling period of y(n) in Fig. 7.1.
PP −1 Mp PP −1 M
Note that with p=0 M < 1 we have a redundant TMUX whereas p=0 Mp = 1
gives a critically sampled TMUX.
7.4.2 TMUX Illustration

Figure 7.3 shows the frequency spectrum of y(n) for three cases where P = {2, 3, 4}.
These figures are obtained by an MDFT-based TMUX with M = 20. We only
adjust the commutators to choose the correct polyphase component. The corre-
sponding AFB filters are shown in Fig. 7.5(a). For these filters, ρ = 0.2 and only
the adjacent filters overlap. Further, Dg = 1 and we have offset αp,l (by 1) so that3
the user X0 has all its spectrum in ωT ≥0. In Fig. 7.3(a), Mp = {11, 7} whereas
in Figs. 7.3(b) and (c), we have Mp = {9, 5, 3} and Mp = {7, 3, 5, 1}, respectively.
We can define different multimode setups by the vectors Mp . A multimode setup
is the configuration of a number of users occupying the whole frequency spectrum.
If Mp changes, the multimode setup changes also but the filter coefficients need
not change.
Figure 7.4 illustrates a CMFB-based TMUX with the AFB filters of Fig. 7.5(b).
As ρ = 0.5, the adjacent filters overlap. Furthermore, Dg = 1 and we have also
offset αp,l (by 1). In Fig. 7.4(a), Mp = {10, 7} whereas in Figs. 7.4(b) and (c), we
have Mp = {8, 5, 3} and Mp = {7, 3, 4, 1}, respectively.
7.4.3 Choice of GB
As the proposed TMUX uses maximally decimated FBs, the user spectra can over-
lap as in Case I. In the TMUX illustration, both of the Cases II and III are covered
to show that the proposed TMUX can be reconfigured for any ∆. The GBs affect
the spectrum efficiency but not the reconstruction error or the filter design.
A specific ρ allows Case II with a minimal GB. With Dg = 1, the MDFT-based
TMUX achieves Case II if ωs T = 2π M whereas the CMFB-based TMUX allows a
π
minimal GB if ωs T = M . This can be seen in Figs. 7.3 and 7.4. Figure 7.3 has no
extra GB as for this specific4 MDFT FB, ωs T = 2π M . In contrast, Fig. 7.4 follows
Case III as ρ = 0.5 gives ωs T < 2πM .
7.4.4 Choice of Center Frequency

Some applications, e.g., cognitive radios [44, 59–62, 156–158], may require time-
varying center frequencies. The proposed TMUX allows this by simple adjustments
of the commutators. Assume, for example, Mp = {2, 3, 4}, Dg = 1, and an offset
3 This is for illustration purposes only. The spectrum of the user X , in Fig. 1.1, is partially
0
centered on ωT = 0 and ωT = 2π.
4 The design method for this FB is outlined in [76] where ω T = 2π .
s M
105
(a)
Mag. [dB]
0
−20
X0 X1
−40
0 0.25π 0.5π 0.75π π 1.25π 1.5π 1.75π 2π
ωT [rad]
(b)
Mag. [dB]
0
−20
X0 X1 X2
−40
0 0.25π 0.5π 0.75π π 1.25π 1.5π 1.75π 2π
ωT [rad]
(c)
Mag. [dB]
0
−20
X0 X1 X2 X3
−40
0 0.25π 0.5π 0.75π π 1.25π 1.5π 1.75π 2π
ωT [rad]
Figure 7.3: Illustration of the MDFT-based TMUX where 2, 3, and 4 users have
occupied the frequency spectrum ωT ∈[0, 2π]. Here, ∆ = 0.
(by 1) in αp,l . Then, X0 (z) uses two branches corresponding to F1 (z) and F2 (z)
where α0,0 = 1 and α0,1 = 2. On the other hand, X1 (z) uses three branches F4 (z)−
F6 (z). Further, X2 (z) occupies four frequency bands covered by F8 (z) − F11 (z).
In an M -channel MDFT FB, the center frequency for Fm (z), m = 0, 1, . . . , M − 1,
is 2πm 3π
M . Therefore, the center frequency for X0 (z) is M whereas X1 (z) and X2 (z)
10π 19π
have center frequencies M and M , respectively. If Xp (z) occupies Mp branches,
its center frequency Ωp is5
Mp −1
2π 1 X
Ωp = αp,l . (7.10)
M Mp + 1
l=0
7.5 Implementation Cost

Any FB realization can be adopted for the proposed TMUX. The main aim is thus
to determine the system parameters so as to minimize the arithmetic complexity.
The TMUX may also require some considerations with respect to the number of
multimode setups. This section treats the problem of determining the system pa-
rameters. In Scenario I, we assume that all system parameters are adjustable. We
5 For 2π
a CMFB, the term M
must be modified.
106
(a)
Mag. [dB]
0
−20
X0 X1
−40
0 0.2π 0.4π 0.6π 0.8π π
ωT [rad]
(b)
Mag. [dB]
0
−20
X0 X1 X2
−40
0 0.2π 0.4π 0.6π 0.8π π
ωT [rad]
(c)
Mag. [dB]
0
−20
X0 X1 X2 X3
−40
0 0.2π 0.4π 0.6π 0.8π π
ωT [rad]
Figure 7.4: Illustration of the CMFB-based TMUX where 2, 3, and 4 users have
occupied the frequency spectrum ωT ∈[0, π]. Here, ∆ > 0.
outline a general optimization problem to minimize the overall per-sample arith-

metic complexity. The Scenario II assumes a fixed transition band for G(z) and a
different optimization problem is discussed. We consider the MDFT-based TMUX
but the discussion can be extended to CMFBs after appropriate modifications.
Scenario I
To process M complex-valued input samples, the M -channel MDFT FB requires
M (4K + log2 M − 3) + 4 real multiplications and M (4K + 3 log2 M − 1) − 4 real
additions [109]. Thus, the total per-sample arithmetic complexity is 8K+4 log2 M −
4. The order of the prototype filter G(z), in an MDFT FB, can be obtained as [88]
NM DF T ≈ KM (7.11)
2 2π
≈ − log10 (10δc δs ) π(1−ρ)
3 2π
M − M
2 2M
≈ − log10 (10δc δs ) .
3 1+ρ
Consequently, we define the total per-sample arithmetic complexity as
−32 log10 (10δc δs )
C(M, ρ, δc , δs ) = + 4 log2 M − 4. (7.12)
3(1 + ρ)
107
(a)
0
−20
Mag. [dB]
−40
−60
−80
0 0.25π 0.5π 0.75π π 1.25π 1.5π 1.75π 2π
ωT [rad]
(b)
0
−10
Mag. [dB]
−20
−30
−40
−50
0 0.25π 0.5π 0.75π π
ωT [rad]
Figure 7.5: The AFB filters for the TMUXs illustrated in Figs. 7.3 and 7.4 where
M = 20. (a) MDFT FB with ρ = 0.2. (b) CMFB with ρ = 0.5.
As shown in Fig. 7.6, for a fixed ρ, a larger M increases C(M, ρ, δc , δs ). Further,

for a fixed M , decreasing ρ will increase C(M, ρ, δc , δs ). Thus, we can determine M
so that C(M, ρ, δc , δs ) does not increase unnecessarily. Note also the discontinuity
of the curves due to the integer values of K.
We can use other formulae, e.g., Kaiser [89], to obtain NM DF T . This slightly
changes (7.12) but not the overall conclusions. In addition, (7.11) does not account
for the power complementary property of G(z). We found experimentally that if
NM DF T is estimated by (7.11) and G(z) is to satisfy the power complementary
property with an error of δc , we need to increase NM DF T by 20% − 30%.
Scenario II
If the user spectra overlap, we can fix the transition band of G(z). For example, we
can increase M and allow the overlap of several user spectra. Then, (7.11) becomes
Φ − 4π
3 log10 (10δc δs )
NM DF T = KM = = (7.13)
Ψ Ψ
which would give
Φ
K= . (7.14)
MΨ
108
(a)
C(M,ρ,δc,δs)
40
20
0
60 50 40 30 0.5
20 10 1
ρ
M
(b)
C(M,ρ,δc,δs)
80
60
40 0
60 50 40 30 0.5
20 10 1
ρ
M
Figure 7.6: Arithmetic complexity of the M -channel MDFT FB using (7.12). (a)
δs = δc = 0.01. (b) δs = δc = 0.0001.
Here, Ψ represents the transition band of G(z). With constant δc and δs , the value
Φ
of Φ becomes fixed. Then, Ψ can be adjusted and (7.12) becomes
8Φ
C(M, ρ, δc , δs ) = + 4 log2 M − 4. (7.15)
MΨ
To minimize C(M, ρ, δc , δs ), we should have
∂C(M, ρ, δc , δs ) 4 −2Φ
= ( + log2 e) = 0 (7.16)
∂M M MΨ
and the optimum number of channels is
2Φ
Mopt = . (7.17)
Ψ log2 e
As shown in Fig. 7.7, C(M, ρ, δc , δs ) indeed decreases by increasing M . According

to Section 3.3, K is selected to be an integer. With an integer K, the curves in
Fig. 7.7 will be sparser but the general trend will still be preserved. Then, Mopt
may not be valid but a valid (integer) value of Mopt can easily be obtained by
Φ
adjusting Ψ .
Figure 7.7 suggests to choose M to be larger than that required by the applica-
tion. This not only decreases the arithmetic complexity, it increases the flexibility
in terms of Tp , Ωp , and Mp . If M is close to Mopt , the value of C(M, ρ, δc , δs ) does
not change significantly and we can also avoid a (possibly) large Mopt by choosing
109
90
80 Φ/Ψ=105
Φ/Ψ=70
Φ/Ψ=52
70 Φ/Ψ=42
C(M,ρ,δc,δs)
Φ/Ψ=35
60
50
40
30
20 40 60 80 100 120 140 160 180 200

M
Figure 7.7: Arithmetic complexity of the M -channel MDFT FB using (7.15) and
δs = δc = 0.001.
a moderate M < Mopt . Some systems may anyhow favor a large M from the
channel equalization point of view. Then, the channel is divided into very small
frequency bands in which it has a flat frequency response [120, 159].
Although a large M is advantageous due to the minimization of C(M, ρ, δc , δs ),
this may not be useful from another point of view. In hardware realization, the
implementation complexity could be affected by the number of connections, etc. In
the case of the proposed TMUX, these connections are directly proportional to M .
Consequently, one should bear this in mind so as not to increase the complexity of
these connections [92].
7.5.1 Choice of M and ρ

Consider different time instants with P (t) users. Assume also that each user occu-
pies Mp (t), p = 0, 1, . . . , P (t) − 1, branches and Dg (t) branches are used as GBs.
Thus,
P (t)−1
X
(P (t) − 1)Dg (t) + Mp (t) ≤ M (7.18)
p=0
meaning that Mp (t), P (t), and Dg (t) determine M . Then, an appropriate ρ can
be chosen to meet the criterion on C(M, ρ, δc , δs ). With fixed ripples δc and δs , we
110
65
60
55
C(M,ρ,δc,δs)
50
45
40
35
0 0.2 0.4 0.6 0.8 1

ρ
Figure 7.8: Trend of C(M, ρ, δc , δs ), in (7.12), for different values of ρ where P = 3,

M = 10, and δc = δs = 0.001.
can solve an optimization problem as
min C(M, ρ, δc , δs ) (7.19)

M,ρ,t
P (t)−1
X
subject to (P (t) − 1)Dg (t) + Mp (t) ≤ M.
p=0
With a maximum value for P (t), the flexibility can be defined. To exemplify, as-
sume P = 3, M = 10, and Dg = 1. Thus, we can obtain all possible combinations
P2
of Mp so that (7.18) holds. In other words, p=0 Mp ≤8 gives the number of multi-
mode setups defined by the vectors Mp . Here, between one and three users occupy
the whole frequency spectrum. This gives 41 multimode setups. Letting Dg = 0
P2
results in p=0 Mp ≤10 and it allows 67 multimode setups. Here, Mp = {2, 3, 5}
has been considered equivalent to Mp = {3, 2, 5} or Mp = {3, 5, 2}. However,
these three multimode setups assign different center frequencies Ωp to each user.
Figure 7.8 shows the trend of C(M, ρ, δc , δs ) in an MDFT-based TMUX with
P = 3, M = 10, and δc = δs = 0.001. As can be seen, we can choose a ρ to satisfy
the constraints on C(M, ρ, δc , δs ). The discussion above considers Scenario I. With
Scenario II, ρ is fixed giving a rather different problem. However, both problems
minimize C(M, ρ, δc , δs ) subject to the constraints in (7.18).
Even though the proposed TMUX allows a large number of multimode setups,
an standardized communication system may anyhow require fewer setups. This
TMUX is still advantageous in an standardized system with fewer modes. As an
example, the long term evolution requires 6 scalable bandwidths as 1.4, 3, 5, 10,
111
15, and 20 MHz [160]. For these bandwidths, it may still be advantageous to use
the proposed TMUX rather than using a number, e.g., 6, of separately designed
TMUXs. Generally, this TMUX becomes more advantageous if the number of
multimode setups is large.
7.5.2 Filter Design Restrictions

According to Section 3.3.1, G(z) has a passband edge which is proportional to
1
M . For moderate M , the well-known estimation formulae, viz., Hermann, Kaiser,
and Bellanger, can be used [69]. At very small (or large) passband edges, these
estimation formulae suffer from inaccuracies. To overcome this, we must consider
other order estimation formulae, e.g., [90]. Irrespective of the order estimation,
the numerical challenges to design very sharp G(z) will always exist if M increases.
However, we can also use efficient design techniques, e.g., frequency response mask-
ing [79], to obtain very narrow band filters.
7.6 Comparison with Existing Multimode TMUXs

This section compares the proposed TMUX to the existing multimode TMUXs
discussed in Chapters 5 and 6. Issues such as flexibility, realization regularity,
spectrum efficiency, and filter design are considered.
7.6.1 Flexibility
In the proposed TMUX, the bandwidth of each user is an integer multiple of the
GRB. This GRB is fixed and it is equal to the passband width of G(z). This is
similar to [64–66] as they also allow bandwidths equal to integer Bp multiples of
the GRB with a spectral width of 2π(1+ρ) Ap . In contrast to the fixed GRB in the
proposed TMUX, the width of the GRB in [64–66] can have different values. The
TMUXs in [46, 63] start from a fixed GRB with a spectral width of 2π(1+ρ)L . Then,
they allow each user to occupy bandwidths which are rational Rp multiples of the
GRB.
From the flexibility point of view, [46, 63] are superior to the proposed TMUX
as well as [64–66]. However, the flexibilities of the proposed TMUX and [64–66]
may be comparable if we do not consider the flexibility in the center frequencies.
One can increase the flexibility of the proposed TMUXs by increasing the number
of branches but it then brings challenges with filter design. Note that any of these
TMUXs can be configured to support a given multimode setup. To do so, proper
A
values of L, ρ, Bpp , M , and Mp need to be determined so that
2π(1 + ρ)Rp 2π(Mp + 1) 2π(1 + ρ)Bp

= = . (7.20)
L M Ap
Table 7.1 compares these parameters for some of the multimode setups in Sec-
tion 7.4.2. Here, the values for ρ and L are chosen in accordance with those out-
112
Table 7.1: Comparison of parameters to support the same multimode setup.
Ref. Parameters
Proposed M = 20, Mp = {11, 9, 7, 5}
[46, 63] L = 12, ρ = 0.05, Rp = {6.86, 5.71, 4.57, 3.43}
A
[64–66] ρ = 0.2, Bp = {2, 2.4, 3, 4}
p
lined in earlier chapters. We use the MDFT-based TMUX as [46, 63–66] transmit
a complex composite signal. Furthermore, we do not compare the reconstruction
errors and the complexity as the proposed TMUX can be PR whereas those in
[46, 63–66] are NPR. Further, the frequency shifters in [46, 63–66] would require
additional hardware. On the other hand, the commutators in the proposed TMUX
may also give some control hardware overhead. Therefore, a detailed complexity
comparison is rather subtle.
7.6.2 Spectrum Efficiency

The proposed TMUX is critically sampled and the user spectra can overlap. This
is in contrast to the redundant TMUXs of [46, 63–66]. Thus, the proposed TMUX
has a better spectrum efficiency. According to Section 7.2, the application mainly
determines the amount of the GB. If the application allows it, the proposed TMUX
can support scenarios with spectra overlap. This increases the spectrum efficiency.
7.6.3 Direct or Indirect Design

The proposed TMUX allows one to use the filter design techniques for general
modulated FBs. These techniques are direct as they include ISI and ICI in their
filter design [118, 119]. However, [46, 63–65] indirectly design the filters by utilizing
their redundancy. This redundancy simplifies the filter design but it has some
overdesign thereby increasing the implementation cost. Therefore, there is a trade-
off between filter design complexity and implementation cost.
Generally and without the knowledge of multimode setups, the indirect design
may be the only option. However, if all Mp (or the most common ones) are known,
we can use direct design [66]. Consequently, the proposed TMUX is superior to
those in [46, 63–65] as it always designs the filters directly.
7.7 Conclusion
In this chapter, a reconfigurable nonuniform TMUX based on uniform modulated
FBs was introduced. It uses polyphase processing and each user is processed by a
number of TMUX branches. Any user can occupy different bandwidths and center
frequencies. This comes at the expense of some adjustable commutators and re-
quires no additional arithmetic complexity. The chapter considered both CMFBs
and MDFT FBs to obtain reconfigurable TMUXs. Issues related to filter design,
113
realization, implementation cost, and parameter selection were discussed. Further-

more, the proposed TMUX was compared to some existing multimode TMUXs.
114
8
Applications to Cognitive Radios
This chapter discusses two approaches for frequency allocation and reallocation
used in baseband processing of cognitive radios. These approaches can be used
depending on the availability of a composite signal comprising several user signals
or the individual user signals. After an introduction in Section 8.1, Section 8.2
outlines Approach I which is based on flexible frequency-band reallocation (FFBR)
networks. Then, Section 8.3 discusses Approach II based on transmultiplexers
(TMUXs). Discussions on reconfigurability with respect to cognitive radios are
also provided. Section 8.4 treats the issues regarding the choice of the center
frequencies. Finally, some concluding remarks are given in Section 8.5.
8.1 Introduction
One perspective in the design of communication systems is to increase the spec-
trum utilization using cognitive radios. A cognitive radio is a network of intelligent
co-existing radios which senses1 the environment to find available frequency slots,
white spaces, or spectrum holes [59, 61]. Then, it modifies its transmission char-
1 Spectrum sensing is mostly done by digital baseband processing. As these algorithms require
a long time to detect an available channel, one can combine radio frequency (RF) and analog
circuits for a faster spectrum sensing [161].
115
8. APPLICATIONS TO COGNITIVE RADIOS
Power Spectrum Density

Overlay Transmission Primary User
0 Frequency f
Figure 8.1: Overlay approach for spectrum sharing in a cognitive radio.
From receiver Radio ADC and Baseband To user

To transmitter Frequency DAC Processing From user
Reconfiguration
Figure 8.2: General block diagram of a cognitive radio composed of digital to analog
converters (DACs), analog to digital converters (ADCs), baseband processing, and
RF part.
acteristics to use that particular frequency slot. Figure 8.1 illustrates the overlay2
spectrum sharing [164] or the opportunistic spectrum access [157, 162] or the dy-
namic spectrum access [60]. Here, secondary users occupy the frequency slots not
used3 by the primary users. One of the main tasks in a cognitive radio is conse-
quently the spectrum mobility [61, 156] or the dynamic frequency allocation [59]
or the dynamic spectrum allocation [62, 157]. This chapter uses the term dynamic
frequency-band allocation (DFBA). Being dynamic means that the transmission
parameters, e.g., bandwidth, center frequency, transmission power, communication
standard, etc., may vary with time [156]. One should at least be able to change the
center frequency and bandwidth although other parameters may also change [60].
This is also referred to as the reconfigurability [4, 6, 59–62]. The general block
diagram of a cognitive radio is shown in Fig. 8.2.
Another perspective in the design of communication systems calls for satellites
to play a complementary role supporting various wideband services accessible to
everybody everywhere [32–39]. For this purpose, the European space agency has
proposed three major network structures for broadband satellite-based communi-
cation systems [37]. This requires an efficient use of the limited available frequency
2 The underlay spectrum sharing or the ultra wideband [162], exploits spread spectrum. Users
transmit at certain portions of spectrum regarded as noise by the primary (licensed) users [163].
3 Under certain conditions, the secondary users need not wait for a vacant channel. This allows
a simultaneous transmission over the same time or frequency [165].
116
spectrum by satellite on-board signal processing [32–57].

Similar to satellite-based communication systems which require both on-ground
DFBA and on-board dynamic frequency-band reallocation (DFBR), the ad hoc-
or infrastructure-based cognitive radios can also utilize these tasks. In the ad
hoc-based networks, individual users can utilize DFBA while DFBR can be per-
formed by the base stations of an infrastructure-based network. However, DFBA
can also be deployed by the individual users in an infrastructure-based network.
Furthermore, DFBR is applicable to composite signals comprising several users
with different but fixed bandwidths. On the other hand, DFBA is applicable to
individual users enabling them to change their bandwidths as well.
Both DFBA and DFBR can be realized using interpolation/decimation with
variable parameters. For large sets of variable conversion factors, the implementa-
tion complexity of this approach increases. Complexity reduction can be achieved
by reconfigurable structures. These solutions must perform various tasks by simple
modifications and without hardware changes. These modifications are applied to
(i) the values of some multipliers, or (ii) the operation of some commutators and
channel switches. Further, the filter coefficients do not change which enables us to
solve the filter design problem only once and offline. Specifically, one must be able
to reprogram the same hardware.
As both DFBR and DFBA process baseband digital signals, this chapter focuses
on the frequency allocation and reallocation for baseband processing of cognitive
radios. We consider operations related to changing the center frequency and band-
width of the user signals. Similar to bentpipe satellite payloads [33, 38], these do
not require any modulation/demodulation, coding/decoding, etc.
This chapter discusses the reconfiguration and parameter selection issues when
adopting the DFBA and DFBR for cognitive radios. Two basically different ap-
proaches, referred to as Approach I and II, are discussed. They are appropriate
depending on the availability of (i) a composite signal comprising several user sig-
nals, or (ii) the individual user signals. An alternative to deal with composite
signals is to first divide, using a filter bank (FB), the composite signal into its cor-
responding user signals and, then, use Approach II on each user signal. However,
it is more efficient to directly use Approach I. Consequently, we will not discuss
this alternative. Combinations of Approaches I and II provide an increased degree
of freedom to allocate and reallocate the user signals.
8.2 Approach I: Use of DFBR Networks

For DFBR, we assume that signals from several users, e.g., mobile handsets in a
cellular network or computers in a wireless local area network (WLAN), have been
added into a composite signal at a main station, e.g., a base station in a cellular
network or an access point in a WLAN. This main station finds available frequency
slots and reallocates each user to one of them. In a dynamic communication system,
users can occupy any bandwidth at any time. Such a main station is similar to a
bentpipe satellite payload [33, 38] with its idea of operation shown in Fig. 8.3. The
117
Output signal 1
3 1
DFBR Network
In 1 Out 1 2p
1 2 3
2p
Out 2 5 4
In 2 2p
4 5 6 Out 3
2p
2 6
wTout [rad]
2p
Figure 8.3: Approach I: DFBR networks process composite signals to reallocate

users from one composite input signal to another composite output signal.
composite signals are processed by the DFBR network and the users are reallocated
to new frequency slots. These slots could be different antenna beams of a satellite
payload or different cells in a cellular network. Multiple antennas of a satellite
payload perform signal filtering in spatial rather than frequency domain. This is
similar to the techniques utilizing multiple antennas for cognitive radios [166]. The
DFBR networks could also be useful for the centralized4 cooperative5 cognitive
radios [167]. They can also be considered as secondary base stations in licensed
band cognitive radios [168]. In licensed band networks, the DFBR can coexist with
the primary networks and opportunistically operate in an overlay transmission.
The DFBR network can be a mutli-input multi-output system as it can have
a number of composite input and output signals. The dynamic nature of the
DFBR networks allows the users to occupy any suitable6 frequency slot in a time-
varying manner. Each user can be sent in contiguous or separate frequency bands
requiring contiguous or fragmented DFBR [62]. The separate frequency bands
can be considered as a multi-spectrum transmission. Specifically, as white spaces
are mostly fragmented [169], the user signals can be transmitted in several non-
contiguous frequency bands.
8.2.1 Structure of the DFBR Network

This chapter uses the term DFBR which is essentially similar to FFBR in Chapter 4.
Consequently, the DFBR network has the structure of the FFBR network as in,
e.g., Fig. 4.3. One can in principle use any of the FFBR networks outlined in
Sections 4.1.1 and 4.1.2.
4A centralized entity, e.g., the DFBR network, controls the spectrum allocation and access.
5 Non-cooperative networks do not share the interference measurements of each user with the
others. Generally, cooperative networks are more accurate in sensing the spectrum [61].
6 The frequency slot depends on spatial and temporal parameters such as the number of avail-
able slots, user movement, and activity of primary users, etc. [156]. The operation of the DFBR
network is independent of these parameters.
118
Granularity Band (BGRB) Guardband (GB)

(a)
1BGRB
wT
0 2p
Additional GB
(b) User Bandwidth
1.75BGRB 3.5BGRB
wT
0 2p
Multiplexing Bandwidth
(c)
2BGRB 3BGRB
wT
0 2p
Figure 8.4: User bandwidth versus multiplexing bandwidth.
8.2.2 User Bandwidth Versus Multiplexing Bandwidth

The DFBR networks divide the user signals into a number of granularity bands
(GRBs) on which the frequency shifts are performed. As the DFBR networks
utilize FBs, the multiplexing bandwidth must be an integer multiple of the GRB.
The DFBR networks perform frequency shifts on users whose bandwidths are,
in general, rational multiples of the GRB. In this way, bandwidth-on-demand is
supported. An important issue is to ensure that the users do not share a GRB which
can be achieved by allowing some extra guardband (GB). However, the extra GBs
affect the spectrum efficiency resulting in a trade-off. As in Fig. 8.4, a multiplexing
bandwidth contains a user bandwidth and some extra GB.
8.2.3 Reconfigurability
A cognitive radio should adjust its operating parameters without hardware mod-
ifications [4]. It is built on the platform for a software defined radio with the
processing mainly in the digital domain [162]. There are several reconfigurable
parameters such as operating frequency, modulation method, transmission power,
communication standard, etc. In the context of adaptable operating frequency, or
flexible frequency carrier tuning [62], a cognitive radio changes its operating fre-
quency. However, this should not restrict the system throughput and hardware.
The DFBR networks can perform any frequency shift of any user having any
bandwidth, using a channel switch. This switch seamlessly directs different FB
channels to their desired outputs and requires no arithmetic complexity. In addi-
tion, the system parameters are determined and fixed only once offline. Then, the
reconfigurable operation is performed by reconfiguring the channel switch online.
119
(a)
|X(ejωT)| [dB]
0
−20
−40 X X1 X2 X3
−60 0
0 0.25π 0.5π 0.75π π 1.25π 1.5π 1.75π 2π
ωT [πrad]
(b)
|Y1(ejωT)| [dB]
0
−20
−40 X2 X0 X3 X1
−60
0 0.25π 0.5π 0.75π π 1.25π 1.5π 1.75π 2π
ωT [πrad]
(c)
|Y2(ejωT)| [dB]
0
−20
−40 X1 X3 X0 X2
−60
0 0.25π 0.5π 0.75π π 1.25π 1.5π 1.75π 2π
ωT [πrad]
Figure 8.5: Input pattern and the reallocated outputs using the channel switch
configurations in Figs. 8.7 and 8.8.
Here, the user bandwidths are predetermined but can be arbitrary. The DFBR
network makes a hand off by changing the operating frequency [61].
Figures 8.5 and 8.6 show two cases where, respectively, four and three users have
occupied the whole frequency band between [0, 2π]. To generate these user signals,
the multimode TMUX of Chapter 5 has been used. In Fig. 8.5(a), the user signals
{X0 , X1 , X2 , X3 } occupy, respectively, user bandwidths of {1, 2.9, 3.6, 1.9} GRBs.
Each GRB has a width of 2π 2ǫπ
Q − Q with Q = 10 and ǫ = 0.125. In Fig. 8.6(a),
the user signals {X0 , X1 , X2 } occupy {1, 6.9, 1.9} GRBs, respectively. As can be
seen, the user signals can occupy any rational number of GRBs. To ensure that
the users do not share a GRB, one can add some extra GB. This difference in the
amount of the GBs between different users can be recognized from Figs. 8.5 and
8.6.
These examples assume the DFBR network to operate on the same antenna
beam. By having several DFBR networks, the users can be reallocated between
different antenna beams as well. This requires a duplication of DFBR networks
and a channel switch capable of directing the user signals between different DFBR
networks. Each branch of the channel switches in Figs. 8.7–8.10 represents the
operation of two FB channels as each GRB contains two FB channels. Specifically,
the values of N , M , and L, in Fig. 4.3, are 20, 10, and 2 respectively.
120
(a)
|X(ejωT)| [dB]
0
−20
−40 X X1 X2
−60 0
0 0.25π 0.5π 0.75π π 1.25π 1.5π 1.75π 2π
ωT [πrad]
(b)
|Y1(ejωT)| [dB]
0
−20
−40 X1 X2 X0
−60
0 0.25π 0.5π 0.75π π 1.25π 1.5π 1.75π 2π
ωT [πrad]
(c)
|Y2(ejωT)| [dB]
0
−20
−40 X2 X0 X1
−60
0 0.25π 0.5π 0.75π π 1.25π 1.5π 1.75π 2π
ωT [πrad]
Figure 8.6: Input pattern and the reallocated outputs using the channel switch
configurations in Figs. 8.9 and 8.10.

Figure 8.7: Scenario I. Figure 8.8: Scenario II.


Figure 8.9: Scenario I. Figure 8.10: Scenario II.
121
User 1 at t1,B1,f1
wT
2p
TMUX
t,B,f User 1 at t2,B2,f2
Out
wT
2p
User 1 at t3,B3,f3
wT
2p
Figure 8.11: Approach II: TMUXs used to perform DFBA. At any time tk , each
user can decide its bandwidth Bk and operating frequency fk .
8.2.4 Modifications
The use of DFBR networks in cognitive radios needs some modifications which
are mainly related to the choice of the system parameters. For different system
parameters, the implementation complexity may be different. However, once the
parameters are chosen, the implementation complexity remains constant and the
system can be easily reconfigured on the same hardware platform.
For the DFBR networks, the width of a GRB must be proportional to that of
the spectrum holes. Thus, one requires to choose a value for the BGRB = 2π(1−ǫ)
Q ,
in Fig. 8.4, such that the bandwidth of any spectrum hole can be represented as a
rational multiple of BGRB .
8.3 Approach II: Use of TMUXs

Using a TMUX-based solution, each user terminal can adjust its operating fre-
quency and bandwidth. The basic idea is depicted in Fig. 8.11 where different band-
widths and center frequencies can be generated using multirate signal processing
techniques. These TMUXs can also be regarded as the time-spectrum blocks [158]
which can transmit any amount of data at any time interval and on any portion of
the frequency spectrum. This finds application if licensed users choose frequency di-
vision multiple access and/or time division multiple access as their spectrum access
mode. Then, the spectrum holes are identified in the time/frequency plane [170].
As shown in Fig. 8.12, the interpolation part represents the transmitter where a
variable filter places the desired user signal at the required center frequency. The
receiver, i.e., the decimation part, is designed accordingly so that the input signal
is recovered with reasonable and controllable levels of error.
Similar to straightforward DFBR solutions, one can use conventional nonuni-
form TMUXs to place users with different bandwidths at different center frequen-
cies. This becomes inefficient when simultaneously considering the increased num-
122
Interpolation Decimation
(a) x1 v1 v2 y
M G(z) H(z) M
T1 T2
(b) X1(ejwT1)
X1
wT1
0 2p/M 4p/M 6p/M 2p
(c) G(ejwT2)
wT2
0 2p/M 4p/M 2p
(d) V2(ejwT2)
X1
wT2
0 2p/M 4p/M 2p
Figure 8.12: Principle of TMUXs using multirate building blocks.
ber of communication scenarios and the desire to support dynamic communications.
In this context, TMUX structures of the general form shown in Fig. 8.13 are
introduced. In the synthesis FB, the system Cp performs interpolation by a rational
ratio Rp whereas the system Ĉp in the analysis FB performs decimation by a
rational ratio Rp . These blocks enable one to transmit and receive baseband signals,
having arbitrary bandwidths, through a common channel.
8.3.1 Structure of the TMUX

Any of the TMUXs in Chapters 5–7 can be used here. Note that the TMUX
in Chapter 7 has a rather different structure. Instead of variable lowpass filters
and frequency shifters as in Chapters 5–6, it performs bandpass rational SRC
using flexible commutators and fixed bandpass filters. However, one can generally
describe it in terms of Fig. 8.13.
8.3.2 Reconfigurability
The cognitive radio must adjust its operating frequency and bandwidth without
hardware modifications. The DFBR networks partially provide this capability but
123
jw0n ^
e e-jw0n
x0(n0) C0 C^0 x^0(n0)
^
ejw1n e-jw1n
^
y(n) y(n)
x1(n1) C1 C^1 ^x (n1)
1
^
ejwP-1n e-jwP-1n
xP-1(nP-1) CP-1 C^P-1 ^x (nP-1)
P-1
Figure 8.13: General structure of a multimode TMUX where systems Cp and Ĉp
perform rational sampling rate conversion (SRC).
they have no control over the user bandwidth. In contrast, the TMUX-based
approaches add reconfigurability to the user bandwidth as well. Furthermore, these
solutions bring flexible receiver signal filtering [62] by changing the transmitter and
receiver filters.
As can be seen from Figs. 8.5 and 8.6, the TMUX allows different numbers
of users, e.g., four and three, with different user bandwidths to occupy the whole
frequency band between [0, 2π]. These TMUXs provide this full reconfigurability
without any hardware changes.
8.3.3 Modifications
Similar to the DFBR networks, one requires certain system parameters to eliminate
the need for any hardware change while having a simple reconfigurability.
Regarding DFBA, there are different ways to perform SRC which could be
useful in different scenarios. The TMUX in [46, 63] generates a GRB through
integer interpolation by, e.g., W , resulting in BGRB = 2π(1+ρ) W where ρ is the
roll-off. Then, rational Rp multiples of BGRB can be created using the Farrow
structure. One can determine BGRB according to the bandwidth of the spectrum
holes. Then, any user may occupy any rational number of spectrum holes.
The TMUX in [64–66] assumes no GRBs and it allows the users to occupy any
portion of the spectrum. It utilizes the Farrow structure to perform general rational
A
SRC by, e.g., Rp = Bpp . This allows one to cover a large set of user bandwidths.
Here, one can also assume a GRB of size BGRB = 2π(1+ρ)
Ap . Then, users can have
bandwidths which are integer Bp multiples of BGRB .
Although [67, 68] propose a slightly different TMUX, one can also assume
BGRB = 2π(1+ρ)
M . Then, users have bandwidths which are integer Mp multiples
of BGRB . Note that this applies to the case with modified discrete Fourier trans-
124
form FBs. For cosine modulated FBs, similar formulae can be derived.
8.4 Choice of Frequency Shifters

To perform a hand off without loss of information, the DFBR network requires
the users not to share a GRB. Consequently, a lossless reallocation requires to (i)
generate appropriate frequency division multiplexed (FDM) input patterns, and
(ii) determine proper parameters for the DFBR networks. To generate the input
patterns, the reconfigurability of the TMUXs in Fig. 8.13 can be used. Specifically,
after generating the user signals with desired bandwidths, the frequency shifters
ωp , p = 0, 1 . . . , P − 1, can be computed to allow some extra GB. Here, an example
using the TMUX in Chapter 5 is provided. Assuming bandwidths that are rational,
e.g., Rp , multiples of BGRB , the subcarrier ωp for user p is
( F
2
0
if p = 0
ωp = Pp−1 Fp (8.1)
k=0 Fk + 2 if p6=0.
where Fp = ⌈Rp ⌉ 2π Q , p = 0, 1, . . . , k, with ⌈x⌉ being the ceiling of x. Here, Fp is

the multiplexing bandwidth and the ceiling operation ensures that the users do not
share a GRB. This formulation applies to the case where DFBA and DFBR are
simultaneously used. Otherwise, formulae similar to (6.16) can be used instead. In
practice, one may anyhow require some extra GB due to the design margins.
In Figs. 8.5 and 8.6, the users occupy Rp = {2.9, 3.6, 1.9, 6.9} GRBs. This ne-
cessitates an extra GB which is Ep = {0.1, 0.4, 0.1, 0.1} multiples of 2π Q . Therefore,
the spectrum efficiency decreases as some parts of the spectrum are not used. For
a set of values Rp , p = 0, 1, . . . , P − 1, about
2π
PP −1 PP −1
Q p=0 (⌈Rp ⌉ − Rp ) p=0 ⌈Rp ⌉ − Rp
ηdec = = (8.2)
2π Q
percent of the spectrum in [0, 2π] is not used by the DFBR network. In the examples
of Figs. 8.5 and 8.6, about 6% and 2% of the total spectrum is not used due to the
extra GBs.
To decrease these percentages, one can increase Q by, e.g., K times, which
would, in turn, decrease ηdec . In this case, (8.2) becomes
PP −1
p=0 ⌈KRp ⌉ − KRp
ηdec = . (8.3)
KQ
However, increasing Q would increase the order of the prototype filter P (z) in (4.4).
For each K, the prototype filter of the DFBR network would have a transition band
2πǫ
of KQ [42, 43]. As the order of a linear-phase finite-length impulse response filter
is inversely proportional to the width of its transition band [69], there is a trade-off
between the spectrum efficiency and the arithmetic complexity.
125
(a)
350
340 R = 1,2.9,3.6,1.9
Real operations
330 R = 1,6.9,1.9
p
R = 1.75,1.25,2,3.5
p
320
310
0 5 10 15
ηdec. (%)
(b)
15
10
ηdec. (%)
0
1 2 3 4 5 6 7 8 9 10
K
Figure 8.14: Trade-off between spectrum efficiency and arithmetic complexity. (a)
Decrease in spectrum efficiency versus per-sample arithmetic complexity. (b) Trend
of spectrum efficiency versus different K in (8.3).
With a K-fold increase in Q, the length of the prototype filter and the number
of FB channels increase proportional to K. Figure 8.14 shows the trend in spec-
trum efficiency with respect to the per-sample arithmetic complexity of the DFBR
network discussed in Section 4.2. Here, the examples of Figs. 8.5 and 8.6 as well
as that of [63] with Rp = {1.75, 1.25, 2, 3.5} are considered and K = 1, 2, . . . , 10.
As can be seen, a larger K increases the per-sample arithmetic complexity but it
decreases ηdec . The values of Rp mainly determine the maximum and minimum
amounts of ηdec . Hence, for every set of Rp , one can determine a K such that ηdec
and the per-sample arithmetic complexity are within the acceptable ranges.
8.5 Conclusion
This chapter discussed two approaches for the baseband processing in cognitive
radios based on DFBR and DFBA. They can support different bandwidths and
center frequencies for a large set of users and are easily reconfigurable.
In DFBR networks, composite FDM signals comprising several users are pro-
cessed and the users are reallocated to new center frequencies. They are applicable
to cognitive radios with multiple antennas [166]; centralized cooperative cognitive
126
radios [167]; and secondary base stations in licensed band cognitive radios [168]. In
DFBA networks, each user controls its operating frequency and bandwidth. These
networks can be regarded as the time-spectrum blocks [158].
The reconfigurability of DFBA and DFBR is performed either by a channel
switch, in DFBR, or by variable multipliers/commutators, in DFBA. The examples
in Figs. 8.5 and 8.6 show the increased flexibility to allocate and reallocate any user
to any center frequency by simultaneous utilization of DFBA and DFBR. In this
case, the individual users can occupy any available frequency slot and be reallocated
by the base station.
Basically, utilizing any of Approaches I and II in cognitive radios only requires
modifications imposed by the special choice of the system parameters. After choos-
ing these parameters once, we must design the filters to satisfy any desired level of
error. Then, the same hardware can be reconfigured in a simple manner.
127
128
9
Conclusion and Future Work
This thesis introduced nonuniform transmultiplexers (TMUXs) which enable dif-

ferent numbers of users, having different bandwidths, to share the whole frequency
band in a time-varying manner. The TMUXs support dynamic communications
and are reconfigurable. The TMUX in Chapter 5 uses a two-stage sampling rate
conversion (SRC) composed of integer and rational SRC schemes. Adjustable fre-
quency shifters, along with SRC of an arbitrary rational ratio, allow the users to
occupy any bandwidth and center frequency. On the other hand, Chapter 6 intro-
duced a TMUX where SRC is performed in the context of a conventional rational
SRC scheme. Here, integer SRC along with integer upsamplers/downsamplers and
adjustable frequency shifters are used. Both of these TMUXs use the Farrow struc-
ture. Further, Chapter 7 proposed TMUXs based on modulated filter banks (FBs).
Here, adjustable commutators are used enabling the users to occupy the desired
bandwidths and center frequencies. All of these TMUXs require neither filter re-
design nor hardware changes. Specifically, the filters are designed only once offline.
Then, the TMUXs are reconfigured by simple modifications without restricting the
system operation. These modifications apply to some multipliers, i.e., the frac-
tional delay of the Farrow structure and the variable parameters of the frequency
shifters, and some commutators.
The thesis also outlined solutions for flexible frequency-band reallocation (FFBR)
networks. The FFBR network in Chapter 4 enables different users, in different
composite multiple frequency/time division multiple access (MF/TDMA) input
129
9. CONCLUSION AND FUTURE WORK
signals, to be reallocated to different positions in different composite MF/TDMA

output signals. Although Chapter 4 considers real signals, the FFBR network can
generally be used for complex signals. The reallocation is done using simple modifi-
cations in the channel switch and it does not require any filter redesign or hardware
changes. The FFBR solutions do not restrict the bandwidth of the users or the
system operation.
In all of these chapters, the problems of filter design were discussed and illustra-
tive examples were provided. The thesis also outlined possible applications of the
proposed TMUXs and FFBR networks in the context of cognitive radios. These
applications are mainly related to spectrum mobility thereby allowing the users
to dynamically occupy any bandwidth and center frequency. As topics of future
research, the following issues are identified:
1. Application of cosine modulated FBs (CMFBs) to derive FFBR networks for

real signals. It would specifically be interesting to compare the flexibility in
frequency-band reallocation between complex modulated FBs and CMFBs.
2. Development of efficient methods to produce the complex periodic sequences

resulting from the variable frequency shifters. This is also referred to as direct
digital frequency synthesis. These methods should be reconfigurable as the
values of the frequency shifters, in the proposed TMUXs, can change online.
3. Realization of the proposed solutions in hardware for some emerging stan-

dards, e.g., long term evolution.
4. Derivation of integer SRC using multi-stage structures to reduce the arith-

metic complexity. Specifically, it would be interesting to propose efficient
realization structures for the TMUX in Chapter 5. Then, one can allow a
smaller granularity band thereby increasing the reconfigurability.
5. Development of methods to design the Farrow structure which further reduce

the implementation cost. A paper has already been published which outlines
an idea on how to reduce the orders of Farrow subfilters. This idea can be
directly applied to the TMUX in Chapter 5. Generalizations of this idea for
application to the TMUX in Chapter 6 would also be possible.
130
A
Derivation of (6.23)
The sequel derives (6.23). Figure A.1 shows the structure of the transmultiplexer
with the desired signal X̂d (z) in the analysis filter bank. Using (3.4), we have
X p (z) = Xp (z Ap )Gp (z), p = 0, 1, . . . , P − 1 (A-1)
ejw0n
-
x ~
x
x0(n0) 0 0 y0(n)
A0 G0(z) B0
jw1n ^
~
e e-jwdn
-
x1 x1 y (n) y(n) xd
g
x
x1(n1) A1 G1(z) B1 1
Bd ^ (z) -d
G Ad ^x (nd)
d d
- ~ ejwP-1n
xP-1 xP-1 yP-1(n)
xP-1(nP-1) AP-1 GP-1(z) BP-1
Figure A.1: The synthesis filter bank and the desired signal for the transmultiplexer
in Chapter 6.
131
APPENDIX A. DERIVATION OF (6.23)
where (3.2) gives

Bp −1
1 X 1
X̃p (z) = X p (z Bp WBk p ) (A-2)
Bp
k=0
Bp −1
1 X 1 1
= Xp ([z Bp WBk p ]Ap )Gp (z Bp WBk p ).
Bp
k=0
With frequency shifts by ωp and −ω̂d and assuming zp = ejωp T and zd = ej ω̂d T ,
we get
z
Yp (z) = X̃p ( ), (A-3)
zp
and
P
X −1
Y (z) = Yp (z). (A-4)
p=0
In the analysis filter bank,
P −1
X zd
X̆d (z) = Y (zzd ) = X̃p (z ), (A-5)
p=0
zp
which using (A-2) becomes

P −1 Bp −1
1 X X zd 1 zd 1
X̆d (z) = Xp ([(z ) Bp WBk p ]Ap )Gp ((z ) Bp WBk p ). (A-6)
Bp p=0 zp zp
k=0
Further,
X d (z) = X̆d (z Bd )Ĝd (z) (A-7)
P −1 BX
p −1
1 X zd B1p k Ap zd 1
= Xp ([(z Bd ) WBp ] )Gp ((z Bd ) Bp WBk p )Ĝd (z)
Bp p=0 k=0
zp zp
P −1 Bp −1
1 X X Ap Bd
zd Ap kA Bd
zd 1
= Xp (z Bp ( ) Bp WBp p )Gp (z Bp ( ) Bp WBk p )Ĝd (z).
Bp p=0 zp zp
k=0
Further, the downsampler by Ad gives

Ad −1
1 X 1
X̂d (z) = X d (z Ad WAmd ). (A-8)
Ad m=0
Ignoring the scaling factors, (A-8) and (A-7) give
P −1 BX
p −1 Ad −1
X X 1 Ap Bd
zd A p
kA
X̂d (z) = Xp ([z Ad WAmd ] Bp
( ) Bp WBp p )×
p=0 k=0 m=0
zp
1 Bd
zd B1p k 1
Gp ([z Ad WAmd ] Bp ( ) WBp )Ĝd (z Ad WAmd ). (A-9)
zp
132
After some manipulations, we have

P d −1 B
−1 AX p −1 mAp Bd
X X Ap Bd
Bp zd BAp
kA
X̂d (z) = Xp (z Ad Bp WAd ( ) p WBp p )×
p=0 m=0 k=0
zp
mBd
Bd
zd B1p k 1
) WBp )Ĝd (z Ad WAmd )
B
Gp (z Ad Bp WAd p ( (A-10)
zp
which is (6.23).
133
134
B
Derivation of (6.35)
Ap
The sequel derives (6.35) using Fig. B.1. For interpolation by Bp , we have
X
y3 (n) = x(k)g(nBp − kAp ). (B-1)
k
As X
y5 (n) = y3 (m)ĝ(n − mBp ) (B-2)
m
and
y(n) = y5 (nAp ), (B-3)
we get XX
y5 (n) = x(k)g(mBp − kAp )ĝ(n − mBp ). (B-4)
m k
Then, XX
y(n) = x(k)g(mBp − kAp )ĝ(nAp − mBp ). (B-5)
m k
Finally X X
y(n) = x(k) g(mBp − kAp )ĝ(nAp − mBp ). (B-6)
k m
Now, replace k with n − k and (6.35) follows.
135
APPENDIX B. DERIVATION OF (6.35)
y1(n) y2(n) y3(n) y4(n) y5(n)

x(n) Ap G(z) Bp Bp ^
G(z) Ap y(n)
Ap
Figure B.1: Cascade of interpolation and decimation by Bp for the transmultiplexer
in Chapter 6.
136
References
[1] J. H. Reed, Ed., Software Radio: A Modern Approach to Radio Engineering.

Prentice Hall; NJ, 2002.
[2] W. H. W. Tuttlebee, “Software-defined radio: facets of a developing technol-

ogy,” IEEE Personal Commun. Mag., vol. 6, no. 2, pp. 38–44, Apr. 1999.
[3] T. Hentschel and G. Fettweis, “Sample rate conversion for software radio,”
IEEE Commun. Mag., vol. 38, no. 8, pp. 142–150, Aug. 2000.
[4] F. K. Jondral, “Software defined radio: Basics and evolution to cognitive

radio,” EURASIP J. Wireless Commun. Networking, 2005.
[5] W. H. T. (Editor), Software Defined Radio: Enabling Technologies. New

York: Wiley, 2002.
[6] U. Ramacher, “Software-defined radio prospects for multistandard mobile

phones,” IEEE Computer, vol. 40, no. 10, pp. 62–69, Oct. 2007.
[7] A. A. Abidi, “Will the multimode wireless transceiver soon resemble a

software-defined radio?” in Proc. IEEE Int. Symp. Radio-Frequency Inte-
gration Techn., Singapore, 2009, pp. 1–4.
137
References
[8] M. Brandolini, P. Rossi, D. Manstretta, and F. Svelto, “Toward multistan-

dard mobile terminals–fully integrated receivers requirements and architec-
tures,” IEEE Trans. Microwave Theory Tech., vol. 53, no. 3, pp. 1026–1038,
Mar. 2005.
[9] A. C. Tribble, “The software defined radio: Fact and fiction,” in Proc. IEEE
Radio Wireless Symp., Orlando, Florida, USA, Jan. 2008, pp. 5–8.
[10] P. Koch and R. Prasad, “The universal handset,” IEEE Spectrum, pp. 36–41,
Apr. 2009.
[11] H. Elwan, H. Alzaher, and M. Ismail, “A new generation of global wireless
compatibility,” IEEE Circuits Devices Mag., vol. 17, no. 1, pp. 7–19, Jan.
2001.
[12] P. P. Vaidyanathan, Multirate Systems and Filter Banks. Englewood Cliffs,
NJ: Prentice-Hall, 1993.
[13] A. N. Akansu, P. Duhamel, L. Xueming, and M. de Courville, “Orthogonal
transmultiplexers in communication: A review,” IEEE Trans. Signal Pro-
cessing, vol. 46, no. 4, pp. 979–995, Apr. 1998.
[14] B. Farhang-Boroujeny and C. H. Yuen, “Cosine modulated and offset QAM
filter bank multicarrier techniques: A continuous-time prospect,” EURASIP
J. Adv. Signal Processing, vol. 2010, Article ID 165654, 2010.
[15] C. B. Ribeiro, M. L. R. de Campos, and P. S. R. Diniz, “Time-varying FIR
transmultiplexers with minimum redundancy,” IEEE Trans. Signal Process-
ing, vol. 57, no. 3, pp. 1113–1127, Mar. 2009.
[16] C. Y.-F. Ho, B. W.-K. Ling, Y.-Q. Liu, K.-L. Teo, and P. K.-S. Tam, “Opti-
mal design of nonuniform FIR transmultiplexer using semi-infinite program-
ming,” IEEE Trans. Signal Processing, vol. 53, no. 7, pp. 2598–2603, July
2005.
[17] X. M. Xie, S. C. Chan, and T. I. Yuk, “Design of linear-phase recombination
nonuniform filter banks,” IEEE Trans. Signal Processing, vol. 54, no. 7, pp.
2809–2814, July 2006.
[18] T. Chen, L. Qiu, and E.-W. Bai, “General multirate building structures
with application to nonuniform filter banks,” IEEE Trans. Circuits Syst.
II, vol. 45, no. 8, pp. 948–958, Aug. 1998.
[19] T. Liu and T. Chen, “Design of multichannel nonuniform transmultiplexers
using general building blocks,” IEEE Trans. Signal Processing, vol. 49, no. 1,
pp. 91–99, Jan. 2001.
[20] L. C. R. de Barcellos, P. S. R. Diniz, and S. L. Netto, “A generalized over-
sampled structure for cosine-modulated transmultiplexers and filter banks,”
Circuits Syst. Signal Processing, vol. 25, no. 2, pp. 131–151, 2006.
138
References
[21] H.-T. Chiang, S.-M. Phoong, and Y.-P. Lin, “Design of nonuniform filter
bank transceivers for frequency selective channels,” EURASIP J. Adv. Signal
Processing, vol. 2007, Article ID 61396, 2007.
[22] J. Kovac̆ević and M. Vetterli, “Perfect reconstruction filter banks with ra-
tional sampling factors,” IEEE Trans. Signal Processing, vol. 41, no. 6, pp.
2047–2066, June 1993.
[23] J. Princen, “The design of nonuniform modulated filter banks,” in Proc. IEEE
Int. Symp. Time-Frequency Time-Scale Anal., 1994, pp. 112–115.
[24] X. M. Xie, S. C. Chan, and T. I. Yuk, “Design of perfect-reconstruction
nonuniform recombination filter banks with flexible rational sampling fac-
tors,” IEEE Trans. Circuits Syst. I, vol. 52, no. 9, pp. 1965–1981, Sept 2005.
[25] S. Akkarakaran and P. P. Vaidyanathan, “New results and open problems
on nonuniform filter banks,” in Proc. IEEE Int. Conf. Acoust. Speech Signal
Processing, vol. 3, Mar 1999, pp. 1501–1504.
[26] X. M. Xie, X. Y. Chen, and G. M. Sho, “A simple design method of linear-
phase nonuniform filter banks with integer decimation factors,” in Proc. IEEE
Int. Midwest Symp. Circuits Syst., Aug. 2005, pp. 7–10.
[27] J. Li, T. Q. Nguyen, and S. Tantaratana, “A simple design method for near-
perfect-reconstruction nonuniform filter banks,” IEEE Trans. Signal Process-
ing, vol. 45, no. 8, pp. 2105–2109, Aug. 1997.
[28] I. Djokovic and P. P. Vaidyanathan, “Results on biorthogonal filter banks,”
Applicat. Comput. Harmonic Anal., vol. 1, no. 4, pp. 329–343, Sept. 1994.
[29] S. Wada, “Design of nonuniform division multirate FIR filter banks,” IEEE
Trans. Circuits Syst. II, vol. 42, no. 2, pp. 115–121, Feb. 1995.
[30] A. S. Mehr, “Alias-component matrices of nonuniform transmultiplexers,”
Circuits Syst. Signal Processing, 2009, DOI 10.1007/s00034-008-9071-6.
[31] C. Y.-F. Ho, B. W.-K. Ling, Y.-Q. Liu, P. K.-S. Tam, and K.-L. Teo, “Ef-
ficient algorithm for solving semi-infinite programming problems and their
applications to nonuniform filter bank designs,” IEEE Trans. Signal Process-
ing, vol. 54, no. 11, pp. 4223–4232, Nov. 2006.
[32] M. Wittig, “Satellite on-board processing for multimedia applications,” IEEE
Commun. Mag., vol. 38, no. 6, pp. 134–140, June 2000.
[33] J. V. Evans, “Satellite systems for personal communications,” Proc. IEEE,
vol. 86, no. 7, pp. 1325–1341, July 1998.
[34] J. Farserotu and R. Prasad, “A survey of future broadband multimedia satel-
lite systems, issues and trends,” IEEE Commun. Mag., vol. 38, no. 6, pp.
128–133, June 2000.
139
References
[35] E. D. Re and L. Pierucci, “Next-generation mobile satellite networks,” IEEE

Commun. Mag., vol. 40, no. 9, pp. 150–159, Sept. 2002.
[36] B. Evans, M. Werner, E. Lutz, M. Bousquet, G. E. Corazza, G. Maral, and

R. Rumeau, “Integration of satellite and terrestrial systems in future multi-
media communications,” IEEE Wireless Commun. Mag., vol. 12, no. 5, pp.
72–80, Oct. 2005.
[37] B. Arbesser-Rastburg, R. Bellini, F. Coromina, R. D. Gaudenzi, O. del Rio,

M. Hollreiser, R. Rinaldo, P. Rinous, and A. Roederer, “R&D directions for
next generation broadband multimedia systems: An ESA perspective,” in
Proc. AIAA Int. Commun. Satellite Syst. Conf. Exhibit, Montreal, Canada,
May 2002.
[38] T. Nguyen, J. Hant, D. Taggart, C.-S. Tsang, D. M. Johnson, and J.-C.

Chuang, “Design concept and methodology for the future advanced wideband
satellite system,” in Proc. IEEE Military Commun. Conf., vol. 1, USA, Oct.
2002, pp. 189–194.
[39] M. Lucente, E. Re, T. Rossi, M. D. Sanctis, C. Stallo, E. Cianca, and M. Rug-

gieri, “Future perspectives for the new European data relay system,” in Proc.
IEEE Aerospace Conf., Mar. 2008, pp. 1–7.
[40] M. N. Abdulazim and H. G. Göckler, “Flexible bandwidth reallocation

MDFT SBC-FDFMUX filter bank for future bent-pipe FDM satellite sys-
tems,” in Proc. Int. Workshop Signal Processing Space Commun., Noordwijk,
Netherlands, 2006.
[41] L. Rosenbaum, H. Johansson, and P. Löwenborg, “Oversampled complex-

modulated causal IIR filter banks for flexible frequency-band reallocation
networks,” in Proc. Eur. Signal Processing Conf., Florence, Italy, Sept. 2006.
[42] H. Johansson and P. Löwenborg, “Flexible frequency-band reallocation

networks using variable oversampled complex-modulated filter banks,”
EURASIP J. Adv. Signal Processing, vol. 2007, Article ID 63714, 2007.
[43] A. Eghbali, H. Johansson, and P. Löwenborg, “Flexible frequency-band re-

allocation: complex versus real,” Circuits Syst. Signal Processing, Jan. 2009,
DOI 10.1007/s00034-008-9090-3.
[44] A. Eghbali, H. Johansson, P. Löwenborg, and H. G. Göckler, “Flexible dy-

namic frequency-band reallocation and allocation: from satellite-based com-
munication systems to cognitive radios,” J. Signal Processing Syst., Feb. 2009,
DOI 10.1007/s11265-009-0348-1.
[45] A. Eghbali, H. Johansson, and P. Löwenborg, “Flexible frequency-band real-

location MIMO networks for real signals,” in Proc. Int. Symp. Image Signal
Processing Anal., Istanbul, Turkey, Sept. 2007.
140
References
[46] ——, “An arbitrary bandwidth transmultiplexer and its application to flexi-
ble frequency-band reallocation networks,” in Proc. Eur. Conf. Circuit The-
ory Design, Seville, Spain, Aug. 2007.
[47] H. Johansson and P. Löwenborg, “Flexible frequency band reallocation net-
work based on variable oversampled complex-modulated filter banks,” in
Proc. IEEE Int. Conf. Acoust. Speech Signal Processing, Philadelphia, USA,
Mar. 2005.
[48] H. G. Göckler and M. N. Abdulazim, “Tree-structured MIMO FIR filter
banks for flexible frequency reallocation,” in Proc. Int. Symp. Image Signal
Processing Anal., Istanbul, Turkey, Sept. 2007.
[49] M. N. Abdulazim, T. Kurbiel, and H. G. Göckler, “Modified DFT SBC-
FDFMUX filter bank systems for flexible frequency reallocation,” in Proc.
Eur. Signal Processing Conf., Poznan, Poland, 2007, pp. 60–64.
[50] M. N. Abdulazim and H. G. Göckler, “Design options of the versatile two-
channel SBC-FDFMUX filter bank,” in Proc. Eur. Conf. Circuit Theory De-
sign, Cork, Ireland, Sept. 2005.
[51] H. G. Göckler and M. N. Abdulazim, “Joint oversampling FDM demultiplex-
ing and perfectly reconstructing SBC filter bank for two channels,” in Proc.
Eur. Signal Processing Conf., Antalya, Turkey, Sept. 2005.
[52] H. G. Göckler and H. Eyssele, “Study of on-board digital FDM-
demultiplexing for mobile SCPC satellite communications (part I and II),”
Eur. Trans. Telecommun., vol. 3, pp. 7–30, Jan. 1992.
[53] M. N. Abdulazim and H. G. Göckler, “Efficient digital on-board de- and
remultiplexing of FDM signals allowing for flexible bandwidth allocation,” in
Proc. Int. Commun. Satellite Syst. Conf., Rome, Italy, Sept. 2005.
[54] H. G. Göckler and B. Felbecker, “Digital on-board FDM-demultiplexing with-
out restrictions on channel allocation and bandwidth,” in Proc. Int. Work-
shop Digital Signal Processing Tech. Space Applicat., Noordwijk, 1999.
[55] M. L. Boucheret, I. Mortensen, and H. Favaro, “Fast convolution filter banks
for satellite payloads with on-board processing,” IEEE J. Select. Areas Com-
mun., vol. 17, no. 2, pp. 238–248, Feb. 1999.
[56] G. Chiassarini and G. Gallinaro, “Frequency domain switching: Algorithms,
performances, implementation aspects,” in Proc. Tyrrhenian Int. Workshop
Digital Commun., Viareggio, Italy, Sept. 1995.
[57] G. Lippolis, L. Simone, M. C. Comparing, D. Gelfusa, V. Piloni, and R. Nov-
ello, “Overview on band-pass sampling approaches for on-board processing,”
in Proc. IEEE Int. Symp. Signal Processing Inf. Theory, Dec. 2004, pp. 543–
548.
141
References
[58] W. Kogler, H. Schlemmer, and O. Koudelka, “Timing synchronization in MF-

TDMA systems for geostationary satellites,” IEEE Commun. Mag., vol. 45,
no. 12, pp. 36–42, Dec. 2007.
[59] S. Haykin, “Cognitive radio: Brain-empowered wireless communications,”

IEEE J. Select. Areas Commun., vol. 23, no. 2, pp. 201–220, Feb. 2005.
[60] M. Sherman, A. N. Mody, R. Martinez, C. Rodriguez, and R. Reddy, “IEEE

standards supporting cognitive radio and networks, dynamic spectrum access,
and coexistence,” IEEE Commun. Mag., vol. 46, no. 7, pp. 72–79, July 2008.
[61] I. F. Akyildiz, W. Y. Lee, M. C. Vuran, and S. Mohanty, “A survey on

spectrum management in cognitive radio networks,” IEEE Commun. Mag.,
vol. 46, no. 4, pp. 40–48, Apr. 2008.
[62] P. Leaves, K. Moessner, R. Tafazolli, D. Grandblaise, D. Bourse, R. Tonjes,

and M. Breveglieri, “Dynamic spectrum allocation in composite reconfig-
urable wireless networks,” IEEE Commun. Mag., vol. 42, no. 5, pp. 72–81,
May 2004.
[63] A. Eghbali, H. Johansson, and P. Löwenborg, “A multimode transmultiplexer

structure,” IEEE Trans. Circuits Syst. II, vol. 55, no. 3, pp. 279–283, Mar.
2008.
[64] ——, “A Farrow-structure-based multi-mode transmultiplexer,” in Proc.

IEEE Int. Symp. Circuits Syst., Seattle, Washington, USA, May 2008.
[65] ——, “A class of multimode transmultiplexers based on the Farrow struc-

ture,” Circuits Syst. Signal Processing, 2010, submitted.
[66] ——, “On the filter design for a class of multimode transmultiplexers,” in
Proc. IEEE Int. Symp. Circuits Syst., Taipei, Taiwan, May 2009.
[67] ——, “Reconfigurable nonuniform transmultiplexers based on uniform filter

banks,” in Proc. IEEE Int. Symp. Circuits Syst., Paris, France, 2010.
[68] ——, “Reconfigurable nonuniform transmultiplexers using uniform modu-

lated filter banks,” IEEE Trans. Circuits Syst. I, 2010, accepted for publica-
tion.
[69] S. K. Mitra, Digital Signal Processing: A Computer Based Approach.

McGraw-Hill, Feb. 2006.
[70] K. K. Parhi, VLSI Digital Signal Processing Systems: Design and Implemen-
tation. John Wiley and Sons, 1999.
[71] A. V. Oppenheim, R. W. Schafer, and J. R. Buck, Discrete-Time Signal

Processing. Prentice Hall, 1999.
142
References
[72] S. K. Mitra and J. F. Kaiser, Eds., Handbook for Digital Signal Processing.
New York: Wiley, 1993.
[73] R. G. Lyons, Undestanding Digital Signal Processing, 2nd ed. Prentice Hall,
Mar. 2004.
[74] S. R. Pillai and G. H. Allen, “Generalized magnitude and power complemen-

tary filters,” in Proc. IEEE Int. Conf. Acoust. Speech Signal Processing, Apr.
1994, pp. 585–588.
[75] T. Saramäki, Handbook for Digital Signal Processing. New York: Wiley,
1993, ch. 4, pp. 155–277.
[76] N. J. Fliege, Multirate Digital Signal Processing. John Wiley & Sons, 1995.
[77] T. Saramäki and Y. Neuvo, “A class of FIR Nyquist (N th-band) filters with
zero intersymbol interference,” IEEE Trans. Circuits Syst., vol. 34, no. 10,
pp. 1182–1190, Oct. 1987.
[78] A. Viholainen, T. Saramäki, and M. Renfors, “Nearly perfect-reconstruction

cosine-modulated filter banks for VDSL modems,” in Proc. IEEE Int. Conf.
Electron. Circuits Syst., Sept. 1999, pp. 373–376.
[79] M. B. J. Furtado, P. S. R. Diniz, S. L. Netto, and T. Saramäki, “On the

design of high-complexity cosine-modulated transmultiplexers based on the
frequency-response masking approach,” IEEE Trans. Circuits Syst. I, vol. 52,
no. 11, pp. 2413–2426, Nov. 2005.
[80] A. Ghasemi. and E. S. Sousa, “Spectrum sensing in cognitive radio net-

works: Requirements, challenges and design trade-offs,” IEEE Commun.
Mag., vol. 46, no. 4, pp. 32–39, Apr. 2008.
[81] D. Cabric, S. M. Mishra, and R. W. Brodersen, “Implementation issues in

spectrum sensing for cognitive radios,” in Proc. Asilomar Conf. Signals Syst.
Comput., Nov. 2004, pp. 772–776.
[82] J. G. Proakis, Ed., Digital Communications. McGraw-Hill, 2000.
[83] B. P. Lathi, Ed., Modern Digital and Analog Communication Systems. Ox-
ford University Press, 2009.
[84] P. A. Regalia, Handbook for Digital Signal Processing. New York: Wiley,
1993, ch. 13, pp. 907–980.
[85] S. L. Hahn, Hilbert Transforms in Signal Processing. Artech House, 1996.
[86] Y. C. Lim and Y. J. Yu, “Synthesis of very sharp Hilbert transformer using
the frequency-response masking technique,” IEEE Trans. Signal Processing,
vol. 53, no. 7, pp. 2595–2597, Aug. 2005.
143
References
[87] C. M. Radger and L. B. Jackson, “Approximating noncausal IIR digital fil-

ters having arbitrary poles, including new Hilbert transformer designs, via
forward/backward block recursion,” IEEE Trans. Circuits Syst. I, vol. 53,
no. 12, pp. 2779–2787, Dec 2006.
[88] M. Bellanger, “On computational complexity in digital filters,” in Proc. Eur.
Conf. Circuit Theory Design, Aug 1981, pp. 58–63.
[89] J. F. Kaiser, “Nonrecursive digital filter design using I0 -sinh window func-
tion,” in Proc. IEEE Int. Symp. Circuits Syst., Apr. 1974, pp. 20–23.
[90] K. Ichige, M. Iwaki, and R. Ishii, “Accurate estimation of minimum filter
length for optimum FIR digital filters,” IEEE Trans. Circuits Syst. II, vol. 47,
no. 6, pp. 1008–1016, Oct. 2000.
[91] G. Cariolaro, P. Kraniauskas, and L. Vangelista, “A novel general formulation
of up/downsampling commutativity,” IEEE Trans. Signal Processing, vol. 53,
no. 6, pp. 2124–2134, June 2005.
[92] C. Y. F. Ho, B. W. K. Ling, and P. K. S. Tam, “Representations of linear
dual-rate system via single SISO LTI filter, conventional sampler and block
sampler,” IEEE Trans. Circuits Syst. II, vol. 55, no. 2, pp. 168–172, Feb.
2008.
[93] C. W. Farrow, “A continuously variable digital delay element,” in Proc. IEEE
Int. Symp. Circuits Syst., vol. 3, Espoo, Finland, June 1988, pp. 2641–2645.
[94] J. Vesma and T. Saramäki, “Interpolation filters with arbitrary frequency
response for all-digital receivers,” in Proc. IEEE Int. Symp. Circuits Syst.,
May 1996.
[95] D. Babic, J. Vesma, T. Saramäki, and M. Renfors, “Implementation of the
transposed Farrow structure,” in Proc. IEEE Int. Symp. Circuits Syst., vol. 4,
May 2002, pp. 5–8.
[96] J. Vesma, “Optimization and applications of polynomial-based interpolation
filters,” Ph.D. dissertation, Department of Information Technology, Tampere
University of Technology, June 1999.
[97] T. I. Laakso, V. Välimaki, M. Karjalainen, and U. K. Laine, “Splitting the
unit delay–tools for fractional delay filter design,” IEEE Signal Processing
Mag., vol. 13, no. 1, pp. 30–36, Jan. 1996.
[98] J. Vesma and T. Saramäki, “Optimization and efficient implementation of
FIR filters with adjustable fractional delay,” in Proc. IEEE Int. Symp. Cir-
cuits Syst., vol. IV, Hong Kong, June 1997, pp. 2256–2259.
[99] H. Johansson and E. Hermanowicz, “Adjustable fractional-delay filters utiliz-
ing the Farrow structure and multirate techniques,” in Proc. Int. Workshop
Spectral Methods Multirate Signal Processing, Florence, Italy, Sept. 2006.
144
References
[100] W. S. Lu and T. B. Deng, “An improved weighted least-squares design for

variable fractional delay FIR filters,” IEEE Trans. Circuits Syst. II, vol. 46,
no. 8, pp. 1035–1040, Aug. 1999.
[101] C. C. Tseng, “Design of variable fractional delay FIR filter using differentiator
bank,” in Proc. IEEE Int. Symp. Circuits Syst., vol. 4, May 2002, pp. 421–
424.
[102] H. Johansson and P. Löwenborg, “On the design of adjustable fractional delay
FIR filters,” IEEE Trans. Circuits Syst. II, vol. 50, no. 4, pp. 164–169, Apr.
2003.
[103] J. Vesma and T. Saramäki, “Design and properties of polynomial-based

fractional delay filters,” in Proc. IEEE Int. Symp. Circuits Syst., Geneva,
Switzerland, 2000.
[104] C. K. S. Pun, Y. C. Wu, S. C. Chan, and K. L. Ho, “On the design and
efficient implementation of the Farrow structure,” IEEE Signal Processing
Lett., vol. 10, no. 7, pp. 189–192, July 2003.
[105] R. J. Tallarida, Pocket Book of Integrals and Mathematical Formulas. CRC

Press LLC, 1999.
[106] T. Saramäki and R. Bregović, “An efficient approach for designing nearly
perfect-reconstruction cosine-modulated and modified DFT filter banks,” in
Proc. IEEE Int. Conf. Acoust. Speech Signal Processing, May 2001.
[107] K.-H. Chen and T.-D. Chiueh, “A cognitive radio system using discrete
wavelet multitone modulation,” IEEE Trans. Circuits Syst. I, vol. 55, no. 10,
pp. 3246–3258, Nov. 2008.
[108] T. Ihalainen, T. H. Stitz, M. Rinne, and M. Renfors, “Channel equalization

in filter bank based multicarrier modulation for wireless communications,”
EURASIP J. Applied Signal Processing, vol. 2007, Article ID 49389, 2007.
[109] A. Viholainen, J. Alhava, and M. Renfors, “Efficient implementation of com-

plex modulated filter banks using cosine and sine modulated filter banks,”
EURASIP J. Applied Signal Processing, vol. 2006, Article ID 58564, 2006.
[110] N. J. Fliege, “Computational efficiency of modified DFT-polyphase filter

banks,” in Proc. Asilomar Conf. Signals Syst. Comput., Asilomar, Nov. 1993,
pp. 1296–1300.
[111] ——, “Modified DFT polyphase SBC filter banks with almost perfect re-
construction,” in Proc. IEEE Int. Conf. Acoust. Speech Signal Processing,
Adelaide, Australia, Apr. 1994, pp. 149–152.
[112] T. Karp and N. J. Fliege, “Computationally efficient realization of MDFT

filter banks,” in Proc Eur. Signal Processing Conf., Trieste, Italy, Sept. 1996.
145
References
[113] R. Bregović and T. Saramäki, “A systematic technique for designing linear-

phase FIR prototype filters for perfect-reconstruction cosine-modulated and
modified DFT filter banks,” IEEE Trans. Signal Processing, vol. 53, no. 8,
pp. 3193–3201, Aug. 2005.
[114] P. N. . Heller, T. Karp, and T. Q. Nguyen, “A general formulation of mod-

ulated filter banks,” IEEE Trans. Signal Processing, vol. 47, no. 4, pp. 986–
1002, Apr. 1999.
[115] T. Karp and N. J. Fliege, “Modified DFT filter banks with perfect recon-
struction,” IEEE Trans. Circuits Syst. II, vol. 46, no. 11, pp. 1404–1414,
Nov. 1999.
[116] S. Mirabbasi and K. Martin, “Overlapped complex-modulated transmulti-

plexer filters with simplified design and superior stopbands,” IEEE Trans.
Circuits Syst. II, vol. 50, no. 5, pp. 1170–1183, May 2003.
[117] P. Martin-Martin, F. Cruz-Roldan, and T. Saramäki, “Optimized transmul-

tiplexers for multirate systems,” in Proc. IEEE Int. Symp. Circuits Syst.,
vol. 2, Kobe, Japan, May 2005, pp. 1106–1109.
[118] T. Saramäki and R. Bregovic, “An efficient approach for designing nearly
perfect-reconstruction cosine-modulated and modified DFT filter banks,” in
Proc. IEEE Int. Conf. Acoust. Speech Signal Processing, vol. 6, 2001, pp.
3617–3620.
[119] P. Martin-Martin, R. Bregovic, A. Martin-Marcos, F. Cruz-Roldan, and

T. Saramäki, “A generalized window approach for designing transmultiplex-
ers,” IEEE Trans. Circuits Syst. I, vol. 55, no. 9, pp. 2696–2706, Oct 2008.
[120] P. S. R. Diniz, L. C. R. de Barcellos, and S. L. Netto, “Design of high-

resolution cosine-modulated transmultiplexers with sharp transition band,”
IEEE Trans. Signal Processing, vol. 52, no. 5, pp. 1278–1288, May 2004.
[121] P. P. Vaidyanathan and B. Vrcelj, “Transmultiplexers as precoders in mod-

ern digital communications: A tutorial review,” in Proc. IEEE Int. Symp.
Circuits Syst., vol. 5, May 2004, pp. 405–412.
[122] J. Alhava and M. Renfors, “Exponentially-modulated filter bank transmul-

tiplexer with fine-coarse adaptive filtering,” in Proc. Int. Symp. Commun.
Control Signal Processing, Malta, Mar. 2008.
[123] M. Ziolko and M. Nowak, “Integer-to-integer filters in transmultiplexer sys-

tem,” in Proc. IEEE Symp. Comput. Commun., July 2007, pp. 637–641.
[124] Y. S. F. Ding and T. Chen, “2-norm based recursive design of transmultiplex-

ers with designable filter length,” Circuits Syst. Signal Processing, vol. 25,
no. 4, pp. 447–462, Aug. 2006.
146
References
[125] F. D. Beaulieu and B. Champagne, “Design of prototype filters for perfect

reconstruction DFT filter bank transceivers,” Signal Processing, vol. 89, no. 1,
pp. 87–98, Jan. 2009.
[126] B. Borna and T. N. Davidson, “Biwindowed discrete multitone transceiver
design,” IEEE Trans. Signal Processing, vol. 55, no. 8, pp. 4217–4226, Aug.
2007.
[127] T. Bianchi and F. Argenti, “SVD-Based techniques for zero-padded block
transmission over fading channels,” IEEE Trans. Signal Processing, vol. 55,
no. 2, pp. 594–604, Feb. 2007.
[128] Y.-P. Lin and S.-M. Phoong, “ISI-free FIR filterbank transceivers for
frequency-selective channels,” IEEE Trans. Signal Processing, vol. 49, no. 11,
pp. 2648–2658, Nov. 2001.
[129] T. Bianchi, F. Argenti, and E. D. Re, “Performance of filterbank and wavelet
transceivers in the presence of carrier frequency offset,” IEEE Trans. Com-
mun., vol. 53, no. 8, pp. 1323–1332, Aug. 2005.
[130] B. Farhang-Boroujeny and G. Mathew, “Nyquist filters with robust perfor-
mance against timing jitter,” IEEE Trans. Signal Processing, vol. 46, no. 12,
pp. 3427–3431, Dec. 1998.
[131] T. N. Davidson, T. Luo, and K. M. Wong, “Design of orthogonal pulse shapes
for communications via semidefinite programming,” IEEE Trans. Signal Pro-
cessing, vol. 48, no. 5, pp. 1433–1445, May 2000.
[132] M. Liu, C. J. Zarowski, and F. W. Fairman, “Factorable FIR Nyquist filters
with least stopband energy under sidelobe level constraints,” IEEE Trans.
Signal Processing, vol. 48, no. 5, pp. 1495–1498, May 2000.
[133] T. N. Davidson, “Efficient design of waveforms for robust pulse amplitude
modulation,” IEEE Trans. Signal Processing, vol. 49, no. 12, pp. 3098–3111,
Dec. 2001.
[134] L. E. Franks, “Further results on Nyquist’s problem in pulse transmission,”
IEEE Trans. Comput., vol. 16, no. 4, pp. 337–340, Apr. 1968.
[135] J. L. Sullivan, J. W. Adams, R. A. Reisner, and R. L. Armstrong, “New
optimization algorithm for digital communication filters,” in Proc. Asilomar
Conf. Signals Syst. Comput., vol. 1, Nov. 2002, pp. 323–327.
[136] P. R. Chevillat and G. Ungerboeck, “Optimum FIR transmitter and receiver
filters for data transmission over band-limited channels,” IEEE Trans. Com-
put., vol. 30, no. 8, pp. 1909–1915, Aug. 1982.
[137] A. C. Salazar and V. B. Lawrence, “Design and implementation of transmitter
and receiver filters with periodic coefficient nulls for digital system,” in Proc.
IEEE Int. Conf. Acoust. Speech Signal Processing, 1982, pp. 306–310.
147
References
[138] H. Samueli, “On the design of optimal equiripple FIR digital filters for data
transmission applications,” IEEE Trans. Circuits Syst., vol. 35, pp. 1542–
1546, Dec. 1988.
[139] A. K. Wang, R. Ligmanowski, J. Castro, and A. Mazzara, “EVM simulation

and analysis techniques,” in Proc. IEEE Military Commun. Conf., USA, Oct.
2006, pp. 1–7.
[140] G. Technical Specification Group Access Network, “Base station conformance

testing,” 2000, (TS 25.141 V3.2.0).
[141] A. Eghbali, “On filter bank based MIMO frequency multiplexing and demul-
tiplexing,” Master’s thesis, Linköping University, Sweden, Sept. 2006, Report
Number: LITH-ISY-EX–06/3911–SE.
[142] Z. Wang, “On computing discrete Fourier and cosine transforms,” IEEE
Trans. Acoust., Speech, Signal Processing, vol. ASSP-33, pp. 1341–1344, Oct.
1985.
[143] P. Duhamel, “Implementation of split-radix FFT algorithms for complex,

real, and real-symmetric data,” IEEE Trans. Acoust., Speech, Signal Pro-
cessing, vol. ASSP-34, pp. 285–295, Apr. 1986.
[144] M. F. Mansour, “On the odd-DFT and its applications to DCT/IDCT com-
putation,” IEEE Trans. Signal Processing, vol. 54, pp. 2819–2822, July 2006.
[145] M. Vetterli and H. J. Nussbaumer, “Simple DFT and DCT algorithms with
reduced number of operations,” Signal Processing, vol. 6, pp. 267–278, Aug.
1984.
[146] H. V. Sorensen, D. L. Jones, M. T. Heideman, and C. S. Burrus, “Real-valued

fast Fourier transform algorithms,” IEEE Trans. Acoust., Speech, Signal Pro-
cessing, vol. ASSP-35, pp. 849–863, June 1987.
[147] P. Duhamel and M. Vetterli, “Fast Fourier transforms: A tutorial review and
state of the art,” Signal Processing, vol. 19, pp. 259–295, Apr. 1990.
[148] H. Murakami, “Real-valued decimation-in-time and decimation-in-frequency

algorithms,” IEEE Trans. Circuits Syst. II, vol. 41, pp. 808–816, Dec. 1994.
[149] B. R. Sekhar and K. M. M. Prabhu, “Radix-2 decimation-in-frequency al-

gorithm for the computation of the real-valued FFT,” IEEE Trans. Signal
Processing, vol. 47, pp. 1181–1184, Apr. 1999.
[150] G. Haitao, G. A. Sitton, and C. S. Burrus, “The quick Fourier transform: An

FFT based on symmetries,” IEEE Trans. Acoust., Speech, Signal Processing,
vol. 46, no. 2, pp. 335–341, Feb. 1998.
148
References
[151] M. T. Heideman and C. S. Burrus, “On the number of multiplications nec-

essary to compute a length-2n DFT,” IEEE Trans. Acoust., Speech, Signal
Processing, vol. ASSP-34, pp. 91–95, Feb. 1986.
[152] H. Johansson and O. Gustafsson, “Linear-phase FIR interpolation, decima-

tion, and M -th band filters utilizing the Farrow structure,” IEEE Trans.
Circuits Syst. I, vol. 52, no. 10, pp. 2197–2207, Oct. 2005.
[153] D. Babic, V. Lehtinen, and M. Renfors, “Discrete-time modeling of

polynomial-based interpolation filters in rational sampling rate conversion,”
in Proc. IEEE Int. Symp. Circuits Syst., vol. 4, Bangkok, Thailand, May
2003, pp. 321–324.
[154] H. Johansson, “Farrow-structure-based reconfigurable bandpass linear-phase

FIR filters for integer sampling rate conversion,” IEEE Trans. Circuits Syst.
II, 2010, submitted.
[155] H. Johansson and P. Löwenborg, “Reconstruction of nonuniformly sam-

pled bandlimited signals by means of time-varying discrete-time FIR filters,”
EURASIP J. Adv. Signal Processing, vol. 2006, Article ID 64185, 18 pages,
2006.
[156] I. F. Akyildiz, W. Y. Lee, M. C. Vuran, and S. Mohanty, “Next genera-

tion/dynamic spectrum access/cognitive radio wireless networks: A survey,”
Comput. Networks J. (Elsevier), vol. 50, pp. 2127–2159, Sept. 2006.
[157] Q. Zhao and A. Swami, “A survey of dynamic spectrum access: Signal

processing and networking perspectives,” in Proc. IEEE Int. Conf. Acoust.
Speech Signal Processing, vol. 4, 2007, pp. 1349–1352.
[158] Y. Yuan, P. Bahl, R. Chandra, T. Moscibroda, S. Narlanka, and Y. Wu,

“Allocating dynamic time-spectrum blocks in cognitive radio networks,” in
Proc. ACM Int. Symp. Mobile Ad Hoc Networking Comput., Sept. 2007.
[159] K. M. Tsui, S. C. Chan, and Y. C. Lim, “Design of multi-plet perfect recon-

struction filter banks using frequency-response masking technique,” IEEE
Trans. Circuits Syst. I, vol. 55, no. 9, pp. 2707–2715, Oct. 2008.
[160] “Long term evolution,” http://www.3gpp.org/LTE.
[161] B. Razavi, “Cognitive radio design challenges and techniques,” IEEE J. Solid-
State Circuits, vol. 45, no. 8, pp. 1542–1553, Aug. 2010.
[162] Q. Zhao and B. M. Sadler, “A survey of dynamic spectrum access,” IEEE

Signal Processing Mag., vol. 24, no. 3, pp. 79–89, May 2007.
[163] J. Huang, R. A. Berry, and M. L. Honig, “Spectrum sharing with distributed

interference compensation,” in Proc. IEEE Int. Dynamic Spectrum Access
Networks, Nov. 2005, pp. 88–93.
149
References
[164] D. Cabric, I. D. ÓDonnell, M. S. W. Chen, and R. W. Brodersen, “Spectrum

sharing radios,” IEEE Circuits Syst. Mag., vol. 6, no. 2, pp. 30–45, 2006.
[165] N. Devroye, P. Mitran, and V. Tarokh, “Achievable rates in cognitive radio

channels,” IEEE Trans. Inform. Theory, vol. 52, no. 5, pp. 1813–1827, May
2006.
[166] D. Cabric and R. W. Brodersen, “Physical layer design issues unique to cogni-
tive radio systems,” in Proc. IEEE Personal Indoor Mobile Radio Commun.,
Sept. 2005.
[167] G. Ganesan and Y. G. Li, “Cooperative spectrum sensing in cognitive radio

networks,” in Proc. IEEE Int. Dynamic Spectrum Access Networks, Nov.
2005, pp. 137–143.
[168] http://ieee802.org/22/.
[169] Y. Yuan, P. Bahl, R. Chandra, P. A. Chou, J. I. Ferrell, T. Moscibroda,

S. Narlanka, and Y. Wu, “KNOWS: Cognitive radio networks over white
spaces,” in Proc. IEEE Int. Dynamic Spectrum Access Networks, Apr. 2007,
pp. 416–427.
[170] F. K. Jondral, “Cognitive radio: A communications engineering view,” IEEE

Wireless Commun. Mag., vol. 14, no. 4, pp. 28–33, Aug. 2007.
150

Fulltext01 PDF

Hochgeladen von

Dokumentinformationen

Originaltitel

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

Fulltext01 PDF

Hochgeladen von

Copyright:

Verfügbare Formate

Linköping Studies in Science and Technology

Dissertation No. 1344

Contributions to Reconfigurable Filter Banks

Division of Electronics Systems

c 2010 Amir Eghbali

Department of Electrical Engineering,

Printed by LiU-Tryck, Linköping, Sweden 2010

2 Basics of Digital Filters 9

3 Basics of Multirate Signal Processing 21

3.2.1 Design of the Farrow Structure . . . . . . . . . . . . . . . . . 28

4 Flexible Frequency-Band Reallocation For Real Signals 37

5 A Multimode Transmultiplexer Structure 57

6 A Class of Multimode Transmultiplexers Based on the Farrow

7 Reconfigurable Nonuniform Transmultiplexers Using Uniform

8 Applications to Cognitive Radios 115

9 Conclusion and Future Work 129

A Derivation of (6.23) 131

B Derivation of (6.35) 135

Acronyms and Abbreviations

1.1 Motivation and Problem Formulation

Standard Bit Rate No. of Users Channel Spacing

To support multimode communications, we thus need a system which allows

this thesis we will mostly assume ∆ = 0.

Case II: D=0

Case III: D>0

broadband satellite-based systems in which satellites communicate with the users

1. A. Eghbali, H. Johansson, and P. Löwenborg, “Flexible frequency-band

2. A. Eghbali, H. Johansson, and P. Löwenborg, “Flexible frequency-band

3. A. Eghbali, H. Johansson, and P. Löwenborg, “An arbitrary bandwidth

4. A. Eghbali, H. Johansson, and P. Löwenborg, “A multimode transmulti-

5. A. Eghbali, H. Johansson, and P. Löwenborg, “A Farrow-structure-based

6. A. Eghbali, H. Johansson, and P. Löwenborg, “A class of multimode trans-

7. A. Eghbali, H. Johansson, and P. Löwenborg, “On the filter design for a

8. A. Eghbali, H. Johansson, and P. Löwenborg, “Reconfigurable nonuniform

9. A. Eghbali, H. Johansson, and P. Löwenborg, “Reconfigurable nonuniform

10. A. Eghbali, H. Johansson, and P. Löwenborg, and H. G. Göckler, “Dy-

1. A. Eghbali, O. Gustafsson, H. Johansson, and P. Löwenborg, “On the com-

2. G. Mehdi, N. Ahsan, A. Altaf, and A. Eghbali, “A 403-MHz fully differential

3. A. Eghbali, H. Johansson, T. Saramäki, and P. Löwenborg, “On the design

1.2 Thesis Outline

2.1 FIR Filters

Figure 2.1: Direct form realization of an N th-order FIR filter.

Figure 2.2: Transposed direct form realization of an N th-order FIR filter.

2.2 IIR Filters

With b(n) = 0, n = 0, 1, . . . , N, an IIR filter reduces to an FIR filter. Realization

2.2.1 Note on Stability

|h(n)|≤M1 K1n , n≥0, (2.5)

then (2.4) converges for

2.3 Polyphase Decomposition

which can be rewritten as [12, 69, 70]

Here, Hi (z) are the polyphase components and

hi (n) = h(nL + i), i = 0, 1, . . . , L − 1. (2.11)

This decomposition is frequently referred to as the Type I polyphase decomposition.

2.4 Special Classes of Filters

2.4.1 Complementary Filters

In general, Hk (z) are complementary of order p if [74]

2.4.2 Linear-Phase FIR Filters

Symmetric : h(n) = h(N − n), n = 0, 1, . . . , N (2.19)

where HR (ωT ) is the real zero-phase frequency response with c = 0 and c = π2

In general, the linear-phase response can be of the form [75]

Φ(ωT ) = −αωT + β. (2.23)

Depending on h(n) being symmetric or antisymmetric and N being odd or even,

Type I : h(n) = h(N − n), N even