You are on page 1of 159

Relaying in Interference Limited Networks: Models, Bounds, and Strategies

DISSERTATION
for the Degree of

Doctor of Philosophy (Electrical Engineering)

Onur S ahin January 2010

Relaying in Interference Limited Networks: Models, Bounds, and Strategies DISSERTATION


Submitted in Partial Fulllment of the Requirement for the Degree of Doctor of Philosophy (Electrical Engineering) at the POLYTECHNIC INSTITUTE OF NEW YORK UNIVERSITY by Onur S ahin October 2009

Approved:

Department Head 20 Copy No.

ii Approved by the Guidance Committee:

Major: Electrical Engineering

Elza Erkip Associate Professor of Electrical and Computer Engineering

Shivendra S. Panwar Professor of Electrical and Computer Engineering

Osvaldo Simeone Assistant Professor of Electrical and Computer Engineering New Jersey Institute of Technology Newark, NJ

iii

Microlm or other copies of this dissertation are obtainable from

UMI Dissertation Publishing Bell & Howell Information and Learning 300 North Zeeb Road P. O. Box 1346 Ann Arbor, Michigan 48106-1346

iv

VITA

Onur S ahin was born in Denizli, Turkey, in 1981 and received his B.S. degree in Electrical and Electronics Engineering from Middle East Technical University, Ankara, Turkey in 2003. Since then, he has been working towards his PhD degree at Polytechnic Institute of New York University. In 2005 and 2007, he worked as a summer intern at Philips Research North America, Briarcli, NY. During his internships, he worked on cooperative protocols for ad-hoc systems and cognitive radio systems for high-denition video transmission over wireless channels. His research interests include information theory and communication theory with the emphasis on cooperative systems and multi-user networks.

To my beautiful family for their love

vi

ACKNOWLEDGEMENT

It is impossible to distinguish the eort spent on the technical outcome of my PhD study from the daily life in many aspects. Several people have been in the center of this process and made tremendous eects mostly easing the harsh road of research which desperately pulls one to obsession and loneliness. I want to foremost express my appreciation to my advisor Professor Elza Erkip. It is always well beyond expectations to work with such a nice and understanding person who is also a great scholar. It has always been a pleasure to be her student bearing great condence at all times. I am indebted to her for guiding my study with patience and support for the problems I encountered beyond school. I am extremely grateful to Prof. Osvaldo Simeone whom I had chance to know in the last year of my study. I could never imagine his visit to Poly would pave the way towards completing my work. He has been a great friend and I enjoyed all our discussions particularly the ones at West Village cafes on Sundays. I will always remember his endless help when getting stuck in various points at my research. Throughout all these years Poly introduced great friends. I would like to thank Zinan Lin, Melda Y uksel and Deniz G und uz for their support during my rst years at school and in New York. I appreciate their endless support while I nd myself getting lost at various things. I extremely enjoyed spending all these years next to u Alay, Ka Mehmet Turan, Ozg gan Bakano glu, Sintayehu Dehnie, Ya gz S utc u and H useyin Akcan with whom I experienced unique friendships and amusing moments. Getting PhD is already a dicult task, and it becomes even harder in New York City. However, one person existed at any time, from most joyful times to over depressed moods. I am fortunate to have Yi git Atlgan as a great friend and roommate for more than 4 years. Whether we should have spent more energy for our thesis than our out-of-school life may be an interesting question to consider but in any case we enjoyed the city together. He is one of the best gifts to me from New York City.

vii My dear friends Mustafa Avc, Duygu Parmakszoglu and Mert G urb uzbalaban have always made me feel relieved. We formed one of the most interesting academic backgrounds together; PhDs at musicology, anthropology and mathematics, respectively. I am grateful for the splendid times we spent together and sure to come more. My dear Cece, Ceren Erdem, shined in the darkest times and showed me beautiful colors I couldnt even imagine before. Thanks for warming my life through Bosphorus and physically in New York. This thesis might not have nished without her day-care service in Cihangir where a signicant part of it is written. Thanks for pure love that made everything more meaningful. Since the rst day in New York, my beautiful family has always been in my heart and mind. I always felt the luckiest person to have such a great family whom I miss more and more everyday. This thesis is a very basic gift to them. I sincerely appreciate my dad and moms eorts and struggles all these years to raise all of us without hesitating any sacrices from their life. My only wish is that they raised us as they desired mostly with joy and pride. Finally, thanks to my dear G ulnaz, Deniz, Taylan, Utku, and Bari s for making me happy anytime and completing life in all aspects.

viii

AN ABSTRACT Relaying in Interference Limited Networks: Models, Bounds, and Strategies by Onur S ahin Advisor: Elza Erkip Submitted in Partial Fulllment of the Requirements for the Degree of Doctor of Philosophy (Electrical Engineering) October 2009

Interference is inherent in wireless communication systems and can severely degrade communication rates. Relaying, on the other hand, is a well-known technique primarily used to provide signicant rate improvements to point-to-point communications. In this dissertation, the impact of relaying in interference-limited networks is investigated. Depending on the relays transmission and reception bands, various congurations are explored: the relay reception, or both reception and transmission, take place over orthogonal links (out-of-band relay reception and/or transmission ), and the relay receives and transmits in the same band as the sources (in-band relay reception/ transmission ). Novel relaying techniques, denoted by signal relaying, interference forwarding and interference cancelation, to manage interference in interference-limited multi-terminal networks are proposed. To model and explore these interference-limited systems, initially, an interference channel (IC) with a relay is considered where the relay operates over an

ix orthogonal band with respect to the underlying IC. The overall system is referred to as IC with an out-of-band relay (IC-OBR). The conditions under which signal relaying and/or interference forwarding operations ensure optimality in terms of capacity are identied. The optimality of signal relaying is shown to entail separability among the IC and the relay channel, exhibiting substantial reduction in encoding/decoding complexity. The essence of interference forwarding in addition to signal relaying in improving the communication rates is demonstrated and the capacities are established in special cases. Next, the relay is assumed to transmit in the band of IC while reception is employed over the channels with orthogonal bands and nite capacities. This model is denoted by IC with an out-of-band reception/in-band transmission relay (IC-OIR). It is shown that in-band transmission by the relay uniquely facilitates the reduction of interference at the terminals through interference cancelation feature at the relay, which cannot be exploited by IC-OBR. An achievable scheme based on partial decode-and-forward transmission and dirty-paper coding at the relay is proposed. Outer bounds on the capacity region are obtained and shown to be tight for relatively large source-to-relay channel gains and the channel conditions named as mixed relay-interference conditions and very strong relay-interference conditions. Finally, an interference channel assisted by a full-duplex relay that operates in the transmission band of the source-destination pairs is explored. An achievable scheme with rate-splitting at the sources and decode-and-forward (DF) transmission at the relay is obtained and compared with its time-division counterpart. Interference forwarding and cancelation operations are demonstrated to provide signicant throughput gain depending on the network topology.

Contents

List of Figures 1 Introduction 1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2 Key Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.3 Dissertation Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Background and Related Work 2.1 Interference Channel . . . . . . . . . . . . 2.1.1 Han-Kobayashi Achievable Scheme 2.1.2 Capacity Results . . . . . . . . . . 2.2 Relay Channel . . . . . . . . . . . . . . . . 2.2.1 Decode-and-Forward Transmission 2.3 Related Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

xiii 1 1 5 6 8 8 9 11 13 13 15 18 18 19 20 22 22 23 23 24 24 26 33 35 36 38 46

3 Interference Channel with an Out-of-band Reception/Out-of-band Transmission Relay 3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2 System Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2.1 IC-OBR Type-I . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2.2 IC-OBR Type-II . . . . . . . . . . . . . . . . . . . . . . . . . 3.2.3 Symmetric IC-OBR Type-I and Type-II . . . . . . . . . . . . 3.2.4 Achievable Rates for Fixed and Variable OBRC Bandwidth Allocation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3 Analysis of IC-OBR Type-I . . . . . . . . . . . . . . . . . . . . . . . 3.3.1 Outer Bound . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3.2 Achievable Rate Region . . . . . . . . . . . . . . . . . . . . . 3.3.3 Capacity Results for Fixed OBRC Bandwidth Allocation . . . 3.3.4 Capacity Results for OBRC Variable Bandwidth Allocation . 3.4 Analysis of IC-OBR Type-II . . . . . . . . . . . . . . . . . . . . . . . 3.4.1 Outer Bounds . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4.2 Capacity Results for Fixed OBRC Bandwidth Allocation . . . 3.4.3 Capacity Results for OBRC Variable Bandwidth Allocation .

xi 3.5 Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

4 Interference Channel with an Out-of-band Reception/In-band Transmission Relay 60 4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60 4.2 System Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61 4.3 A General Achievable Region and Outer Bounds for IC-OIR . . . . . 63 4.4 An Achievable Region . . . . . . . . . . . . . . . . . . . . . . . . . . 64 4.4.1 Outer Bounds for the IC-OIR and IC-CR . . . . . . . . . . . . 67 4.5 Very Strong Capacity Region of IC-OIR . . . . . . . . . . . . . . . . 71 4.6 Sum-Capacity of IC-CR under Mixed Relay-Interference Conditions . 74 4.7 Capacity Region of a Degraded IC-CR . . . . . . . . . . . . . . . . . 75 4.8 Illustration of the Results . . . . . . . . . . . . . . . . . . . . . . . . 76 4.9 Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83 5 Interference Channel with an In-band sion Relay 5.1 Introduction . . . . . . . . . . . . . . 5.2 System Model . . . . . . . . . . . . . 5.3 Rate Splitting at the Sources . . . . . 5.4 Orthogonal Transmission . . . . . . . 5.5 Sum-Rate Maximization . . . . . . . 5.6 Discussion of the Results . . . . . . . 5.7 Chapter Summary . . . . . . . . . . 6 Conclusion A A.1 Proof A.2 Proof A.3 Proof A.4 Proof A.5 Proof A.6 Proof A.7 Proof A.8 Proof A.9 Proof A.10 Proof A.11 Proof A.12 Proof A.13 Proof A.14 Proof of of of of of of of of of of of of of of Proposition 5 . Proposition 6 . Proposition 8 . Proposition 9 . Proposition 10 Proposition 11 Proposition 12 Propositions 13 Propositions 15 Proposition 16 Proposition 17 Proposition 19 Proposition 20 Proposition 21 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . and 14 and 18 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Reception/In-band Transmis. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85 85 86 86 89 90 91 93 97 100 100 102 103 105 106 107 107 108 111 112 114 114 115 116

xii A.15 Proof A.16 Proof A.17 Proof A.18 Proof Bibliography of of of of Proposition Proposition Proposition Proposition 22 23 25 28 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118 126 129 131 133

xiii

List of Figures

2.1 2.2 2.3 3.1

Gaussian interference channel in standard form. . . . . . . . . . . . . Gaussian relay channel. . . . . . . . . . . . . . . . . . . . . . . . . . . Block-Markov encoding scheme for the relay channel. . . . . . . . . . Interference Channel (IC) with an out-of-band relay (OBR). The OBR channel (OBRC) has channel uses for each channel use of IC. (i ) IC-OBR Type-I: The OBRC is divided into four Gaussian orthogonal channels with iR , Ri , i = 1, 2, channel uses each; (ii ) IC-OBR Type II: The OBRC is divided into two orthogonal channels with M AC , BC channel uses. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Achievable sum-rate from Proposition 6 for IC-OBR Type-I with symmetric IC, for xed OBRC bandwidth allocation versus a for various OBRC link capacities, namely i) = 0 (no relay), ii) 1R = 2R = 2 2 2 R1 = R2 = 1, b2 1 = b2 = 1.5, c1 = c2 = 0.1, (which satises the conditions in Proposition 7), and iii) 1R = 2R = R1 = R2 = 1, 2 2 2 b2 2 = c1 = 6.3, b1 = c2 = 1.5 (which satises the conditions in Proposition 8). All powers are set to 10 dB. . . . . . . . . . . . . . . . . . . Achievable sum-rate (from Proposition 6, (A.35) with signal relaying (only common information transmission over the IC), and outer bound (from Proposition 5 (A.36)) versus b2 for IC-OBR Type-I with variable OBRC bandwidth allocation ( = 1, a21 = a12 = 2, P1 = P2 = P1R = P2R = PR1 = PR2 =10dB, c2 = 3, c1 = b1 = 2). . . . . . . . . . . . . . . Achievable sum-rate (from Proposition 8, (A.23)-(A.26)) with signal relaying and interference forwarding (only common information transmission over the IC) and outer bound (from Proposition 5, (A.36)) versus b2 for IC-OBR Type-I with variable OBRC bandwidth allocation 2 i=1 iR + Ri = , = 1, a12 = 3, a21 = 2, P1 = P2 = P1R = P2R = PR1 = PR2 =10dB, b1 = 1.5, c1 = 3, and c2 = 1. . . . . . . . . . Illustration of the OBRC conditions leading to the sum-capacity in 2 Proposition 16: c1 c2 , M AC C (b2 1 P1R ) BC C (c1 PR ). . . . . . . . . .

10 14 14

49

3.2

50

3.3

51

3.4

52 53

3.5

xiv 3.6 Illustration of the OBRC conditions leading to the sum-capacity in 2 2 Proposition 16: c1 c2 , M AC C (b2 1 P1R ) < BC C (c1 PR ), M AC C (b2 P2R ) 2 c P BC C 1+2c2 R where where is the optimal power allocation that PR
2

maximizes the sum-rate in Proposition 16 ( = 1 ). . . . . . . . . 3.7 Achievable sum-rate ((3.22a)-(3.22c)) and outer bound ((3.18a)-(3.18d)) for a symmetric IC-OBR Type II channel with respect to relay-todestination channel gains, c (a = 2, b = 1, P = Ps = 10). . . . . . . . 3.8 Achievable sum-rate and outer bound for an IC-OBR Type-II channel with respect to R D1 channel gain, c1 and S2 D1 channel gain a21 {0.1, 0.9.1.8} (a12 = 0.5, b1 = 1, b2 = 10, c2 = 1 and all node powers are equal to 10). . . . . . . . . . . . . . . . . . . . . . . . . . 3.9 Achievable sum-rates and outer bound for a symmetric IC-OBR TypeII with respect to the relay-to-destination gain c (b = 1, M AC = BC = 0.5, P1 = P2 = P1R = P2R = PR =10dB; Fig. i) a = 3, Fig. ii) a = 10). 3.10 Achievable sum-rate (from Proposition 16, (3.19)) with signal relaying (DF) and outer bound (3.25) and optimal parameters (M AC , BC , ) of the DF scheme (3.19) for an IC-OBR Type-II with respect to S1 R channel gain b1 (b2 = 2, c1 = 2, c2 = 0.3, = 1, all node powers are equal to 10 dB, a21 = 1.8, a12 = 0.5.) . . . . . . . . . . . . . . . . . . 3.11 Achievable sum-rate (from Proposition 16, (3.19)) with signal relaying (DF) and outer bound (3.25) for an IC-OBR Type-II with respect to S1 R channel gain, b1 and (b2 , c1 ) {(3, 2), (10, 5), (20, 10)} (c2 = 1, M AC + BC = with = 1, all node powers are equal to 10 dB, a21 = 1.8, a12 = 0.5.) . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.12 Achievable sum-rate (from Proposition 19, (3.23a)-(3.23d)) with signal relaying and interference forwarding (DF, only common information transmission by S2 and only private information transmission by S1 over the IC) and outer bound (3.25) versus b2 for IC-OBR Type-II with variable OBRC bandwidth allocation 2 i=1 iR + Ri = , ( = 1, a21 = 1, a12 = 0.5, P1 = P2 = P1R = P2R = PR =10dB, c1 = 4, c2 = 1.5, b1 = 1). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.1 4.2 Gaussian interference channel with an out-of-band reception/in-band transmission relay (IC-OIR). . . . . . . . . . . . . . . . . . . . . . . . Parallel source-destination pairs with a common orthogonal relay. The 2 2 total relay power satisfy 1 c + 2c 1. This is an equivalent system for an IC-OIR under very strong interference. . . . . . . . . . . . . . . . Achievable region of a symmetric IC-OIR with dierent C1 , C2 values. The system parameters are P1 = P2 = PR = 10, a11 = a22 = 1, c1 = c2 = 1, a21 = a12 = 2. The C1 , C2 curve also corresponds to the IC-CR model. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

53

54

55

56

57

58

59 64

72

4.3

77

xv 4.4 Achievable sum-rate of a symmetric IC-CR for dierent transmission techniques. The system parameters are P1 = P2 = PR = 10, a11 = a22 = 1, c1 = c2 = 1, = a21 /a22 = a12 /a11 . . . . . . . . . . . . . . . . Achievable sum-rate of a one-sided IC-CR for dierent transmission techniques. The system parameters are P1 = P2 = PR = 10, a11 = a22 = 1, c1 = 1, a12 = c2 = 0, = a21 /a22 . . . . . . . . . . . . . . . . . Optimal relay power allocations that maximize achievable sum-rate of the one-sided IC-CR in Fig. 4.5. The system parameters are P1 = P2 = PR = 10, a11 = a22 = 1, c1 = 1, a12 = c2 = 0, = a21 /a22 . . . . . Achievable sum-rate (4.6a)-(4.6f) and outer bound (4.18a-4.18d) of a symmetric IC-CR with respect to = a21 /a22 = a12 /a11 . The system parameters are P1 = P2 = PR = 10, a11 = a22 = 1, and c1 = c2 = 0.2 . Achievable sum-rate and sum capacity of an IC-CR operating under mixed relay-interference regime. The system parameters are P1 = P2 = PR = 10, a11 = 1, c1 = 1, a21 = 1, = a22 /a21 = a12 /a11 = c2 /c1 . . . . Gaussian Interference Relay Channel with two source-destination pairs Total optimized sum-rate (bits/channel use) of the symmetric channel with a21 = 1/8, P1 = P2 = PR = 10. When the relay is not present, we have P1 = P2 = 15. . . . . . . . . . . . . . . . . . . . . . . . . . . Optimal power allocation variables of the symmetric channel for rate splitting with a21 = 1/8, P1 = P2 = PR = 10. . . . . . . . . . . . . . Total optimized Rate (bits/channel use) of the symmetric channel as a function of a21 with b1 = 100, c1 = 1.2, P1 = P2 = PR = 10. When the relay is not present, we have P1 = P2 = 15. . . . . . . . . . . . . Optimal power allocation variables of the symmetric system for rate splitting as a function of a21 with b1 = 100, c1 = 1.2, P1 = P2 = PR = 10.

78

4.5

79

4.6

80

4.7

82

4.8

83 87

5.1 5.2

93 94

5.3 5.4

95 96

5.5

A.1 Equivalent model for IC-OBR Type II channel for c1 , c2 . . . . . 111

Chapter 1 Introduction

1.1

Motivation
Modern wireless networks are characterized by the coexistence of multiple

systems and devices. The increasing demand for the utilization of various services in a network simultaneously, such as Wi-Fi, cellular, and bluetooth systems, necessitates an overall design that judiciously incorporates interactions among these systems and the devices within each system. In a communication medium consisting of independent transmitters and receivers, interference between dierent transmitter and receiver pairs is an unavoidable and fundamental phenomenon that limits communication. In the current wireless systems, common approaches to overcome the limitation due to interference can be stated as orthogonalizing the communication links in time/frequency domain or treating the interfering signal as noise while the desired signal is decoded. Despite the substantial reduction in complexity, it is well-known that these schemes can be strictly suboptimal in terms of transmission rates [9]. In the last decade, multiple-input multiple-output (MIMO) scheme has emerged as the foremost technique to improve the rate and reliability of communication systems [18] [19]. The gains oered by MIMO systems rely on their potential to combat fading, introduce spatial diversity and increase spatial multiplexing. The relay channel, on the other hand, is another basic communication model that forms the building

2 block of larger networks [6]. Relaying mimics MIMO communications by establishing interactions among the distributed nodes that serve as multiple antennas both at the transmitter and receiver sides. Despite the inherit loss arising from distributed nature of the nodes, relaying oers substantial gains similar to MIMO communications. Even though the strategies, corresponding achievable rates and the capacities of relaying systems in a point-to-point communication scenario are investigated extensively, relaying in interference limited systems is less understood. In this dissertation, we explore the eect of relaying on the communication rates in multi-terminal networks where the system is prone to interference due to simultaneous transmission at the transmitter nodes. Our aim is to develop novel transmission techniques by incorporating joint source and relay transmission schemes and to get insight on the implications of these schemes on the practical systems such as next generation wireless systems, i.e. cellular networks with Long Term Evolution (LTE) standards. These techniques are not merely straightforward extensions of the standard pointto-point relaying schemes which basically aim at improving the decodability of the desired message, but also involve the functions to eectively ease the degradation in the network due to interference at the destination terminals. To understand the ultimate transmission rates one can achieve in the systems of interest and determine the optimal transmission strategies, another goal in this thesis is to establish the capacity of various systems that are investigated under interference limited system framework. Specically, the optimal scenarios under which the system complexity is reduced are investigated. Motivated by these points, various communication systems with multi-terminal network aided by a relay node are explored in this thesis. Initially, to elaborate fundamental relaying operations in multi source-destination systems, we explore a model where the relay transmits and receives via orthogonal bands with respect to the band the multiple sources and destination operate in. This scenario of interest is modeled by a two-user interference channel aided by a relay with orthogonal re-

3 ception/transmission bands to the underlying interference channel, and denoted by interference channel with an out-of-band relay (IC-OBR), [72] [78]. We investigate two specic models: (i) IC-OBR Type-I: The out-of-band relay channel (OBRC) is operated by assuming orthogonal transmissions over the four Gaussian links connecting sources to relay and relay to destinations; (ii) ICOBR Type-II: The OBRC is more generally operated by orthogonalizing the Gaussian multiple access channel between the two sources and the relay on one side, and the broadcast channel from relay to destinations, on the other. For a total relay bandwidth constraint, both xed and variable bandwidth allocations at the relay channel are explored. Besides its essence in modeling distinct communication systems that operate in the same network concurrently, such as Wi-Fi and cellular systems operating simultaneously in the coverage of a base-station, this model helps us to understand fundamental relaying features in a general interference limited communication network. We denote such relaying operations as: i) Signal relaying, i.e. the relay convey additional information from the sources to the destinations, ii) Interference forwarding, i.e the relay helps the decodability of interference at the destinations, and iii) Interference cancelation, the relay reduces the strength of interference signal at the destinations [66] [72] [78]. As a crucial design parameter, the impact of such relaying operations on the overall system complexity is a striking question that we further elaborate in this thesis. In this manner, IC-OBR lends itself in demonstrating signicant reduction in encoding/decoding complexity through optimality of particular relaying operations such as signal relaying. Next, our aim is to discover the impact of relaying on the communication rates of multiple terminals in the case where relay transmission is employed in the same band of the terminals, whereas the relay reception is in orthogonal bands. Compared with the out-of-band transmission of the relay node as mentioned above, this assumption at the relay yields considerable utilities in the practical network design and improves the capability of the relay to further manage the interference in the

4 network. The fundamental reason behind such utilities relies on the relays capability to interact with the received signals at the destinations directly as a consequence of transmission in the bands of the interfering terminals. Hence, one of our primary interests in examining such a model is to reveal which relaying features bear particular importance in treating the interference for utmost benet. Similar to IC-OBR, we model such a scenario with a two user interference channel assisted by a relay node where the relay is connected to the sources via orthogonal and nite capacity links and the relay transmission is performed in the band of the interfering source terminals. Due to its structure, we denote this system as an interference channel with an out-of-band reception/in-band transmission relay (IC-OIR), [41] [59] [66]. Finally, we explore a situation that is of particular interest in relaying/cooperation among the terminals that communicate on the same transmission band in a network. Such situation arises in almost all networks with multiple terminals wishing to communicate with the others in the network, regardless of the communication technique/standard used. Similar to the previously mentioned systems, a natural model to explore such a scenario is through a two user interference channel assisted by an in-band reception/in-band transmission relay. We denote the model in this thesis as an interference channel with a relay [43] [66]. The inherit dierence of this model with respect to IC-OBR and IC-OIR is the reception band used by the relay which is indeed the only band used in the network. Despite its practical importance, this property limits the relaying features and interference management capability substantially. Yet, determining and analyzing the relaying functions suitable for this scenario further indicates the eectiveness and importance of interference management in the communication networks.

1.2

Key Contributions
In this part, we summarize our contributions based on the system models

mentioned in Sec. 1.1. For IC-OBR Type-I with xed bandwidth allocation at the OBRC, the optimality of signal relaying is shown for various situations such as when the relayto-destination channels of the OBRC form the performance bottleneck with respect to the source-to-relay. Moreover, we demonstrate that signal relaying attains optimal separation of IC and OBRC and therefore under such conditions, the encoding/decoding complexity is shown to be reduced substantially. Optimality of signal relaying and interference forwarding jointly for IC-OBR Type-I is also established under some conditions, which requires inseparable operation over the IC and OBRC and therefore joint encoding/decoding over these channels are essential. For IC-OBR Type-II, achievable regions due to decode-and-forward, compress-and-forward and lattice coding are given and compared for dierent channel conditions. The optimality of these schemes incorporated with signal relaying and interference forwarding are established under specic scenarios. For both IC-OBR Type-I and Type-II, variable bandwidth allocation at OBRC leads to similar conclusions with its xed bandwidth allocation counterpart. For instance, interference forwarding is shown to be crucial under some cases, whereas signal relaying is optimal for various channel conditions, hence yielding separable IC and OBRC operations. In an IC-OIR, our primary aim is to explore the eect of in-band transmission by the relay. Initially, relying on coding for states known at the transmitter techniques such as partial decode-and-forward transmission [6], dirty paper coding [12], and rate splitting at the sources [9], we give a general achievable rate region. Our results suggest that while relaying is usually utilized as an eective technique for boosting received signal quality, such as signal relaying and/or interference forwarding, in interference networks, an equally important task of the relay is to reduce the

6 interference signal strength at the destinations. Interference cancelation, where the relay transmits negatively beamforms with the relay signal, is shown to be a very crucial technique for IC-OIR to improve communication rates in the network. In particular the results suggests that, under some set of channel gains, the relay dedicates most of its resources in reducing the interference strength at the destinations through interference cancelation. On the other hand, we establish the optimality of signal relaying and interference forwarding by demonstrating the capacity region of IC-OIR under very strong relay-interference conditions. Moreover, for very large source-to-relay channel capacities, we establish the sum-capacity of IC-OIR under the specied mixed relay-interference conditions. Both capacity results entail the necessity of inseparability of IC and the relay channel for optimal performance. Finally, in an interference channel with an in-band reception/in-band transmission relay, we incorporate full-duplex feature to reveal the maximum benet due to relaying. We give an achievable rate region obtained by incorporating rate splitting at the sources [9] and decode-and-forward relaying [6]. In a symmetric system, optimal power allocations for maximum total throughput show that the relay employs interference forwarding only via common information transmission for most of the channel gains. We also demonstrate the throughput gain due to in-band reception and transmission of the relay by comparing the system with an orthogonal counterpart and show that in-band transmission provide signicant throughput improvement.

1.3

Dissertation Outline
The dissertation is organized as follows. In Section 2, we give some back-

ground information about the model of interest and discuss related works. In Section 3, we introduce our rst model, interference channel with an out-of-band relay (IC-OBR). Depending on the bandwidth allocation at the out-of-band relay channel(OBRC), we focus on two sub-models: IC-OBR Type-I and IC-OBR Type-II. For

7 both systems, achievable regions and outer bounds are given for xed and variable bandwidth allocation at the OBRC. Capacity results demonstrating the optimality of relaying operations as well as separability of IC and OBRC are demonstrated. In Section 4, we extend the model to an in-band transmission relay which still receives in orthogonal bands from the sources (IC-OIR). A general achievable scheme including non-linear relaying schemes is given. For large source-to-relay link, we make the analogy of the system to a cognitive model and determine outer bounds. We give the sum capacity of this model as well as the capacity region of the nite-link system under very strong and mixed relay interference conditions. In Section 5, we explore an interference channel with a relay where the IC and the relay channel operate in the same band. We demonstrate a general achievable region and the optimal total throughput is compared with an orthogonal system. In Section 6, we conclude the dissertation.

Chapter 2 Background and Related Work

Relaying in interference limited systems inherits two building blocks of communication systems: i) Interference channel, and ii) relay channel. Both channels have been investigated extensively in the literature and the capacities are established in some special cases. However, complete understanding of any of the two channels has not been succeeded yet. In this section, we briey review these channels in terms of achievable schemes developed and established capacity results. We also provide literature review on relaying in interference limited systems and consider related works on our models of interest.

2.1

Interference Channel
Interference is a central phenomenon in any communication system com-

posed of multiple transmitter and receiver nodes sharing a common medium. To account and understand the role of interference, interference channel (IC) model is used extensively in the literature. In an interference channel, the source signals interfere with each other at the destinations, hence the point-to-point communication performances are degraded. The best strategy and corresponding achievable rate region is obtained by Han and Kobayashi [9]. The scheme involves splitting of source messages as common and private parts where the common part is decoded at both

9 destinations to reduce the eect of interference. Recently, for a Gaussian interference channel, Etkin et.al. [62] demonstrated that a very simple version of Han-Kobayashi scheme can achieve the rates within 1 bits/s/Hz of capacity region. However, the capacity region of a Gaussian IC is still unknown for most of the cases except under the strong [10] and noisy interference [64] [44] [67]. In this section we briey mention about the Han-Kobayashi scheme as well as capacity results known for a Gaussian IC. In Figure 2.1, a two-user Gaussian interference channel in its standard form is shown. The signals received on the IC by the two receivers D1 and D2 in channel use t = 1, ..., n are given as, respectively, Y1,t = X1,t + a21 X2,t + Z1,t Y2,t = a12 X1,t + X2,t + Z2,t , (2.1) (2.2)

where Xi,t R represents the (real) input symbol of source Si , which satises the power constraint
1 n n t=1 2 E[Xi,t ] Pi , and Zi,t are independent identically distributed

(i.i.d.) zero-mean Gaussian noise processes. Due to its convenience in characterizing achievable and capacity regions, symmetric interference channel is quite important. The following assumptions determine a symmetric Gaussian interference channel as well as symmetric rate of IC where a a21 = a12 , P P1 = P2 , R R1 = R2 .

2.1.1

Han-Kobayashi Achievable Scheme


As mentioned, Han-Kobayashi scheme gives the best known achievable re-

gion for an IC to date. In the scheme, rate-splitting at the sources is performed to facilitate interference cancelation at the destinations. Each transmitter splits its message as Wi = (Wic , Wip ), i = 1, 2, such that the common splits (W1c , W2c ) are

10

Z1

W1

X1

1 a12 a21

Y1

W1

W2

X2

+
Z2

Y2

W2

Figure 2.1: Gaussian interference channel in standard form. decoded at both destinations, and hence part of the interference can be removed at
n the destinations by stripping-o the corresponding interfering codeword, X1 c (W1c ) n or X2 c (W2c ). The private split Wip is decoded at Di only, i = 1, 2. The encoding n n is performed by generating two codebooks with codewords Xic (Wic ) and Xip (Wip ),

i = 1, 2, at the sources. Then, each source employs superposition coding to transmit


n n the codeword Xin = Xic (Wic ) + Xip (Wip ). Han-Kobayashi scheme involves joint de-

coding of the message splits (W1c , W1p , W2c ) at D1 and (W2c , W2p , W1c ) at D2 . The following proposition demonstrates that a simple Han-Kobayashi scheme achieves within 1 bit/s/Hz of a symmetric Gaussian IC capacity. Proposition 1 ( [62]): The following rate R = min C (P + a2 P ) + C 1 + 1 a2 1, 2C a2 P + 1 a2 1 (2.3)

is within 1 bit/s/Hz of a symmetric capacity of a Gaussian interference channel. Proof 1 The proof is based on the rate-splitting idea used in the general Han-Kobayashi scheme as discussed above. However, sequential decoding is performed at the destinations such that the common splits (W1c , W2c ) are decoded rst and the codewords
n n X1 c (W1c ) and X2c (W2c ) are stripped-o. Then, each destination simply decodes its

own private message Wip , i = 1, 2. The details are given in [62] and omitted here.

11

2.1.2

Capacity Results
The capacity region of interference channel is known under strong interfer-

ence conditions only [9] [10]. Recently, the sum-capacity of a Gaussian IC is characterized under the so-called noisy interference regime [64] [44] [67]. In obtaining the capacity region under strong regime, [10] uses the decodability of the interference signals at the destinations to obtain the corresponding outer bounds. In particular, suppose that (R1 , R2 ) is in the capacity region of a Gaussian IC as given in (2.1)-(2.2). Then, for the generated codebook structure, D1 is able
n to decode W1 , and hence reconstruct the codeword X1 (W1 ). Using the regenerated

codeword and the received signal Y1n , D1 is able to form a new signal
n n Y1n = a12 X1 + X2 + n Z1 . a21

(2.4)

Clearly, for a21 1, D1 observes a less noisy version of the signal observed at D2 , that is Y2n , and hence is able to decode W2 as long as D2 decodes its desired message, W2 . Using similar approach it is possible to show that for a12 1, D2 decodes the message W1 whenever D1 is able to decode it. Then, for a12 1 and a21 1, both destinations can decode W1 , W2 and the overall system becomes a compound multiple-access channel whose capacity is the intersection of the two multiple-access channels. The following proposition gives the capacity region of a Gaussian IC operating under strong interference conditions. Proposition 2 ( [9] [10]): The capacity region of a Gaussian interference channel with a21 1 and a12 1 is given by R1 C (P1 ) R2 C (P2 )
2 R1 + R2 min C (P1 + a2 21 P2 ), C (a12 P1 + P2 ) .

(2.5) (2.6) (2.7)

In establishing the sum-capacity of a Gaussian IC under noisy-interference conditions, [64] [44] [67] propose giving genie signals to the destinations which help

12 in obtaining meaningful outer bounds for small interfering gains, a12 and a21 . In addition the following Proposition and Corollary play a crucial role in demonstrating the eectiveness of these outer bounds and is also used in this thesis in proving various results. Proposition 3 ( [36]): Consider the following optimization problem
n n W = max h(X n + Z1 ) h(X n + Z2 ) p(x)

subject to Cov(X n )

(2.8)

n n are Gaussian vectors with strictly positive denite covariance matrix and Z2 where Z1

KZ1 and KZ2 , respectively, and the maximization is over all random vector X n inden n pendent of Z1 and Z2 . Then, for any 1 and positive semidenite S, a Gaussian

X n is an optimal solution of this optimization problem. The following Corollary explicitly solves the optimization problem (2.8) for
n n iid. noises Z1 , Z2 with covariance matrices N1 I and N2 I, respectively and assuming

= 1. Corollary 1 ( [67]): For = 1, the optimal solution of (2.8) is iid Gaussian and the optimum solution denoted by W is n log P +N1 2 P +N2 W = n log N1 2 N2

if N1 N2 if N1 > N2

(2.9)

Using Proposition 3 and Corollary 1, the noisy sum-capacity of a symmetric Gaussian IC is characterized as follows. Proposition 4 ( [44] [64] [67]): In a symmetric Gaussian interference channel satisfying the condition a + a3 P 0.5, (2.10)

13 the sum capacity is given by Csum = C 1 + P 1 + a2 P . (2.11)

Note that the sum-capacity obtained in (2.11) for the noisy-interference conditions given in (2.10) is obtained by simply treating the interference as noise.

2.2

Relay Channel
The relay channel was introduced by Van der Meulen [2]. Cover and El

Gamal [6] demonstrated various achievable schemes such as decode-and-forward and compress-and-forward transmission. The capacity of the relay channel is established for a physically degraded scenario and decode-and-forward transmission is shown to be the optimal scheme [6] under the physically degradedness condition. Fig. 2.2 shows a Gaussian relay channel. The signals received by the relay and the destination in channel use t = 1, ..., n are given as, respectively, YR,t = bX1,t + ZR,t Y1,t = X1,t + cXR,t + Z1,t , (2.12) (2.13)

where Xi,t R represents the (real) input symbol of the source with the power constraint
1 n n t=1 2 E[X1 ,t ] P1 , 1 n n t=1 2 E[XR,t ] PR and ZR,t and Z1,t are independent

identically distributed (i.i.d.) zero-mean Gaussian noise processes. As a key component and technique of several achievability and capacity results established in this thesis, we briey review the decode-and-forward scheme in the next section.

2.2.1

Decode-and-Forward Transmission
The main idea in the decode-and-forward transmission is that, the relay

decodes the signal it receives from the source terminal and reencodes the source

14

Z1

X1
ZR

1
c

Y1

b
b=1 X1n(1,W11) XRn(1)

YR

XR

Figure 2.2: Gaussian relay channel.


b=2 X1n(W11,W12) XRn(W11) . b=B X1n(W1B-1,W1B) XRn(W1B-1) b = B+1 X1n(W1B,1) XRn(W1B)

Figure 2.3: Block-Markov encoding scheme for the relay channel. message and transmits to the destination. The decoding and transmission structure of the relay together with the source transmission is performed through a block strategy as shown in Fig. 2.3. In Fig. 2.3, Block-Markov (BM) encoding scheme is shown. In BM transmission, the source encodes its messages by incorporating memory such
b1 n b that at each block b, the source encodes and transmits the codeword X1 (W1 , W1 ).

The relay knows the messages transmitted in the previous block b 1, and removes
b1 n the eect of XR (W1 ) from the received signal. Therefore, the following rate is

achievable, R I (X1 ; YR |X2 ). (2.14)

n b ) After decoding the corresponding message, the relay transmits the codeword XR (W1

in block b + 1. At the destination, sliding window technique is employed such that the source the signals received at block b and b + 1 are used in decoding the messages

15
b+1 b , W1 ). Then, it is possible to show that the following rates are achievable, (W1

R I (X1 , X2 ; YR ). Overall, the following rate is achievable, R max min{I (X1 ; YR |X2 ), I (X1 , X2 ; YR )}.
pX1 XR

(2.15)

(2.16)

For a Gaussian system, one can choose the distribution as well as the codewords such that X1 = P1 XR + X1 PR (2.17)

where XR N (0, PR ) and X1 N (0, P1 ). Then, the following gives an achievable rate, R min C b2 P1 , C P1 + c2 PR + 2c P1 PR . (2.18)

2.3

Related Work
In this part, we discuss the related work on relaying in interference limited

systems. As a closely related area, we also consider relaying and cooperation in networks with multiple nodes which lead to various cooperation models. Vast majority of work on cooperation among multiple nodes assume single source-destination pair assisted by multiple relay nodes [7] [26]. Aref [7] generalized the techniques developed for the regular relay channel [6] to account for multiple relays. Moreover, in [7], capacity achieving schemes for deterministic relay networks are demonstrated. More recently, by incorporating irregular encoding/successive decoding and decode-and-forward relaying, Gupta and Kumar [20] give an achievable scheme for a large network with multiple relays. Extension of single source-destination pair to multiple pairs assisted by multiple relays is also investigated in [20]. In [26], using decode-and-forward and

16 compress-and-forward techniques, achievable regions for multiple relay, source and destination nodes are demonstrated. In particular, a multiple-access relay channel (MARC) model is explored. In the MARC model, multiple sources wish to communicate with a common destination are assisted by a relay which can be full-duplex or half-duplex. Using cut-set bound argument, upper bounds on the capacity region are given and compared with the achievable regions due to DF and CF schemes for various node topologies in the network. It is shown that DF scheme performs very close to the upper bound when the relay is located close to the sources. However, when the relay operates in the vicinity of the destination, CF is the shown to be the best candidate for in terms of achievable rates. As a counterpart to the MARC model, a broadcast relay channel (BRC) is also introduced in [26]. In a BRC, a dedicated relay helps a source node that wishes to transmit its signal to multiple destinations in broadcast fashion. In [39], a partially cooperative relay broadcast channel is studied where one of the destinations acts as a relay in forwarding the source signal in spite of a dedicated relay node. The capacity region of this channel is established for various scenarios such as semideterministic and orthogonal partially cooperative BRC. As a natural extension of MARC and BRC, the IC aided by a relay has been rst studied in [43]. The discrete memoryless and Gaussian IC with a relay is further investigated in [49] [55], in which simplied channel models are considered where the relay only receives from one source. DF-based strategies at the relays with joint decoding at the destinations are proposed without rate splitting and shown to exhaust the capacity region under some conditions. These works emphasize the fact that forwarding the interference of even a single source may improve the rates of both users. Related work is also presented in [53], where the relay is assumed to be aware a priori of the users messages (cognitive relay) and sophisticated achievable strategies are investigated. [77] derives an upper bound on the capacity region by allocating innite power to the relay. The eect of relaying for one-sided interference channel is explored in [57]. Both cases where the interference-free destination assists

17 the interfered destination (Type I) and vice versa (Type II) are investigated. Based on Han-Kobayashi rate-splitting technique, DF and CF relaying schemes, achievable rate regions are given for both cases. The optimality of interference-forwarding is established for asymptotically high signal-to-noise and interference-to-noise ratios for Type I system where it is shown that each relay bit asymptotically improves the sum-capacity one-bit. For Type II system, CF is shown to achieve sum-capacity for asymptotically high relay to destination link. Due to its crucial role in the system complexity, separability of IC has been taken attention recently in the literature. As shown in [76] [73] [71], optimal operation over parallel ICs, unlike scenarios with a single source or destination, typically entails joint encoding over the parallel channels. In other words, the signals sent over the parallel ICs need to be generally correlated to achieve optimality, and thus a separable approach, whereby the parallel channels are treated independently, is in general not sucient. The original work [76] derives conditions under which correlated transmission of private messages is optimal, whereas in [71] scenarios are found for which sending both correlated private and common messages is optimal. As a related set of work, [70] explores the sum-capacity of parallel Gaussian interference channel in noisy interference regime and shows that the sum-capacity is achieved by treating interference as noise in each sub-channel as well as independent (separable) transmission at each sub-channel.

18

Chapter 3 Interference Channel with an Out-of-band Reception/Out-of-band Transmission Relay

3.1

Introduction
Modern wireless communication networks are characterized by the coexis-

tence of an increasing number of interfering devices and systems. While this often leads to an overall system performance that is limited by mutual interference, the presence of many independent wireless devices may also potentially oer new opportunities and performance benets by allowing cooperation. Opportunities for cooperation are further enhanced for multistandard terminals that are able to communicate simultaneously over multiple radio interfaces, and thus to interact and cooperate with devices belonging to dierent systems and networks. For instance, many current wireless terminals are equipped with a 3G cellular transceiver along with Wi-Fi interface. This chapter focuses on investigating the advantages of cooperation in interference-limited scenarios where cooperation is enabled by orthogonal radio interfaces and multistandard terminals. To gain insight on the general picture, we model the system of interest as an interference channel (IC) and an out-of-band relay (OBR) assisting the source-destination pairs via its orthogonal channels, and denote the system as an interference channel with an out-of-band relay (IC-OBR). We are

19 interested in understanding the optimal operation on the overall system, in particular the role of an orthogonal relay in interference-limited systems in terms of signal relaying and interference forwarding. Also, our aim is to further elaborate the impact of using orthogonal bands for the relay on the overall system design. Therefore, we are interested in revealing under which scenarios separable operation over the IC and OBRC, such that encoding and decoding operations are employed independently for IC and OBRC, lead to the optimal performance. As explained above, the two components of an IC-OBR model are an IC and the OBR channel (OBRC). We model the latter in two dierent ways: (i ) ICOBR Type-I: The OBRC is operated by assuming orthogonal transmissions (e.g., via TDMA or FDMA) over the four Gaussian links connecting sources to relay and relay to destination; (ii ) IC-OBR Type-II: The OBRC is more generally operated by orthogonalizing the Gaussian multiple access channel between the two source and the relay and the broadcast channel from relay to destinations. The chapter is organized as follows. In Section 3.2, we give the system models of IC-OBR Type-I and Type-II channels. Section 3.3 gives a general outer bound and achievable region for IC-OBR Type-I for both xed and variable OBRC bandwidth allocations. Capacity results are established for both cases. In Section 3.4, we give outer bounds on the capacity region for IC-OBR Type-II. Conditions for the optimality of signal relaying and/or interference forwarding for IC-OBR Type-II are established in this section. Finally, we conclude the chapter in Section 3.5. Notation : We dene C (x) = 1/2 log2 (1 + x).

3.2

System Models
We investigate the IC-OBR models shown in Fig. 3.1, which we refer to as

IC-OBR Type-I and Type-II. In both models, the sources S1 and S2 communicate to their respective destinations D1 and D2 via two orthogonal channels, namely a

20 Gaussian IC and the out-of-band relay channel (OBRC), where the latter is characterized by 0 channel uses per channel use of the IC. In pratice, parameter can be thought of as the ratio between the bandwidth of the OBRC and of the IC. Specically, each source Si , i = 1, 2, wishes to send a message index Wi , uniformly drawn from the message set [1, 2nRi ] 1 , to its destination Di , with the help of an OBR, which operates half-duplex. Notice that n is the number of channel uses of the IC available for communication of the given messages (which yields n channel uses for the OBRC), so that Ri is the rate of the ith pair (Si , Di ) in terms of bits per IC channel use. The signals received on the IC by the two receivers D1 and D2 in channel use t = 1, ..., n are given as, respectively, Y1,t = X1,t + a21 X2,t + Z1,t Y2,t = a12 X1,t + X2,t + Z2,t , (3.1a) (3.1b)

where Xi,t R represents the (real) input symbol of source Si , which satises the power constraint
1 n n t=1

x2 i,t Pi , and Zi,t are independent identically distributed

(i.i.d.) zero-mean Gaussian noise processes with unit power. The two IC-OBR models studied in the following dier in the way the OBRC is operated, as discussed below. Both models assume a half-duplex relay.

3.2.1

IC-OBR Type-I
In the IC-OBR Type-I model, shown in Fig.3.1-(i ), the OBRC bandwidth (or

equivalently the set of channel uses n) is partitioned into four orthogonal Gaussian channels, corresponding to dierent source-to-relay and relay-to-destination pairs. This can be realized by orthogonal access schemes such as TDMA or FDMA. Specifically, we have two Gaussian channels from sources Si to relay R with fraction of
As it is common in the literature, we consider the number of messages 2nRi rounded o to the smallest larger integer. We will use the same convention wherever integer quantities are needed.
1

21 channel uses iR , i = 1, 2, and two Gaussian channels from the relay R to destinations Di , i = 1, 2, with fraction of channel uses Ri . We have are given by YiR,t = bi XiR,t + ZiR,t , (3.2)
2 i=1 (iR

+ Ri ) = .

The signals received by the relay R over the OBRC on the source-to-relay channels

for t = 1, ..., iR n and i = 1, 2, whereas the signals received at the destination over the OBRC on the relay-to-destination channels are given by YRi,t = ci XRi,t + ZRi,t , (3.3)

for t = 1, ..., Ri n, i = 1, 2, where (ZiR,t , ZRi,t ) are i.i.d. zero-mean Gaussian noise processes with unit power. We assume power constraints
1 n Ri n t=1 1 n iR n t=1

x2 iR,t PiR , and

x2 Ri,t PRi , i = 1, 2. The rationale behind this power constraint assumption

arises for the scenarios such as the relay transmits using TDMA with per-symbol power constraints, or employs FDMA transmission with spectral mask constraints. Another model that encompasses this assumption is where the relay communicates with the destinations using two distinct radio interfaces with dierent transceivers. A (2nR1 , 2nR2 , n) code for the IC-OBR type-I is dened by: (a ) The encoding functions at the sources Si , i = 1, 2, given by fi : [1, 2nRi ] Rn RiR n
(n) (n)

(3.4)

iR n which maps a message Wi [1, 2nRi ] into the codewords (Xin , XiR ) = fi (Wi ) to

be transmitted on the IC and OBRC, respectively; (b ) The encoding function g (n) at the relay given by g (n) : R1R n R2R n RR1 n RR2 n , (3.5)

nR1 nR2 1R n 2R n (n) which maps the received signal to the codewords (XR (YR 1 , XR 2 ) = g 1 , YR2 ),

i = 1, 2, that are sent to the destinations; (c ) The decoding functions at the destinations Di , denoted by hi , i = 1, 2, with, hi : Rn RRi n [1, 2nRi ]
(n) (n)

(3.6)

22
Ri n which maps the received signal via the IC, Yin and OBRC YRi into the estimated Ri n message Wi = hi (Yin , YRi ). (n)

3.2.2

IC-OBR Type-II
The IC-OBR Type-II model, shown in Fig.3.1-(ii ), the OBRC is orthogonal-

ized into two channels, one being a multiple-access channel (MAC) from S1 and S2 to R, with fraction of channel uses M AC , and the other being a broadcast channel (BC) from R to D1 and D2 , with fraction of channel uses BC . We have M AC + BC = . The received signal at the relay R over the OBRC is given by YR,t = b1 X1R,t + b2 X2R,t + ZR,t (3.7)

for t = 1, ..., M AC n; and the signal received at destination Di over the OBRC is YRi,t = ci XR,t + ZRi,t , i = 1, 2, for t = 1, ..., BC n, i = 1, 2. We have the power constraints
1 n BC n t=1 1 n iR n t=1

x2 iR,t PiR , and

x2 R,t PR . A (2nR1 , 2nR2 , n) code for the IC-OBR Type-II is dened similar to codes

for IC-OBR Type-I with the dierence that the encoding function g (n) at the relay is modied as g (n) : RM AC n RBC n ,
M AC n BC n which maps the received signal YR into the transmitted codeword XR,t = M AC n g (n) (YR ).

Note that IC-OBR Type-I is a special case of Type-II, obtained by orthogonalizing the MAC and the BC.

3.2.3

Symmetric IC-OBR Type-I and Type-II


Throughout the paper an IC-OBR Type-I with symmetric IC is dened by a a12 = a21 , P P1 = P2 , (3.8)

23 whereas we will refer to a IC-OBR Type-II as symmetric if it satises (3.8) and b b1 = b2 , c c1 = c2 , P s P1R = P2R . (3.9)

3.2.4

Achievable Rates for Fixed and Variable OBRC Bandwidth Allocation


Following conventional denitions, we dene the probability of error as the

probability that any of the two transmitted messages is not correctly decoded at the intended destination. Achievable rates (R1 , R2 ) are then dened for two dierent scenarios: (a ) Fixed OBRC bandwidth allocation : Here, the bandwidth allocation parameters over the OBRC, namely (R1 , R2 , 1R , 2R ) for IC-OBR Type-I and (M AC , BC ) for Type-II, are considered to be given and xed. Therefore, a rate pair (R1 , R2 ) is achievable if a coding scheme can be found that drives the probability of error to zero for the given (feasible2 ) bandwidth allocation parameters; (b ) Variable OBRC bandwidth allocation : Here the bandwidth allocation can be optimized, so that a rate pair (R1 , R2 ) is said to be achievable if a coding scheme exists that drives the probability of error to zero for some feasible bandwidth allocation parameters. The capacity region is in both cases the closure of the set of all achievable rates.

3.3

Analysis of IC-OBR Type-I


In this section, we investigate the IC-OBR Type-I system described in Sec.

3.2. We consider outer bounds and inner bounds to the achievable rate regions for both xed and variable OBRC bandwidth allocation. It is noted that the results for xed OBRC were partly presented in [72].
Bandwidth allocation parameters are feasible if M AC + BC = for Type-II.
2 2 i=1 (Ri

+ iR ) = for IC-OBR Type-I and

24

3.3.1

Outer Bound
In this section, we rst present a general outer bound to the capacity region

of an IC-OBR in terms of multi-letter mutual informations (Proposition 1). This bound is then specialized to a number of special cases of interest, allowing the identication of the capacity region of IC-OBR for various scenarios. Proposition 5 (Outer bound for IC-OBR Type-I): For xed OBRC bandwidth allocation, the capacity region of the IC-OBR Type-I is contained within the set of rates (R1 , R2 ) satisfying lim closure
n n n p(xn 1 ,x2 )=p(x1 )p(x2 )

(R1 , R2 ):

R1

1 n 2 2 I (X1 ; Y1n ) + min 1R C (b2 1 P1R ) + 2R C (b2 P2R ), R1 C (c1 PR1 ) n 1 n n 2 R1 I (X1 ; Y1n |X2 ) + min 1R C (b2 1 P1R ), R1 C (c1 PR1 ) n 1 n 2 2 R2 I (X2 ; Y2n ) + min 1R C (b2 1 P1R ) + 2R C (b2 P2R ), R2 C (c2 PR2 ) n 1 n n 2 R2 I (X2 ; Y2n |X1 ) + min 2R C (b2 2 P2R ), R2 C (c2 PR2 ) n

(3.10)

n where the union is taken with respect to all multi-letter input distributions p(xn 1 )p(x2 )

that satisfy the power constraints 1/n

n t=1

2 E[Xi,t ] Pi , i = 1, 2. With variable

OBRC bandwidth allocation, an outer bound is given as above but with the union in (3.10) taken also with respect to all parameters iR , Ri , i = 1, 2, such that Ri ) = . Proof 2 Appendix A.1.
2 i=1 (iR +

3.3.2

Achievable Rate Region


In this section, we derive an achievable rate region for the IC-OBR Type-

I. We propose to use a rate splitting scheme similar to the standard approach for

25 ICs [5] [9]. Specically, we split the message of each user into private and common messages, where the private message of each source is to be decoded only by the intended destination and the common is to be decoded at both intended and interfered destinations. However, private and common parts are further split into two (independent) messages as follows. One of the private message splits is sent over the IC and the other one over the OBRC. As for the common message, both parts are sent over the IC, but one of the two is also sent over the OBRC to the interfered destination for interference cancellation. More specically, we have the following four-way split of each message Wi , Wi = (WiR , Wip , Wic , Wic ), i = 1, 2, where: (i ) WiR [1, ...2nRiR ] is a private message that is transmitted via the OBRC only, directly to Di . Notice the since the OBR has orthogonal channels to the IC, this message is conveyed interference-free to Di ; (ii ) Wip [1, ...2nRip ] is a private message that is transmitted over the IC, decoded at Di and treated as noise at Dj , j = i; (iii ) Wic [1, 2nRic ] is a common message that is transmitted over the IC and OBRC. Specically, the relay conveys Wic to Dj only, j = i, to enable interference cancellation; (iv ) Wic [1, ...2nRic ] is a common message that is transmitted over IC only and decoded at both destinations. Remark 1 (Separability and Private vs. Common Messages) It should be noted that the considered transmission scheme is in general not separable, in the sense that correlated messages are sent over the IC and OBRC. Specically, while the private message splits (WiR , Wip ) are sent separately over the two parallel channels IC and OBRC, part of the common messages, Wic , is sent over both IC and OBRC in order to allow interference mitigation. This is apparently a reasonable choice for a IC-OBR Type-I, since the private information sent on the OBRC is conveyed without interference to the intended destination, and thus there is no need for transmission also over the IC. Notice that transmission of the private messages over the OBRC amounts to signal relaying, while transmission of the common parts can be seen as interference forwarding.

26 Proposition 6 (Achievable Rate Region for IC-OBR Type-I): For xed OBRC bandwidth allocation, the convex hull of the union of all rates (R1 , R2 ) with Ri = Ric + Ric + Rip + RiR , i = 1, 2, that satisfy the inequalities Rj C
j S j S1 2 kj 1 Pj

N1
j S2 2 kj 2 Pj

(3.11a) (3.11b) (3.11c) (3.11d) (3.11e) (3.11f)

Rj C
j S2

N2

R1c + R1R 1R C (b2 1 P1R ) R2c + R2R 2R C (b2 2 P2R ) R2c + R1R R1 C (c2 1 PR1 ) R1c + R2R R2 C (c2 2 PR2 ),

provides an achievable rate region for the Gaussian IC-OBR Type-I, where conditions (3.11a)-(3.11b) must hold for all subsets S1 T1 = {1c, 1p, 2c } and S2 T2 = {2c, 2p, 1c }, except S1 = {2c } and S2 = {1c }; and we dene Ric = Ric + Ric ,
2 N1 = a 2 21 P2p +1, N2 = a12 P1p +1, and the parameters kj 1 = 1, kj 2 = a12 if j {1c, 1p},

and kj 1 = a21 , kj 2 = 1 if j {2c, 2p}. Moreover, we use the convention Pjc = Pjc , j = 1, 2, and the power allocations must satisfy the power constraints P1c + P1p P1 and P2c + P2p P1 . With variable OBRC bandwidth allocation, rates (3.11a)-(3.11f) can be evaluated for all bandwidth allocations satisfying Proof 3 Appendix A.2.
2 i=1 (iR

+ Ri ) = .

3.3.3

Capacity Results for Fixed OBRC Bandwidth Allocation


Consider xed OBRC bandwidth allocation. Here, we discuss some condi-

tions under which signal relaying or a combination of signal relaying and interference

27 forwarding are optimal. According to the discussion above, it is noted that optimality of signal relaying implies that separable operation on the IC-OBR is optimal, whereas, if both signal relaying and interference forwarding are needed, a separable scheme (at least in the class of strategies we considered) is not sucient to attain optimal performance. Optimality of Signal Relaying The rst result below shows that, if, on the path Si R Di , the relay to des2 tinations links form the performance bottleneck, i.e., R1 C (c2 1 PR1 ) 1R C (b1 P1R ) and 2 R2 C (c2 2 PR2 ) 2R C (b2 P2R ), signal relaying (and thus separable operation) achieves

capacity. The proposition below is expressed in terms of the capacity region CIC of a regular IC, which is generally unknown in single-letter formulation apart from special cases. Proposition 7 (Optimality of Signal Relaying Only): For a IC-OBR Type-I
2 2 with xed OBRC bandwidth allocation and R1 C (c2 1 PR1 ) 1R C (b1 P1R ), R2 C (c2 PR2 )

2R C (b2 2 P2R ), the capacity region C is given by the capacity region CIC of the IC, en2 hanced by (R1 C (c2 1 PR1 ), R2 C (c2 PR2 )) along the individual rates as

C = {(R1 , R2 ): (R1 R1 , R2 R2 ) CIC },


2 for R1 R1 C (c2 1 PR1 ), R2 R2 C (c2 PR2 ). Equivalently, the capacity region C is given

by C = lim closure
n
n n n p(xn 1 ,x2 )=p(x1 )p(x2 )

(R1 , R2 ):

1 n ; Y1n ) + R1 C (c2 I (X1 1 PR1 ) n 1 n R2 I (X2 ; Y2n ) + R2 C (c2 2 PR2 ) n R1 the power constraints 1/n by signal relaying only.
n t=1

(3.12)

n where the union is taken with respect to the input distribution p(xn 1 )p(x2 ) that satises 2 ] Pi , i = 1, 2. The capacity region is achieved E [Xi,t

28 Proof 4 The converse follows immediately from Proposition 5 with the conditions
2 Ri C (c2 i PRi ) iR C (bi PiR ), i = 1, 2 and by Ahlswedes multi-letter characterization

of the interference channel capacity region [3]. Achievability follows by sending independent messages (W1R , W2R ) in the notation of Proposition 6, of rates R1 C (c2 1 PR1 ) and R2 C (c2 2 PR2 ) from sources S1 and S2 , respectively, on the OBRC, and then using the Gaussian IC as a regular IC, stripped of the OBRC. 2 Remark 2 One bit on the OBRC links from relay to Di can clearly increase the end-to-end rates Ri by at most one bit (see also [57]). Under the assumptions of Proposition 7, signal relaying is able to fully capitalize on such gains by performing signal forwarding (i.e., by sending independent, private, information bits) over the OBRC. In other words, the relay-to-destination links can be used here solely for signal forwarding in a way that every carried bit increases the achievable rates by one bit, and this result cannot be improved by interference forwarding. Remark 3 Due to Proposition 7, in any scenario where a single-letter capacity region is known for the regular IC, the single-letter capacity region immediately carries
2 2 over to the IC-OBR Type-I with R1 C (c2 1 PR1 ) 1R C (b1 P1R ) and R2 C (c2 PR2 )

2R C (b2 2 P2R ). Therefore, for instance, we can obtain a single-letter capacity region expression for an IC-OBR Type-I in the strong interference regime (a21 1 and a12 1) [9] [10] or the sum-capacity in the noisy or mixed interference regime [64]
2 2 2 [44] [67], as long as R1 C (c2 1 PR1 ) 1R C (b1 P1R ) and R2 C (c2 PR2 ) 2R C (b2 P2R ).

Remark 4 Both Proposition 6 and 7 apply also to a general discrete memoryless IC-OBR Type-I (with the caveat of eliminating the power constraint). Optimality of Interference Forwarding While Proposition 7 provides a general capacity result for the case where the relay-to-destination links set the performance bottleneck on the paths Si R Di ,

29
2 2 2 i.e., R1 C (c2 1 PR1 ) 1R C (b1 P1R ) and R2 C (c2 PR2 ) 2R C (b2 P2R ), we next investigate

the capacity region for the complementary scenario in which such condition is not sat2 ised. We focus specically on two cases characterized by 1R C (b2 1 P1R ) R1 C (c1 PR1 ) 2 and R2 C (c2 2 PR2 ) 2R C (b2 P2R ), (Propositions 8, 9, and 10). Under the assumption

at hand, the following rate


2 2 2 Rex21 = min R1 C (c2 1 PR1 ) 1R C (b1 P1R ), 2R C (b2 P2R ) R2 C (c2 PR2 )

(3.13)

plays a key role. This can be interpreted as the excess rate from S2 to D1 on the OBR links once user 1 and 2 allocate the maximum possible rate on the OBRC channel for
2 2 signal relaying, i.e R1R = min{R1 C (c2 1 PR1 ), 1R C (b1 P1R )} = 1R C (b1 P1R ) and R2R = 2 2 min{R2 C (c2 2 PR2 ), 2R C (b2 P2R )} = R2 C (c2 PR2 ). Notice that the rationale behind this

maximum rate allocation for signal relaying follows from the considerations in Remark 2. The following propositions provide conditions under which both signal relaying and interference forwarding, and thus non-separable operation, are optimal. Propositions 8 and 9 assume conditions similar to strong interference for standard ICs [10], while Propositions 10 and 11 are formulated under conditions akin to the mixed interference regime [67]. Proposition 8 (Optimality of Joint Signal Relaying and Interference Forwarding, 1): For an IC-OBR Type-I with xed OBRC bandwidth allocation, and
2 channel conditions a12 1 and Rex21 max {0, C (a2 12 P1 + P2 ) C (P1 + a21 P2 )}, the

following inequalities characterize the capacity region, R1 C (P1 ) + 1R C (b2 1 P1R ) R2 C (P2 ) + R2 C (c2 2 PR 2 )
2 2 R1 + R2 C a2 12 P1 + P2 + 1R C (b1 P1R ) + R2 C (c2 PR2 ),

(3.14a) (3.14b) (3.14c)

which is achieved by joint signal relaying and interference forwarding.

30 Proof 5 Appendix A.3. Remark 5 To interpret the results of Proposition 8, allocate the maximum possible rate on the OBRC channel for signal relaying, i.e., R1R = R1 C (b2 1 P1R ) and R2R = R2 C (c2 2 PR2 ). After this, what we are left with is the IC plus an orthogonal link from S2 to D1 of capacity Rex21 . Under the conditions of Proposition 8, D2 is in the strong interference regime (a12 1) for this channel, and thus it can be proved that transmission of only common information by S1 over the IC is optimal. Moreover,
2 condition Rex21 max {0, C (a2 12 P1 + P2 ) C (P1 + a21 P2 )} essentially guarantees that

also D1 is in the strong interference regime in the channel at hand (IC plus capacity Rex21 ). As a result, S2 can transmit only common information with the caveat that part of it will be sent also over the OBRC (non-separable operation). To be specic, the assumptions in Proposition 8 encompass two dierent situations. In the rst,
2 we have the channel conditions (a2 12 1)P1 + (1 a21 )P2 0, so that the sum-rate

bound (3.14c) to receiver D2 forms the performance bottleneck in terms of sum-rate irrespective of a positive excess rate Rex21 (which increases the sum-rate at D1 as per (A.26)). Therefore, it can be seen that the capacity region of Proposition 8 is attained without performing interference forwarding, R2c = 0. In the second case, we have
2 (a2 12 1)P1 + (1 a21 )P2 > 0, so that, conversely, the sum-rate bound at D1 (see

(A.26) in Appendix A.3) may be more restrictive than (A.25). Here, it can be seen that it is optimal to exploit the excess rate Rex21 to perform interference forwarding
2 from S2 to D1 with rate equal to R2c = C (a2 12 P1 + P2 ) C (P1 + a21 P2 ).

In Fig. 3.2, we show the maximum achievable sum-rate of Proposition 6 for dierent congurations of the OBRC channel gains and bandwidths. We have a symmetric IC with P1 = P2 = 10 and a21 = a12 = a. For comparison, we show the case 1R = 2R = R1 = R2 = 0. Moreover, we rst consider a scenario where the relay-to-destination links form bottleneck with respect to source-to-relay links, i.e.
2 2 2 1R C (b2 1 P1R ) = 2R C (b2 P2R ) = 2, R1 C (c1 PR1 ) = R2 C (c2 PR2 ) = 1, thus falling within

31 the assumptions of Proposition 7. It can be seen that the sum-rate increases by 2 bits/ch use of IC for all values of a. Moreover, from Proposition 7, it is known that in the noisy [44] (a(1 + 10a2 ) 0.5, i.e, a 0.28) and strong (a 1) [10] interference regimes, the sum-rate shown in the gure is the sum-capacity. Finally, we consider
2 2 2 a situation with 2R C (b2 2 P2R ) = R1 C (c1 PR1 ) = 3, 1R C (b1 P1R ) = R2 C (c2 PR2 ) = 2,

which falls under the conditions of Proposition 8 for a 1. As stated in the Proposition, for a 1, the sum-rate shown in the sum-capacity is 4 bits/channel use of IC larger than the reference case of zero OBRC capacities. Proposition 8 assumes that the excess rate satises a given lower bound. The next result considers a scenario where the complementary upper bound is assumed. Proposition 9 (Optimality of Joint Signal Relaying and Interference Forwarding, 2): For a IC-OBR Type-I with xed OBRC bandwidth allocation, and the
2 2 conditions a21 1, 0 Rex21 = R1 C (c2 1 PR1 ) 1R C (b1 P1R ) C (a12 P1 + P2 ) C (P1 +

a2 21 P2 ), the following conditions characterize the capacity region R1 C (P1 ) + 1R C (b2 1 P1R ) R2 C (P2 ) + R2 C (c2 2 PR 2 )
2 2 R1 + R2 C P1 + a2 21 P2 + R1 C (c1 PR1 ) + R2 C (c2 PR2 ),

(3.15a) (3.15b) (3.15c)

which is achieved by joint signal relaying and interference forwarding. Proof 6 Appendix A.4. Remark 6 Similarly to the second case described in Remark 5, under the assumptions of Proposition 9, interference forwarding is useful in increasing the capacity region. However, unlike Proposition 8, here the excess rate is not large enough to make the sum-rate constraint (3.14c) at D2 the performance bottleneck with respect to (A.26). In fact, under the assumptions of Proposition 9, the sum-rate bound (A.26) for D1 is always more restrictive than (3.14c) for D2 in terms of the sum-rate, given

32
2 the constraint Rex21 C (a2 12 P1 + P2 ) C (P1 + a21 P2 ). The capacity region of Proposition 2 9 is attained by setting R2c = Rex21 = R1 C (c2 1 PR1 ) 1R C (b1 P1R ).

Remark 7 The assumptions in Proposition 9 imply the strong interference conditions a12 1 and a21 1. In contrast, the assumptions of Proposition 8 in general do not imply strong interference. This is, however, the case if we further assume the
2 condition (a2 12 1)P1 + (1 a21 )P2 0 (as in the rst case discussed in Remark 5).

Proposition 10 (Optimality of Joint Signal Relaying and Interference Forwarding, 3): For an IC-OBR Type-I with xed OBRC bandwidth allocation and the conditions a12 < 1, Rex21 max 0, C (P1 ) + C the following condition characterizes the sum capacity R1 + R2 C (P1 ) + C P2 1 + a2 12 P1
2 + 1R C (b2 1 P1R ) + R2 C (c2 PR2 ), P2 1+a2 12 P1

C (P1 + a2 21 P2 ) ,

which is achieved by joint signal relaying and interference forwarding. Proof 7 Appendix A.5. Remark 8 Unlike Propositions 8 and 9 where the overall IC-OBR operates in conditions similar to the strong interference regime of a standard IC, here D2 observes weak interference while D1 , thanks to the excess rate, is able to decode and strip-o the interference. Accordingly, the optimal coding strategy for S1 is to transmit private messages only, separably over both IC and OBRC, whereas S2 transmits only common messages in a non-separable way over both IC and OBRC (interference forwarding). We nally remark that, from the conditions in Proposition 10 and Appendix A.5, for a21
1+P1 , 1+a2 12 P1

the constraint on the excess rate in Proposition 8 becomes Rex21 0,

showing that all interference is decoded and removed over the IC at D1 without resorting to interference forwarding. Proposition 11 (Optimality of Joint Signal Relaying and Interference Forwarding, 4): For an IC-OBR Type-I with xed OBRC bandwidth allocation

33
2 and the conditions a21 1, 0 Rex21 = R1 C (c2 1 PR1 ) 1R C (b1 P1R ) C (P1 ) +

P2 1+a2 12 P1

C (a2 21 P2 + P1 ), the following condition gives the sum capacity


2 2 R1 + R2 C P1 + a2 21 P2 + R1 C (c1 PR1 ) + R2 C (c2 PR2 ),

which is achieved by joint signal relaying and interference forwarding. Proof 8 Appendix A.6. Remark 9 The situation in Proposition 11 is similar to the one in Proposition 9 in the sense that, even with the help of the excess rate, Rex21 , the achievable sumrate is limited by the rate at D1 in (A.26). However, as opposed to Proposition 9, the assumptions in Proposition 11 imply the conditions 1 a21
1+P1 1+a2 12 P1

which

in turn leads to 0 a12 1. Therefore, Proposition 11 essentially establishes the optimality of signal relaying and interference forwarding when one of the interfering links observes weak interference and the other strong interference, and hence the IC operates in mixed interference regime.

3.3.4

Capacity Results for OBRC Variable Bandwidth Allocation


In this part, we discuss optimality of the considered strategies under variable

OBRC bandwidth allocation. The next proposition shows that, thanks to the ability to allocate bandwidth among the source-to-relay and relay-to-destination channels, under certain conditions bandwidth can be allocated only to the best channels and in a way that only signal relaying is used. Proposition 12 (Optimality of Signal Relaying Only): Consider an IC-OBR Type-I channel with variable OBRC bandwidth allocation and a12 = a21 1, P1 = P2 ,
2 2 2 2 2 C (b2 2 P2R ) = C (c2 PR2 ), C (b2 P2R ) C (b1 P1R ) and C (c2 PR2 ) 2C (c1 PR1 ). The sum

34 capacity is given by 2 R1 + R2 C P1 + a2 21 P2 + C (c2 PR2 ) 2 and is achieved by signal relaying and transmitting only common information on the IC. Proof 9 Appendix A.7. Remark 10 Proposition 12 states that for a symmetric and strong IC, when the path S2 R D2 is better than S1 R D1 (see conditions on bi and ci ), under appropriate conditions, it is optimal to allocate all the bandwidth to the better path to perform signal relaying. Signal relaying increases the sum-rate by the number of bits carried on C (c2 the OBRC links, namely 2 PR2 ). Notice that it is possible to generalize Proposition 2 12 to arbitrary b2 and c2 , i.e. not necessarily restricting the channels to satisfy the
2 condition C (b2 2 P2R ) = C (c2 PR2 ). However, here we have focused on this simple case,

to illuminate more clearly the gist of the main result. To obtain further insight into the result of Proposition 12, in Fig. 3.3 we compare the sum-rate achievable by transmitting only common information on the IC and using signal relaying (with optimized bandwidth allocation), obtained from Proposition 6 (see (A.35) in Appendix A.7), and the upper bound obtained from Proposition 5 (see (A.36) in Appendix A.7). Parameters are xed as = 1, a21 = a12 = 2, c2 = 3, c1 = b1 = 2, P1 = P2 = P1R = P2R = PR1 = PR2 =10dB and b2 is varied. Numerical results show for this example that the achievable rate matches the upper bound for most of the channel gains (except 1.2 b2 1.8). The result in Proposition 12 begs the question as to whether interference forwarding can be ever useful in the case of variable bandwidth allocation. The following example answers this question in the armative. Consider an IC-OBR Type-I channel with b1 = 1.5, c1 = 3, c2 = 1, a12 = 3, a21 = 2, = 1 and all transmission powers set to 10dB. Here, b2 is varied. Fig. 3.4 shows the achievable

35 sum-rate obtained from Proposition 8 (which was obtained from Proposition 6) by assuming transmission of common messages only over the IC and either signal relaying only ((Wic , WiR ), i = 1, 2) or both signal relaying and interference forwarding ((W1c , W1R ),(W2c , W2c , W2R )) (see (A.23)-(A.26) in Appendix A.3) along with optimized bandwidth allocation (iR , Ri ), i = 1, 2. Also shown is the outer bound from Proposition 5, given in (A.36) for the channel parameters at hand. For b2 c2 = 1, the excess rate Rex21 is positive and interference forwarding is potentially useful as discussed in Sec. 3.3, i.e. Propositions 8 and 9 for the xed bandwidth case. For the variable bandwidth case, Fig. 3.4 shows similarly that interference-forwarding is instrumental in improving the achievable sum-rate for b2 2. However, for b2 < 2, interference forwarding is not needed (Proposition 12 shows that indeed signal forwarding is optimal for b2 0.75).

3.4

Analysis of IC-OBR Type-II


In this section, we investigate the IC-OBR Type-II channel described in Sec.

3.2. This channel diers from IC-OBR Type-I in that on the OBRC: (i ) The signals transmitted from S1 and S2 are superimposed at the relay; (ii ) Relay broadcasts to the destinations. These aspects allow novel, and more general, transmission strategies at sources and relay, thus making the design of optimal schemes more complex than on the IC-OBR Type-I. This is reected by the results presented below that encompass scenarios and techniques that have a counterpart in the IC-OBR Type-I analysis and others that stand out as specic to IC-OBR Type-II. To simplify the analysis, we still consider the same class of strategies described in Sec.3.3.2 for source and destinations. Namely, we assume a four-way message split into two private and two common parts, such that, as discussed in Remark 1, private signals are transmitted in a separable way over IC and OBRC, while one of the common messages is sent over both IC and OBRC (and, hence, transmitted

36 in a non-separable way). Even within this class of strategies, the variety of possible approaches is remarkable. For instance, the relay may perform Decode-and-Forward (DF) or Compress-and-Forward (CF) and the sources may encode by using random codes or structured (e.g., lattice) codes [48] [47]. For this reason, unlike IC-OBR Type-I, we will not give a general achievable rate region but rather focus on specic conditions under which optimality of certain design choices can be assessed. We rst give three outer bounds on the capacity region for IC-OBR TypeII. Then, we investigate conditions under which signal relaying (and thus separable operation, as per the discussion above) or joint signal relaying/ interference forwarding (and thus non-separable strategies) are optimal.

3.4.1

Outer Bounds
In the following, we give three outer bounds on the capacity region of IC-

OBR Type-II. The rst bound requires weak interference conditions on at least one of the links over the IC, while the second holds for any channel gains. The third bound is valid under strong IC conditions. Proposition 13 (Outer Bound for IC-OBR Type-II, 1): For an IC-OBR Type-II with xed OBRC bandwidth allocation, for a12 1 and c1 c2 , the capacity

37 region is included in the following region lim closure


n n n p(xn 1 ,x2 )=p(x1 )p(x2 ),0 + 1

(R1 , R2 ): (3.16a) (3.16b) (3.16c) (3.16d) (3.16e) (3.16f)

R1 R1 R1 R2 R2 R2

1 n 2 I (X1 ; Y1n ) + M AC C b2 1 P1R + b2 P2R n 1 n n I (X1 ; Y1n |X2 ) + M AC C b2 1 P1R n 1 n n I (X1 ; Y1n |X2 ) + BC C (c2 1 PR ) n 1 n 2 I (X2 ; Y2n ) + M AC C b2 1 P1R + b2 P2R n 1 n n I (X2 ; Y2n |X1 ) + M AC C b2 2 P2R n 1 c2 2 PR n I (X2 ; Y2n ) + BC C n 1 + c2 2 PR

n where the union is taken with respect to multi-letter input distributions p(xn 1 )p(x2 )

that satisfy the power constraints 1/n

n t=1

2 E[Xi,t ] Pi , i = 1, 2, and with respect

to parameters , with 0 1 and = 1 . With variable OBRC bandwidth allocation, an outer bound is given as above but with the union in (3.16a)-(3.16f) taken also with respect to all parameters M AC , BC , i = 1, 2, such that M AC + BC = . Proof 10 Appendix A.8. Proposition 14 (Outer Bound for IC-OBR Type-II, 2): For the IC-OBR Type-II with xed and variable OBRC bandwidth allocation and any channel gains, an outer bound to the capacity region is given as in Proposition 13 by replacing the bounds in (3.16c) and (3.16f) by, 1 n n ) + BC C c2 ; Y1n |X2 I (X1 1 PR n 1 n n R2 I (X2 ) + BC C c2 ; Y2n |X1 2 PR , n R1 respectively. (3.17a) (3.17b)

38 Proof 11 Appendix A.8. Proposition 15 (Outer Bound for IC-OBR Type-II, 3): For the IC-OBR Type-II with xed OBRC bandwidth allocation, an outer bound to the capacity region for a12 1, a21 1 is given by, R1 C (P1 ) + M AC C (b2 1 P1R ) R2 C (P2 ) + M AC C (b2 2 P2R )
2 2 R1 + R2 C (P1 + a2 21 P2 ) + M AC C (b1 P1R + b2 P2R ) 2 2 R1 + R2 C (a2 12 P1 + P2 ) + M AC C (b1 P1R + b2 P2R ).

(3.18a) (3.18b) (3.18c) (3.18d)

With variable OBRC bandwidth allocation, an outer bound is given as above but with M AC = in (3.18a)-(3.18d). Proof 12 Appendix A.9.

3.4.2

Capacity Results for Fixed OBRC Bandwidth Allocation


In this section, we consider xed OBRC bandwidth allocation.

Optimality of Signal Relaying In this section, we consider conditions under which signal relaying is found to be sum-rate optimal for xed bandwidth allocation. Decode-and-Forward (DF) Relaying We start by focusing on Decode-and-Forward (DF) strategies at the relay, as implicitly also done for IC-OBR Type-I. For the previous model, Proposition 7 showed that, in case the relay-to-destination links form the bottleneck on the OBRC, signal relaying achieves the entire capacity region. The next propositions extend this condition to a IC-OBR Type-II. However, due to the more

39 complex structure of the IC-OBR Type-II, the results below only establish sum-rate optimality. Moreover, they are limited to mixed interference [67] (Proposition 16) and strong interference conditions (Proposition 18), rather than to any interference channel. Proposition 16 (Optimality of Signal Relaying Only, DF): In an IC-OBR Type-II with xed bandwidth allocation, a12 < 1, a21 capacity is given by the inequality R1 + R2 C (P1 ) + C
2 if M AC C (b2 1 P1R ) BC C (c1 PR ); and by 1+P1 1+a2 12 P1

and c1 c2 , the sum

P2 1 + a2 12 P1

+ BC C c2 1 PR

(3.19)

R1 + R2 max + BC C

0 + 1

C (P1 ) + C c2 2 PR 1 + c2 2 PR

P2 1 + a2 12 P1 ,

2 + min M AC C b2 1 P1R , BC C c1 PR

(3.20)
c2 2 PR 1+c2 2 PR

2 2 if M AC C (b2 1 P1R ) < BC C (c1 PR ) and M AC C (b2 P2R ) BC C

, where is

the optimal power allocation that maximizes the sum-rate (3.20) with = 1 . Proof 13 Appendix A.10.

Remark 11 Proposition 16 involves two separate set of channel conditions for the OBRC. In both cases, the interference conditions on the IC are mixed and the optimal transmission strategy over the IC turns out to prescribe transmission of only private information by S1 (given that D2 is in weak interference) and of only common information by S2 (given that D1 is in a strong interference condition). The operation over the OBRC is instead dierent under the two set of channel conditions, but entails in both cases only signal forwarding (of independent information according to our separable approach, recall Remark 1).

40 The rst set of conditions (under which the sum-capacity is (3.19)) is char2 acterized by c1 c2 and M AC C (b2 1 P1R ) BC C (c1 PR ) and is illustrated in Fig. 3.5.

It corresponds to the case where the sources-to-relay MAC S1 R constitutes the bottleneck with respect to the relay-to-destinations BC. Notice that this view is consistent with the assumption of DF at the relay. This set of conditions is thus akin to the one of Proposition 7 for the IC-OBR Type-I. In this case, the optimal strategy in terms of sum-rate is for only user 1 to transmit over the OBRC using signal forwarding. Notice that this operating point on the OBRC (see dot in the gure) is sum-rate optimal if one focuses on the OBRC alone and on DF, since the corresponding achievable rate region is given by the intersection of the MAC and BC regions in Fig. 3.5. Proposition 16 shows that such operating point is also optimal for communications over the IC-OBR under the given conditions. The second set of conditions (under which the sum-capacity is (3.19)) is given
2 2 by c1 c2 , M AC C (b2 1 P1R ) < BC C (c1 PR ) and M AC C (b2 P2R ) BC C c2 2 PR 1+c2 2 PR

, and

is illustrated in Fig. 3.6. Here, the sum-rate optimal operation of the OBRC entails signal relaying for both sources, not just source 1 as above, and transmission over the OBRC at rates given by the operating point indicated in the gure. This rate pair is characterized by the power split in Proposition 16. Notice that the same considerations given above regarding optimality of this point for the OBRC alone with DF apply here. Remark 12 Observing Fig. 3.6 and recalling Propositions 8-10, one could guess (erroneously, as Proposition 16 shows) that signal relaying is suboptimal under the conditions of Fig. 3.6. These conditions seem to suggest an excess rate, in the sense of Proposition 8 and Remark 5, between S2 and D1 , since the maximum rate S2 R
2 (i.e., M AC C (b2 2 P2R )) is larger than the maximum rate R D2 (i.e., BC C (c2 PR )),

while the opposite is true for the path S1 R D1 . The fact that such excess rate is not to be exploited for interference forwarding by the optimal scheme of Proposition 16

41 can be interpreted in light of Remark 8, since the condition a21 need for interference forwarding. For a21 <
1+P1 , 1+a2 12 P1 1+P1 , 1+a2 12 P1

assumed

in Proposition 16, already guarantees strong interference condition at D1 , and thus no we will show that interferenceforwarding is instead instrumental in maximizing the sum-rate, see Sec. 3.4.2. We now consider strong interference conditions on the IC. Proposition 17 (Optimality of Signal Relaying Only, DF): In a symmet ric IC-OBR Type-II with xed bandwidth allocation, for 1 a 1 + P and BC C (c2 PR ) M AC C (2b2 Ps ), the sum capacity is given by R1 + R2 C (P + a2 P ) + M AC C (2b2 Ps ). Proof 14 Appendix A.11. Remark 13 The sum-rate (3.21) is achieved by transmission of only common information over the IC (due to the strong interference conditions) and symmetric-rate private messages (signal relaying) over the OBRC. Compress-and-Forward (CF) Relaying Here, we briey show that signal relaying may be (asymptotically) optimal also in combination with CF at the relay. Proposition 18 (Optimality of Signal Relaying Only, CF): In an IC-OBR Type-II with xed bandwidth allocation, the following rates are achievable via signal relaying and CF at the relay: R1 C (P1 ) + M AC C R2 C (P2 ) + M AC C b2 1 P1R 1 + 2 b2 2 P2R 1 + 2
2 b2 1 P1R + b2 P2R 1 + 2

(3.21)

(3.22a) (3.22b) , (3.22c)

2 R1 + R2 min C (P1 + a2 21 P2 ), C (a12 P1 + P2 ) + M AC C

42 where 2 satises 2
2 1 + b2 1 P1R + b2 P2R
BC BC

min (1 +

M AC c2 1 PR )

, (1 +

M AC c2 2 PR )

The above provides the capacity region, given by (3.18a)-(3.18d), for a12 1, a21 1 and c1 , c2 . Proof 15 See Appendix A.9. The rate (3.22a)-(3.22c) is achieved by transmitting common information over the IC, which leads to its optimality in the strong interference regime, and private information (signal relaying) over the OBRC. The relay performs CF which becomes optimal asymptotically as the relay-to-destinations BC quality improves. Fig. 3.7 shows the comparison of achievable sum-rate obtained from (3.22a)-(3.22c) and the outer bound (3.18a)-(3.18d) on the sum-rate given in Proposition 15 for a symmetric IC-OBR Type-II channel as a function of c with a = 2, b = 1, P = Ps = 10, = 1. We observe that the achievable sum-rate and outer bound are not only asymptotically equal for large c, as shown by Proposition 18, but in practice become very close already for c 6. Optimality of Interference Forwarding In this section, we discuss conditions under which strategies based on interference forwarding are optimal for IC-OBR Type-II. Decode-and-Forward (DF) Relaying We start from the scenario discussed in Remark 11 following Proposition 16. Specically, consider the conditions in Proposition 16, illustrated in Fig. 3.6 (also recall that we assume a12 < 1). As discussed in Remark 11, if we further have that a21
1+P1 , 1+a2 12 P1

as in Proposition 16, signal

relaying is optimal in terms of sum capacity and there is no need to exploit the excess rate between S2 and D1 . However, when the interfering link from S2 to D1 ,

43 a21 , is not strong enough to allow decoding and cancellation of the interference on the IC to D2 , i.e, a21 <
1+P1 , 1+a2 12 P1

then interference forwarding is potentially useful.

This is shown here by rst providing an achievable region that involves joint signal relaying and interference forwarding, then establishing its asymptotic optimality under given conditions that include the scenario discussed above, and nally discussing some numerical results. Proposition 19 (Asymptotic Optimality of Joint Signal Relaying and Interference Forwarding, DF): In an IC-OBR Type-II with xed bandwidth allocation, the following conditions give an achievable region R1 C (P1 ) + M AC C (b2 1 P1R ) R2 C P2 1 + a2 12 P1 + BC C c2 2 PR 1 + c2 2 PR c2 2 PR 1 + c2 2 PR (3.23a) (3.23b) (3.23c) (3.23d)

2 R1 + R2 C (P1 + a2 21 P2 ) + BC C (c1 PR ) + BC C 2 2 R1 + R2 C (P1 + a2 21 P2 ) + M AC C (b1 P1R + b2 P2R )

with + 1 via DF and joint signal relaying and interference forwarding. Moreover, for a12 < 1, and b2 , c1 , the sum-capacity is achieved by such scheme and given by R1 + R2 C (P1 ) + C Proof 16 Appendix A.12. Remark 14 The scheme achieving (3.23a)-(3.23d) is based on transmitting only private information over the IC and OBRC (signal relaying) by S1 , while S2 transmits common information over the IC and both common and private on the OBRC (joint signal relaying and interference forwarding). This scheme is shown to be sum-rate optimal if weak interference is seen at D2 and a large excess rate (in the sense of Remark 11) is available between S2 and D1 so as to essentially drive D2 in the very strong interference regime. P2 1 + a2 12 P1
2 + M AC C (b2 1 P1R ) + BC C (c2 PR ).

44 To investigate the role of interference forwarding in a non-asymptotic regime, Fig. 3.8 shows the sum-rate obtained from (3.23a)-(3.23d), by assuming that source 2 either uses only signal relaying (i.e., R2c = 0 in the achievable region given in Appendix A.12) or also interference forwarding, and the sum-rate upper bound obtained from Proposition 13 and given in (A.74), Appendix A.10. The OBRC gains are set to b1 = 1, b2 = 10, c2 = 1 and c1 is varied, all node powers are equal to 10 dB and M AC = BC = 1. We also have a12 = 0.5 and S2 D1 channel gain takes the values a21 {0.1, 0.9.1.8}. Note that for a21 = 1.8
1+P1 1+a2 12 P1

= 1.78 the condi-

tions given in Proposition 12 are satised and signal relaying alone is optimal. For a21 {0.1, 0.9} 1.78, the advantages of interference forwarding becomes substantial with increasing c1 , which is due to the fact that the S2 D1 pair can exploit more excess rate. The asymptotic optimality derived in Proposition 19 is here shown to be attained for nite values of b2 , c1 . Lattice Coding We nally briey provide a second scheme that is asymptotically (in the sense dened below) optimal and is based on interference forwarding. Consider a symmetric IC-OBR Type-II with a . Due to the latter assumption, S1 and D2 , can be equivalently seen as a single terminal, since communication between the two can take place with vanishingly small power. Similar conclusion applies to S2 and D1 . Therefore, the equivalent system now has only two terminals wishing to communicate with one another over two parallel channels: The rst (i.e., the IC in the original model) is a two-way Gaussian channel [1], while the second (the OBRC) is a two-way relay channel [38] [48] [47]. As shown in [48] [47], communications over a two-way relay channel can benet from structured, specically lattice, codes, rather than the usual randomly-generated codes. The proposition below shows that using structured codes to transmit common message splits Wic on the OBRC and transmitting private and common messages (Wip , Wic ) on the IC is asymptotically optimal in terms of equal rates R = R1 = R2 for a symmetric IC-OBR Type-II.

45 Using structured codes for IC-OBR Type-II incorporates interference forwarding which is structurally dierent than its counterparts as discussed in Type-I (Propositions 8, 9, 10 and 11) and Type-II (Proposition 19). As discussed in [48] [47] and mentioned above briey, the lattice coding allows the OBR to broadcast the modulo-sum of the source signals. On the other hand, the destinations decode the messages received over the relay-to-destination links by using the messages received and decoded via the IC. Therefore, as opposed to the schemes discussed before where the OBRC is exploited in decoding and canceling the messages transmitted via the IC, in the structured codes discussed, the received signals over the IC helps the destinations decode the messages broadcasted by the relay. Proposition 20 (Asymptotic optimality of Interference Forwarding Only): The symmetric capacity of a symmetric IC-OBR Type-II channel for BC C (c2 PR ) and a is given by R C (P ) + BC C c2 PR and is achieved by interference forwarding via structured codes. Proof 17 Appendix A.13. Fig. 3.9 compares the rate (3.21) achievable with the DF scheme of Proposition 13, rate (3.22a)-(3.22c) achievable with CF scheme of Proposition 18 and rate (3.24) achievable by lattice coding with the outer bounds (Proposition 14, (3.17a)(3.17b)) and (Proposition 15, (3.18a)-(3.18d)) for a symmetric IC-OBR Type-II. In Fig. 3.9 i) we set a = 3 and in Fig. 3.9 ii) a = 10 and study rates as a function of c. We have b = 1, M AC = BC = 0.5 and all node powers are set to 10 dB. We observe that DF achieves the sum-capacity for c 1.5 as given in (3.21). Also, as the interference gain a increases, lattice coding yields an achievable sum-rate that approaches the outer bounds, conrming the results in Proposition 20. (3.24)
M AC 2

log

1 2

+ b2 Ps

46

3.4.3

Capacity Results for OBRC Variable Bandwidth Allocation


Here, we briey investigate the eect of variable bandwidth allocation for

the IC-OBR Type-II via numerical results. We consider a mixed interference scenario with a12 = 0.5, a21 = 1.8 and c1 c2 , which satisfy all the conditions of Proposition 16 except the ones that depend on the bandwidth allocation (M AC , BC ). Recall that for given conditions on xed allocations (M AC , BC ), Proposition 16 shows the optimality of DF with signal relaying (separable operation). We compare the performance of the DF scheme in Proposition 16 (separable transmission) with an outer bound obtained from Proposition 13. In particular, from the conditions (3.16b) and (3.16c) for R1 and (3.16d) and (3.16f) for R2 , with a12 < 1, we obtain the following outer bound, R1 + R2 C (P1 ) + C P2 1 + a2 12 P1
2 + min M AC C (b2 1 P1R ), BC C (c1 PR )

2 + min M AC C (b2 1 P1R + b2 P2R ), BC C

c2 2 PR 1 + c2 2 PR

(3.25)

for = 1 . In both cases, bandwidth allocation (M AC , BC ) is optimized. In Fig. 3.10, the sum-rate discussed above are shown for variable S1 R gain, b1 , and the other channel gains are set to b2 = 2, c1 = 2, c2 = 0.3 and all nodes powers are 10 dB and = 1. The right part of the gure also shows the optimal bandwidth (M AC and BC ) and power ( ) allocation. We know from the rst part of Proposition 12 that if b1 is suciently larger than c1 , for xed bandwidth allocation, the DF rate (3.19) where the relay helps the S1 D1 pair only, is optimal. A similar conclusion is drawn here for b1 2 as it can be seen from the optimal power allocation . Moreover, the total bandwidth is balanced between the S1 R and R D1 channels. This result is akin to Proposition 12 for IC-OBR Type-I. The second part of Proposition 16 proves that, for xed bandwidth allocation, if c1 and b2 are suciently large, it is optimal to use DF by letting the relay

47 help both source-destination pairs (see Remark 11). Fig. 3.11 shows that a similar conclusion holds also when optimizing the bandwidth allocation. Specically, Fig. 3.11 compares the achievable sum-rate (3.20) (attained by the DF scheme just discussed) with the outer bound (3.25) for variable b1 , and b2 , c1 with values from the set (b2 , c1 ) {(3, 2), (10, 5), (20, 10)}. We also have c2 = 1, = 1, and other conditions as above. It is seen that for (b2 , c1 ) large enough, the outer bound and the achievable sum-rate match for b1 7. A natural question that arises is to understand the eect of interference forwarding for IC-OBR Type-II with variable bandwidth allocation, similar to its IC-OBR Type-I counterpart as discussed in Sec. 3.3.4. To observe this eect, we consider a Type-II channel with the parameters set to = 1, a21 = 1, a12 = 0.5, c1 = 4, c2 = 1.5, b1 = 1, P1 = P2 = P1R = P2R = PR =10dB and b2 is varied. Note that since a21
1+P1 1+a2 12 P1

= 1.78, D1 can be aided by the relays interference

forwarding, since the interference cannot be decoded and removed over the IC only at D1 . Fig. 3.12 essentially shows that this is indeed the situation especially for b2 2. In the gure, the achievable scheme follows from Proposition 19, (3.23a)-(3.23d), where S1 transmits only private information W1p via the IC whereas S2 transmits common information (W2c , W2c ) only. The relay facilitates interference forwarding by broadcasting W2c which is used at D1 to remove part of the interference. As shown in the gure, the increase in b2 gain helps the OBR forward more interference to D1 , and hence interference forwarding is crucial for larger b2 gains.

3.5

Chapter Summary
Operation over parallel radio interfaces is bound to become increasingly com-

mon in wireless networks due to the large number of multistandard terminals. This enables cooperation among terminals across dierent bandwidths and possibly standards. In this paper, we have studied one such scenario where two source-destination

48 pairs, interfering over a given bandwidth, cooperate with a relay over an orthogonal spectral resource (out-of-band relaying, OBR). We have focused on two dierent models that correspond to distinct modes of transmission over the out-of-band relay channel (OBRC). As discussed in previous work, relaying can assist interfering communications via standard signal relaying but also through interference forwarding, which eases interference mitigation. For both considered models, this paper has derived analytical conditions under which either signal relaying or interference forwarding are optimal. These conditions have also been related to the problem of assessing optimality of either separable or non-separable transmission over parallel interference channels. Overall, the analysis shows that, in general, joint signal relaying and interference forwarding, and thus non-separable transmission, is necessary to attain optimal performance. This clearly complicates the design. Moreover, this is shown to be the case for both xed and variable (i.e., optimized) bandwidth allocation over the OBRC. However, in some special scenarios of interest, a separable approach has been shown to be optimal. An example of such cases, for both considered models, is the case where the relay has better channel conditions from the sources than to the destination. Moreover, in the presence of optimized OBRC bandwidth allocation, separable schemes are often (but not always) optimal. The analysis in this paper leaves open a number of problems related to interference management via cooperation and through multiple radio interfaces. In particular, scenarios that extend the current model to more than two sources and one relay are of interest, and expected to oer new research challenges in light of the results of [75].

49

Z1

X1

X1 X1R
b1

1
Z1R

Y1 D1
+
ZR1 ZR2

a12 Y1R R Y2R a21 1


+

+ +

XR1 XR2

c1

YR1

b2

Z2R

c2
+

X2

X2R X2

YR2 D2 Y2

Z2 Z1

X1

X1 X1R
b1
ZR

1 a12
c1
+

+ +
ZR1 ZR2

Y1 D1 YR1

YR

R XR
c2

b2

X2

X2R X2

a21 1
+
Z2

YR2 D2 Y2

Figure 3.1: Interference Channel (IC) with an out-of-band relay (OBR). The OBR channel (OBRC) has channel uses for each channel use of IC. (i ) IC-OBR Type-I: The OBRC is divided into four Gaussian orthogonal channels with iR , Ri , i = 1, 2, channel uses each; (ii ) IC-OBR Type II: The OBRC is divided into two orthogonal channels with M AC , BC channel uses.

50

8 7

Rate (bits/ch use of IC)

Noisy 6 SumCapacity 5 4 3 2 1 0 0

Strong Capacity Regime

Achievable SumRate with =0 Achievable SumRate with =2, b2=b2=1.5, c2=c2=0.1


1 1 2 1 2

Achievable SumRate with =2, b2=c2=6.3, b2=c2=1.5


2 2 1

0.5

1.5

2.5

3.5

Figure 3.2: Achievable sum-rate from Proposition 6 for IC-OBR Type-I with symmetric IC, for xed OBRC bandwidth allocation versus a for various OBRC link capacities, namely i) = 0 (no relay), ii) 1R = 2R = R1 = R2 = 1,
2 2 2 b2 1 = b2 = 1.5, c1 = c2 = 0.1, (which satises the conditions in Proposition 7), 2 2 2 and iii) 1R = 2R = R1 = R2 = 1, b2 2 = c1 = 6.3, b1 = c2 = 1.5 (which satises the

conditions in Proposition 8). All powers are set to 10 dB.

51

3.25 3.2 3.15 Rate (bits/ch use of IC) 3.1 3.05 3 2.95 2.9 2.85 0 Achievable SumRate Outer Bound

0.5

1.5

2 b

2.5

3.5

Figure 3.3: Achievable sum-rate (from Proposition 6, (A.35) with signal relaying (only common information transmission over the IC), and outer bound (from Proposition 5 (A.36)) versus b2 for IC-OBR Type-I with variable OBRC bandwidth allocation ( = 1, a21 = a12 = 2, P1 = P2 = P1R = P2R = PR1 = PR2 =10dB, c2 = 3, c1 = b1 = 2).

52

3.05 3 Rate (bits/channel use of IC) 2.95 2.9 2.85 2.8 2.75 2.7 0

Achievable SumRate(Sig. Rel.&Int. Forw.) Outer Bound Achievable SumRate(Sig. Rel.)

0.5

1.5

2 b2

2.5

3.5

Figure 3.4: Achievable sum-rate (from Proposition 8, (A.23)-(A.26)) with signal relaying and interference forwarding (only common information transmission over the IC) and outer bound (from Proposition 5, (A.36)) versus b2 for IC-OBR Type-I with variable OBRC bandwidth allocation
2 i=1

iR + Ri = , = 1, a12 = 3, a21 = 2,

P1 = P2 = P1R = P2R = PR1 = PR2 =10dB, b1 = 1.5, c1 = 3, and c2 = 1.

53

R2

S1,S2 to R MAC

 

45o line

R to D1,D2 BC

R1  

Figure 3.5: Illustration of the OBRC conditions leading to the sum-capacity in Propo2 sition 16: c1 c2 , M AC C (b2 1 P1R ) BC C (c1 PR ).

R2

 
45o line

S1,S2 to R MAC

R to D1,D2 BC

R1  

Figure 3.6: Illustration of the OBRC conditions leading to the sum-capacity in Propo2 2 sition 16: c1 c2 , M AC C (b2 1 P1R ) < BC C (c1 PR ), M AC C (b2 P2R ) BC C c2 2 PR 1+c2 2 PR

where where is the optimal power allocation that maximizes the sum-rate in Proposition 16 ( = 1 ).

54

3.6 3.4

Rate (bits/ch use of IC)

3.2 3 2.8 2.6 2.4 2.2 2 1.8 0 2 4 6 8 10 Achievable SumRate (CF) Outer Bound

Figure 3.7: Achievable sum-rate ((3.22a)-(3.22c)) and outer bound ((3.18a)-(3.18d)) for a symmetric IC-OBR Type II channel with respect to relay-to-destination channel gains, c (a = 2, b = 1, P = Ps = 10).

55

4.4 4.2

Rate (bits/channel use of IC)

4 3.8 3.6 3.4 3.2 3 2.8 2.6 1 3 5 7 Ach. SumRate (Sig.Rel. & Int.Forw.), a21=0.1 Ach. SumRate (Sig.Rel.), a21=0.1 Ach. SumRate (Sig.Rel. & Int.Forw.), a21=0.9 Ach. SumRate (Sig.Rel.), a21=0.9 Ach. SumRate (Sig.Rel.), a21=1.8 Outer Bound 9 11 13 15 17 19 20

Figure 3.8: Achievable sum-rate and outer bound for an IC-OBR Type-II channel with respect to R D1 channel gain, c1 and S2 D1 channel gain a21 {0.1, 0.9.1.8} (a12 = 0.5, b1 = 1, b2 = 10, c2 = 1 and all node powers are equal to 10).

56

3.6

Rate (bits/channel use of IC)

3.4 3.2 3 2.8 2.6 2.4 2.2 2

0.5

1.5

2 c

2.5

3.5

3.6

Rate (bits/channel use of IC)

3.4 3.2 3 2.8 2.6 2.4 2.2 2 0 0.5 1 1.5 2 2.5 3 3.5 4 DecodeandForward (Prop. 16) CompressandForward (Prop. 18) Lattice Coding (Prop. 20) Outer Bnd (Prop. 14) Outer Bnd (Prop. 11)

Figure 3.9: Achievable sum-rates and outer bound for a symmetric IC-OBR Type-II with respect to the relay-to-destination gain c (b = 1, M AC = BC = 0.5, P1 = P2 = P1R = P2R = PR =10dB; Fig. i) a = 3, Fig. ii) a = 10).

57

1.6 Rate (bits/channel use of IC) 1.4 1.2 1 0.8 0.6 0.4 0.2 0 Achievable SumRate Outer Bound

4 b1

10

1 Parameters (Achievable SumRate)

0.8

BC

MAC

0.6

0.4

0.2

0 0

4 b1

10

Figure 3.10: Achievable sum-rate (from Proposition 16, (3.19)) with signal relaying (DF) and outer bound (3.25) and optimal parameters (M AC , BC , ) of the DF scheme (3.19) for an IC-OBR Type-II with respect to S1 R channel gain b1 (b2 = 2, c1 = 2, c2 = 0.3, = 1, all node powers are equal to 10 dB, a21 = 1.8, a12 = 0.5.)

58

Achievable SumRate: b =3, c =2


2 1

Rate (bits/channel use of IC)

2.2 2 1.8 1.6 1.4 1.2 1 0.8

Outer Bound: b =3, c =2


2 1

b2=20,c1=10
2 1

Achievable SumRate: b =10, c =5 Outer Bound: b2=10, c1=5 Achievable SumRate: b =20, c =10
2 1

b2=10,c1=5

Outer Bound: b =20, c =10


2 1

b2=3,c1=2
0 1 2 3 4 5 6 7 8

Figure 3.11: Achievable sum-rate (from Proposition 16, (3.19)) with signal relaying (DF) and outer bound (3.25) for an IC-OBR Type-II with respect to S1 R channel gain, b1 and (b2 , c1 ) {(3, 2), (10, 5), (20, 10)} (c2 = 1, M AC + BC = with = 1, all node powers are equal to 10 dB, a21 = 1.8, a12 = 0.5.)

59

Rate (bits/channel use of IC)

3.4 3.2 3 2.8 2.6 2.4 2.2 0

Achievable SumRate (Sig. Relay & Int. Forw.) Achievable SumRate (Sig. Relay) Outer Bound

4 b2

Figure 3.12: Achievable sum-rate (from Proposition 19, (3.23a)-(3.23d)) with signal relaying and interference forwarding (DF, only common information transmission by S2 and only private information transmission by S1 over the IC) and outer bound (3.25) versus b2 for IC-OBR Type-II with variable OBRC bandwidth allocation
2 i=1

iR + Ri = , ( = 1, a21 = 1, a12 = 0.5, P1 = P2 = P1R = P2R = PR =10dB,

c1 = 4, c2 = 1.5, b1 = 1).

60

Chapter 4 Interference Channel with an Out-of-band Reception/In-band Transmission Relay

4.1

Introduction
In heterogenous networks, cooperation among independent terminals that

belong to dierent standards/subsystems is a possible way to improve the system performance as discussed in the previous section. However, multistandard nodes are equipped with dierent standards that are in general operated on dierent bands. This feature gives multistandard nodes the exibility of choosing among various bands in helping other nodes to maximize the benet obtained through cooperation in the network. In the previous chapter, we considered a scenario with an out-of-band relay helping multiple source-destination pairs and coined the investigated system with the name IC-OBR. In this part, we are interested in a model where the relay (multistandard node) transmits in the band of multi-terminal nodes however receives in orthogonal bands. In conjunction with the practical example given for IC-OBR, one can consider such a case where the relay receives through Wi- frequencies from the cellular nodes, and transmits to the base-stations using the frequency bands of the cellular nodes. We model this scenario with an interference channel assisted by an an out-of-band reception/in-band transmission relay, IC-OIR in short.

61 Our aim, similar to the IC-OBR model, is to explore various relaying operations and their eect on the overall system performance. In particular, we are interested in determining relaying operations which can not be utilized by the out-of-band transmission at the relay but in-band transmission and explore their signicance on improving the transmission rates. To gain more insight on the model, we investigate optimal operations at the sources and the relay. The rest of the section is organized as follows. The system model is introduced in Section 4.2. In Section 4.4.1, we give a general achievable region IC-OIR and two outer bounds on the capacity region and sum-capacity, respectively. Section 4.5 is devoted in determining the very strong relay-interference conditions and the corresponding capacity region for IC-OIR. In Section 4.6, we investigate interference channel with a cognitive relay (IC-CR) under mixed relay-interference regime and determine its sum capacity. Section 4.7 investigates a degraded IC-CR and provides its capacity region. Numerical results illustrating the general achievable region for the IC-OIR and capacity results under very-strong and mixed relay-interference conditions are given in Section 4.8. We conclude this section in Section 4.9.

4.2

System Model
We consider a Gaussian interference channel (IC) with a relay shown in Fig.

4.1. Each source Si , i = 1, 2, wishes to send a message index Wi , uniformly drawn from the set [1, 2nRi ], to its destination Di , with the help of the relay R. The sources S1 and S2 communicate simultaneously with their respective destinations D1 and D2 via the Gaussian IC. The (real) input symbols of source Si are denoted as (Xi,t R, t = 1, ..., n ), on which we enforce the power constraint 1/n
n t=1 2 Pi for each Xi,t

codeword, and {Zi,t } is independent identically distributed (i.i.d.) Gaussian noise with unit variance. Also Z1 , Z2 are independent. In the model of interest, the relay receives the source signals over links that

62 are orthogonal to each other and to the underlying IC, hence it can be classied as an interference channel with an out-of-band relay reception/in-band relay transmission relay [66]. The orthogonal link from Si to the relay has capacity Ci bits/channel use, i = 1, 2. We implicitly assume that Si R channels are used n times before transmission over IC takes place. We dene Vi [1, 2nCi ], i = 1, 2, as the messages sent by the source Si over the nite-capacity link to the relay. When Ci , i = 1, 2, the model reduces to the cognitive scenario of [41] [59] [53], where the messages Wi , i = 1, 2, are known at the relay non-causally. This scenario will be called interference channel with a cognitive relay (IC-CR). We assume in-band relay transmission where the relay transmits its symbols XR,t , t = 1, ..., n to the destinations simultaneously with the sources. The relay power constraint is represented as 1/n
n i=1 2 XR,t PR . Note that XR,t is a function of

(V1 , V2 ). Hence, the received signals at the destinations at time t = 1, ..., n are given by, Y1,t = a11 X1,t + a21 X2,t + c1 XR,t + Z1,t Y2,t = a12 X1,t + a22 X2,t + c2 XR,t + Z2,t . where the channel coecient aij are real. A (2nR1 , 2nR2 , n) code for the interference channel with a nite capacity relay (IC-OIR) is dened by the encoding function at the source Si , i = 1, 2: fi : [1, 2nRi ] Rn [1, 2nCi ], (4.2) (4.1a) (4.1b)

which maps a message Wi [1, 2nRi ] into a codeword X n Rn for the IC and a message to the nite-capacity relay link Vi [1, 2nCi ]. The encoding function at the relay R is given by, fr : [1, 2nR1 ] [1, 2nR2 ] Rn , (4.3)

n which maps the messages V1 and V2 into a relay codeword XR . Finally, decoding

63 function at destination Di , i = 1, 2, is, gi : Rn [1, 2nRi ], (4.4)

i. which maps the received signal over the IC, Yin , into the estimated message W As a special case, we dene a degraded interference channel with a cognitive relay. Denition 1: An interference channel with a cognitive relay is said to be degraded if it satises the conditions, a12 = a11 a21 = a22 c1 = c2 . (4.5a) (4.5b) (4.5c)

Denition 2: The average probability of error for this code, is dened as


n n Pen = max{Pe, 1 , Pe,2 }, where n Pe,i

1 2n(R1 +R2 )
(W1 ,W2 )W1 W2

i = Wi Pr W

for i = 1, 2.

Denition 3: A rate pair is said to be achievable for the IC-OIR if there exists a sequence (2nR1 , 2nR2 ) codes with Pen 0 as n . Denition 4: The capacity region C for the IC-OIR is the closure of the set of all achievable rate pairs. Denition 5: In this chapter, Cl(S ) and Co(S ) dene the closure and convex hull operations over the non-empty set S , respectively. Also the notation C (x) = 1/2 log(x) is used throughout the chapter.

4.3

A General Achievable Region and Outer Bounds for IC-OIR


In this section, we obtain a general achievable rate region for IC-OIR and

outer bounds on the capacity region of IC-OIR and IC-CR.

64

Z1

X1

X1 X1R
C1

1
c1

Y1

D1

a12 R XR

C2

c2

X2

X2R X2

a21 1

+
Z2

Y2

D2

Figure 4.1: Gaussian interference channel with an out-of-band reception/in-band transmission relay (IC-OIR).

4.4

An Achievable Region
To obtain an achievable region for IC-OIR, we perform message splitting to

serve the following purposes: (i) to reduce the eect of interference at the destinations as in the Han-Kobayashi scheme [9], (ii) to exploit the relay via the nite-capacity links using partial decode and forward [25], and (iii) to use partial knowledge of the interference at the relay in transmitting additional information as in dirty-paper coding [12]. Therefore, each source message is split into common and private parts, where the common parts are decoded at both destinations and the private parts are decoded at the desired destination only. To incorporate the relay into the transmission, we further split the common and private parts as follows. The common messages are split into two parts such that rst part is transmitted directly to the destination and the second part is transmitted with the help of the relay. On the other hand, the private part is divided into three splits; the rst split is transmitted directly, the second split is transmitted with the help of the relay, and the third part is transmitted only by the relay to the destinations using partial knowledge of the interference. Note that the messages transmitted directly to the destinations

65 need not be decoded at the relay. The message splitting can be represented as Wi = (Wic , Wic , Wip , Wip , Wip ), i=1,2, where (i) Wic [1, ..., 2nRic ] is the common message transmitted by Si only, (ii) Wic [1, ..., 2nRic ] is the common message transmitted by Si and R jointly, (iii) Wip [1, ..., 2nRip ] is the private message transmitted by Si only, (iv ) Wip [1, ..., 2nRip ] is the private message transmitted by Si and R jointly, and (v ) Wip [1, ..., 2nRip ] is the private message transmitted by R only. In the encoding of Wip , with the assumption that the message splits Wic , Wip , i = 1, 2, are decoded error-free at the relay for the given rate constraints, the relay performs dirty-paper coding and hence transmits Wip to the destinations interference free from the splits Wic , Wip , i = 1, 2. Based on this strategy, we have the following achievable region for IC-OIR. Proposition 21 : An achievable rate region for a Gaussian interference channel assisted by an out-of-band reception/in-band transmission relay (IC-OIR) is given by
ach

= Cl Co R1 , R2 ) , where R1 = R1c + R1c + R1p + R1p + R1p , R2 = R2c + R2c +

R2p + R2p + R2p , with R1c + R1p + R1p C1 R2c + R2p + R2p C2 Rs C
sS1a S1b sS1a

(4.6a) (4.6b)
2

ai1 s Pi + c1 s PR Nt1 ai2 s Pi + c2 s PR Nt2

sS1b

2 a2 i1 s Pi

(4.6c)
2

Rs C
sS2a S2b

sS2a

sS2b

2 a2 i2 s Pi

(4.6d) R1p C R2p C


2 c2 1 1p PR 2 a2 21 2p P2 +

(4.6e) (4.6f)

2 c2 2 2p PR 2 2 2 a2 12 1p P1 + c2 1p PR + 1

and S1a {1c, 1p, 2c}, S1b {1c, 1p, 2c}, S2a {2c, 2p, 1c}, S2b {2c, 2p, 1c},

66 2 2 2 2 2 2 2 Nt 1 = c 2 1 (1p + 2p )PR + a21 2p P2 + (a21 2p P2 +c1 2p PR ) + 1, Nt2 = c2 (1p + 2 2 2 2 ) P + a P + ( a P + c PR )2 + 1. Moreover, the power allocations R 1 12 1 p 1 2 1 p p 12 1 p
2 2 2 2 satisfy ic + ic + ip + ip 1, i=1,2, sB 2 s 1 where B = {1c, 2c, 1p, 2p, 1p, 2p}.

The notation for channel and power allocations are ai1 = a11 , ai2 = a12 , Pi = P1 for s {1c, 1p, 1c, 1p} and ai1 = a21 , ai2 = a22 , Pi = P2 otherwise. Proof 18 Appendix A.14. Remark 15 The rate constraints in (4.6e) and (4.6f) are obtained by using the knowledge of (W2p , W2p ) at the relay in delivering additional information at reduced interference to D1 , while D2 views the power allocation to W1p as noise. We can obtain another achievable region by switching the roles of D1 and D2 . This region can be simply obtained by interchanging R1p and R2p , i.e. 1 2 and 2 1 in (4.6e) and (4.6f). Denoting this region as
ach , ach

ach

gives an extended achievable region.

Remark 16 The achievable region in Proposition 21 includes the achievable regions of the IC-CR system given in [41], where C1 , C2 and the sources transmit private messages only, and IC-CR [53] with C1 , C2 and the sources employ general rate-splitting. Remark 17 Note that the relay allocations 1c , 2c , 1d and 2d allow for the relay to negatively and positively correlate its signal with the sources. Positive correlation for the common splits, i.e. 1c , 2c 0, improves the received power of these splits at both destinations, hence can be attributed to signal-forwarding [55] [72]. On the other hand, negative correlation at the relay for the private splits decreases the received power of the private signals at the destinations, which may be benecial at the interfered destination. We denote this relay operation as interference cancelation. Interference cancelation was used to improve the achievable rates for the asymmetric interference channel with a relay in improving the achievable rates in [59].

67

4.4.1

Outer Bounds for the IC-OIR and IC-CR


In this section, we give two outer bounds on the IC-OIR. The rst bound is

valid for any C1 0 and C2 0, while the second bound is derived for C1 , C2 . These outer bounds are used to establish: (i) the capacity region of IC-OIR under very strong relay-interference conditions, in Sec. 4.5, (ii) the sum capacity of IC-CR under mixed relay-interference conditions, Sec. 4.6, and (iii) the capacity region of a degraded IC-OIR as given in Sec. 4.7. For a regular Gaussian interference channel, the capacity region is known for strong conditions such that, a12 a11 a21 a22 . In obtaining the capacity region under strong regime, [10] uses the decodability of the interference signals at the destinations to obtain the corresponding outer bounds. In an analogous manner, to extend the strong conditions to IC-OIR, we use the vector degradedness1 conditions dened in [33]. Denition 5 [33]: A receive vector ya = Ha x + za is said to be degraded with respect to yb = Hb x + zb if there exists a matrix D such that DHb = Ha and such that DDT I, where ya is ra n, yb is rb n, Ha is ra t, Hb is rb t and x
n Y1 ,

is t n input matrix and za is ra n noise matrix. Noting that for Gaussian IC-OIR, we have ra = rb = 1, t = 2, ya yb
n Y2 , Ha = [a22 c2 ] and Hb = [a21 c1 ] with D = 1 being a constant, we say

n T n that in a Gaussian IC-OIR, for the input signal x = [Xn 2 XR ] , Y2 is a degraded

version of Y1n with respect to (S2 , R) pair if, [a21 c1 ] = [a22 c2 ], 1.


1

(4.7)

The degradedness denition is used alternately in the IC literature. In [22] (and references therein), a Gaussian interference channel is said to be degraded for a12 .a21 = 1 only. However, our degradedness denition is dierent and used in ordering the destinations in terms of signal reception.

68 Similarly, Y1n is said to be a degraded version of Y2n with respect to (S2 , R) pair for an enhanced system if, [a12 c2 ] = [a11 c1 ], 1. (4.8)

In the following, we obtain outer bounds for the IC-OIR, where we employ the degradedness condition in (4.7) particularly in obtaining outer bounds on the sum capacity. Proposition 22 : For an interference channel with out-of-band reception/in-band transmission relay (IC-OIR) satisfying the conditions, [a21 c1 ] = [a22 c2 ], 1, the following gives an outer bound on the capacity region, ROIR out,a =
2 2 2 2 2 2 ic 1, ic 1, 1c 1c +2c 2c 1, i=1,2

(4.9)

(R1 , R2 ) : P1 PR (4.10a) (4.10b) P2 PR (4.10c) (4.10d) P1 PR (4.10e)

2 2 R1 C a2 11 P1 + c1 1c PR + 2a11 c1 1c 1c 2 R1 C1 + C a2 11 (1 1c )P1 2 2 R2 C a2 22 P2 + c2 2c PR + 2a22 c2 2c 2c 2 R2 C2 + C a2 22 (1 2c )P2

2 2 2 2 R1 + R2 C2 + C a2 11 P1 + a21 (1 2c )P2 + c1 1c PR + 2a11 c1 1c 1c

2 2 R1 + R2 C a2 11 P1 + a21 P2 + c1 PR + 2a11 c1 1c 1c

P1 PR (4.10f) . (4.10g)

+ 2a21 c1 2c 2c

P2 PR

2 2 2 R1 + R2 C1 + C2 + C a2 11 (1 1c )P1 + a21 (1 2c )P2

Proof 19 The proof uses the notion of degradedness in Denition 5 adopted to the IC-OIR and given in Appendix A.15.

69 Remark 18 Using symmetry, it is possible to show that another outer bound on the capacity region of IC-OIR can be obtained from Proposition 22 by simply switching the indices in (4.12)-(4.10g) as 1 2 and 2 1. Denote this region as ROIR out,b . Then for the conditions [a12 c2 ] = [a11 c1 ], 1, (4.11)

ROIR out,b is an outer bound on the capacity region of IC-OIR where the union is taken
2 2 2 2 2 2 over all parameters ic 1, ic 1, 1 c 1c + 2c 2c 1, i = 1, 2.

An analogous model to IC-OIR is the one investigated under interference channel with a cognitive relay (IC-CR) [41] [59] [55] [53]. In this case, the relay is assumed to have the sources messages W1 , W2 non-causally, or similarly C1 , C2 . Note that since the relay already knows the sources messages, it is not necessary for the sources to transmit information to the destinations directly as in IC-OIR model, however, all messages are transmitted with the help of the relay. The following corollary will be used in proving the sum capacity of IC-CR under mixed relayinterference conditions as given in Proposition 5. Corollary 2 For an interference channel with out-of-band reception/in-band transmission relay (IC-OIR) and an interference channel with a cognitive relay (IC-CR) satisfying the conditions, [a21 c1 ] = [a22 c2 ], 1, (4.12)

70 the following gives an outer bound on the capacity region, RCR out,a =
2 + 2 1 1 c 2c

(R1 , R2 ) : a11 a22 a11 P1 + c1 1c P2 + c2 2c P1 + c1 1c PR PR PR


2

R1 C R2 C R1 + R2 C

(4.13a) (4.13b) + a21 P2 + c1 2c PR


2

. (4.13c)

Proof 20 Follows immediately from Proposition 22, by noticing that for C1 , C2 , the region (4.10a)-(4.10g) is maximized for ic = 1, i = 1, 2. Remark 19 As discussed in the previous remark, using symmetry, we can obtain another outer bound from Corollary 2 by switching the indices which is valid for the conditions [a12 c2 ] = [a11 c1 ], 1, and the outer bound is denoted by RCR out,b . Now, we give another outer bound on the sum-capacity of IC-OIR which is valid for C1 , C2 , or, namely for IC-CR. The rationale of this outer bound is somewhat aligned with the bound in Proposition 22, however, the desired destination is assumed to be degraded w.r.t. interfered destination instead. Similarly, for the following degradedness conditions, [a22 c2 ] = [a21 c1 ], 1. (4.15) (4.14)

Y1n is said to be degraded with respect to Y2n , from the perspective of S2 , R pair.

71 Proposition 23 For an interference channel with out-of-band reception/in-band transmission relay (IC-OIR) and an interference channel with a cognitive relay (IC-CR) satisfying the conditions, [a22 c2 ] = [a21 c1 ], 1, the following gives an outer bound on the sum capacity, 2 a11 P1 + c1 1c PR R1 + R2 2 max C 2 1 1c +2 1 + a21 P2 + c1 2c PR c + C Proof 21 Appendix A.16. a22 P2 + c2 2c PR
2

(4.16)

(4.17a)

4.5

Very Strong Capacity Region of IC-OIR


In [5], Carleial shows that in an interference channel when the interfering

links are suciently strong, the interference channel capacity is equivalent to a pair of point-to-point channels. Under the so-called very strong interference conditions, the interference can be decoded rst and removed without degrading the point-to-point achievable rates. On the other hand, in an IC-OIR, when the interfering links are suciently strong such that the destinations are not aected by the interference, the system becomes equivalent to two parallel source destination pairs sharing a common relay with orthogonal links, as in Figure 4.2. Note that each parallel relay channel is similar to the model investigated in [25] whose capacity is characterized. Therefore, it is possible to characterize the sum-capacity of this system which further depends on the relay power split among the sources. Proposition 24 The sum capacity of two parallel source-destination pairs sharing a

prl 2 2 2 =max2 common relay with orthogonal links as shown in Figure 4.2 is given by Csum 1c 1,2c 1,1c +2c

72

Z1

W1

S1

X1 R R

a11
Power: 1c2PR

Y1

D1

a R1

Power: 2c2PR

aR2
+
Z2

W2

S2

X2

a22

Y2

D2

Figure 4.2: Parallel source-destination pairs with a common orthogonal relay. The
2 2 total relay power satisfy 1 c + 2c 1. This is an equivalent system for an IC-OIR

under very strong interference. (R1 + R2 ), where


2 2 R1 C a2 11 P1 + c1 1c PR + 2a11 c1 1c 1c 2 R1 C1 + C a2 11 1c P1 2 2 R2 C a2 22 P2 + c2 2c PR + 2a22 c2 2c 2c 2 R2 C2 + C a2 22 2c P2 2 2 2 2 with 1 c = 1 1c , and 2c = 1 2c . 2 Proof 22 Follows directly from [25], with the relay allocating ic PR power for each

P1 PR

(4.18a) (4.18b)

P2 PR

(4.18c) (4.18d)

(Si , Di ), i = 1, 2 source-destination pair.


prl Now, we denote by 1 c , 2c , 1c , 2c the optimal allocations in Csum . The

following denition determines the channel conditions in an IC-OIR which guarantee the decodability of the interference at the destinations without reducing the point to point rates, hence transforming an IC-OIR to the system with parallel sourcedestination pairs with a common relay as in Figure 4.2. We denote these conditions as very strong relay-interference conditions. The proposition, on the other hand, gives the sum capacity of IC-OIR under very strong relay-interference conditions.

73 Denition 6: An IC-OIR is in very strong relay-interference regime if, a12 a12 a21 a21 1 (a11 1 c P1 a11 D2 1 (a22 2 c 2c P2 a22 D1
1 c P1 + c1 1 c

PR ) D2 c2 1 c PR ) D1 c1 2 c

PR

(4.19a) (4.19b)

P2 + c2 2 c

PR

(4.19c) (4.19d)

2 where Di = 1 + a2 ii ic Pi + (aii ic Pi + ci ic PR ) , i = 1, 2. Proposition 25 In an IC-OIR satisfying the very strong relay-interference condiOIR prl . tions, the sum capacity is given by Cvsi = Csum

Proof 23 Appendix A.17.

Remark 20 It is possible to obtain the capacity region of IC-OIR under very strong relay-interference conditions following similar steps used in obtaining the sum capacity. Note that it is possible to characterize the whole capacity region of the system shown in Figure 4.2 by obtaining the optimal source and relay power allocations that maximize R1 + R2 , [0, 1]. Similarly, one can obtain the corresponding channel conditions for IC-OIR evaluated at these optimal allocations, which transform the system into the system in Figure, hence giving the capacity region.
OIR Remark 21 For C1 = C2 = 0, the capacity region of Cvsi given in (4.18a)-(4.18d)

is clearly maximized for ic = ic = 0, i = 1, 2, and the region becomes the capacity region of a Gaussian interference channel under very strong interference regime. Also, for ic = ic = 0, the conditions (4.19a)-(4.19d) give us the very strong interference conditions of a Gaussian interference channel [5].

74

4.6

Sum-Capacity of IC-CR under Mixed RelayInterference Conditions


Proposition 23, (4.17a), gives an upper bound to the sum-rate of the IC-

CR where the interference is treated as noise at one of the destinations (D1 , for the conditions of Proposition 23). Proposition 22, on the other hand, provides an outer bound which is obtained by considering decodability of the interference (again at D1 for the conditions of Proposition 22). Combining the conditions of the two propositions, namely conditions in Proposition 23 and conditions in Proposition 22 adopted for D2 , leads us to the channel conditions which can be called as mixed relay-interference regime. Denition 7: An interference channel with a cognitive relay is in mixed relay-interference regime if, [a12 c2 ] = [a11 c1 ] [a22 c2 ] = [a21 c1 ], 1. (4.20a) (4.20b)

The following proposition gives us the sum capacity of IC-CR under mixed relay interference conditions. Proposition 26 The sum-capacity of an interference channel with a cognitive relay (IC-CR) operating under mixed interference channel conditions is given by, R1 + R2 + min C C max 2 2 a22 P2 + c2 2c ,C PR
2

2 a11 P1 + c1 1c PR 1 + a21 P2 + c1 2c PR

1c +2c 1

2 a12 P1 + c2 1c PR 1 + a22 P2 + c2 2c PR

(4.21) Proof 24 Achievability: Achievability follows directly from the general achievable region in Proposition 21. Note that since the relay is cognitive, C1 , C2 , all source signals are

75 transmitted with the help of the relay, i.e. s = 0 where s {1c, 1p, 2c, 2p} in (4.6a)(4.6f). Moreover, in this particular achievable scheme, S1 transmits the common message, W1c , only, whereas S1 transmits the private message, W1p , only, so that we have 1c = 1 and 2p = 1. Also, no dirty-paper coding is employed at the relay, and 1p = 2p = 0. Again, for notational convenience, we assume 1c Converse: The rst outer bound follows from Corollary 2 and Remark 19, which is denoted by RCR out,b and valid for (4.20a). Another outer bound which is valid for (4.20b) follows from Proposition 23 and given in (4.17a). Moreover, since RCR out,b can be written equivalently as, R1 + R2 C a22 P2 + c2 2c PR
2

1c and 2c

2p .

Then, from (4.6a)-(4.6f), with R1 + R2 = R1c + R2p , we obtain the sum-rate (4.21).

2 a12 P1 + c2 1c PR +C 1 + a22 P2 + c2 2c PR

(4.22)
2 2 with 1 c + 2c 1, the minimum of (4.22) and (4.17a) give us the rate in (4.21),

hence the sum-capacity. 2

4.7

Capacity Region of a Degraded IC-CR


In this section, we establish the capacity region of a degraded IC-CR, where

a degraded IC-CR is dened Sec. 4.2. In a regular Gaussian interference channel, the decodability of the interfering signal at the destinations leads to the determination of the capacity region of the channel [10]. When a cognitive relay is incorporated into the system, the conditions in (4.5a)-(4.5c) similarly give necessary conditions for the decodability of the interfering signals at the destinations. For a degraded IC-CR, the following proposition gives the capacity region.

76 Proposition 27 : The capacity region of a degraded IC-CR is given by, C CR =


2 + 2 1 1 c 2c

(R1 , R2 ) : P1 PR P2 PR P1 PR (4.23c) P1 PR (4.23d) (4.23a) (4.23b)

2 2 R1 C a2 11 P1 + c1 1c PR + 2a11 c1 1c 2 2 R2 C a2 22 P2 + c2 2c PR + 2a22 c2 2c

2 2 R1 + R2 C a 2 11 P1 + a21 P2 + c1 PR + 2a11 c1 1c

+ 2a21 c1 2c

P2 PR

2 2 R1 + R2 C a 2 12 P1 + a22 P2 + c2 PR + 2a12 c2 1c

+ 2a22 c2 2c

P2 PR

Proof 25 Achievability: The achievability follows directly from the general achievable region in Proposition 21, with the parameters set to ic 1, ip = ip = 0, i = 1, 2
2 2 and 1 c + 2c 1. This particular achievable scheme corresponds to the sources trans-

mitting common messages only, i.e. Wic , i = 1, 2, followed by the relay beamforming with the messages. From Proposition 21, we obtain the rate region in (4.23a)-(4.23d).
CR Converse: The converse uses the outer bound given in Corollary 2, Rout,a and CR the bound Rout,b as discussed in the Remark 19 of Corollary 2. Taking the minimum

of these two outer bounds we obtain the expression in (4.23a)-(4.23d). 2

4.8

Illustration of the Results


In this section, we illustrate the results in Section 4.3-4.5 through some nu-

merical examples. We rst illustrate the capacity region under strong relay-interference conditions.

77

2 1.8 1.6 C1R,C2R=0.2 C1R,C2R 2

R2 (bits/ch use)

1.4 1.2 1 0.8 0.6 0.4 0.2 0 0 0.2 0.4 0.6 0.8 1 C1R,C2R=0

1.2

1.4

1.6

1.8

R1 (bits/ch use)

Figure 4.3: Achievable region of a symmetric IC-OIR with dierent C1 , C2 values. The system parameters are P1 = P2 = PR = 10, a11 = a22 = 1, c1 = c2 = 1, a21 = a12 = 2. The C1 , C2 curve also corresponds to the IC-CR model. In Figure 4.3, we consider a symmetric IC-OIR with a11 = a22 = 1, c1 = c2 = 1, a12 = a21 = 2, and P1 = P2 = PR = 10. The gure shows the achievable region of this IC-OIR for source-to-relay capacities of C1 = C2 = 0, C1 = C2 = 0.2 and C1 = C2 2, respectively. The achievable region follows from Proposition 21, (4.6a)(4.6f), however with the sources transmitting common messages only, i.e. (Wic , Wic ) and ip = ip = 0, i = 1, 2. For C1 = C2 = 0, we obtain a Gaussian interference channel operating under strong interference, whose capacity region is given by the corresponding compound multiple-access channel. As the source-to-relay link capacities are increased to C1 = C2 = 0.2, the relay starts helping both source-destination pairs with partial decode-and-forward transmission, and the capacity region is increased compared with C1 = C2 = 0. If we

78

2.5

Rate (bits/ch use)

Private only Common only Private+dpc Rate splitting+dpc

1.5

0.5 0 0.5

1.5

Figure 4.4: Achievable sum-rate of a symmetric IC-CR for dierent transmission techniques. The system parameters are P1 = P2 = PR = 10, a11 = a22 = 1, c1 = c2 = 1, = a21 /a22 = a12 /a11 . further increase C1 , C2 , we observe that for C1 = C2 2, the capacity region remains the same. Hence, for C1 = C2 2, IC-OIR is equivalent to IC-CR. In fact, from the achievable region (4.6a)-(4.6f), we see that for large enough C1 and C2 values, the rate expressions with no C1 and C2 become bottleneck, which are clearly maximized for ic = ip = 0, i = 1, 2. Physically, this corresponds to the sources transmitting no message splits directly, i.e. (Wic , Wip ), i = 1, 2, instead, all message splits are conveyed via the relay. In Figure 4.4, we let C1 , C2 and therefore investigate an IC-CR channel. In the gure, various transmission techniques used in characterizing the achievable region in Proposition 21 are demonstrated in terms of their maximum sum-rate for a symmetric IC-CR. The system parameters are P1 = P2 = PR = 10,

79

9 8.5

Rate (bits/channel use)

8 7.5 7 6.5 6 5.5 5 Private only Rate splitting Rate splitting+dpc

10

15

20

25

30

Figure 4.5: Achievable sum-rate of a one-sided IC-CR for dierent transmission techniques. The system parameters are P1 = P2 = PR = 10, a11 = a22 = 1, c1 = 1, a12 = c2 = 0, = a21 /a22 . a11 = a22 = 1, c1 = c2 = 1, and also = a21 /a22 = a12 /a11 which essentially determine the relative strength of the interference in the system. In the gure, private only corresponds to ic = ic = ip = 0, ic = ip = 0, i = 1, 2, common only corresponds to ip = ic = ip = 0, ip = ip = 0, i = 1, 2, private+dpc corresponds to ic = ic = ip = 0, ic = 0, i = 1, 2, and rate splitting along with dirty-paper coding corresponds to fully optimizing the achievable region in Proposition 21. In order to understand the role of relaying in an interference limited system more clearly, we also consider a one-sided interference channel, with a cognitive relay helping the interfered destination only. We have, P1 = P2 = PR = 10, a11 = a22 = 1, c1 = 1, a12 = c2 = 0, hence D1 is corrupted by the interference from S2 , and the relay

80

1 0.8 0.6

Power allocations

0.4 0.2 0 0.2 0.4 0.6 0.8 1 0 5 10 15 20 1p 1phat 2c 2p

25

30

a /a

Figure 4.6: Optimal relay power allocations that maximize achievable sum-rate of the one-sided IC-CR in Fig. 4.5. The system parameters are P1 = P2 = PR = 10, a11 = a22 = 1, c1 = 1, a12 = c2 = 0, = a21 /a22 . has link to D1 only. Therefore, the only aim of the relay is to help S1 D1 pair. We set P1 = P2 = PR = 10. Similar to Fig. 4.4, Fig. 4.5 investigates maximum sumrate of this network for various transmission techniques. For this gure we include the power allocations such as ic = ic = ip = 0, ic = ip = 0, i = 1, 2, that is denoted by private only, ic = ip = 0, ip = 0, i = 1, 2, denoted by rate splitting, and including all power constraints, denoted by rate splitting+dpc. We observe that rate-splitting at S2 (since S1 does not have link to D2 , we do not need rate splitting at this node) has substantial impact in improving the sum-rate. Note that, also, the eect of dirty-paper transmission at the relay is seen to improve the sum-rate for moderate interference link ratios particularly. In comparison with the one sided IC

81 without the relay, where private information transmission is optimal for < 1 and transmitting common information is optimal for 1, we observe that presence of the relay necessities rate splitting for a large range of interference levels. Figure 4.6 shows the optimal relay power allocations corresponding to the best achievable scheme in Fig. 4.5. In the gure, 1p corresponds to the relay power allocation for S1 signal, 1p is the relay power allocation used in transmitting dirtypaper coded signal to D1 , 2c is the relay power allocated in beamforming with the common information of S2 , and 2p is the relay power allocation for beamforming with the private information of S2 . We observe that for weak to moderate interference ratios, i.e. for 0 14, the relays functioning can be categorized as signal forwarding and interference cancelation. Signal forwarding is mostly performed by transmitting fresh information of S1 , which is not transmitted by S1 to its destination directly using dirty-paper coding. Interference cancelation, on the other hand, is observed to be as crucial as signal forwarding in these regimes, and is performed by negative beamforming with S2 s private message split at the relay. Note that in these regimes, the relay only uses a small amount of power to forward S1 a signal transmitted over the IC. As the interference ratio increases, the relay switches its functioning to interference forwarding , as well as signal forwarding. Interference forwarding takes places by relaying the common information of S2 , whereas signal forwarding is done by improving the strength of S1 s signal. Fig. 4.7 considers a symmetric IC-CR with the system parameters P1 = P2 = PR = 10, a11 = a22 = 1, and c1 = c2 = 0.2. Achievable sum-rate is obtained from Proposition 21, (4.6a)-(4.6f), which is specialized for cognitive relay, i.e. C1 , C2 , which leads to the parameters set to ic = ip = 0, i = 1, 2. The outer bound is also shown which is given in (4.18a-4.18d). For the given system parameters, the conditions (4.19a)-(4.19d) are given by, a12 = a21 = 4.5252. The gure also shows that for a12 = a21 5, the outer bound matches with the achievable sum-rate. An interesting property observed for very strong interference regime is also observed from

82

3 2.8 Achievable SumRate Outer Bound

Rate (bits/ch use)

2.6 2.4 2.2 2 1.8 1.6 0

Figure 4.7: Achievable sum-rate (4.6a)-(4.6f) and outer bound (4.18a-4.18d) of a symmetric IC-CR with respect to = a21 /a22 = a12 /a11 . The system parameters are P1 = P2 = PR = 10, a11 = a22 = 1, and c1 = c2 = 0.2 Fig. 4.7. Note that, for no interference case, i.e. a12 = a21 = 0, the achievable sumrate is given by R1 + R2 = 2.5 bits/ch use, whereas for a12 = a21 5, the achievable sum-rate and the capacity is R1 + R2 = 2.75 bits/ch use. Interestingly, interference helps the IC-CR to improve its sum-rate around 0.25 bits/ch use. The reason behind this reason is the fact that, for no interference case, the destinations still suer by the interference occurred by the relays broadcast signal, however for very-strong interference, each destination is able to decode the interference fully, hence observes interference-free signals from the sources and the relay. Finally, we study the mixed relay-interference regime. Figure 4.8 shows two scenarios; S1 transmits common message only, while S2 transmits private, and S1 transmits private message only, S2 common message only. According to Proposition

83

3.5 3 2.5 2 1.5 S1 common msg. only, S2 priv. msg. only 1 0.5 0 S1 priv. msg. only, S common msg. only 2 Sum Capacity

Rate (bits/ch use)

0.5

1.5

Figure 4.8: Achievable sum-rate and sum capacity of an IC-CR operating under mixed relay-interference regime. The system parameters are P1 = P2 = PR = 10, a11 = 1, c1 = 1, a21 = 1, = a22 /a21 = a12 /a11 = c2 /c1 . 26, the former is optimal (in terms of sum rate) when > 1, while the latter is optimal for < 1. These results are also conrmed by Figure 4.8.

4.9

Chapter Summary
Interference is a main paradigm in communication systems and often un-

avoidable. Understanding the role and benet due to cooperation/relaying in interferencelimited systems is therefore essential. In this chapter, we have investigated an interference channel with a relay where the relay is connected to the sources via nite capacity links. Similar to our previous work, we characterized relevant relaying functions for the model such as interference forwarding, interference cancelation and signal relaying by considering the nite-capacity links from the sources to the relay. Basi-

84 cally, in-band transmission of the relay and the sources lead to additional interference cancelation opportunity, which is not possible for the relay operating out-of-band to the sources. Also, interference forwarding is shown to be fundamental as the interference at the destinations become larger, and the optimality is shown for very-strong relay-interference conditions. Finite capacity source-relay links entail additional message-splitting at the sources to maximize the achievable region. However, we showed that for large enough nite-capacities, the system becomes equivalent to a cognitive relay scenario such that the relay has a-priori messages of the sources. This substantially decreases the system complexity by reducing the message splitting. This model is open to various optimality conclusions such as strong or noisy capacity regimes. Also, it may be interesting to understand the situations where the relaying operations discussed lead to the optimal performances.

85

Chapter 5 Interference Channel with an In-band Reception/In-band Transmission Relay

5.1

Introduction
In this chapter, we investigate another model for relaying in interference-

limited systems with the emphasis on single band allocation for the whole network. As opposed to the previously analyzed models IC-OBR and IC-OIR, in this case, the relay receives and transmits in the communication band of the multiple sourcedestination pairs. The in-band reception/tranmission relaying operation can be example of cooperation in an ad-hoc network, where the destinations are prone to the interference as a result of simultaneous transmission at the transmitter nodes. To uncover the benets of relaying, we assume a Gaussian channel with a full-duplex relay that employs decode-forward (DF) type strategy. This chapter is organized as follows. In Section 5.2, the system model is given. We present an achievable rate region for the Gaussian interference relay channel using rate splitting in Section 5.3. In Section 5.4, the orthogonal transmission is considered. Section 5.5 investigates sum-rate maximization. Numerical results and discussions are presented in Section 5.6 and we conclude the chapter in Section 5.7.

86

5.2

System Model
We consider a Gaussian interference relay channel with two source-destination

pairs as shown in Fig. 5.1. Terminal Si , i = 1, 2 wishes to communicate with Di , i = 1, 2 simultaneously utilizing the relay. The inputs to the channel at time n are denoted as Xi,n , i = 1, 2 and XR,n , and the outputs as YR,n and Yi,n , i = 1, 2. Here subscript R denotes the relay. We assume that the relay is full-duplex and employs decode-and-forward type strategy [17]. The relation between the channel inputs and outputs can be written as follows, YR,n Y1,n Y2,n b1 X1,n + b2 X2,n + nR,n = X1,n + a21 X2,n + c1 XR,n + nD1,n = a12 X1,n + X2,n + c2 XR,n + nD2,n =

where a12 , a21 are the channel gain between the nodes S1 , D2 and S2 , D1 , respectively, bi is the channel gain between Si and R and ci is the channel gain between R and Di , i = 1, 2. Also, nm,n represents zero mean, unit variance Gaussian noise with N (0, 1) which is independent for dierent m and n. The transmission is performed over N channel uses and each transmitting node has a power constraint Pi , i = 1, 2, R. In the sequel we use the standard denitions of achievable rates and probability of errors [17]. The bold letters dene vectors of length N , i.e. V = [V1 , . . . , Vn , . . . , VN ] log2 (1 + x). and C (x) = 1 2

5.3

Rate Splitting at the Sources


Our achievable region for the interference relay channel follows the rate

splitting technique of Carleial at the sources [5] and incorporates the relay. In [5], it was shown that dividing the message index set as common and private indices and transmitting the superposition of the subcodewords formed by these splits increases

87

Z1

X1
b1
ZR

1 a12 YR a21 1 R XR
c2
c1

Y1

D1

b2

X2

+
Z2

Y2

D2

Figure 5.1: Gaussian Interference Relay Channel with two source-destination pairs

the achievable region of the interference channel for some conditions. Similarly, for the interference relay channel, we allow the sources to form common and private index sets as wic {1, . . . , 2N Ric }, wip {1, . . . , 2N Rip }, i = 1, 2 such that wic is decoded at both destinations whereas wip is decoded only at Di . The relay on the other hand decodes both common and private message indices from the sources and transmits them after properly allocating its power among the common and private messages of S1 and S2 . Under this strategy, the following propositions give an achievable region of Gaussian interference relay channel. Proposition 28 For the Gaussian interference relay channel, the following conditions describe an achievable rate region, Ri
iS c I (XS ; YR |U1c , U1p , U2c , U2p , XS , XR ) c c I (XS1 , US1 ; YD1 |XS , US ) 1 1 c c I (XS2 , US2 ; YD2 |XS , US ) 2 2

(5.1a) (5.1b) (5.1c)

Ri
iS1

Ri
iS2

with Xt =

t Pk /PR Ut , t [1c, 2c, 1p, 2p], Pk = P1 if t [1c, 1p], Pk = P2 if t [2c, 2p] and XR = 1c U1c + 2c U2c + 1p U1p + 2p U2p . The random variables

t + t X

88 have the distributions, X1c , X1p N (0, P1 ), X2c , X2p N (0, P2 ), U1c , U1p , U2c , U2p N (0, PR ). The summation indices are, S {1c, 1p, 2c, 2p}, S1 {1c, 2c, 1p}, S2
c {1c, 2c, 2p}, XS denotes {Xj , j S } and XS denotes the inputs for the complement of

S , that is {Xj , j S c }. The power allocation variables satisfy 1c + 1c + 1p + 1p = 1, 2c + 2c + 2p + 2p = 1 and 1c + 1p + 2c + 2p = 1. Proof 26 Appendix A.18. Proposition 29 For the symmetric interference relay channel where P1 = P2 , b1 = b2 , c1 = c2 and a12 = a21 , the following describe the set of achievable rates R10 R1c R1c R1c 1 C (2b1 1c P1 ) 2 S1 R 1c,tot P1 + c1 1c PR + 1 c C S2 R 1 + a21 1p,tot P1 + c1 1p PR + 1 p C 1 C 2
S2 R a21 1c,tot P1 + c1 1c PR + 1 c S2 R 1 + a21 1p,tot P1 + c1 1p PR + 1 p

(5.2a) (5.2b) (5.2c)

(1 + a21 )1c,tot P1 S2 R 1 + a21 1p,tot P1 + c1 1p PR + 1 p (5.2d) (5.2e) (5.2f) (5.2g)

R1p R1p R1c + R1p

S1 R S2 R 2c1 1c PR + 1 c + 1c + S2 R 1 + a21 1p,tot P1 + c1 1p PR + 1 p 1 C (21p b1 P1 ) 2 S1 R 1p,tot P1 + c1 1p PR + 1 p C S2 R 1 + a21 1p,tot P1 + c1 1p PR + 1 p 1 C (2b1 1c P1 + 2b1 1p P1 ) 2

89
S1 R S1 R P1 + 1 c P + 1 c + 1p 2 1 R S2 R 1 + a21 1p,tot P1 + c1 1p PR + 1 p

R1c + R1p R1c + R1p

C C +

(5.3a)

a21 1c,tot P1 + 1p,tot P1 S2 R 1 + a21 1p,tot P1 + c1 1p PR + 1 p


1 c P 2 1 1c R S2 R S1 R + 1 c + 1p

S2 R 1 + a21 1p,tot P1 + c1 1p PR + 1 p

(5.3b) (5.3c) (5.3d)

R1c + 2R1p 2R1c + R1p 2R1c + R1p

C (b1 1c P1 + 2b1 1p P1 ) C (2b1 1c P1 + b1 1p P1 ) C + (1 + a21 )1c,tot P1 + 1p,tot P1 S2 R 1 + a21 1p,tot P1 + c1 1p PR + 1 p


S2 R 1 + a21 1p,tot P1 + c1 1p PR + 1 p

S1 R S2 R S1 R 2c1 1c PR + c1 1p PR + 1 c + 1c + 1p

(5.3e)

iR where 1c,tot = 1c + 1c , 1p,tot = 1p + 1p , iR 1c = 2 1c 1c ai1 c1 Pi PR and 1p = 2 1p 1p ai1 c1 Pi PR for i = 1, 2. Proof 27 Due to symmetry, the power allocation variables of the sources should be 1c = 2c , 1p = 2p , 1c = 2c , 1p = 2p . Also, the relay helps S1 and S2 equally, thus 1c = 2c and 1p = 2p . One can show that the rate bounds in Proposition 1 can be written as (5.2a)-(5.3e).

5.4

Orthogonal Transmission
For the interference channel even though rate splitting increases the achiev-

able rate region for most channel conditions, orthogonal transmission at the sources outperforms rate splitting in terms of sum rate for moderate interference [5]. Motivated by this, we analyze orthogonal transmission in terms of time division for the interference relay channel as well. We consider the case where in the rst time slot, S1 and R cooperatively transmit to D1 and in the second time slot, S2 and R cooperatively transmit to D2 .

90 Proposition 30 In a time division scheme where the sources S1 and S2 transmit orthogonally and R helps each source in its respective orthogonal channel, the following describe the set of achievable rates for the interference relay channel; b1 1 P1 P1 + c1 1 PR + 1 , C P2 + c2 2 PR + 2 b2 2 P2 , C R2 (1 ) min C 1 1 where 1 = 2 1 1 c1 P1 PR , 2 = 2 2 2 c2 P2 PR , (0, 1) is the time duration R1 min C allocated for the (S1 , R) transmission to D1 , i , i = 1, 2 is the power allocation at the source Si for the transmission of new information and i is the power allocation at the relay for Si . Note that in this case, the power allocations are i + i = 1, i = 1, 2 and 1 + 2 = 1. Proof 28 At each time slot, we have the point-to-point relay channel. Using [17], it is easy to show that each rate expression inside the minimum term of (5.4) and (5.4) is achievable. Thus the rates obtained by time sharing between the orthogonal relay channels is achievable.

5.5

Sum-Rate Maximization
The achievable sum-rate bounds of the symmetric interference relay channel

can be obtained from Proposition 29. The maximum sum-rate of the symmetric interference relay channel is given by,
Rsum = 1c ,1c ,1p ,1p ,1c ,1p

max

2(R1c + R1p )

(5.4)

where R1c , R1p are given in Proposition 29 with 1c + 1c + 1p + 1p = 1 and 1c + 1p = 1/2. On the other hand, for the orthogonal transmission scheme given in Sec. 5.4, the optimal sum-rate will be,
Rsum = ,1 ,2 ,1 ,2

max

(R1 + R2 )

(5.5)

91 where R1 and R2 are the rates given in Proposition 30. Note that obtaining the optimal power allocations in (5.4) requires evaluation of the multiple sum-rate bounds given in Appendix. Obtaining optimal power allocations seems formidable, the next section provides numerical results.

5.6

Discussion of the Results


For the performance comparison, we consider a symmetric network model

such that P1 = P2 , b1 = b2 , c1 = c2 and a12 = a21 . In Figs 5.2 and 5.3 we study the eect of relay locations on total maximum sum rate and power allocation. A path-loss model is considered such that a ,m = d ,m where d ,m is the distance between nodes {S1 , S2 , R} and m {R, D1 , D2 }, and = 2 is the path-loss exponent. We assume that dS1 D1 = dS2 D2 = 1 and dS1 D2 = dS2 D1 = 8. The relay is equidistant from the sources and destinations, and d denotes the distance between the sources and the relay. The node powers are equal, P1 = P2 = PR = 10. Note that without the relay, the interfering channel gain a21 = 1/8 would correspond to the weak interference region and rate-splitting with only private messages outperforms orthogonal transmission [5]. When the relay is added to the system, rate splitting still outperforms time division up to d = 0.6. For d > 0.6, the transmission from the sources to the relay becomes the bottleneck. Thus, the optimal achievable sum-rate is limited by the source-relay multiple-access channel which has the same maximum sum-rate as time division [17]. Hence for d > 0.6, rate splitting is equivalent to time division. The gure also shows the best known achievable rate without the relay, however for a fair comparison the relay power is split among the sources, thus P1 = P2 = 15. For d > 0.7, the advantage of the relay vanishes which is due to the decode-and-forward nature of the relay. Fig 5.3 shows the optimal power allocations for the achievable sum-rate as a function of the relay distance, d. Contrary to the case without relay, the sources

92 allocate power to common information in addition to the private information, i.e. 1c > 0 as well as 1p > 0. Moreover the relay never helps the sources in transmission of private messages and allocates all of its power for the transmission of sources common information such that 1p = 0 and 1c = 1/2. Figures 5.4 and 5.5 depict the maximum sum-rate of the symmetric system for xed source-relay and relay-destination link gains whereas the interference to direct link gain a21 is changed by varying a21 . We have b1 = 100, c1 = 1.2, respectively, hence DF is not limiting. The nodes have powers P1 = P2 = PR = 10. In Fig. 5.4, it can be seen that rate splitting outperforms time division when a21 < 0.4 and a21 > 1.6. However, even for small a21 , Fig. 5.5 shows that the sources never transmit private messages only, as opposed to the weak interference channel. This can be attributed to the relays equal contribution to the signals received at the destinations. Moreover, as in the interference channel, for a21 > 1, the sources do not transmit private information. We note that all numerical results show 1p = 0, that is the relay never forwards common information. We conjecture this result holds for any symmetric Gaussian interference relay channel. Considering the achievable sum-rate obtained from Proposition 29, the sum-rates are in fact monotonically non-increasing functions of 1p . Thus, the sum-rate bounds obtained from these expressions also preserve monotonically non-increasing behavior with 1p . This leads the optimal 1p to be 1p = 0 when these bounds are active. Our numerical results illustrate that for a21 > 1 optimal allocation requires 1p = 0, so that 1p = 0. For a21 < 1, the sum rate bounds obtained from (5.2f) do not dominate the sum-rate, which leads to optimal 1p value such that 1p = 0. It should be noted that even though the relay does not transmit private information, decoding of these messages at the relay improves the achievable rates of the common messages from sources to the relay since the interfering eect of the private messages is removed.

93

2.5 2.4 2.3 Rate (bits/ch use) 2.2 2.1 2 1.9 1.8 1.7 1.6 0.1 0.2 0.3 0.4 0.5 0.6 Relay location, d 0.7 0.8 0.9 Rate Splitting Time Division No Relay

Figure 5.2: Total optimized sum-rate (bits/channel use) of the symmetric channel with a21 = 1/8, P1 = P2 = PR = 10. When the relay is not present, we have P1 = P2 = 15.

5.7

Chapter Summary
We study the interference relay channel where two interfering source-destination

pairs are helped by a common relay. Using rate-splitting idea of Carleial [5] and a decode-and-forward relay, we derive an achievable rate region for the Gaussian interference relay channel. We analyze optimal power allocation among the common and private messages at the sources and relay to maximize the sum-rate. We argue that the presence of the relay aects the transmission and power allocation strategies of the sources and the notions of interference levels, (weak/medium/strong) compared to the interference channel. For the symmetric interference relay channel, the results show that the relay allocates all of its power to transmit common messages. On the other hand, unlike the interference channel, the sources continue to send common messages along with private even when the interference is weak.

94

1 0.9 Power allocation variables 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 Relay location, d 0.7 0.8 0.9 1c 1p 1p 1c

Figure 5.3: Optimal power allocation variables of the symmetric channel for rate splitting with a21 = 1/8, P1 = P2 = PR = 10.

95

3 2.8 2.6 2.4 2.2 2 1.8 1.6 Rate Splitting Time Division No Relay

Rate (bits/ch use)

0.5

a21

1.5

2.5

Figure 5.4: Total optimized Rate (bits/channel use) of the symmetric channel as a function of a21 with b1 = 100, c1 = 1.2, P1 = P2 = PR = 10. When the relay is not present, we have P1 = P2 = 15.

96

1 0.9 0.8 Power allocation variables 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 0 0.5 1 aS 1.5 /aS 2
D

10 11 11 10

2.5

2 1

1 1

Figure 5.5: Optimal power allocation variables of the symmetric system for rate splitting as a function of a21 with b1 = 100, c1 = 1.2, P1 = P2 = PR = 10.

97

Chapter 6 Conclusion

Next generation wireless systems are required to ensure improved qualityof-service (QoS) requirements due to increasing demands by the users. Among the QoS requirements, higher communication rates, increased number of users in a given network and utilization of the coexistence of various systems can be regarded as the primary components. However, due to broadcasting nature of wireless terminals, interference is an unavoidable and foremost phenomenon in the wireless communication networks which contradicts with the demands of these and almost all QoS requirements. Therefore, interference is the fundamental design parameter in the architecture of next generation systems. Recently, cooperation/relaying has emerged as a very instrumental tool in providing substantial improvements for mostly point-topoint systems. In this thesis, we investigated the impact of relaying in interference limited multi-user networks. We examined several models to understand the benets of relaying to explore various practical multi-user networks. First, we considered a two-user Gaussian interference channel aided by a relay where the relay reception and transmission take place over orthogonal bands. Despite the simplicity oered by the orthogonal relay channels, this system reveals the major relaying operations in a general interference limited multi-user network with the emphasis of improving communication rates. We examined two signicant relaying operations, namely, signal relaying and interference

98 forwarding, where the former enables the destination to obtain a stronger desired signal or additional information, and the latter helps removing part of the interference. Our results show that signal relaying, or the combination of signal relaying and interference forwarding can attain optimal transmission rates under some channel conditions. On the other hand, the optimality of these operations have also been related to the problem of assessing optimality of either separable or non-separable transmission over parallel interference channels. For instance, under the conditions where the relay-to-destination channels form the performance bottleneck with respect to the source-to-relay channels, we established that signal relaying is optimal along with the substantial reduction in encoding/decoding complexity as a result of interference and relay channel separation. However, the analysis shows that, in general, joint signal relaying and interference forwarding, and thus non-separable transmission, is necessary to attain optimal performance. This clearly complicates the design. Moreover, we showed that these results are generally valid under the assumption of variable bandwidth allocation among the relay channels, emphasizing the importance of proposed operations. Next, we explored a dierent model with a particular interest in revealing additional relaying functions relevant in interference limited system. In the model, the relay is assumed to receive over orthogonal bands with nite capacities and transmit in the band of the interfering nodes, hence giving relay the opportunity to reduce the interfering signal strength. We denoted this relaying operation as interference cancelation. Our analysis showed that signal relaying and interference forwarding are essential for improved performance, however, interference cancelation can indeed be the primary and most eective relaying operation under some channel conditions. Moreover, the orthogonality feature of the relay reception channels has been demonstrated to provide the opportunity of extra source information transmission via the relay through non-linear techniques, which also brings additional performance improvement.

99 Finally, we studied the generalization of Gaussian relay channel [6] to incorporate multiple source-destination nodes. It is shown that in-band reception at the relay necessitates enhanced decoding and encoding schemes at the relay to maintain the benets oered by proposed relaying operations. Despite the complexity added into the system, our analysis revealed the substantial performance improvement compared with standard approach which is to orthogonalize the channels to combat interference.

100

Appendix A

A.1

Proof of Proposition 5
We consider outer bounds on R1 and the bounds on R2 can be obtained

similarly. We have the following bound nR1 = H (W1 )


R 1 n R 1 n n = I (W1 ; Y1n , YR 1 ) + H (W1 |Y1 , YR1 ) R 1 n I (W1 ; Y1n , YR 1 )+n n

(A.1) (A.2) (A.3) (A.4)


n

R 1 n R 1 n n I (X1 ; Y1n ) + h(YR 1 ) h(ZR1 ) + n n , R 1 n where (A.3) follows from the Fano inequality H (W1 |Y1n , YR 1 ) n

and (A.4) is

n from chain rule, the Markovity W1 X1 Y1n and conditioning decreases entropy. R1 n Now, the rst bound on R1 in (3.10) can be obtained by noting that, h(YR 1 ) R1 n R 1 n R 1 n R 1 n 2 h(ZR + ZR 1 ) = h(c1 XR1 1 ) h(ZR1 ) nR1 C (c1 PR1 ), and therefore n nR1 I (X1 ; Y1n ) + nR1 C (c2 1 PR1 ) + n n .

101 Also, from (A.3), we get the bound the other bound in (3.10) from the following series of inequalities
R1 n nR1 I (W1 ; Y1n , YR 1 )+n n n n n

(A.5) (A.6) (A.7) (A.8) (A.9)

2R n 1R n I (W1 ; Y1n , Y1 R , Y2R ) + n

2R n 1R n 2R n n 1R n I (X1 ; Y1n ) + h(Y1 R , Y2R ) h(Z1R ) h(Z2R ) + n

2R n 1R n 2R n n 1R n I (X1 ; Y1n ) + h(Y1 R ) + h(Y2R ) h(Z1R ) h(Z2R ) + n n 2 I (X1 ; Y1n ) + n1R C (b2 1 P1R ) + n2R C (b2 P2R ) + n n ,

2R n R 1 n 1R n where (A.6) is from the Markov chain W1 Y1 YR and data processR , Y2R 1 n ing inequality, (A.7) is from chain rule, W1 X1 Y1n and conditioning decreases
2R n 1R n and Y2 entropy and (A.8) is from independence of Y1 R . Finally, the last bound R

on R1 in (3.10) is obtained as, nR1 = H (W1 |W2 )


R 1 n R 1 n n = I (W1 ; Y1n , YR 1 |W2 ) + H (W1 |Y1 , YR1 , W2 ) R 1 n I (W1 ; Y1n , YR 1 |W 2 ) + n n n

(A.10) (A.11) (A.12) (A.13) (A.14)

R1 n n = I (W1 ; Y1n |W2 ) + I (W1 ; YR 1 |Y 1 , W 2 ) + n

R 1 n R 1 n n I (X1 ; Y1n |X2 ) + h(YR 1 |W2 ) h(ZR1 ) + n n .

R 1 n R1 n R 1 n R 1 n 2 Now, since h(YR 1 |W2 ) h(ZR1 ) h(YR1 ) h(ZR1 ) nR1 C (c1 PR1 ), we have n n nR1 I (X1 ; Y1n |X2 ) + nR1 C (c2 1 PR1 ) + n n .

Moreover, following (A.13), we have,


2R n n n n 1R n ) + I (W1 ; Y1 ; Y1n |X2 nR1 I (X1 R , Y2R |Y1 , W2 ) + n n n

(A.15) (A.16) (A.17) (A.18)

2R n 1R n 2R n n n 1R n ) + h(Y1 ; Y1n |X2 I (X1 R , Y2R |W2 ) h(Z1R ) h(Z2R ) + n 1R n n n 1R n ) + h(Y1 ; Y1n |X2 I (X1 R ) h(Z1R ) + n n n ) + n1R C (b2 ; Y1n |X2 I (X1 1 P1R ) + n n n

102
2R n R 1 n 1R n where (A.15) is from W1 Y1 YR R , Y2R 1 , (A.16) is due to conditioning 2R n decreases entropy and (A.17) is from the fact that X2 is a function of W2 and R 2R n 1R n independence of Y1 and Z2 R R .

The general outer bound is then obtained by taking the union of all rates
n (R1 , R2 ) satisfying the constraints for all n 0 and input distributions p(xn 1 , x2 ) = n p(xn 1 )p(x2 ), with power constraints (P1 , P2 ), which can be proved to coincide with the

limiting region in (3.10), (see, e.g., [8], Remark 1). 2

A.2

Proof of Proposition 6
Codeword Generation and Encoding: The sources divide their messages

as W1 = (W1R , W1p , W1c , W1c ), and W2 = (W2R , W2p , W2c , W2c ) as explained in
n n Sec.3.3.2. Messages Wip and (Wic ,Wic ) are encoded into codewords Xip and Xic

with rates Rip , Ric + Ric for i = 1, 2, respectively, and sent over the IC in n channel uses. Such codewords are generated i.i.d. from independent Gaussian distributions with zero-mean and powers Pip , Pic , respectively. Overall, we have the transmitted codewords over the IC:
n n n X1 (W1 ) = X1 p (W1p ) + X1c (W1c , W1c ) n n n X2 (W2 ) = X2 p (W2p ) + X2c (W2c , W2c ).

(A.19a) (A.19b)

Message WiR is transmitted to Di via the OBRC only. Moreover, to facilitate interference cancellation, source Si transmits message Wic to the interfered destination Dj , j = i, via the OBRC. The messages (Wic , WiR ) are jointly encoded by Si into the
iR n codewords XiR (Wic , WiR ) which are generated i.i.d. with rate Ric + RiR from inde-

pendent Gaussian distributions with zero-mean and power PiR , i = 1, 2. On the other hand, after successfully decoding the messages (W1R , W2R , W1c , W2c ), the OBR enRi n codes these messages into the codewords XRi (Wjc , WiR ) with rate Rjc + RiR which

are also generated i.i.d. from independent Gaussian distributions with zero-mean and

103 power PRi , j = i, i, j = 1, 2. Decoding: The destination Di initially decodes the messages (Wjc , WRi ) using the R Di channel which leads to the achievable rates (3.11e) and (3.11f). The signals received on the IC are given by (3.1) with (A.19). Moreover, since the
n destination D1 decodes W2c , it thus sees an equivalent codebook X2 c (W2c , W2c )

with only 2nR2c codewords (and power Pic ). Similarly, D2 sees an equivalent coden book X1 c (W1c , W1c ) with rate R1c . Decoding of the messages (W1c , W1c , W1p , W2c )

at destination D1 (and (W2c , W2c , W2p , W1c ) at destination D2 ) is then performed jointly as over a multiple access channel with three sources of rates R1c = R1c + R1c , R1p and R2c (and R2c = R2c + R2c , R2p and R1c for D2 ), by treating the private messages as noise, thus with equivalent noise power Ni = a2 ji Pjp + 1 for i, j = 1, 2, i = j hence giving the achievable rates (3.11a) and (3.11b). It is also noted that, as explained in [51], error events corresponding to erroneous decoding of only message W2c at destination D1 and W1c at destination D2 do not contribute to the probability of error and thus can be neglected. The relay decodes the messages (Wic , WiR ), i = 1, 2 using the orthogonal source-to-relay links as given in (3.2). Therefore, it is possible to show that the rates in (3.11c) and (3.11d) which are the point-to-point rates in decoding the messages (WiR , Wic ) are achievable. 2

A.3

Proof of Proposition 8
The converse follows from Proposition 5. Namely, the upper bounds on

individual rates (3.14a) and (3.14b) are a consequence of the second bounds on both R1 and R2 , while the upper bound on the sum rate (3.14c) follows by summing second

104 and rst bounds on R1 , R2 , respectively and accounting for the condition a12 1 as R1 + R2 1 1 n n n 2 I (X1 ; Y1n |X2 ) + I (X2 ; Y2n ) + min 1R C (b2 1 P1R ), R1 C (c1 PR1 ) n n (A.20) 1 1 1 n n n n n n n n ) + h(a12 X1 ) h(a12 X1 ) h(Z1 + Z2 ) + Z1 + X2 + Z2 h(X1 n n n (A.21) (A.22)

2 2 + min 1R C (b2 1 P1R ) + 2R C (b2 P2R ), R2 C (c2 PR2 )

2 + 1R C (b2 1 P1R ) + R2 C (c2 PR2 ) 2 2 C (a2 12 P1 + P2 ) + 1R C (b1 P1R ) + R2 C (c2 PR2 )

2 2 where (A.21) is due to the conditions 1R C (b2 1 P1R ) R1 C (c1 PR1 ) and R2 C (c2 PR2 ) n n 2R C (b2 2 P2R ), (A.22) is from the worst-case noise result [36], i.e., h(X2 + Z2 ) n n h(a21 X2 + Z1 ) n log(1) for a21 1, and the rst entropy is maximized by i.i.d.

Gaussian inputs. For achievability, we use the general result of Proposition 6, where the sources transmit common messages (W1c , W2c ) over the IC which are decoded at both destinations. In addition, S2 transmits also the message W2c to be decoded at D2 . The other rates are set to R1c = R1p = R2p = 0. The OBRC is used to transmit independent messages W1R , W2R with rates R1R = 1R C (b2 1 P1R ) and R2R = R2 C (c2 2 PR2 ), but also message W2c of rate R2c to D1 in order to facilitate interference cancellation. From Proposition 2, and applying Fourier-Motzkin elimination, we obtain the following achievable region R1 C (P1 ) + 1R C (b2 1 P1R ) R2 C (P2 ) + R2 C (c2 2 PR 2 )
2 2 R1 + R2 C (a2 12 P1 + P2 ) + 1R C (b1 P1R ) + R2 C (c2 PR2 ) 2 2 R1 + R2 C (P1 + a2 21 P2 ) + Rex21 + 1R C (b1 P1R ) + R2 C (c2 PR2 ),

(A.23) (A.24) (A.25) (A.26)

2 so that for Rex21 max{0, C (a2 12 P1 + P2 ) C (P1 + a21 P2 )}, the claim is proved.2

105

A.4

Proof of Proposition 9
The converse is again a consequence of Proposition 5. Specically, the single

rate bounds (3.15a) and (3.15b) follow immediately from the second bounds on R1 and R2 , while the bound on the sum-rate (3.15c) is obtained from the summation of rst and second bound on R1 , R2 , respectively, and the condition a12 1 as R1 + R2 1 1 n n n 2 2 I (X1 ; Y1n ) + I (X2 ; Y2n |X1 ) + min(1R C (b2 1 P1R ) + 2R C (b2 P2R ), R1 C (c1 PR1 )) n n (A.27) 1 1 1 n n n n n n n n ) h(a21 X2 ) h(X1 + a21 X2 + Z1 + Z1 ) + h(X2 + Z2 ) h(Z2 n n n (A.28) (A.29)

2 + min(2R C (b2 2 P2R ), R2 C (c2 PR2 ))

2 + R1 C (c2 1 PR1 ) + R2 C (c2 PR2 ) 2 2 C (P1 + a2 21 P2 ) + R2 C (c2 PR2 ) + R1 C (c1 PR1 ),

2 2 where (A.28) is due to the conditions R1 C (c2 1 PR1 ) + R2 C (c2 PR2 ) 1R C (b1 P1R ) + 2 2 2R C (b2 2 P2R ) and R2 C (c2 PR2 ) 2R C (b2 P2R ), (A.29) is from the worst-case noise 1 n n n n log(1) for a21 1, and the fact result [36], i.e., h(X2 + Z2 ) h(a21 X2 + Z1 ) n2

that rst entropy is maximized by i.i.d. Gaussian inputs. For the achievability, consider the achievable rate region given in the proof of Proposition A.3 ((A.23)-(A.26)) (3.14a)-(3.14c). Clearly, when the conditions in
2 2 Proposition 9 which can also be written as R1 C (c2 1 PR1 )1R C (b1 P1R ) 2R C (b2 P2R ) 2 2 R2 C (c2 2 PR2 ) and Rex21 C (a12 P1 + P2 ) C (P1 a21 P2 ) are satised, (A.29) is achievable,

hence gives the sum capacity. 2

106

A.5

Proof of Proposition 10
The converse is obtained from Proposition 5 by adding the second constraint

on R1 and rst constraint on R2 in (3.10) such that, R1 + R2 1 1 n n n 2 I (X1 ; Y1n |X2 ) + I (X2 ; Y2n ) + min(1R C (b2 1 P1R ), R1 C (c1 PR1 )) n n (A.30) 1 1 1 1 n n n n n n n n h(X1 + Z1 ) h(Z1 ) + h(a12 X1 + X2 + Z2 ) h(a12 X1 + Z2 ) n n n n (A.31) (A.32) P2 1 + a2 12 P1

2 2 + min(1R C (b2 1 P1R ) + 2R C (b2 P2R ), R2 C (c2 PR2 ))

2 + 1R C (b2 1 P1R ) + R2 C (c2 PR2 )

C (P1 ) + C

2 + 1R C (b2 1 P1R ) + R2 C (c2 PR2 )

2 2 where (A.31) is due to the conditions R1 C (c2 1 PR1 ) 1R C (b1 P1R ) and 2R C (b2 P2R ) n n R2 C (c2 2 PR2 ), (A.32) is from the worst-case noise result of [36], i.e. h(X1 + Z1 ) n n h(a12 X1 + Z2 ) nC (P1 ) nC (a2 12 P1 ) for a12 < 1.

Achievability follows directly from Proposition 6 by letting transmitter S1 transmit private message only, i.e., W1p over the IC and W1R over the OBR, whereas user S2 transmits common information both on the IC and OBR (W2c , W2c ) as well as private message via OBR, W2R . The other rates are set to zero R1c = R1c = R2p = 0. Then, using Fourier-Motzkin elimination, it is possible to show that the following sum-rate is achievable,
2 2 R1 + R2 C (P1 + a2 21 P2 ) + Rex21 + 1R C (b1 P1R ) + R2 C (c2 PR2 )

(A.33) (A.34)

R1 + R2 C (P1 ) + C

P2 1 + a2 12 P1

2 + 1R C (b2 1 P1R ) + R2 C (c2 PR2 )

where Rex21 is given in (3.13). Then, for Rex21 C (P1 ) + C (A.34) is achievable, hence gives the sum capacity.2

P2 1+a2 12 P1

C (P1 + a2 21 P2 ),

107

A.6

Proof of Proposition 11
The converse is directly from (A.29) in Appendix A.6, which is true for

a21 1. For the achievability, consider (A.33) and (A.34) in Appendix A.5. Hence, for 0 Rex21 C (P1 ) + C 1R C (b2 1 P1R ). 2
P2 1+a2 12 P1

C (a2 21 P2 + P1 ), the achievable sum-rate is

given by (A.33), which is equal to the upper bound (A.29) for Rex21 = R1 C (c2 1 PR1 )

A.7

Proof of Proposition 12
For the achievable scheme, we consider a special case of Proposition 6,

where sources operate separately over the IC and OBRC, by sending only message (W1R , W2R ) over the OBRC (signal relaying) and only common information (W1c , W2c ) over the IC. From Proposition 2, i.e. using (3.11a)-(3.11f), we obtain that the following sum-rate is achievable
2 2 R1 + R2 C P1 + a2 21 P2 + min 1R C (b1 P1R ), R1 C (c1 PR1 )) 2 + min 2R C (b2 2 P2R ), R2 C (c2 PR2 ))

(A.35) = . Optimizing over

to be maximized over (iR , Ri ) with constraint

2 i=1 (iR + Ri )

2 2 the bandwidth allocation, and recalling that C (b2 2 P2R ) C (b1 P1R ) and C (c2 PR2 ) C (c2 1 PR1 ), the optimal allocations are 2R = R2 = /2, so that the optimal achievable

sum-rate is R1 + R2 C P1 + a2 21 P2 + C (c2 2 PR2 ) . 2

For the outer bound, using Proposition 5 with a12 = a21 1, P1 = P2 , from (3.10), we obtain the upper bounds
2 2 2 R1 + R2 C P1 + a2 21 P2 + min{1R C (b1 P1R ) + 2R C (b2 P2R ), R1 C (c1 PR1 )} 2 + min{2R C (b2 2 P2R ), R2 C (c2 PR2 )}

(A.36)

108 which should be maximized over (iR , Ri ) with


2 i=1 (iR + Ri )

= . Optimizing over

2 (iR , Ri ), i = 1, 2, using C (b2 2 P2R ) C (b1 P1R ), we obtain

R1 + R2 max C P1 + a2 21 P2 + C P1 + a2 21 P2

2 2 C (c2 1 PR1 )C (c2 PR2 ) , 2 C (c2 2 PR2 ) + 2C (c1 PR1 ) C (c2 2 PR 2 ) + , 2 C (b2 P
1

(A.37)
)

2R where the rst term in the max corresponds to the choice R1 = 2R C (c2 , R 2 = 2P R1 )

2R , 1R = 0, R1 + 2R + R2 = , whereas the second to 2 R = R2 = 2 , R1 = 0. It

is possible to show that for


2 C (c2 2 PR2 ) 2C (c1 PR1 )

the outer bound obtained in (A.37) becomes equal to the optimal achievable sumrate.2

A.8

Proof of Propositions 13 and 14


We start with the bound (3.16a):

nR1 H (W1 )
BC n BC n ) ) + H (W1 |Y1n , YR I (W1 ; Y1n , YR 1 1 BC n I (W1 ; Y1n , YR )+n 1 n n n n n

(A.38) (A.39) (A.40) (A.41) (A.42) (A.43) (A.44) (A.45)

BC n = I (W1 ; Y1n ) + I (W1 ; YR |Y1n ) + n 1

M AC n n I (X1 ; Y1n ) + I (W1 ; YR |Y1n ) + n

M AC n M AC n n I (X1 ; Y1n ) + h(YR ) h(ZR )+n

M AC n M AC n M AC n M AC n n I (X1 ; Y1n ) + h(b1 X1 + b2 X2 + ZR ) h(ZR )+n R R 2 n ; Y1n ) + M AC nC b2 I (X1 1 P1R + b2 P2R + n n

where (A.40) is from Fanos inequality, (A.42) is from the Markov relations W1
BC n M AC n n Y1n . and W1 X1 YR , Y1n YR 1

109 From cut-set bound around S1 , we obtain the bound (3.16b) as


M AC n M AC n M AC n n n nR1 I (X1 ; Y1n |X2 ) + I (X1 ; YR |X2 ) R R n n ; Y1n |X2 ) + M AC nC b2 I (X1 1 P1R .

(A.46) (A.47)

Using similar steps we obtain the corresponding bounds on R2 (3.16d) and (3.16e) as
n 2 nR2 I (X2 ; Y2n ) + M AC nC b2 1 P1R + b2 P2R n n nR2 I (X2 ; Y2n |X1 ) + M AC nC b2 2 P2R .

(A.48) (A.49)

We now focus on the remaining two bound (3.16c) and (3.16f). For (3.16f), we have (from (A.41) modied for R2 ),
BC n n nR2 I (X2 ; Y2n ) + I (W2 ; YR |Y2n ) 2 BC n BC n BC n n I (X2 ; Y2n ) + h(YR ) h(c2 XR + ZR |Y2n , W2 ) 2 2

(A.50) (A.51)

where (A.51) is from conditioning decreases entropy. Now, consider the following,
BC n BC n BC n BC n BC n ) +ZR |Y2n , W2 ) h(c2 XR +ZR ) h(c2 XR h(ZR 2 2 2

BC n log(2e(1+c2 2 PR )). 2

Hence, without loss of generality, one can assume


BC n h(YR |Y2n , W2 ) = 2

BC n log(2e(1 + c2 2 PR )), 2

(A.52)

for some 0 1. Then, (A.51) becomes


n nR2 I (X2 ; Y2n ) +

BC n BC n log(2e(1 + c2 log(2e(1 + c2 2 PR )) 2 PR )) (A.53) 2 2 c2 2 PR n = I (X2 ; Y2n ) + BC nC (A.54) 1 + c2 2 PR

where we have used the fact that Gaussian distribution maximizes the entropy term for a given variance constraint. Now, consider (A.40) for the bound on R1 given in (3.16c),
BC n |W 2 ) + n nR1 I (W1 ; Y1n , YR 1 n

(A.55) (A.56)
n

BC n n n ) + I (W1 ; YR |Y1n , W2 ) + n n ; Y1n |X2 I (X1 1 c2 BC n n n n ) + I (W1 ; YR ; Y1n |X2 = I (X1 |Y 1 , W 2 ) + n c1 1

(A.57)

110 where (A.142) is due to conditioning decreases entropy and independence of W1 and
n n W2 , (A.56) is from Markovity W1 X1 Y1n and X2 is a function of W2 , (A.57) is

since scaling does not change the mutual information. Since the capacity region of BC depends on the conditional marginal distriBC n butions and noting that c1 c2 , we can write YR = 2 BC n c2 BC n Y + ZR c1 R 1 BC n where ZR c2 2 . c2 1

is an iid. Gaussian noise with variance 1 Inequality, we now have 2 BC n


2 BC |Y n ,W ) h(YR2 2 1 n

From the conditional Entropy Power

2 BC n

c2 BC n n ,W Y |Y1 2 c1 R1

n 2 n ,W h Z BC |Y1 2) + 2 BC n ( R .

(A.58)

Also, for the condition a12 1, we have,


BC n BC n n n |X1 + Z1 , W2 ) |Y1n , W2 ) = h(YR h(YR 2 2 BC n n n |X1 + Z2 , W2 ) = h(YR 2 BC n n n |a12 X1 + Z2 , W2 ) h(YR 2 BC n |Y2n , W2 ) = h(YR 2 BC n = log(2e(1 + c2 2 PR )) 2

(A.59) (A.60) (A.61) (A.62) (A.63)

n where (A.59) is from the fact that a21 X2 is a function of W2 , (A.60) follows from BC n n n n , Z1 the independence of ZR , and Z2 , (A.61) is due to the Markov chain, a12 X1 + 1 BC n n n n n Z2 X1 + Z2 for the fact that a12 1, (A.62) is true since X2 , W 2 YR is 2

a function of W2 , and (A.63) from (A.52). Then, using (A.58), (A.63), and noticing
BC n that h ZR |Y1n , W2 = BC n 2

log 2e(1

c2 2 ) c2 1

, we obtain, . (A.64)

c2 BC n n Y |Y 1 , W 2 c1 R 1

BC n c2 2 log 2e c2 P + R 2 2 c2 1

So that, recalling (A.57) and considering the inequality (A.64), we get,


n n nR1 I (X1 ; Y1n |X2 ) + h(

c2 BC n n c2 BC n Y R 1 |Y 1 , W 2 ) h ( Z R )+n c1 c1 1

(A.65) (A.66)

n n I (X1 ; Y1n |X2 ) + BC nC c2 1 PR + n n ,

111

Z1

X1 X1R
b1R b1R

+ +
a12 ZR

Y1 D1 YR

b2R

a21

ZR

X2R X2

b2R

YR D2 Y2

+
Z2

Figure A.1: Equivalent model for IC-OBR Type II channel for c1 , c2 . which recovers (3.16c) and completes the proof. For Proposition 9, we obtain the bound (3.17a) as nR1 H (W1 |W2 )
BC n |W2 ) + npen I (W1 ; Y1n , YR 1 BC n n n |Y1n , W2 ) I (X1 ; Y1n |X2 ) + I (W1 ; YR 1 BC n BC n n n ) ) h(ZR I (X1 ; Y1n |X2 ) + h(YR 1 1 n n I (X1 ; Y1n |X2 ) + nBC C (c2 1 PR ).

(A.67) (A.68) (A.69) (A.70) (A.71)

n where (A.68) is from Fanos inequality, (A.69) is from the Markov chain W1 X1 n Y1n and X2 is a function of W2 , (A.70) is since conditioning decreases entropy and

(A.71) is from the fact that Gaussian distribution maximizes the entropy for given variance constraints. Using similar steps, we obtain the rate for R2 .2

A.9

Proof of Propositions 15 and 18


We derive rst the outer bound of Proposition 15. Assume c1 , c2 , so

that we obtain an equivalent model as shown in Fig. A.1. The equivalent model

112 is in fact equivalent to a 2 2 MIMO interference channel matrices, whose channel 1 0 a 0 , H2 = 21 , H3 = following the notation in [74] are given by H1 = 0 b1 0 b2 a 0 1 0 12 , and H4 = . Notice that such equivalence is due to the fact that 0 b1 0 b2 noise correlations are immaterial in terms of the capacity region. For this channel, the assumed conditions a12 1 and a21 1 imply the strong interference regime H 2 H2
H 4 H4 and H3 H3

H 1 H1 , so that the capacity region can be found

from [74] as given by (3.18a)-(3.18d). The achievable rate for Proposition 14 follows from standard arguments assuming Wyner-Ziv compression at the relay with Gaussian test channels (see, e.g., [42]).2

A.10

Proof of Proposition 16
The converse follows from (3.16b), (3.16c), and (3.16f) as

R1 + R2

1 1 n n n 2 I (X1 ; Y1n |X2 ) + I (X2 ; Y2n ) + min{M AC C (b2 1 P1R ), BC C (c1 PR )} n n c2 2 PR + BC C (A.72) 1 + c2 2 PR
2 C (a2 12 P1 + P2 ) + C (P1 ) C (a12 P1 )

2 + min M AC C (b2 1 P1R ), BC C (c1 PR ) + BC C

c2 2 PR 1 + c2 2 PR

(A.73)

= C (P1 ) + C + BC C

P2 1 + a2 12 P1 c2 2 PR 1 + c2 2 PR

2 + min M AC C (b2 1 P1R ), BC C (c1 PR )

(A.74)

where (A.73) is from the worst-case noise result of [36] applied for a12 1. For
2 M AC C (b2 1 P1R ) BC C (c1 PR ), (A.74) is maximized for = 1 since c1 c2 , and hence

113 the outer bound becomes, R1 + R2 C (P1 ) + C P2 1 + a2 12 P1 + BC C (c2 1 PR ). (A.75)

2 Instead, for M AC C (b2 1 P1R ) < BC C (c1 PR ), denote as the optimal parameter that

maximizes (A.74) so that we get R1 + R2 C (P1 ) + C + BC C where + = 1. For the achievability, we use a special case of the coding scheme (A.19) in which, over the IC, S1 transmits private information only via a Gaussian coden n book X1 (W1 ) = X1 p (W1p ), and S2 transmits common information using a Gaussian n n codebook X2 (W2 ) = X2 c (W2c ). Only private messages (W1R , W2R ) are sent over

P2 1 + a2 12 P1

2 + min M AC C (b2 1 P1R ), BC C (c1 PR )

c2 2 PR 1 + c2 2 PR

(A.76)

the OBRC by using standard Gaussian codebooks and MAC decoding at the relay, and superposition coding at the relay. The proof then follows similarly to Appendix E, by accounting for the capacity regions of MAC and BC Gaussian channels (see, e.g., [42]). Specically, setting R1 = R1p + R1R , R2 = R2c + R2R , we obtain an achievable sum-rate, R1 + R2 C (P1 ) + C P2 1 + a2 12 P1
2 + min M AC C (b2 1 P1R ), BC C (c1 PR )

+ min M AC C (b2 2 P2R ), BC C

c2 2 PR 1 + c2 2 PR

(A.77)

2 Now, for = 1 and M AC C (b2 1 P1R ) BC C (c1 PR ), the outer bound (A.75) is achiev-

able, hence we obtain the sum capacity. On the other hand, with the power split of PR allocated at the relay for the transmission of W1R (and PR for W2R ), for the
2 2 conditions M AC C (b2 1 P1R ) < BC C (c1 PR ) and M AC C (b2 P2R ) BC C c2 2 PR 1+c2 2 PR

, the

achievable sum-rate (A.77) is equal to the outer bound (A.76).2

114

A.11

Proof of Proposition 17
For a symmetric IC-OBR Type-II, we have the following achievable region

n when the sources transmit common messages over the IC (Xin (Wi ) = Xic (Wic ),

i = 1, 2) and independent messages (W1R , W2R ) via OBRC, as discussed in Appendix A.10. Then, the following rate region can be easily shown to be achievable R1c C (P ) R1c + R2c C (P + a2 P ) R1R M AC C (b2 Ps ) R1R + R2R M AC C (2b2 Ps ) R1R + R2R BC C (c2 PR ) We now set R1c = R2c , R1R = R2R and R1 + R2 = 2(R1c + R1R ), so that R1 +R2 min 2C (P ), C (P + a2 P ) +min 2M AC C (b2 Ps ), M AC C (2b2 Ps ), BC C (c2 PR ) . Then, for a 1 + P and c2 PR (1 + 2b2 Ps )
M AC BC

(A.78) (A.79) (A.80) (A.81) (A.82)

1, the achievable sum-rate is

equal to the outer bound (3.18a), hence giving the sum capacity.2

A.12

Proof of Proposition 19
The achievable region is obtained similarly to Appendix A.10. Source S1

n n transmits private message only, i.e., X1 (W1 ) = X1 p (W1p ) over the IC and indepen-

dent private message over the OBRC W1R via Gaussian codebooks. Source S2 transn n (W2 ) = X2 mits common messages (W2c , W2c ) over the IC (X2 c (W2c , W2c )), and

the private message W2R is transmitted via the OBR along with W2c (interference forwarding). Then, the following conditions are easily seen to provide an achievable

115 region R1p C (P1 ) R2c + R1p C (P1 + a2 21 P2 ) R2c C P2 1 + a2 12 P1 (A.83) (A.84) (A.85) (A.86) (A.87) (A.88) (A.89) (A.90)

R1R M AC C (b2 1 P1R ) R2c + R2R M AC C (b2 2 P2R )


2 R1R + R2c + R2R M AC C (b2 1 P1R + b2 P2R )

R2c + R1R BC C (c2 1 PR ) R2R BC C c2 2 PR 1 + c2 2 PR

Using Fourier-Motzkin elimination method, with the fact that R1 = R1p + R1R , R2c = R2c + R2c , and R2 = R2c + R2R , the achievable region in Proposition 15 can be obtained. Now, for b2 , c1 , the achievable region becomes R1 C (P1 ) + M AC C (b2 1 P1R ) R2 C P2 1 + a2 12 P1 + BC C c2 2 PR (A.91) (A.92)

since the overall region is maximized for = 0 for b2 , c1 . For the outer bound, we use the bound on IC-OBR Type-II given in (A.74) Proposition A.10, which is obtained from Proposition 9 for a12 < 1. Again, as b2 , c1 , the outer bound is maximized for = 0, and the achievable sum-rate R1 + R2 obtained by the summation of (A.91) and (A.92) is equal to the outer bound (A.74), thus concluding the proof.2

A.13

Proof of Proposition 20
n n )= ; Y1n |X2 The converse follows directly from Proposition 2 by noticing that I (X1

n n ) = C (P ) under the given assumptions. The achievability follows from ; Y2n |X1 I (X2

116 the discussion above since on the equivalent channel described before Proposition 20, one can achieve a symmetric rate R C (P )+min{BC C (c2 PR ) ,
M AC 2

log

1 2

+ b2 Ps },

where the rst term (C (P )) is the capacity over the two-way channel (i.e., the IC) and the second term follows from achievable rates on the two-way relay channel in [48] [47]. 2

A.14

Proof of Proposition 21
Codebook generation and encoding: Each source performs message splitting as discussed in Sec. 4.4, i.e. Wi =

(Wic , Wic , Wip , Wip , Wip ), i=1,2. Messages Ws , s {ic, ic, ip, ip}, i = 1, 2, are then
n n (Wic ), encoded into codewords Xs with rates Rs , respectively. The IC codewords Xic n n n Xic (Wip ), Xip (Wip ), i = 1, 2, are generated using i.i.d. Gaussian distribu(Wic ), Xip 2 2 2 2 tions with zero mean and variances ic Pi , ic Pi and ip Pi , ip Pi , i = 1, 2, respectively.

Furthermore, all codewords are independent. Then the input signals for the IC can be written as (omitting dependency on the message splits),
n n n n n X1 = X1 c + X1c + X1p + X1p n n n n n X2 = X2 c + X2c + X2p + X2p

(A.93) (A.94)

To maintain power constraint at each source, the allocations should satisfy,


2 2 2 2 ic + ic + ip + ip 1, i = 1, 2.

Moreover, each source Si transmits messages (Wic , Wip , Wip ), i = 1, 2, to the relay via the nite capacity links in n uses of these links, that is, using V1 = (W1c , W1p , W1p ) V2 = (W2c , W2p , W2p ) (A.95) (A.96)

With the assumption that the relay decodes these splits with arbitrarily

117 small probability of error, then, its transmitted signal can be written as,
n n n n XR (V1 , V2 ) = XRc (W1c , W2c ) + XRp (W1p , W2p ) + XR p (W1p , W2p ) n where XRc = 1c 1c 2c PR n X1 c + 2c P1 PR n X2 c, P2 n XRp = 1p 1p 2p PR n X1 p + 2p P1 PR n X2 p, P2

(A.97)

n and XR p is

n n n the relay signal used to transmit the message splits (W1p , W2p ), with XR p = X1p + X2p

with E[Xi2p ] i2p PR , i = 1, 2. To form X1p , the relay performs Costa encoding
2p n [12] by considering the interference signal SD 1 = (a21 + c1 2p PR n )X2 p P2

+ c1 X2p as

the state known non-causally at the transmitter. As given in [12], in encoding of


n nI (U1p ;Y1 ) n X1 number of codewords denoted as U1 p such that U1p p , we generate 2 2 2 N (0, 1 p PR + 1p PSD1 ) where PSD1 is power of SD 1 and 1p is the Costa parameter. In

forming the codebook, these sequences are uniformly placed into 2nR1p bins. To obtain
n n the codeword to be transmitted, the encoder searches for the sequence (U1 p 1p SD1 ) n that is nearly orthogonal to SD 1 . After nding this sequence, the codeword to be n n n transmitted is obtained by X1 p = U1p 1p SD1 . The codebook generation follows

exactly same steps for X2p , except the fact that the non-causally known state at D2
1p n is SD 2 = (a12 + c2 1p

PR n )X1 p P1

n + a12 X1 p , i.e. the interference due to X1p is treated as

noise. Overall, to satisfy the relay power constraint, the allocations should hold
2 2 2 2 2 2 1 c + 2c + 1p + 2p + 1p + 2p 1.

Note that the achievable region discussed in Remark 1, by following same steps in encoding of
1p c2 1p ach , 2p n the states are given by SD 1 = (a21 + c1 2p PR n )X1 p P1 n + a12 X1 p + c2 X1p . PR n )X2 p P2

ach ,

can be obtained

however in encoding of X1p and X2p ,


n n + a21 X2 p and SD2 = (a12 +

Decoding: The relay immediately recovers corresponding splits from each source, (A.95), (A.96), using the nite-capacity links. On the other hand, at each destination, we assume joint decoding of the desired message splits. For example, at D1 , the splits (W1cd , W1cr , W2cd , W2cr , W1pd , W1pr ) are decoded jointly by treating
n n the signals Xin p (Wip ), i = 1, 2 and X2pd (W2pd ), X2pr (W2pr ) as noise and similarly for

118 D2 . The decoding of Wip , at Di , i = 1, 2 follows from Costas decoding [12]. Hence, the message splits W1p and W2p , are decoded at their desired destination without observing any interference from the remaining message splits. However, since in the
n n n encoding of X2 p at the relay, the signal X1p is treated as noise, D2 simply treats X1p

as noise as well in decoding W1p . 2

A.15

Proof of Proposition 22
We dene as Vi Vi , the messages transmitted from Si to R, with Vi being

a function of Wi , i = 1, 2, whereas XR,t is the relay signal transmitted to both destinations at time t, and is a function of (V1 , V2 ). Therefore the following Markov chains, X1,t V1 XR,t and X2,t V2 XR,t hold. We use the following denitions,
2 Var(XR,t |V1 ) = (1 1 c,t )PR,t 2 Var(XR,t |V2 ) = (1 2 c,t )PR,t

(A.98) (A.99)

Var(X1,t |V1 ) = (1 2 1c,t )P1,t Var(X2,t |V2 ) = (1 2 2c,t )P2,t


2 2 where Pi,t = E {Xi,t }, i = 1, 2, PR,t = E {XR,t } satisfy 1/n n t=1

(A.100) (A.101) P1,t P1 , 1/n


n t=1

P2,t

P2 and 1/n

n t=1

2 PR,t PR . Moreover, since Var(X ) 0, we have ic,t 1 and

2 ic,t 1, for i = 1, 2. We also let,


2 1 c PR

1 = n 1 n

n 2 1 c,t PR,t t=1 n 2 2 c,t PR,t t=1

(A.102) (A.103)

2 2 c PR =

119 1 = n 1 n
n

2 1 c P1

2 1c,t P1,t
t=1 n

(A.104) (A.105)

2 2 c P2 =

2 2c,t P2,t
t=1

2 2 to be used throughout the proof and hence ic 1 and ic 1, i = 1, 2.

The following lemma, which demonstrates relations among correlations of Markov chain elements, is essential in proving the outer bounds. Lemma 1 [46]: Given a Markov triple X1 U X2 with X1 and X2
2 2 with nite second moments, i.e., E [X1 ] and E [X2 ] , then

E [X1 X2 ]

2 E [X1 ] Var(X1 |U )

2 E [X2 ] Var(X2 |U )

(A.106)

Then using Lemma 1 for the above Markov chains we obtain the correlations as, E[X1,t XR,t ] P1,t (1 2 1c,t )P1,t P1,t PR,t
2 PR,t (1 1 c,t )PR,t

(A.107) (A.108)

= 1c,t 1c,t and similarly,

E[X2,t XR,t ] 2c,t 2c,t We start with the constraint on R1 , nR1 = H (W1 )
n = H (W1 |V2 , X2 )

P2,t PR,t .

(A.109)

(A.110) (A.111) (A.112) (A.113) (A.114)


n n

n n = I (W1 ; Y1n |V2 , X2 ) + H (W1 |Y1n , V2 , X2 )

n I (W1 ; Y1n |V2 , X2 )+n n h(Y1n |V2 , X2 )+n n

n n n n = h(a11 X1 + c1 XR + Z1 |V2 , X2 )+n n

(A.115) (A.116)

t=1

h(a11 X1,t + c1 XR,t + Z1,t |V2 ) + n

120
n where (A.111) is from independence of W1 and V2 , X2 , (A.113) is from Fanos in-

equality and (A.116) is from chain rule and removing conditioning does not reduce entropy. Continuing, we have, R1 1 n 1 n
n

EV2 h(a11 X1,t + c1 XR,t + Z1,t |V2 = v2 ) + n


t=1 n 2 EV2 C a2 11 P1,t + c1 Var(XR,t |V2 = v2 ) t=1

(A.117)

+ 2a11 c1 Cov(X1,t XR,t |V2 = v2 ) 1 n 1 n


n

+n

(A.118)

2 C a2 11 Var(X1,t ) + c1 EV2 [Var(XR,t |V2 = v2 )] t=1 n

+ 2a11 c1 EV2 [Cov(X1,t XR,t |V2 = v2 )] + n


n

(A.119)

2 C a2 n 11 P1,t + c1 Var(XR,t |V2 ) + 2a11 c1 Cov(X1,t XR,t |V2 ) + n(A.120) t=1

1 n C

n 2 2 C a2 11 P1,t + c1 (1 2c,t )PR,t + 2a11 c1 1c,t 1c,t t=1

P1,t PR,t + n PR,t P1,t

(A.121) +

a2 11 P1

1 c2 1

(1
t=1

2 2 c,t )PR,t

1 + 2a11 c1 n

1c,t 1c,t
t=1

2 C a2 11 P1 + c1

1 PR n

n 2 2 c,t PR,t i=1

+ 2a11 c1

1 n

1 2 1c,t P1,t

(A.122)
2 1 c,t PR,t

t=1

t=1

(A.123)

where (A.118) is from the fact that entropy is maximized by Gaussian inputs for given variance constraints, (A.119) is from Jensens inequality. In (A.122), we incorporate (A.99) and use further the equalities such that, Cov(X1,t XR,t |V2 ) = EV2 E [X1,t XR,t |V2 = v2 ] E [X1,t |V2 = v2 ]E [XR,t |V2 = v2 ] =E [X1,t XR,t ] E [X1,t ]EV2 E [XR,t |V2 = v2 ] which follows from the independence of X1,t and V2 , and since E [X1,t ] = 0, we obtain Cov(X1,t XR,t |V2 ) = E [X1,t XR,t ] 1c,t 1c,t P1,t PR,t by (A.108). Moreover, (A.122)

121 is from Jensens inequality, (A.123) is from Cauchy-Schwartz inequality. Finally incorporating (A.102) and (A.104), with
n

0, we obtain the bound in (4.10a).

Using similar steps, we obtain the bound on R2 given in (4.10c). Another bound on R1 can be obtained as follows,
n ) nR1 = H (W1 |V2 , X2 n n ) ) + H (W1 |Y1n , V1 , V2 , X2 = I (W1 ; Y1n , V1 |V2 , X2 n I (W1 ; Y1n , V1 |V2 , X2 )+n n n

(A.124) (A.125) (A.126) (A.127)

n n I (W1 ; V1 |V2 , X2 ) + I (W1 ; Y1n |V1 , V2 , X2 )+n

n n n = H (V1 |V2 , X2 ) H (V1 |W1 , V2 , X2 ) + I (W1 ; Y1n |V1 , V2 , X2 ) + n (A.128) n n nC1 + I (W1 ; Y1n |V1 , V2 , X2 )+n n

(A.129)

n n n nC1 + h(Y1n |V1 , V2 , X2 ) h(Y1n |W1 , X1 , V 1 , V2 , X 2 )+n n n n nC1 + h(Y1n |XR , V1 , V 2 , X2 ) h(Z1 )+n n n

(A.130) (A.131) (A.132) (A.133) (A.134)

n n n n n nC1 + h(a11 X1 + Z1 |XR , V 1 , V 2 , X2 ) h(Z1 )+n n

nC1 +
t=1 n

h(a11 X1,t + Z1,t |V1 ) nh(Z1,t ) + n


2 C a2 11 1 1c,t P1,t + n t=1

nC1 +

n where (A.124) is from independence of W1 and V2 , X2 , (A.126) is from Fanos inn equality, (A.129) is from independence of V1 and V2 , X2 , with H (V1 ) nC1 , and n from the fact that H (V1 |W1 , V2 , X2 ) = 0, since V1 is a function of W1 , (A.130) is from n conditioning decreases entropy, (A.131) is from the denition XR = fR (V1 , V2 ) and

(A.133) is from chain rule and the fact that conditioning decreases entropy. Using (A.100) with Jensens inequality as done in (A.119), we get the expression (A.134).

122 We further proceed from (A.134) as, R1 1 C1 + n C1 + C


n 2 C a2 11 1 1c,t P1,t + t=1 n

(A.135) (A.136) (A.137)

a2 11 P1

1 a2 11

2 1c,t P1,t
t=1 n

2 = C1 + C a2 11 (1 1c )P1 +

where (A.136) is from Jensens inequality, and (A.137) follows by using (A.104). For
n

0, we obtain the bound (4.10b). Using similar steps, we obtain

the other bound for R2 given in (4.10d). Next, we consider the outer bound on the sum rate R1 + R2 . Now, using denition 1 to obtain degradedness conditions of Y2n w.r.t Y1n for the system in (4.1a) and (4.1b), we get, [a21 c1 ] = [a22 c2 ], 1. Hence,
n n(R1 + R2 ) = H (W1 ) + H (W2 |W1 , X1 )

(A.138)

(A.139)

n n = I (W1 ; Y1n ) + I (W2 ; Y2n |W1 , X1 ) + H (W1 |Y1n ) + H (W2 |Y2n , W (A.140) 1 , X1 ) n I (W1 ; Y1n ) + I (W2 ; Y2n |W1 , X1 ) + 2n n

(A.141)

n n h(Y1n ) h(Y1n |W1 , X1 ) + h(Y2n |W1 , X1 ) n n n h(Y2n |W1 , X1 , W2 , X 2 , XR ) + 2n n

(A.142)

n n n n n n n n ) |W 1 , X 1 + Z2 + c2 XR ) + h(a22 X2 |W 1 , X 1 + Z1 + c1 XR = h(Y1n ) h(a21 X2 n ) + 2n h(Z2 n

(A.143)

n ), (A.141) is from Fanos where (A.139) is from the independence of W2 and (W1 , X1 n a function of inequality and (A.142) is from conditioning decreases entropy and X1

W1 .

123 Incorporating the condition (4.12) into (A.143) we get,


n n n n n(R1 + R2 ) h(Y1n ) h(a22 X2 + c2 XR + Z1 /|W1 , X1 ) 1 n n n n n ) + 2n (A.144) + h(a22 X2 + c2 XR + Z2 |W 1 , X 1 ) log(2 ) h(Z2 n 2 n h(Y1n ) h(Z1 ) + 2n n (A.145) n

i=1

h(a11 X1i + c1 XRi + a21 X2i + Z1i ) h(Z1i ) + 2n

(A.146)

where (A.145) is a consequence of worst-case result in [36] which holds for 1. Similar to (A.117)-(A.123), using Jensens inequality, (A.146) becomes (omitting the intermediate steps), R1 + R2 C 1 a2 11
n

P1,t +
t=1 n

1 a2 21

P2,t +
t=1

1 c2 1

PR,t
t=1 n

1 + 2a11 c1 n

1c,t 1c,t
t=1

1 P1,t PR,t + 2a21 c1 n 1 n


n

2c,t 2c,t
t=1 n

P2,t PR,t (A.147) +2 n

2 2 C a2 11 P1 + a21 P2 + c1 PR + 2a11 c1

2 1c,t P1,t
t=1 t=1

2 1 c,t PR,t

1 + 2a21 c1 n

2 2c,t P2,t
t=1 t=1

2 2 c,t PR,t

+2

(A.148)

2 2 = C a2 11 P1 + a21 P2 + c1 PR + 2a11 c1 1c 1c

P1 PR (A.149)

+ 2a21 c1 2c 2c

P2 PR

+ 2 n.

where in (A.147), we use the correlations given in (A.108) and (A.109), (A.148) is from Cauchy-Schwartz inequality, and (A.148) is obtained by using (A.102)-(A.105). Moreover, since jointly Gaussian rvs, X1i , X2i , and XRi maximize the sum-rate in (A.147), the correlations (A.108) and (A.109) satisfy the following,
2 2 2 2 1c,t 1c,t + 2c,t 2c,t 1,

(A.150)

124 which lead to


2 2 2 2 1 c 1c + 2c 2c 1.

(A.151)

Therefore, for

0, we obtain the bound in (4.10f).

We give another bound on R1 + R2 as follows,


n n(R1 + R2 ) = H (W1 ) + H (W2 |W1 , X1 ) n )+2 I (W1 ; Y1n , V1 ) + I (W2 ; Y2n , V2 |W1 , X1 n

(A.152) (A.153)

I (W1 ; V1 ) + I (W1 ; Y1n |V1 ) + I (W2 ; V2 )


n + I (W2 ; Y2n ,ea |W1 , X1 , V2 ) + 2 n n

(A.154) (A.155)

n , V 1 , V2 ) + 2 nC1 + I (W1 ; Y1n |V1 , V2 ) + nC2 + I (W2 ; Y2n |X1

n n nC1 + nC2 + h(Y1n |V1 , V2 ) h(Y1n |W1 , X1 , V1 , V2 ) + h(Y2n |X1 , V 1 , V2 ) n n n h(Y2n |X1 , W 2 , X2 , XR , V1 , V 2 ) + 2 n

(A.156) (A.157)

n n = nC1 + nC2 + h(Y1n |V1 , V2 ) h(Y1n |W1 , X1 , V1 , V2 ) + h(Y2n |W1 , X1 , V 1 , V2 ) n h(Z2 )+2 n

(A.158)

n n n n = nC1 + nC2 + h(Y1n |V1 , V2 ) h(a21 X2 + c1 XR + Z1 |W1 , X1 , V1 , V 2 ) n n n n n + h(a22 X2 + c2 XR + Z2 |W 1 , X 1 , V1 , V2 ) h(Z2 )+2 n

(A.159)

n where (A.152) is from independence of W2 and W1 , X1 , (A.153) is from Fanos in-

equality, (A.155) is from I (Wi ; Vi ) H (Vi ) nCi and independence of Wi and Vj , for i = j = 1, 2, and (A.156) is from conditioning decreases entropy. Similar to (A.145), using [a21 c1 ] = [a22 c2 ] with 1, we get,
n n n n n(R1 + R2 ) nC1 + nC2 + h(Y1n |V1 , V2 ) h(a22 X2 + c2 XR + Z1 /|W1 , X1 , V 1 , V2 ) 1 n n n n n + h(a22 X2 + c2 XR + Z2 |W 1 , X 1 , V1 , V2 ) log(2 ) h(Z2 )+2 (A.160) n 2 n nC1 + nC2 + h(Y1n |V1 , V2 ) h(Z1 )+2 n (A.161) n

nC1 + nC2 +
t=1

h(Y1,t |V1 , V2 , XR,t ) nh(Z1,t ) + 2

(A.162)

125 where (A.161) is from worst-case result of [36], and (A.162) is from chain rule and
n conditioning decreases entropy, and from XR = fR (V1 , V2 ).

Proceeding from (A.162), R1 + R2 1 C1 + C2 + n C1 + C2 + 1 n


n

h(a11 X1,t + a21 X2,t + Z1,t |V1 , V2 ) h(Z1,t ) + (A.163) 2n


t=1 n 2 2 2 C a2 11 (1 1ri )P1i + a21 (1 2ri )P2i + 2 n (A.164) t=1 n

2 2 2 C1 + C2 + C a2 11 (1 1c )P1 + a21 (1 2c )P2 + 2

(A.165)

where (A.163) is from conditioning decreases entropy, (A.164) is from Jensens inequality (following similar steps used in showing (A.134)) and (A.165) is from Jensens inequality and by using (A.104) and (A.105). For
n

0, we obtain the rate in (4.10g).

Next, consider the last bound on the sum-rate given below,


n n(R1 + R2 ) = H (W1 |V2 ) + H (W2 |W1 , X1 ) n I (W1 ; Y1n |V2 ) + I (W2 ; Y2n , V2 |W1 , X1 ) + 2n n

(A.166) (A.167)

n = h(Y1n |V2 ) h(Y1n |W1 , V2 ) + I (W2 ; V2 |W1 , X1 ) n + I (W2 ; Y2n |W1 , X1 , V 2 ) + 2n n

(A.168)

n h(Y1n |V2 ) h(Y1n |W1 , V2 ) + nC2 + h(Y2n |W1 , X1 , V2 ) n h(Y1n |W1 , X1 , V 2 , W2 ) + 2n n

(A.169)

n nC2 + h(Y1n |V2 ) h(Y1n |W1 , X1 , V2 ) n + h(Y2n |W1 , X1 , V2 ) + 2n n

(A.170) (A.171)
n

nC2 + h(Y1n |V2 ) + 2n


n

nC2 +
t=1

h(Y1,t |V2 ) + 2n

(A.172)

n ), (A.167) from where (A.166) is from independence of W1 and V2 and W2 , (W1 , X1 n ) H (V2 ) nC2 , (A.170) is from Fanos inequality, (A.169) is from I (W2 ; V2 |W1 , X1

126 conditioning decreases entropy and non-negativity of entropy, (A.171) follows similar steps leading (A.145) using worst-case result in [36] for 1, that is true for the condition (4.12). From (A.171), we have, R1 + R2 C2 + C2 + 1 n 1 n
n

h(Y1,t |V2 ) + 2n
t=1 n

(A.173)

2 2 2 2 2 C a2 11 E [X1,t ] + a21 E [X2,t |V2 ] + c1 E [XR,t |V2 ] t=1

+ 2a11 c1 Cov(X1,t XR,t |V2 ) + 2a21 c1 Cov(X2,t XR,t |V2 ) + 2n n (A.174) 1 a2 21


n

C2 + C + 2a11 c1 1 n

a2 11 P1
n

(1
t=1

2 2c,t )P2,t + 2n
n

1 c2 1

n 2 (1 2 c,t )PR,t t=1

1c,t 1c,t
t=1

P1,t PR,t

(A.175) P1 PR (A.176)

2 2 2 2 C2 + C a2 11 P1 + a21 (1 2c )P2 + c1 1c PR + 2a11 c1 1r 1c

+ 2n

where (A.174) follows from Jensens inequality, (A.175) is also from the Jensens inequality and incorporating the correlations given in (A.98)-(A.109). We have Cov(X2i XRi |V2 ) = EV2 E [X2i XRi |V2 = v2 ] E [X2i |V2 = v2 ]E [XRi |V2 = v2 ] , and using the Markovity X2i V2 XRi holds, then EV2 E [X2i XRi |V2 = v2 ] = E [X2i |V2 = v2 ]E [XRi |V2 = v2 ], which leads to Cov(X2i XRi |V2 ) = 0. Finally, (A.176) is obtained by CauchySchwartz inequality as well as using (A.102), (A.104) and (A.105). For 2
n

0, we obtain the outer bound (4.10e).

A.16

Proof of Proposition 23
In this proof, we use the denitions in (A.98)-(A.109).

127 Consider the following bound on rate-sum, i.e.,


n n(R1 + R2 ) = H (W1 ) + H (W2 |X1 )

(A.177)

n n = I (W1 ; Y1n ) + H (W1 |Y1n ) + I (W2 ; Y2n |X1 ) + H (W2 |Y2n , X(A.178) 1) n I (W1 ; Y1n ) + I (W2 ; Y2n |X1 ) + 2n n

(A.179)

n n h(Y1n ) h(Y1n |W1 , X1 ) + h(Y2n |X1 ) n n n h(Y2n |X1 , W2 , X 2 , XR ) + 2n n

(A.180)

n n n n h(Y1n ) h(a21 X2 + c1 XR + Z1 |X1 ) n n n n n + h(a22 X2 + c2 XR + Z2 |X1 ) h(Z2 ) + 2n n

(A.181)

n where (A.177) is from independence of W2 and X1 , (A.178) is from Fanos inequality,

(A.180) is from conditioning decreases entropy and (A.181) is due to cognitive relay
n n with W1 X1 XR .

Now, by incorporating the conditions in (4.16), we continue as,


n n n n n(R1 + R2 ) h(Y1n ) h(a22 X2 + c2 XR + Z1 |X1 ) n n n n n + h(a22 X2 + c2 XR + Z2 |X1 ) h(Z2 ) + 2n n

(A.182)

with < 1. We let,


n n n n n n n n = h(a22 X2 + c2 XR + Z2 |X1 ) h(a22 X2 + c2 XR + Z1 |X1 ).

From Proposition 8, [36], we know that is maximized by Gaussian inputs, hence 1 2 n + c KX n |X n + 2a22 c2 KX n X n |X n + I log a2 2 22 KX2 1 2 1 R R 2 1 2 2 2 n + c KX n |X n + 2 a22 c2 KX n X n |X n + I log 2 a2 2 22 KX2 1 2 1 R R 2

(A.183)

n n is the covariance matrix of X , KX n |X n is the conditional covariance where KX2 2 1 R n n n n X n |X n is the conditional correlation matrix of X , KX2 matrix of XR given X1 2 and 1 R

128
n XR , and I is the n n identity matrix. Note that the corresponding power constraints

lead to the trace constraints,


n } nP2 tr{KX2

(A.184)
n 2 (1 1 c,t )PR,t

n |X n } tr{KXR 1

(A.185) (A.186)

t=1 n

tr{K

n X n |X n X2 1 R

}
t=1

2c,t

P2,t PR,t

n ) = E [X2,t XR,t ] where the (A.185) and (A.186) are due to the fact that Cov(X2,t XR,t |X1

and follow from (A.109) with optimal 2c,t = 1, due to cognitive relay.
2 n + c KX n |X n + 2a22 c2 KX n X n |X n and For simplicity, denote KXtn = a2 2 22 KX2 1 2 1 R R n denote Xt ,t as the tth eigenvalue of KXtn , t = 1, ..., n. Note that tr{KX } is nite t

due to the nite individual trace constraints given above. We argue that (as in
n [67]), for a given trace constraint on KX , is maximized for equal eigenvalues, i.e. t n n n t = u , u = t = 1, ..., n and therefore iid. inputs X1 , X2 and XR maximize 2 21 (A.183). Then, optimal eigenvalue is given by t = a22 P2 + c2 n 1 2a22 c2 n n t=1 n 2 t=1 (1 1c,t )PR,t

2c,t

P2,t PR,t , and follows,


2 nC t nC t .

(A.187)

Also note that, since < 1, is an increasing function of t , and we may further upper bound by increasing t , such as, nC a2 22 P2 + 1 c2 2
n

(1
t=1 n

2 1 c,t )PR,t

+ 2a22 c2

P2

1 n

n 2 2 c,t PR,t t=1

nC

2 a 2 22 P2

1 2 c 2 2

(1
t=1

2 1 c,t )PR,t

+ 2 a22 c2

P2

1 n

n 2 2 c,t PR,t . t=1

(A.188) where (A.188) is due to Cauchy-Schwartz inequality.

129 Now, for incorporating the notation in (A.102) and (A.103) (however with a slight change of notation as 1c 1c and 2c 2c ), we obtain that, P2 PR P2 PR (A.189)

2 2 nC a 2 22 P2 + c2 (1 1c )PR + 2a22 c2 2c 2 2 nC a 2 21 P2 + c1 (1 1c )PR + 2a21 c1 2c

Therefore, the overall outer bound on the sum-rate becomes, R1 + R2 1 1 n h(Y1n ) + h(Z2 )+2 n n n n 1 1 n h(Y1i ) + h(Z2 )+2 n t=1 n C (a11 P1 + c1 1c (A.190)
n

(A.191) P2 + c1 2c P2 PR P2 PR +2
n

PR )2 + (a21

PR )2

2 2 + C a2 22 P2 + c2 (1 1c )PR + 2a22 c2 2c 2 2 C a2 21 P2 + c1 (1 1c )PR + 2a21 c1 2c

(A.192)

where (A.191) is from chain rule and conditioning decreases entropy, (A.192) is using Jensens inequality along with using the notations in (A.102) and (A.103).
2 2 Now it is easy to check that (A.192) is an increasing function of 1 c + 2c 1, 2 2 hence incorporating 1 c = 1 2c and since n

0, we obtain the outer bound (4.17a).

A.17

Proof of Proposition 25
Achievability: The essential part in the achievability is to obtain the condi-

tions such that destinations can decode the interference without limiting the interferencefree capacity region of the system. Since in our achievability scheme, the destinations are able to decode the interference, we assume s = s = 0 where s {1p, 1p, 2p, 2p, 2p, 1p} in (4.6a)-(4.6f). On the other hand, in IC-OIR, since the sources transmit their signals both via relay and directly to the destinations, i.e. (W1c , W1c ) messages for S1 , we need two-fold conditions to obtain interference-free region.

130 Now consider the achievable region in (4.6a)-(4.6f). We modify this achievable scheme to involve sequential-decoding as opposed to simultaneous decoding used to obtain (4.6a)-(4.6f). Then, we have the following achievable rates, R1c C1 R1c C R1c C (A.193) (A.194) (A.195) (A.196)

(a12 1c P1 + c2 1c PR )2 2 1 + a2 22 2c P2 + (a22 2c P2 + c2 2c PR ) 2 a2 12 1c P1 2 1 + a2 22 2c P2 + (a22 2c P2 + c2 2c PR ) P1 + c1 1c PR )2

R1c C (a11 1c

2 R1c C a2 11 1c P1

(A.197) (A.198)
2

R2c C2 R2c C R2c C

P2 + c1 2c PR ) (a11 1c P1 + c1 1c PR )2 1+ 2 a2 21 2c P2 2 1 + a2 11 1c P1 + (a11 1c P1 + c1 1c PR ) P2 + c2 2c PR )2

(a21 2c 2 a11 1c P1 +

(A.199) (A.200)

R2c C (a22 2c
2 R2c C a2 22 2c P2

(A.201) (A.202)

Then, for (A.194) (A.196), (A.195) (A.197) and (A.199) (A.201), (A.200) (A.202), we obtain the very strong relay-interference conditions for all parameters of ic , ic , ic , i = 1, 2, and the new rate region, with Ri = Ric + Ric , i = 1, 2, gives us
the region in (4.18a)-(4.18d). Note that 1 c , 2c , 1c , 2c are the optimal power alloca-

tions giving us the sum capacity. Finally, evaluating the very strong relay-interference conditions for these optimal parameters, we obtain the conditions in (4.19a)-(4.19d). Converse: The converse follows directly from the Proposition 24 with the
prl serves as an outer bound on the sum capacity of IC-OIR. fact that Csum

131

A.18

Proof of Proposition 28
The sources transmit B message indices using superposition block Markov

encoding in B+1 blocks. Since B , the achievable rates are are not reduced, i.e. BRt /(B + 1) Rt , t [1c, 2c, 1p, 2p]. Each block consists of N channel uses. Codebook Generation: We rst generate randomly four independent codebooks. The codewords of each codebook are denoted as u1c (s1c ), u2c (s2c ), u1p (s1p ), u2p (s2p ) where st = {1, . . . , 2N Rt }, t [1c, 2c, 1p, 2p] denote the message indices. The codewords are generated using distributions
N n=1 N n=1

p(u1c,n ),

N n=1

p(u2c,n ),

N n=1

p(u1p,n ),

p(u2p,n ) where U1c , U2c , U1p , U2p are i.i.d. Gaussian r.v.s with N (0, PR ). Then, p(x1c,n |u1c,n (s1c )), p(x2c,n |u2c,n (s2c )),

for each index s1c , s2c , s1p , s2p , respectively we generate 2N R1c , 2N R2c , 2N R1p , 2N R2p i.i.d. sequences x1c , x2c , x1p and x2p using distributions
n n n

p(x1p,n |un (s1p )),

p(x2p,n |un (s2p )) where X1c , X1p are i.i.d. with N (0, P1 ), and

X2c , X2p are i.i.d with N (0, P2 ). Each sequence is labeled as x1c (s1c , w1c ), x2c (s2c , w2c ), x1p (s1p , w1p ) and x2p (s2p , w2p ), where wt 1, . . . , 2N Rt , t [1c, 2c, 1p, 2p]. Encoding: In order to exploit cooperation with the relay, the sources transmit the previous message indices as well as current message indices. Let t and t , t [1c, 2c, 1p, 2p], be the power allocation coecients used for transmitting new message indices and previous message indices, respectively such that 1c + 1c + 1p + 1p = 1 and 2c +2c +2p +2p = 1. The source transmission corresponding to index t at block (b) (b) (b) (b) b is xt = t x t (st , wt ) + t Pk /PR ut (st ), t [1c, 2c, 1p, 2p] with Pk = P1 for t [1c, 1p], Pk = P2 for t [2c, 2p] and st = wt
(b) (b) (b1)

. Note that the cooperative

codeword ut (st ) is properly scaled to satisfy the power constraints of the sources. (b) (b) For index t, the relay on the other hand transmits the signal xR = 1c u1c (s1c ) + (b) (b1) (b) (b) (b) (b) and wt is the esti2c u2c (s2c ) + 1p u1p (s1p ) + 2p u2p (s2p ) where st = wt mate of wt at the relay. The power allocation variables t , t [1c, 2c, 1p, 2p] perform proper power scaling of the cooperative signals ut (sb t ), such that 1c + 1p + 2c + 2p = 1. Overall, both the sources and the relay transmit a superposition of codewords for each
(b)

132 t corresponding to, (b) (b) (b) (b) (b) (b) x1 (b) = 1c x1c (s1c , w1c ) + 1c u1c (s1c )+ 1p x1p (s1p , w1p ) + 1p u1p (s1p ), (b) (b) (b) (b) (b) (b) x2 (b) = 2c x2c (s2c , w2c )+ 2c u2c (s2c ) + 2p x2p (s2p , w2p ) + 2p u2p (s2p ), (b) (b) (b) (b) xR (b) = 1c u1c (s1c ) + 2c u2c (s2c ) + 1p u1p (s1p )+ 2p u2p (s2p ), at blocks b = 1, . . . , B + 1. Note that st
(1)

= 1 and wt

(B +1)

= 1, t

[1c, 2c, 1p, 2p] corresponding to the rst and last blocks, respectively. Decoding: Relay decodes the messages after each received block b. At the end of b=1, from the relays point of view, the system is a multiple-access channel. Reliable decoding of wt , t = [1c, 1p, 2c, 2p] is possible, i.e. P (wt = wt ) = , if;
c , XR ) Ri I (XS ; YR |U1c , U1p , U2c , U2p , XS iS c is satised, where YR is the relay output, S {1c, 1p, 2c, 2p} and XS denotes the

(A.203)

complement set {Xj , j S c }. Proceeding forward, at the end of block b = 2, . . . , B , assuming the messages transmitted in b 1 are decoded correctly, i.e. wt
(b) (b1)

= wt

(b1)

wt can be decoded with arbitrarily small probability of error if (A.203) is satised. The destinations start decoding after the last block, b = B + 1, is received. At the last block, no new information is transmitted. Note that D1 decodes the messages w1c , w2c , w1p and D2 decodes the messages w1c , w2c , w2p . Thus D1 considers the codeword with message w22 as noise, such as D2 considers w1p message. In this case the corresponding messages can be decoded with low probability of error if, Ri
iS1 c c I (XS1 , US1 ; YD1 |XS , US ) 1 1 c c I (XS2 , US2 ; YD2 |XS , US ) 2 2

(A.204) (A.205)

Ri
iS2

where S1 {1c, 2c, 1p}, S2 {1c, 2c, 2p}. The decoding proceeds backwards and assuming the messages are decoded correctly in block b + 1, the messages in block b are decoded with low probability of error if (A.204) is satised. Exact probability of error calculations follow directly from [11] and hence omitted here. 2

133

Bibliography

[1] C. E. Shannon, Two-way communication channels, Proc. Fourth Berkeley Symp. on Math. Statist. and Prob., Berkeley, California, June 1960. [2] E. C. van der Meulen, Three-terminal communication channels, Advances in Applied Probability, vol. 3, pp. 120-154, 1971. [3] R. Ahlswede, Multi-way communication channels, in Proc. IEEE International Symposium on Information Theory, Tsahkadsor, Armenian S.S.R., Sept. 1971. [4] R. Ahlswede, The capacity region of a channel with two senders and two receivers, Annals Probabil., vol. 2, no. 5, 1974. [5] A. Carleial, Interference channels, IEEE Trans. Inform. Theory, vol. 24, no. 1, pp. 60-70, Jan. 1978. [6] T. Cover and A. E. El Gamal, Capacity theorems for the relay channel, IEEE Trans. Inf. Theory, vol. 25, no. 5, pp. 572- 584, Sep. 1979. [7] M. R. Aref, Information ow in relay networks, Ph.D. dissertation, Stanford Univ., Stanford, CA, Oct. 1980 [8] T. Cover, A. El Gamal and M. Salehi, Multiple access channels with arbitrarily correlated sources, IEEE Trans. Inform. Theory, vol. 26, no. 6, pp. 648-657, Nov. 1980. [9] T. Han and K. Kobayashi, A new achievable rate region for the interference channel, IEEE Trans. Inform. Theory, vol. 27, no. 1, pp. 49-60, Jan. 1981.

134 [10] H. Sato, The capacity of the Gaussian interference channel under strong interference, IEEE Trans. Inform. Theory, vol. 27, no. 6, pp. 786-788, Nov. 1981. [11] F. M. J.Willems, Informationtheoretical Results for the Discrete Memoryless Multiple Access Channel, Doctor in de Wetenschappen Proefschrift dissertation, Katholieke Univ. Leuven, Leuven, Belgium, Oct. 1982. [12] M. H. M. Costa, Writing on dirty paper, IEEE Trans. Inf. Theory, vol. 27, pp. 439-441, 1983. [13] F. Willems and E. van der Meulen, The discrete memoryless multiple-access channel with cribbing encoders, IEEE Trans. Inform. Theory, vol. 31, no. 3, pp. 313-327, May 1985. [14] M. H. M. Costa, On the Gaussian interference channel, IEEE Trans. Inform. Theory, vol. 31, no. 5, pp. 607-615, Sep. 1985. [15] M. H. M. Costa, A. E. Gamal, The capacity region of the discrete memoryless interference channel with strong interference, IEEE Trans. Inform. Theory, vol. 33, no. 5, pp. 710-711, Sept. 1987. [16] J. A. Thomas, Feedback can at most double Gaussian multiple access channel capacity, IEEE Trans. Inform. Theory, vol. 33, no. 5, pp. 711-716, Sept. 1987. [17] T. Cover and J. Thomas, Elements of Information Theory, New York: Wiley, 1991. [18] G. J. Foschini, Layered space-time architecture for wireless communication in a fading environment when using multi-element antennas, Bell Laboratories Technical Journal, p. 41, Oct. 1996. [19] I. E. Telatar, Capacity of multiple-antenna Gaussian channels, Europian Transactions on Telecommunications, vol. 10, p. 585, Nov. 1999.

135 [20] P. Gupta and P. R. Kumar, Toward an information theory of large networks: An achievable rate region, IEEE Trans. Inf. Theory, vol. 49, no. 8, pp. 18771894, Aug. 2003 [21] A. Sendonaris, E. Erkip and B. Aazhang, User cooperation diversity-Part I: System description, IEEE Trans. Comm., vol. 51, no. 11, pp. 1927-1938, Nov. 2003. [22] I. Sason, On achievable rate regions for the Gaussian interference channel,IEEE Trans. Inform. Theory, vol. 50, no. 6, pp. 1345 - 1356, June 2004. [23] J. N. Laneman, D. N. C. Tse, and G. W. Wornell, Cooperative diversity in wireless networks: Ecient protocols and outage behavior, IEEE Trans. Inform. Theory, vol. 50, no. 12, pp. 3062-3080, Dec. 2004. [24] B. Bhargava, X. Wu, Y. Lu and W. Wang, Integrating heterogeneous wireless technologies: a cellular aided mobile ad hoc network (CAMA), Journal on Mobile Networks and Applications, vol. 9, no. 4, pp. 393-408, Aug. 2004. [25] A. El Gamal, S. Zahedi, Capacity of a class of relay channels with orthogonal components, IEEE Trans. Inform. Theory, vol. 51, no. 5, pp. 1815-1817, May 2005. [26] G. Kramer, M. Gastpar and P. Gupta, Cooperative strategies and capacity theorems for relay networks, IEEE Trans. Inform. Theory, vol. 51, no.5, pp. 3037-3063, May 2005. [27] Y. Steinberg, S. Shamai, Achievable rates for the broadcast channel with states known at the transmitter in Proc. IEEE International Symposium on Information Theory, Adelaide, Australia, Sept. 2005. [28] G. Kramer, Review of rate regions for interference channels,Proc. Intl Zurich Seminar on Communications, Zurich, Switzerland, Feb. 2006.

136 [29] D. Tuninetti, The interference channel with generalized feedback (IFC-GF), in Proc. Information Theory and Applications Workshop, UCSD, California, Feb. 2006. [30] 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. [31] A. Jovicic, P. Viswanath, Cognitive radio: An information-theoretic perspective, Proc. IEEE International Symposium on Information Theory, Seattle, USA, July 2006. [32] H. Weingarten, Y. Steinberg and S. Shamai, The capacity region of the Gaussian multiple-input multiple-output broadcast channel, IEEE Trans. Inform. Theory, vol. 52, no. 9, pp. 3936-3964, Sept. 2006. [33] H. Weingarten, T. Liu and S. Shamai, The capacity region of degraded multiple input multiple output compound broadcast channel, submitted to IEEE Trans. Inform. Theory, Jan. 2007. [34] O. Sahin, E. Erkip, Dynamic resource allocation for multi source-destination relay networks, Proc. 41st Annual Conference on Information Sciences and Systems, Baltimore, MD, Feb. 2007. [35] Y. Liang and V.V. Veeravalli, Cooperative relay broadcast channels, IEEE Trans. Inform. Theory, vol. 53, no. 5, pp. 900-928, May 2007. [36] T. Liu and P. Viswanath, An extremal inequality motivated by multi terminal information theoretic problems,, IEEE Trans. Inform. Theory, vol. 53, no. 5, pp. 1839-1851, May 2007. [37] T.M. Cover and Y.H.Kim, Capacity of a class of deterministic relay channels,in Proc. IEEE International Symposium on Information Theory, Nice, France, June 2007.

137 [38] B. Nazer and M. Gastpar, The case for structured random codes in network communication theorems, Proc. Information Theory Workshop, Lake Tahoe, California, Sept. 2007. [39] Y. Liang and G. Kramer, Rate Regions for Relay Broadcast Channels, IEEE Trans. Inform. Theory, vol. 53, no. 10, pp. 3517-3535, Oct. 2007. [40] W. Wei, S. Vishwanath, A. Arapostathis, Capacity of a class of cognitive radio channels: Interference channels with degraded message sets, IEEE Trans. Inform. Theory, vol. 53, no. 11, pp. 4391-4399, Nov. 2007. [41] O. Sahin and E. Erkip, On achievable rates for interference relay channel with interference cancellation, Proc. 41st Annual Asilomar Conference on Signals, Systems, and Computers, Pacic Grove, CA, Nov. 2007. [42] G. Kramer, Topics in Multi-User Information Theory, Foundations and Trends in Communications and Information Theory, vol. 4, no. 4-5, pp. 265-444, 2007. [43] O. Sahin and E. Erkip, Achievable rates for the Gaussian interference relay channel,Proc. Global Telecommunications Conference, Washington, DC, Nov. 2007. [44] S. Annapureddy and V. Veeravalli, Sum capacity of the Gaussian interference channel in the low interference regime, in Proc. Information Theory and Applications Workshop, UCSD, California, Feb. 2008. [45] G. Kramer, Topics in Multi-User Information Theory, now Publishers, 2008. [46] M. Wigger, Cooperation on the multiple-access channel, PhD thesis, ETH Zurich, Sept. 2008. [47] W. Nam, S.Y. Chung and Y.H. Lee, Capacity bounds for two-way relay channel, Proc. Intl Zurich Seminar on Communications, Zurich, Switzerland, March 2008.

138 [48] M. P. Wilson, K. Narayanan, H. Pster and A. Sprintson, Joint physical layer coding and network coding for bi-directional relaying, submitted to IEEE Trans. Inform. Theory, April 2008. [49] R. Dabora, I. Maric and A. Goldsmith, Relay strategies for interferenceforwarding, in Proc. IEEE Information Theory Workshop, Porto, Portugal, May 2008. [50] A. Sanderovich, S. Shamai, Y. Steinberg and G. Kramer, Communication via decentralized processing, IEEE Trans. Inform. Theory, vol. 54, no. 7, pp. 30083023 , July 2008. [51] H. Chong, M. Motani, H. Garg and H. El Gamal, On the Han-Kobayashi region for the interference channel, IEEE Trans. on Inform. Theory, vol. 54, no. 7, pp. 3188-3195, July 2008. [52] I. Maric, R. Dabora and A. Goldsmith, On the capacity of the interference channel with a relay, in Proc. IEEE International Symposium on Information Theory, Toronto, Canada, July 2008. [53] S. Sridharan, S. Vishwanath, S.A. Jafar and S. Shamai, On the capacity of cognitive relay assisted Gaussian interference channel in Proc. IEEE International Symposium on Information Theory, Toronto, Canada, July 2008. [54] S. I. Bross, A. Lapidoth and M. A. Wigger, The Gaussian MAC with conferencing encoders, in Proc. IEEE International Symposium on Information Theory, Toronto, Canada, July 2008. [55] I. Maric, R. Dabora and A. Goldsmith, On the capacity of the interference channel with a relay, in Proc. IEEE International Symposium on Information Theory, Toronto, Canada, July 2008.

139 [56] S. Sridharan, S. Vishwanath, S.A. Jafar and S. Shamai, On the capacity of cognitive relay assisted Gaussian interference channel in Proc. IEEE International Symposium on Information Theory, Toronto, Canada, July 2008. [57] W. Yu and L. Zhou, Gaussian Z-interference channel with a relay link: achievability region and asymptotic sum capacity, submitted to IEEE Trans. on Inform. Theory, Sept. 2008. [58] I. Maric, N. Liu and A. Goldsmith, Encoding against an interferers codebook, Proc. Allerton Conference on Communications, Control and Computing, Monticello, IL, Sept. 2008. [59] O. Sahin and E. Erkip, Cognitive relaying with one-sided interference, in Proc. 42nd Annual Asilomar Conference on Signals, Systems, and Computers, Pacic Grove, CA, Nov. 2008. [60] R. Dabora, I. Maric and A. Goldsmith, Interference forwarding in multiuser networks, Proc. Global Telecommunications Conference, New Orleans, LA, Nov. 2008. [61] I. Maric, R. Dabora and A. Goldsmith,Generalized relaying in the presence of interference, Proc. 42nd Annual Asilomar Conference on Signals, Systems, and Computers, Pacic Grove, CA, Nov. 2008. [62] R.H. Etkin, D. N. C. Tse and H. Wang, Gaussian interference channel capacity to within one bit, IEEE Trans. Inform. Theory, vol. 54, no. 12, pp. 5534-5562, Dec. 2008. [63] M. Abouelseoud and A. Nosratinia, The gateway channel: Outage analysis, in Proc. IEEE Global Telecommunications Conference, New Orleans, LA, Dec. 2008.

140 [64] X. Shang, G. Kramer and B. Chen, A new outer bound and noisy-interference sum-rate capacity for the Gaussian interference channels, IEEE Trans. Inform. Theory, vol. 55, no. 2, pp. 689-699, Feb. 2009. [65] D. Gunduz, O. Simeone, A. Goldsmith, H.V. Poor and S. Shamai, Multiple multicasts with the help of a relay, submitted to IEEE Trans. Inform. Theory, Feb. 2009. [66] O. Sahin, E. Erkip and O. Simeone, Interference channel with a relay: models, relaying strategies, bounds, in Proc. Information Theory and Applications Workshop, UCSD, California, Feb. 2009. [67] A. S. Motahari and A. K. Khandani, Capacity bounds for the Gaussian interference channel, IEEE Trans. Inform. Theory, vol. 55, no. 2, pp. 620-643, Feb. 2009. [68] D. G und uz, A. Yener, A. Goldsmith and H. V. Poor, Multi-way relay channel, in Proc. IEEE International Symposium on Information Theory, Seoul, South Korea, June 2009. [69] H. Ghozlan, Y. Mohasseb, H. E. Gamal and G. Kramer, The MIMO wireless switch: Relaying can increase the multiplexing gain, in Proc. IEEE International Symposium on Information Theory, Seoul, South Korea, June 2009. [70] X. Shang, B. Chen, G. Kramer and H.V. Poor, Noisy-interference sum-rate capacity of parallel Gaussian interference channels, submitted to IEEE Trans. on Inform. Theory, March 2009. [71] S. W. Choi and S. Chung, On the separability of parallel Gaussian interference channels, Proc. IEEE International Symposium on Information Theory, Seoul, South Korea, June 2009.

141 [72] O. Sahin, O. Simeone and E. Erkip, Interference channel aided by an infrastructure relay, in Proc. IEEE International Symposium on Information Theory, Seoul, Korea, June 2009. [73] L. Sankar, X. Shang, E. Erkip and H. V. Poor, Ergodic fading interference channels: sum-capacity and separability, submitted to IEEE Trans. on Inform. Theory, June 2009. [74] X. Shang, B. Chen, G. Kramer and H. V. Poor, Capacity regions and sum-rate capacities of vector Gaussian interference channels, submitted to IEEE Trans. on Inform. Theory, July 2009. [75] V.R. Cadambe and S.A. Jafar, Interference alignment and degrees of freedom of the K -user interference channel, IEEE Trans. Inform. Theory, vol. 54, no. 8, pp. 3425-3441, Aug. 2009. [76] V. R. Cadambe and S. A. Jafar, Parallel Gaussian interference channels are not always separable, IEEE Trans. on Inform. Theory, vol. 55, No. 9, pp. 3983-3990, Sep. 2009. [77] Y. Tian and A. Yener, The Gaussian interference relay channel with a potent relay, to appear Proc. IEEE Global Telecommunications Conference, Honolulu, Hawaii, Dec. 2009. [78] O. Sahin, O. Simeone and E. Erkip, Interference channel with an out-of-band relay, to be submitted to IEEE Trans. on Inform. Theory.

142

List of Publications

1. O. Sahin, O. Simeone and E. Erkip, Gaussian interference channel aided by a relay with out-of-band reception and in-band transmission, under review, IEEE Trans. on Communications. 2. O. Sahin, O. Simeone and E. Erkip, Interference channel with an out-of-band relay, under review, IEEE Trans. on Inform. Theory. 3. O. Sahin, O. Simeone, E.Erkip, Interference channel with a half-duplex out-ofband relay, To appear, Proc. IEEE International Symposium on Information Theory, Austin, TX, June 2010. 4. O. Sahin, O. Simeone and E. Erkip, Interference channel aided by an infrastructure relay, in Proc. IEEE International Symposium on Information Theory, Seoul, Korea, June 2009. 5. O. Sahin, E. Erkip and O. Simeone, Interference channel with a relay: models, relaying strategies, bounds, in Proc. Information Theory and Applications Workshop, UCSD, California, Feb. 2009. 6. O. Sahin and E. Erkip, Cognitive relaying with one-sided interference, in Proc. 42nd Annual Asilomar Conference on Signals, Systems, and Computers, Pacic Grove, CA, Nov. 2008. 7. O. Sahin and E. Erkip, Cooperative interference management: The role of cognitive relaying, IEEE Communication Theory Workshop (invited talk), St. Croix, USVI, May 2008.

143 8. O. Sahin and E. Erkip, Achievable rates for the Gaussian interference relay channel,Proc. Global Telecommunications Conference, Washington, DC, Nov. 2007. 9. O. Sahin and E. Erkip, On achievable rates for interference relay channel with interference cancellation, in Proc. 41st Annual Asilomar Conference on Signals, Systems, and Computers, Pacic Grove, CA, Nov. 2007. 10. O. Sahin, E. Erkip, Dynamic resource allocation for multi source-destination relay networks, in Proc. 41st Annual Conference on Information Sciences and Systems, Baltimore, MD, Feb. 2007. 11. O. Sahin, E. Erkip and D. Goodman. Iterative power control for wireless multimedia communications, in Proc. IEEE International Conference on Acoustics, Speech, and Signal Processing, Toulouse, France, May 2006. 12. O. Sahin, J. Yang and M. Ghosh, An ecient relaying scheme for multiple users in spectral ecient cooperative networks, Workshop on Cooperative Communications, Polytechnic University, Brooklyn, September 2005.

You might also like