Performance Analysis and Optimization of Wireless Heterogeneous Networks. Yicheng Lin

Transcription

1 Performance nalysis and Optimization of Wireless Heterogeneous Networks by Yicheng Lin thesis submitted in conformity with the requirements for the degree of Doctor of Philosophy Graduate Department of Electrical and Computer Engineering University of Toronto c Copyright 2015 by Yicheng Lin

2 bstract Performance nalysis and Optimization of Wireless Heterogeneous Networks Yicheng Lin Doctor of Philosophy Graduate Department of Electrical and Computer Engineering University of Toronto 2015 Heterogeneity is becoming a key characteristic of future generations of wireless cellular networks. By deploying different low-power access points in addition to the traditional macro base-stations (BSs), various types of heterogeneous networks (HetNets) can be formed to improve the coverage and transmission rate. Through developing efficient algorithms, analyzing performance and optimizing systems design parameters, this thesis studies three representative HetNet architectures, namely shared relays, multi-tier HetNets, and distributed antenna systems (DS). Relays utilize an in-band backhaul and have low deployment cost. The first part of the thesis studies a sectorized cellular system with relays and, in particular, focuses on a novel architecture in which each relay is shared by M adjacent cells. We develop a joint user scheduling, resource allocation and beamforming algorithm for shared relaying to maximize the proportionally fair utility, taking into account practical in-band backhaul constraints. The benefit of shared relaying in terms of rate and utility is shown in a system-level simulation. The use of shared wireless medium for the backhaul of relays induces half-duplex loss in relay system, which fundamentally limits the performance gain. Motivated by this fact, the second part of this thesis studies multi-tier HetNets with dedicated wired backhaul. Considering random deployment of BSs, we derive the closed-form mean proportionally fair utility of multi-tier HetNets in both downlink and uplink. The optimal user association and spectrum allocation schemes are obtained by optimizing this utility, and are validated with simulations. In a multi-tier HetNet, BSs typically transmit independently. The third part of the thesis studies a DS in which distributed antennas are connected via wired backhaul to the central controller and can cooperatively transmit to users. This thesis analyzes the downlink spectral efficiency of DS where antennas are randomly deployed. We show that for a regular DS with fixed cell boundaries, selective transmission outperforms blanket transmission in most cases, and the cell-edge spectral efficiency under blanked transmission is upper bounded by a constant. ccounting for random user distribution, the user-centric DS layout, where no explicit cell boundaries exist, is also analyzed, which is shown to outperform the regular DS from a network perspective. ii

3 cknowledgements This thesis not only represents a summary of my research work, but also remarks a milestone of four years of continuous effort at University of Toronto. Upon completion of this work, I would like to use this opportunity to thank many who supported me and who shared the four-year memory with me. First and foremost, I would like to express sincere gratitude to my supervisor Professor Wei Yu for taking me on as a PhD student four years ago and for constantly providing me invaluable guidance. It was due to his cheerful enthusiasm and encouragement that I was able to accomplish this research work in a timely and respectable manner. His knowledge, insights, and rigorous attitude towards research projects have greatly influenced me, and I am forever grateful for this. I thank Professor Ben Liang for being my committee member and for his collaboration with me on part of my research topic. His suggestions significantly improved the quality of Chapter 3 of this thesis. Thanks to my committee member Professor Raviraj dve for his careful proof reading and insightful suggestions on this thesis. Professor Matthew C. Valenti graciously agreed to be my external examiner, and his feedbackin the final stagewasinvaluable. I alsowould liketo thank Wei Baoforhis collaboration with me on the uplink utility derivation in multi-tier HetNet, and others who offered helpful discussions: Yuhan Zhou, Gokul Sridharan, Hayssam Dahrouj, and Kianoush Hosseini, and others. The joy times I had every Friday and holiday no matter the badminton games or the card games will become a special memory of mine when looking back to my PhD life. Thanks to Siyu Liu for organizing all these events, and those who attended regularly: Yuhan Zhou, Binbin Dai, Dan Fang, Wei Bao, Qiang Xiao, Chu Pang, Gokul Sridharan, Pratik Patil, Wei Wang, Huiyuan Xiong, Caiyi Zhu, Wanyao Zhao, Chunpo Pan, Wenwen Gao, lice Gao, Cecilia Liu, Louis Tan, etc. I also thank Wei Bao and Zhi Zeng for organizing many board-game nights and I really enjoyed the fun we had together. My very special gratitude belongs to my dearest wife Yu Xiao. With her accompany my life has become more organized and my spare time has become more colorful. I owe her for my life time. Finally, I acknowledge the University of Toronto Fellowship program for providing funding for me and all my other friends during the four years time. iii

4 Contents 1 Introduction Heterogeneous Network rchitecture Design Challenges of HetNets Thesis Outline and Main Contributions Fair Scheduling and Resource llocation for Shared Relaying Introduction Related Work Spatial Degrees of Freedom (DoF) nalysis System Model Transmit and Receive Strategies Utility Maximization Framework lgorithm Design for Shared Relaying Problem Formulation Resource llocation and Scheduling Update of the Lagrangian Dual Variables Complexity and Implementation Issues lgorithm Design for Separate Relaying Performance Evaluation Summary User ssociation and Spectrum llocation in Multi-Tier HetNets Introduction Basic System ssumptions Related Work Main Contribution System Model Mean User Utility Mean Logarithm of Per-User Spectrum Mean Logarithm of Coverage Probability in the Downlink Mean Logarithm of Coverage Probability in the Uplink Utility Optimization Spectrum Sharing in the Downlink Orthogonal Spectrum Partition in the Downlink iv

5 3.4.3 Spectrum Sharing in the Uplink Orthogonal Spectrum Partition in the Uplink Numerical Results Validation of the Optimization of User ssociation and Spectrum Partition Optimal Bias and Spectrum Partition Ratio under Different System Parameters Summary Downlink Spectral Efficiency of Distributed ntenna Systems Introduction Related Work Main Results System Model for DS Spectral Efficiency nalysis under MISO Channel Spectral Efficiency of Regular DS from a Location-Specific Perspective Blanket Transmission Selective Transmission Comparison of Blanket and Selective Transmissions Spectral Efficiency of Regular and User-Centric DS from a Network Perspective Regular DS with Fixed Cell Boundaries User-Centric Selective DS Comparison of ll Schemes Summary Conclusion 77 Upper Bound of Mean Per-User Spectrum 79 B Optimal Uplink User ssociation under Spectrum Sharing 81 Bibliography 83 v

6 List of Tables 2.1 Comparison of Different Schemes in terms of Per-SectorPerformanceMetric, with P r = 1 3 P s 26 vi

7 List of Figures 1.1 Relay scenarios. (a) Separate relays at edge of each sector. (b) Coordinated relay shared by three adjacent sectors of different cells Multi-Tier Hetnet scenarios. (a) Three separated tiers of access points: macro, micro, and pico BSs. (b) Superposition of three tiers n illustration of the distributed antenna system PPP topology of a HetNet. (a) One tier. (b) Three tiers: squares are macro BSs, circles are pico BSs, diamonds are femto BSs n example of a 2-tier HetNet with cell range expansion Relay scenarios. (a) Separate relays at edge of each sector. (b) Coordinated relay shared by three adjacent sectors of different cells Half-duplex two-phase relay frame structure bstract model of a shared relaying system, where S stands for source or basestation, and D stands for destination or user Block diagram of the resource allocation and scheduling process (lgorithm 2) of the shared relaying scheme CDF comparison of the separate and the shared relaying schemes, with P r = 1 3 P s and (a) K = 20, (b) K = Per-sector performance in terms of (a) utility, (b) 5% cell edge rate, as a function of P r, with K = Convergence of rate supply and rate demand of the shared relay (a) with subgradient method (b) with discrete λ-search and subgradient method. K = 40, P r = 1 3 P s, s = n example of a 2-tier HetNet with cell range expansion Two-tier cell topology with and without biasing Downlink performances vs. bias of tier-1 under spectrum sharing. α = 4, {P 1,P 2 } = {46,20}dBm, {λ 1,λ 2 } = {0.01,0.09}λ u, {τ 1,τ 2 } = {2,2} Uplink performances vs. bias of tier-1 under spectrum sharing. α = 4, {P 1,P 2 } = {46,20}dBm, P u = 20dBm, ǫ = 1, {λ 1,λ 2 } = {0.01,0.09}λ u, {τ 1,τ 2 } = {2,2} Downlink performances vs. inter-tier spectrum partition. The x-axis is the number of subcarriers (out of 2048) allocated to tier-1. Other subcarriers are allocated to tier-2. α = 4, {P 1,P 2 } = {46,20}dBm, {λ 1,λ 2 } = {0.01, }λ u, {τ 1,τ 2 } = {2,2} vii

8 3.6 Uplink performances vs. inter-tier spectrum partition. The x-axis is the number of subcarriers (out of 2048) allocated to tier-1. Other subcarriers are allocated to tier-2. α = 4, {P 1,P 2 } = {46,20}dBm, P u = 20dBm, ǫ = 1, {λ 1,λ 2 } = {0.01, }λ u, {τ 1,τ 2 } = {2,2} Optimal downlink bias of tier-1 vs. intensity of tier-2 under spectrum sharing. α = 4, {P 1,P 2 } = {46,20 26}dBm, {λ 1,λ 2 } = {0.01, }λ u, {τ 1,τ 2 } = {2,2} Optimal uplink bias of tier-1 vs. power control factor under spectrum sharing. α = , {P 1,P 2 } = {46,20}dBm, P u = 20dBm, ǫ = 0 1, {λ 1,λ 2 } = {0.01,0.09}λ u Optimal downlink spectrum proportion and bias of tier-1 vs. intensity of tier-2. α = 4, {P 1,P 2 } = {46,20 26}dBm, {λ 1,λ 2 } = {0.01, }λ u, {τ 1,τ 2 } = {2,2} Optimal uplink spectrum proportion and bias of tier-1 vs. power control factor. α = 4, {P 1,P 2 } = {46,20 26}dBm, P u = 20dBm, ǫ = 0 1, {λ 1,λ 2 } = {0.01,0.09}λ u, {τ 1,τ 2 } = {2,2} n illustration of the distributed antenna system The regular DS layout. The boundary of the cell under consideration is modeled as a circle, and the boundaries of interfering cells can be arbitrary Intersection area of the cell and user disc with radius D Grid cellular vs. regular DS: spectral efficiency as a function of the cell-center-to-user distance and the antenna intensity (average antenna number per cell). α = 4, K = 1, R = Regular DS: spectral efficiency as a function of the cell-center-to-user distance and the antenna number per port. α = 4, N = 24, R = Regular DS: spectral efficiency as a function of the cell-center-to-user distance and the path loss exponent. N = 6, K = 1, R = Regular DS: cell-edge spectral efficiency as a function of the antenna intensity (average antenna number per π(1000m) 2 area) and the cell radius. K = 1, r = R Network spectral efficiency of regular DS as a function of the cell radius. K = 1, λ U = 1 π(1000m) 2, and (a) α = 4, (b) α = Network spectral efficiencies of all schemes as a function of the antenna intensity (average antenna number per π(1000m) 2 area). Regular DS is plotted with the optimal cell radius. K = 1, λ U = 1 π(1000m) viii

9 Chapter 1 Introduction Future cellular networks are characterized by ubiquitous coverage and significantly higher data rates in comparison to conventional cellular networks. Wireless operators and service providers face challenges in providing innovative solutions in order to satisfy the demands for consistent quality-of-service (QoS) and real-time data services. Traditional approaches include adding new radio spectrum, deploying multiple antennas at transmitters and receivers, and implementing more efficient modulation and coding schemes. However, these methods have practical constraints, especially in crowded urban environments. First, remote users in the cellular wireless systems are exposed to multiple base-stations (BSs) which are expected to fully reuse the frequency bands, and thus may experience low received signal-to-interferenceand-noise ratio (SINR). Such a cell-edge performance deterioration problem has fundamentally limited traditional cellular topology in achieving the aforementioned goals. Secondly, in a dense urban area with various artificial and natural structures (e.g., buildings, vegetation), users may be located at coverage holes where no line-of-sight (LOS) propagation is available, but only reflected and diffracted signals. The resulting shadowing effect and penetration loss further limit the performance of wireless systems. One way to improve the performance of a cellular system while maintaining it as a homogeneous network is to densify it by adding more macro BSs in the network. However, reducing the inter-site distance can only be pursued to a certain extent as it is more and more difficult and expensive to find proper site locations for new macro BSs, especially in places like city centers. n alternative approach is to overlay the conventional macro cells with a diverse set of low-power and low-cost access points to form the so-called heterogeneous networks(hetnets). These small access points include cooperative relays [1], pico-bss [2], femto-bss [3], and distributed antennas [4, 5]. Cooperative relays utilize in-band backhauls and hence has low deployment cost and the advantage that no new infrastructure extension is needed. On the contrary, femto-bss, pico-bss, and distributed antennas use a dedicated backhaul, which may be wired or wireless. In particular, pico-bss are typically deployed by wireless operators to extend coverage to public indoor areas where outdoor signals cannot reach, and to add capacity in places, known as hot-spots, where very dense voice and data usage is required; while femto-bss are typically designed and installed privately for indoor coverage in home and small business environments. To further achieve ubiquitous coverage, distributed antennas can be deployed remotely and scattered across the cell, connecting with the serving BSs via dedicated high-speed fronthaul links, and can cooperatively transmit to user terminals. Heterogeneity is assumed to be a key feature of next generation wireless protocols [6 8]. This thesis investigates various heterogeneous wireless architectures. 1

10 Chapter 1. Introduction 2 Transmission in the 1st phase BS Transmission in the 2nd phase Transmission in the 1st phase BS Transmission in the 2nd phase MS MS MS MS RS MS MS MS MS MS RS MS RS MS RS MS MS MS BS MS MS BS BS MS MS BS (a) (b) Figure 1.1: Relay scenarios. (a) Separate relays at edge of each sector. (b) Coordinated relay shared by three adjacent sectors of different cells. 1.1 Heterogeneous Network rchitecture HetNet is a mixture of different types of wireless access points. The architecture of HetNets can be classified according to how the backhaul is provisioned when low-power access points are added to the macro network. To minimize the deployment cost, the most convenient option is to deploy and connect two-hop fixed relays to BSs through an in-band backhaul [9]. Relaying has been adopted as a cost-effective option in standards such as long-term evolution (LTE), and various types of relays are considered, e.g., amplify-and-forward (F) relays and decode-and-forward(df) relays. Fixed relays can be deployed in the cell to enhance coverage. However, they are typically not designed with intercell interference in mind. In fact, relays from different neighboring cell sectors are often close to each other in distance, and consequently can create more intercell interference than that in a conventional cellular network. The traditional relay system is shown in Fig. 1.1(a) and is referred as the separate relaying. The concept of the shared relay is introduced as one way to tackle this intercell interference problem. The basic idea, which was firstly proposed in [10] and [11], is to place a multi-antenna relay at the intersection of adjacent cells, which can be thought of as a coordinated version of multiple separate relays from different cells. The shared relay architecture is shown in Fig. 1.1(b). The shared relay can jointly process receive/transmit signals, avoiding inter-relay interference in both the BS-to-relay transmission and the relay-to-user transmission. This thesis addresses the resource allocation aspects of shared relaying, including the spectrum partition between feeder links (i.e., BS-relay links) and access links (i.e., BS-user or relay-user links), the power allocation and user scheduling at BSs and relays. For both conventional relaying and shared relaying, the use of the in-band backhaul inevitably induces a half-duplex loss in transmission. Take a single carrier and point-to-point communication system as an example: a single bit is transmitted through the backhaul and access links in two phases, and consequently the average transmitted bits per phase is reduced to half. This fundamentally limits the gain achieved by the relaying approach.

11 Chapter 1. Introduction 3 macro macro micro micro micro pico pico pico pico pico macro micro pico micro macro pico pico micro (a) (b) Figure 1.2: Multi-Tier Hetnet scenarios. (a) Three separated tiers of access points: macro, micro, and pico BSs. (b) Superposition of three tiers. To remove this half-duplex bottleneck, independent backhaul traffic should be provided to each lowpower access point either via wired connections or dedicated out-of-band spectrum. Various types of independent low-power access points can form small cells, overlaid onto macro cells, to offload macro cell users. In this architecture, the macro cells offer basic long-range coverage, while the small cells provide short-range but high-quality communication to nearby users. While the conventional cellular system can be regarded as a single tier network, the hierarchical overlaid access points in HetNets form a so-called multi-tier network. Fig. 1.2 shows how a HetNet can be seperated into three homogeneous networks with macro, micro, and pico BSs, where circles in Fig. 1.2(a) represent each separate tiers and the circle in Fig. 1.2(b) represents the superposition of the three tiers. Different tiers of access points in HetNets have distinct system parameters such as transmission power, path loss and shadowing attenuation, and spatial deployment intensity (or density). For example, macro BSs usually have higher power and are more sparsely deployed than pico BSs, while pico BSs are deployed in greater number but with smaller power. Because of these different parameters across tiers, the traditional solution to the user-bs association problem in a single-tier network may not be optimal in a multi-tier network. This thesis studies the influence of distinct system parameters on HetNet performance, and addresses the problem of optimal user-bs association and spectrum allocation among tiers in a multi-tier layout. With dedicated and independent backhaul, each small-cell BS in a multi-tier HetNet operates independently, and hence creating a more complex interference environment as interfering nodes get closer. In this dense network, it is natural to consider the option of allowing neighboring access points to cooperatively transmit to users to reduce interference. Cooperation between BSs can be traced back to 3G CDM systems [12], where it is used to maintain communication quality during soft handover. Recently, practical interest for transmitter cooperation is still ongoing. With this scheme, the transmitters share control signal, channel state information (CSI) and data symbols for all users via central controller, at the cost of the inevitable information exchange among coordinated multiple sites. Cooperation can be implemented at different levels, e.g., [13]: MIMO cooperation and Interference coordination. The MIMO cooperation is also known as network MIMO or coordinated multi-point (CoMP) [14, 15], where transmitters share CSI and data of their users via dedicated high-capacity delay-free backhaul links, at

12 Chapter 1. Introduction 4 Figure 1.3: n illustration of the distributed antenna system. the cost of high burden upon channel estimation and backhaul design. In the interference coordination, transmitters share the CSI and have knowledge of both the direct and interfering links. The availability of CSI allows BSs to coordinate strategies such as power allocation, beamforming, and user scheduling in time and frequency [16]. This category has relatively light backhaul burden. It is obvious that the key difference between the two cooperation categories is whether user data is shared amongst BSs or not. s a heterogenous architecture, the distributed antenna system (DS) [4, 5, 17] is one kind of MIMO cooperation scheme under the principle of user data sharing, where the notion of independent BSs is replaced with numerous remote antenna ports 1. In DS, instead of deploying antennas centrally at the BSs as in the traditional cellular systems, by deploying multiple remote antenna ports in coverage holes and connecting them with a central processor via dedicated high-speed backhaul links, detrimental effects such as shadowing and indoor penetration loss can be effectively mitigated. The main advantage of DS is that it statistically reduces the distance of a user to its nearest access point, thereby enhancing the received signal quality. In this way, the user performance in DS is less sensitive to the user location within each cell, and the cell-edge performance degradation becomes less severe. Fig. 1.3 shows an example of the DS layout, where antenna ports are randomly deployed within each cell boundary to serve users. One of the objectives of this thesis is to quantify the benefit of DS by characterizing the downlink spectral efficiency of DS schemes. 1.2 Design Challenges of HetNets Interference Management The main challenge that limits the achievable spectral efficiency in a multi-cell network is the intercell interference (ICI) problem. Interference management is more crucial in HetNets as compared to conventional cellular single tier networks, as the deployment of these low-power infrastructure induces more interference sources, and consequently creates a more complex wireless environment. s a result, more efficient interference management techniques are required. Ways to mitigate inter-cell interferences include power control, scheduling, and frequency planning. There are some works on joint power control and user scheduling, e.g., see [16, 18] as examples. The power control and scheduling schemes, however, 1 n antenna port is a collection of multiple physically co-located antennas.

13 Chapter 1. Introduction 5 (a) (b) Figure 1.4: PPP topology of a HetNet. (a) One tier. (b) Three tiers: squares are macro BSs, circles are pico BSs, diamonds are femto BSs. are dynamic and depends on instantaneous channel information, and its average performance is thus hard to analyze in a theoretical framework. The most representative static and semi-static strategies include the fractional frequency reuse (FFR) [19] and the soft frequency reuse (SFR) [20], which aim to increase the frequency reuse factor for cell-edge users and reduce the transmission power for cell-center users. nother approach is orthogonal spectrum partition [8, 21, 22], where the transmission of the macro tier is periodically muted on a certain fraction of resources. Muting is implemented by the use of almost blank subframes in the 3rd Generation Partnership Project (3GPP) LTE. These two approaches essentially use the same principle and both reduce the spectrum usage, but is simple to implement. Shared relaying and DS are two effective schemes for mitigating interference: shared relaying can eliminate inter-relay interference by replacing several adjacent separate relays with a coordinated relay shared by several cells; while DS is able to improve cell-edge performance by distributing antennas across the coverage of the cell. This thesis studies these two architectures. Besides, orthogonal spectrum partitioning among tiers is adopted in multi-tier HetNets as one way to eliminate the inter-tier interference. This thesis analyzes the optimal way to partition the spectrum in multi-tier HetNets to achieve a tradeoff between spectrum usage and interference mitigation. Irregular Node Placement Traditional cellular networks rely on the assumption of regular hexagonal for optimization and analysis. The coverageprobabilityand rateofsuch a planned topologyis shownto be an upper bound forpractical irregular site deployments [23], as it can maximize the average inter-site distance and thus minimize the average interference. Besides, the system-level evaluation of the hexagonal model usually only considers two or three tiers of interfering BSs with wrap-round method and neglect other interference, adding another source of optimism. The performance analysis of practical networks, however, are more pessimistic due to the irregular site placements. The introduction of hierarchical access points in practical

14 Chapter 1. Introduction 6 Biased coverage of the pico-bs Unbiased coverage of the pico-bs user user user user macro user pico Figure 1.5: n example of a 2-tier HetNet with cell range expansion. HetNets intensifies such irregularity. This thesis uses a stochastic network model instead of the hexagonal model for the analysis of multitier HetNets and DS, which can be useful in performance analysis and optimization. Unlike our work on relaying, which assumes hexagonal cells with sectorization, we assume that in the stochastic model cells are not sectorized and each BS has an omni-directional antenna. The HetNet topology appears almost random in practice[24], as access points are placed at arbitrary locations, e.g., hot-spots(pico-cell), deadspots (DS), and end-user defined installations (femto-cell). Such randomness introduces uncertainties into network performance analysis. To account for the randomness in the HetNet topology, we can model the transmitter locations as Poisson point processes (PPPs) with certain intensities. Fig. 1.4(a) shows a single-tier network under the PPP model, where the two dimensional Euclidean space is partitioned into a Poisson Voronoi tessellation [25]; while multi-tier PPP cell topology forms a multiplicatively weighted Poisson Voronoi tessellation as shown in Fig. 1.4(b). The performance of this model serves as the lower bound of the practical deployment as two neighboring transmitter can be arbitrarily close, but makes the analysis tractable by utilizing the theoretical results of the stochastic geometry [26, 27]. Prior studies have used this model to analyze the performance of ad hoc networks [28, 29], traditional downlink cellular networks [23], downlink cellular networks with fractional frequency reuse [30], uplink cellular systems [31], downlink multi-tier HetNets [22, 24, 32 35], uplink two-tier HetNets [36] and uplink multi-tier HetNets [37]. In this thesis, we utilize this tool in the analysis of the performance of multi-tier HetNets and DS. Load Balancing & User-BS ssociation Load balancing, or the user/cell association problem is a long-standing problem in wireless access networks for congestion control. The coverage of a BS is defined by the users it serves. n appropriate user association scheme should jointly consider the signal quality from the users perspective and load balancing from the BSs perspective, since user rates are related to both the spectrum efficiency and the fraction of resources it gets, and the latter of which are limited and shared with other users. In multi-tier HetNets, besides the load balancing within each tier, the cross-tier load balancing problem emerges as users from macro cells are often off-loaded to small cells with femto- or pico-bss in order to relieve the congestion of the macro-bss. Such an inter-tier load balancing is more complicated as system parameters are usually distinct across BS tiers. Since small-cell tier transmitters usually have lower transmission power and higher deployment density, its served population may often be limited compared to the macro tiers, while on the other hand small cells often share the same frequency band

15 Chapter 1. Introduction 7 as that of the macro tiers. This load disparity not only leads to suboptimal rate distribution across the networks, but these lightly loaded cells may cause bursty interference. The user/cell association problem hence is more crucial in multi-tier HetNets. The most common user association scheme is based on the received signal strength, which ignores the radio resources availability at the BS side. n alternative approach to this problem is the greedy association, i.e., add users that improve a certain metric to the BS, as in [38] and [39] for single-tier and HetNets, respectively. Other approaches includes utility and pricing based association, see [40] for single-tier and [41, 42] for heterogeneous multi-tier networks, and a game theoretical approach where the association problem is jointly considered with resource allocation [43, 44]. These solutions are dynamic and rely on channel and topology realizations, and may require iterations and real-time computation. This thesis adopts a simpler cell range expansion approach, also known as the biased user association for the HetNets [8,45], in which the traffic can be effectively off-loaded to lower-power nodes by setting a power bias term towards them, which artificially expands the association areas of small cells. However, these bias factors are often heuristically set in current approaches, without theoretical guideline. Besides, it is not clear if the same user association bias factors should be applied in both downlink and uplink. In multi-tier HetNets, under the biased user association mechanism and full frequency reuse, the users off-loaded to neighboring cells, due to an unbalanced load, may experience SINR degradation, as the access nodes which offer stronger receiver power may now contribute as interference. See Fig. 1.5 as an example in a two-tier network. Hence, user/cell association and interference management are closely intertwined and need to be jointly tackled. The author of [34] investigates a special case of two-tier networks, and uses resource partitioning to mitigate interferences and jointly studies load balancing and user association. This thesis studies this problem in a more general multi-tier scenario. 1.3 Thesis Outline and Main Contributions This thesis studies three types of HetNet architectures, namely a shared relaying system, a multi-tier HetNet system, and a distributed antenna system in Chapters 2, 3, and 4, respectively. The first part of this thesis studies the resource allocation and scheduling aspects of a downlink shared relaying system. We consider the half-duplex decode-and-forward relay strategy, and assume that in the first phase BSs transmit to users and relays on orthogonal channels, and in the second phase BSs and relays simultaneously transmit to separate sets of users on all subchannels. The minimum-meansquared-error (MMSE) receiver and zero-forcing (ZF) beamforming are used in the first and the second phase, by the shared relay, respectively. We first prove that the number of degrees of freedom of shared relaying is M times that of separate relaying where a single relay is deployed in each cell sector, where M is the number of cell sectors coordinated by the shared relay. We then adopt the proportionally fair (PF) utility and formulate a utility maximization problem, subject to the in-band backhaul constraint such that the total rate demand of the shared relay in the access link of one sector should be satisfied by the total rate supplied in the feeder link from the BS in the same sector. By optimizing this utility using duality theory, an iterative algorithm is proposed to obtain the optimal channel allocation between feeder links and access links in the first phase, the optimal relay power allocation in the second phase, and the optimal user scheduling in both phases. We further propose a fast algorithm to significantly reduce the number of iterations without losing performance gain. System-level simulation shows the benefit of shared relaying in terms of utility and rate.

16 Chapter 1. Introduction 8 Wireless system transmitters in realistic scenarios are irregular distributed, which can be better modeled with random deployment. However, due to the difficulty of incorporating the in-band backhaul constraint into the theory of stochastic geometry, hexagonal topology is adopted in the modeling of the relay system. For HetNets with out-of-band backhaul, access points locations are modeled as a random spatial process. The second part of this thesis studies the user association and spectrum allocation of multi-tier HetNets in both downlink and uplink. We assume each BS tier forms an independent homogeneous PPP, and users form another PPP. Users are associated to BSs with the maximum biased received power, and spectrum can be either shared by or orthogonally partitioned among different BS tiers. In the uplink, fractional power control is implemented to partially compensate for the path loss. The proportionally fair utility is derived in a closed form for both downlink and uplink using the theory of stochastic geometry 2. Maximizing this utility, we can obtain the optimal bias factor used in user association as well as the optimal proportion of spectrum allocated to each tier under orthogonal partition assumption. The optimization results also reveal that users may choose to associate with different BSs in downlink and uplink for better performance, which corresponds to the downlink-uplink decoupling advocated for HetNets [46 48]. It is shown that under orthogonal spectrum partition in both downlink and uplink, the optimal proportion of spectrum of each tier should match the user association probability, or equivalently the proportional of users in each tier. With spectrum sharing in the uplink and assuming the same target SINR in all tiers, the optimal user association is distance based, i.e., each user connects to the closest BS. By numerical evaluation we also observe that orthogonal spectrum partition and spectrum sharing perform better for downlink and uplink, respectively. Stochastic deployment assumption is again assumed in the third part of this thesis, where the downlink of DS is studied with remote antenna ports modeled as a PPP. Different from the multi-tier HetNets where utility is focused, the downlink spectral efficiency of DS is characterized. n antenna port can contain one or multiple antennas. The channel state information is assumed to be not available at the transmitter side and each antenna has an individual power constraint. Multiple antennas may be co-located within one antenna port. Two types of DS architectures are considered: regular DS with fixed cell boundaries, and user-centric DS where all cell boundaries are removed. For regular DS two transmission schemes are considered: blanket transmission where the user is served by all the antenna ports within each cell, and selective transmission where only the closest antenna port to the user within each cell is selected. User-centric DS is studied with selective transmission. We first study regular DS from a location-specific perspective, assuming a single user per cell, and derive tractable spectral efficiency formula using stochastic geometry. We analytically show that the cell-edge spectral efficiency of blanket transmission is upper bounded by α 2, where α is the path loss exponent, while there is no such bound for selective transmission. We further model user locations as an independent PPP, and derive the spectral efficiency of a randomly chosen user from a network perspective, assuming time-division multiple-access (TDM) among users. Based on numerical results, we found that with the same total number of antennas, it is more beneficial to fully distributed antennas (one antenna per port) than to partially distributed antennas (grouping multiple antennas per port). For regular DS, selective transmission achieves higher spectral efficiency that blanket transmission in most cases; user-centric selective DS achieves a higher spectral efficiency than regular DS schemes when averaged over the network. 2 Thanks Wei Bao for contributing ideas in the uplink utility derivation.

17 Chapter 2 Fair Scheduling and Resource llocation for Shared Relaying This chapter examines the shared relay architecture for the sectorized wireless cellular network, where instead of deploying multiple separate relays within each cell sector, a single relay with multiple antennas is placed at the cell edge and is shared by multiple cells. The advantage of shared relaying is that the joint processing of signals at the relay enables the mitigation of intercell interference. To maximize the benefit of shared relaying, the resource allocation and the scheduling of users among adjacent cells need to be optimized jointly. We first provide a degree-of-freedom analysis as a motivation for the shared relay architecture, then formulate a network utility maximization problem for the shared relay system that considers the practical in-band backhaul constraint of matching the relay-to-user rate demand with the BS-to-relay rate supply using a set of pricing variables. In addition, zero-forcing beamforming is used at the shared relay to separate users spatially; multiple users are scheduled in the frequency domain to maximize frequency reuse. heuristic but efficient scheduling and resource allocation algorithm is proposed accordingly. System-level simulations quantify the effectiveness of the proposed approach, and show that the incorporation of the shared relay can improve the overall network performance and in particular significantly increase the throughput of cell edge users as compared to separate relaying. 2.1 Introduction Fig. 2.1 illustrates a cellular relay network with hexagonal layout, where three downlink users are served in three adjacent sectors belonging to different cells respectively. Fig. 2.1(a) shows a separate relay architecture. Fig. 2.1(b) shows a shared relay architecture, where a shared relay is deployed at the intersection of three sectors belonging to different cells, providing coverage to three users simultaneously. The shared relay is capable of maintaining connections to multiple BSs by resolving control messages from them, and acquiring the channel knowledge of the BS-to-relay and relay-to-user links. In the downlink, the shared relay receives signals from all of its donor BSs in adjacent sectors, and spatially separates these signals via receive beamforming. fter receiving the data packets for each user, the shared relay then retransmits the decoded signals to multiple users via spatial multiplexing using transmit beamforming techniques. s compared to separate relaying, although the shared relay is placed further away from the BS, its interference mitigation capability can potentially compensate for the increased BS-to-relay 9

18 Chapter 2. Fair Scheduling and Resource llocation for Shared Relaying 10 Transmission in the 1st phase BS Transmission in the 2nd phase Transmission in the 1st phase BS Transmission in the 2nd phase MS MS MS MS RS MS MS MS MS MS RS MS RS MS RS MS MS MS BS MS MS BS BS MS MS BS (a) (b) Figure 2.1: Relay scenarios. (a) Separate relays at edge of each sector. (b) Coordinated relay shared by three adjacent sectors of different cells. distance, leading to an improved overall network performance. Shared relaying has a clear advantage for cell edge users, which would otherwise suffer from severe intercell interference. To truly quantify the benefit of shared relaying for the entire network, it is important to evaluate the network performance from a system-level perspective. Toward this end, this chapter focuses on the scheduling and resource allocation aspects of shared relaying, while adopting the following assumptions and definitions: Downlink transmission with a half-duplex decode-and-forward strategy [49] is assumed. The BS-to-user and relay-to-user links can both be used to schedule users, and they are called the access links. The BS-to-relay links, which are called the feeder links, provide the in-band backhaul connection. Users can choose either direct transmission from the BS or indirectly via the relay. Depending on this routing choice, the users are classified as one-hop users or two-hop users, respectively. The half-duplex multi-channel frame structure in Fig. 2.2 is adopted. In the first phase, BSs transmit to users and the relay on orthogonal subchannels. In the second phase, BSs and the relay simultaneously transmit to separate sets of users on all subchannels to maximize frequency reuse. The reuse of frequencies by all serving nodes in the second phase inevitably induces more interference, which heightens the importance of scheduling and resource allocation. Perfect channel estimation is assumed in every time slot for both phases across subchannels, so that BSs know both the channel in the access links and the channel in the backhaul links. The main contribution of this chapter is as follows. First we show that shared relaying can achieve more degrees of freedom than separate relaying. We then adopt a network utility maximization framework with a proportionally fair (PF) objective[50], and design a heuristic but efficient resource allocation

19 Chapter 2. Fair Scheduling and Resource llocation for Shared Relaying 11 Frequency N orthogonal subchannels Base-station to User Base-station to Relay Base-station to User Base-station to User Base-station to Relay Base-station to User Relay to User Time Phase 1 Phase 2 Figure 2.2: Half-duplex two-phase relay frame structure. and scheduling algorithm for shared relaying to address questions such as how the frequencies should be allocated among the different links in an OFDM system, and how the frequencies should be reused within each sector. We characterize how much performance gain can be obtained from shared relaying as compared to separate relaying for the entire network from a system-level perspective. In this chapter, we use upper-case bold letters (e.g., I) for matrices, and lower-case bold letters (e.g., w) for column vectors. The conjugate transpose and Euclidean norm of vector w are denoted as w H and w, respectively. Calligraphy letters (e.g., K) are used to denote sets. The subscripts s and r refer to source (BS) and relay respectively. In particular, s m and r m refer to the BS and the separate relay in sector m, whereas r without the subscript refers to the shared relay Related Work There have only been a limited number of works in the literature on the shared relay architecture after the initial qualitative description of the concept in IEEE m [10, 11]. Most notably, [51] shows that shared relaying can approach the gains of local BS coordination at reduced complexity. In [51], multiple-input multiple-output (MIMO) multiple-access and broadcast techniques are used at the shared relay; the time durations of the two phases are optimally adjusted. The shared relay concept is further studied in [52], where zero-forcing (ZF) methods are used in combination with partial or full BS coordination for both one-way and two-way shared relaying. In [53], a joint processing scheme that improves the shared relaying strategy of [51] is proposed by letting the BS and the relay send the same message to the corresponding user in the second phase. However, none of these works consider the impact of scheduling: [51] and [52] assume arbitrary scheduling on a single subchannel, while round robin scheduling for cell-edge users is used in [53]. In [54], a scheduling scheme based on the Hungarian algorithm 1 is proposed for shared relaying under static orthogonal subchannel segmentation among neighboring sectors. The present chapter advances this line of work in addressing fairness and dynamic resource allocation issues of the shared relay architecture [55]. For the conventional separate relay system, resource allocation and scheduling have been studied extensively. In [56], fair resource utilization of relay nodes is considered as an integer optimization 1 combinatorial optimization algorithm that solves the assignment problem in polynomial time.

20 Chapter 2. Fair Scheduling and Resource llocation for Shared Relaying 12 problem. The work in [57] uses a cross-layer optimization framework for centralized resource allocation of OFDM-based relay networks. However, both [56] and [57] assume that the source and the relay always transmit on the same subchannel, an assumption which is removed in this chapter. In [58, 59] the sum-rate maximization problem is formulated with the constraint that the receiving rate (rate supply) and transmitting rate (rate demand) of relays are approximately equal when resources are allocated optimally. Proportional fairness is considered in [60] in formulating the subchannel and rate allocation problem and the relay s rate demand and rate supply constraint is considered on a per-user basis. In [61, 62], a queue-aware resource management algorithm is proposed, and the Hungarian algorithm is used to solve the joint routing and scheduling problem. ll of the above works assume that in the second phase the source and relays use orthogonal resource partition, which limits the performance gain. In this chapter, a proportionally fair (PF) scheduling algorithm based on the relay frame structure in Fig. 2.2 is used for the shared relay system. Differing from [60], our algorithm strikes a balance between the rate demand of the access link for all scheduled users in one sector and the rate supply from the feeder link. Frequency reuse is allowed between the relay and the BS in the second phase, offering maximum flexibility in the overall design. 2.2 Spatial Degrees of Freedom (DoF) nalysis lthough it is clear that shared relays are capable of mitigating interference in both the feeder and access links, a quantitative analysis is challenging under complex wireless channels and topologies. We present an asymptotic degree-of-freedom 2 analysis to illustrate the theoretical motivation of shared relaying [63]. To this end, we consider an abstract two-hop channel model in Fig. 2.3, where the shared relay has M antennas to coordinate the transmission of M source-destination pairs, while all other nodes have a single antenna each. The direct source-to-destination links are ignored since two-hop users typically have weak direct links from the source. Cellular Network Without Relay The cellular network without relaying can be modeled as an M-user interference channel. Using TDM, the achievable DoF is η s,d = 1. If advanced interference alignment techniques are allowed, a DoF of of M/2 can be achieved for this M-user interference channel [64]. Two-Hop Relay Network Let N s,r and N r,d be the number of frequency resources used in the two hops respectively, and let C s,r, C r,d be the corresponding spectral efficiency. To satisfy flow conservation, we have N s,r C s,r = N r,d C r,d, i.e., the throughput of the first hop should match the throughput of the second hop to make the best use of the channel. For a half-duplex system, the equivalent spectral efficiency of the two-hop link is the ratio of the total data rate and total consumed resources C s,r,d = N s,rc s,r N s,r +N r,d = N s,r C s,r = N s,r + Ns,rCs,r C r,d 1 1 C s,r + 1. (2.1) C r,d 2 channel has η degrees of freedom if its capacity can be expressed as C(SNR) = ηlog(snr)+o(log(snr)).