RADIO RESOURCE MANAGEMENT

Transcription

1 DESIGN AND PERFORMANCE EVALUATION OF RADIO RESOURCE MANAGEMENT IN OFDMA NETWORKS Javad Zolfaghari Institute for Theoretical Information Technology RWTH Aachen University

2

3 DESIGN AND PERFORMANCE EVALUATION OF RADIO RESOURCE MANAGEMENT IN OFDMA NETWORKS Javad Zolfaghari A thesis submitted to the Department of Signal Theory and Communications, Technical University of Catalonia-Barcelona Tech, reviewed by Dr. Michael Reyer and Dr. Anna Umbert on 14th December 2013 in partial fulfillment of the requirements for the degree of master in Information and Communications Technology Master of Science in Electrical Engineering, Information and Communications Technology Institute for Theoretical Information Technology RWTH Aachen University

4

5 I assure the single handed composition of this master s thesis only supported by declared resources. Information derived from those resources has been acknowledged in the text and references are given in the bibliography. Aachen, 14th December 2013

6

7 Abstract Inter-cell interference is of great interest and one of the essential issues for wireless operators who want to provide full coverage within their service area and guarantee a certain Quality of Service (QoS) to all users. In this thesis the resource allocation problem in an OFDMA based multi-cell network is formulated. Since no algorithm with polynomial running time exists due to non-linearity and combinatorial nature of underlying problem we adopted a heuristic approach with low computational cost and acceptable performance loss. We address specifically sharing available radio resources among users in terms of bandwidth allocation in order to suppress inter-cell interference. Besides, cell assignment and power allocation is foreseen. Performance of proposed schemes were evaluated by simulation network properly aligned with 3GPP LTE network. vii

8

9 Acknowledgements I would like to express my gratitude and special thanks to my supervisor, Dr. Micheal Reyer for his enthusiasm, patience and kindness. I have learned many aspects of carrying out research in the field of wireless communications in the Institute of Theoretical Technology from his comments, advises and suggestions, and from our fruitful discussions. Without his encouraging and enlightening guidance, knowledge, persistent support this work would not have been successful. Many thanks are given to Dr. Anna Umbert from Technical University of Catalonia who reviewed this thesis and all staffs who have helped me in one way or another and made the time of my master thesis pleasurable and memorable at the institute of Ti. Finally, I would like to give my heartfelt gratitude to my parents and specially my brother for his endless love, encouragement and support throughout my life. ix

10

11 Contents 1 Introduction Background Motivations Organization of Thesis Long Term Evolution (LTE) Overview Multiple access technology in the downlink: OFDM and OFDMA OFDM Technique OFDM Transmitter and Receiver Comparison of OFDM and Code Division Multiple Access (CDMA) OFDMA Technique Multiple access technology in the uplink: SC-FDMA Multiple antenna techniques Radio access modes Transmission bandwidths FDD and TDD LTE frequency bands LTE Air Interface Protocol Aspects Physical channels and modulation Frame structure Resource element and resource block System Architecture Evolution (SAE) Self Organizing Network (SON) Literature review of radio resource allocation in OFDMA networks Single cell multi-user system model Rate Adaptive (RA) Radio Resource Allocation (RRA) Schemes for single-cell OFDMA systems Distributed RA RRA schemes in OFDMA relay networks Margin adaptive (MA) radio resource allocation Adaptive resource allocation in cellular networks Network Optimization Single user water-filling Multi-user water-filling Rate region maximization xi

12 Contents 4 Resource allocation in downlink multi-cell OFDMA networks Problem formulation and system model System model Cell assignment and initialization Frequency Reuse Schemes Subcarrier Assignment Modifications in pre-assigned subcarriers Decrease pre-assigned subcarriers Cost benefit analysis of adding subcarriers Increase pre-assigned subcarriers Release subcarriers Performance evaluation Simulation Framework User Profile Numerical results Conclusion and Future work Future work UE Mobility Load balancing Femtocell and Macrocell Deplpyment Bibliography 63 xii

13 1 Introduction 1.1 Background Increasing demand of advanced mobile services such as high-speed internet access, multimedia online gaming, mobile TV, wireless Digital Subscriber Line (DSL), and integration of voice, video, text in recent years is motivation for technological evolution in mobile communication. Indispensably, delivery of wireless broadband services will become potential and widespread in today s wireless communication systems. Therefore, one of the challenges for designing next-generation wireless systems is to provide wireless broadband at better cost and performance, while maintaining seamless mobility, service control and Quality of Service (QoS) provisioning [4]. Furthermore, the globalization of markets, increasingly serious competence of vendors and especially universality of IEEE 802 wireless technologies in wireless communication domain are also considered as essential motivations for technical evolution in wireless communication systems. As a result, 3GPP, which is a collaboration agreement between a number of telecommunications standards bodies and wireless manufacturers, has already launched the project Long Term Evolution (LTE) regarding the evolution of the existed Third Generation (3G) systems. The stated targets of the project LTE include support for high peak data rates, low latency, improved system capacity and coverage, reduced operating costs, multi-antenna support, flexible bandwidth operations and easy integration to existing systems. LTE is also referred to as EUTRA (Evolved Universal Terrestrial Radio Access) or E-UTRAN (Evolved Universal Terrestrial Radio Access Network). LTE employs multiple access technologies on the air interface: specifically, Orthogonal Frequency Division Multiple Access (OFDMA) in downlink and Single Carrier Frequency Division Multiple Access (SC-FDMA) in uplink. OFDM technique that can be combined with multiple access using time, frequency or coding separation of the users are referred to as OFDMA technique. The OFDM technology has become a common advanced technology for wide-band digital communication and also been considered one of prime transmission technologies of the next generation networks, e.g., LTE Advanced. The basic idea of OFDM is to use a large number of closely spaced orthogonal subcarriers and these subacrriers are used in parallel. The key advantages of employing OFDM over singlecarrier schemes are its robustness against multi-path delay spread and frequency selective fading, elimination of the Inter-Symbol Interference (ISI). All aforementioned advantages 1

14 1 Introduction demonstrates that OFDM technique is chosen as a greatly promising candidate for next generation networks. 1.2 Motivations Interference mitigation in OFDM systems is one of the major challenges, particularly for cell edge users. Although intra-cell interference inside a cell can be eliminated basing on the orthogonal characteristic among the subcarriers in the performance of OFDM technology, inter-cell interference exists and hinges seriously on the performance of system in an OFDM based multi-cell wireless network. More specifically, inter-cell interference in OFDM based network occurs when the same subcarriers are re-used in neighbor cells. It decreases the Signal to Interference and Noise Ratio (SINR) on these subacarriers. Obviously, interference has the greatest impact on cell-edge users. Hence, the inter-cell interference is of great interest and one of the essential issues for wireless operators who want to provide full coverage within their service area and guarantee a certain Quality of Service (QoS) to all users including cell-edge users regardless of their positions inside a cell. In the scope of this thesis, we explore the problem of resource allocation in an OFDMA based multi-cell network or in other words, how to share the available radio resources among users in terms of bandwidth allocation in order to suppress inter-cell interference, enhance throughput of the cell-edge user and spectral efficiency. 1.3 Organization of Thesis The rest of the thesis is organized as follows: Chapter 2 we present an overview of 3GPP LTE networks because the simulation framework that we will work on, in the next chapters is well aligned with the current standards of the LTE. We focus on the description of LTE multiple access technologies such as Orthogonal Frequency Division Multiplexing (OFDM), Orthogonal Frequency Division Multiple Access (OFDMA) for downlink and also address Single Carrier Frequency Division Multiple Access (SC-FMDA) for LTE uplink. Frequency bandwidth and different radio access modes such as FDD and TDD that LTE supports will be presented in this chapter. LTE air interface protocol with the Radio Resource Control (RRC) layer which is responsible for setting up and management of radio resource blocks will be elaborated. We will address self organizing networks as new approaches to the network structure of which LTE is taking advantage. Chapter 3 In this chapter we focus on resource allocation in wireless cellular networks employing OFDMA as a promising air-interface for wireless systems. We present an over- 2

15 1.3 Organization of Thesis view of previous work in the context of radio resource allocation in OFDMA networks. We provide optimization frameworks based on dynamic and adaptive resource allocation schemes studied in the literature for power, bandwidth and rate allocation in the downlink. Resource allocation problems from different aspects such as centralized and distributed scenarios with single- or multi-cell topologies will be considered. Chapter 4 In this chapter resource allocation problem in the context of downlink multicell OFDMA network closely aligned with 3GPP LTE networks will be formulated. Since there is no computationally efficient (polynomial time) algorithm known for this problem we propose an iterative heuristic approach for user allocation and specifically subcarrier assignment with low computational cost and acceptable performance loss. In addition we will make some assumptions to cast the problem, which we want to deal with, from the general problem. The main focus of this chapter will be on providing mechanisms for adding or removing subcarriers and keeping the interference under control. Chapter 5 In this chapter the algorithms proposed in chapter 4 will be evaluated. We take advantage of a simulated OFDMA network which is properly aligned with LTE 3GPP network. Simulation setup, Network layout and user s profile will be addressed in this chapter. We will assess system performance in terms of number of served users and spectral efficiency while investigating the number of pre-assigned subcarriers. Numerical simulation results and proposed schemes will be analyzed and compared to each other. Chapter 6 We summarize the main conclusions of this thesis and make recommendations for possible future work. 3

16

17 2 Long Term Evolution (LTE) In this chapter we present an overview of 3GPP LTE networks because the simulation framework that we will work on, in the next chapters is well aligned with the current standards of the LTE. We focus on the description of LTE multiple access technologies such as Orthogonal Frequency Division Multiplexing (OFDM), Orthogonal Frequency Division Multiple Access (OFDMA), and Single Carrier Frequency Division Multiple Access (SC-FMDA). This chapter will provide the analysis of OFDM technique and a comparison of this technique to Code Division Multiple Access (CDMA) which Universal Mobile Telecommunications System (UMTS) is based on. We will present multiple antenna techniques which are used in LTE with aim at improving Signal to Interference plus Noise Ratio (SINR). The frequency bandwidth and different radio access modes such as FDD and TDD that LTE supports are presented in this chapter. LTE air interface protocol with the Radio Resource Control (RRC) layer which is responsible for setting up and management of radio resource blocks will be elaborated. After that, resource blocks as the smallest information unit which can be transmitted and different frame structures used in LTE will be defined. Finally, we will discuss about self organizing networks as new approaches to the network structure of which LTE is taking advantage. 2.1 Overview LTE stands for Long Term Evolution and it is a project name of a new, high performance air interface for mobile communication systems. It started in 2004 by third Third Generation Partnership Project(3GPP). An earlier 3GPP system known as the Universal Mobile Telecommunication System (UMTS) and evolved UMTS Terrestrial Radio access Network (E-UTRAN) which evolved from the Global System for Mobile Communications(GSM) towards an all-ip broadband network. They are known as the related specifications of LTE. LTE s E-UTRA technology provides a framework for increasing data rates and overall system capacity, reducing latency, improving spectral efficiency and cell-edge performance. A rapid increase of mobile data traffic and emergence of applications like MMOG (Multimedia Online Gaming), mobile TV, web 2.0 streaming contents have motivated the 3GPP to work on LTE. LTE s goals on the way towards Fourth-Generation (4G) are to provide high data rate, low latency and packet optimized radio access technology supporting flexible bandwidth deployments. LTE s network architecture is also able to support packet switched traffic with seamless mobility and great quality of service [2]. 5

18 2 Long Term Evolution (LTE) 2.2 Multiple access technology in the downlink: OFDM and OFDMA Transmission in LTE is based on Multiple Access (MA) technologies which are in particular OFDMA which is a variant of OFDM for the downlink and SC-FDMA for the uplink. To overcome the effect of multi path fading problem available in UMTS, LTE uses OFDM for the downlink that is, from the base station to the terminal to transmit the data over many narrow band carriers instead of spreading one signal over the complete carrier bandwidth OFDM Technique OFDM or Orthogonal Frequency Division Multiplexing is a modulation technique which is used commonly in many radio communication systems like wireless local area network (IEEE versions a, g and n) and WiMAX (IEEE ), as well as in digital television and radio broadcasting. Rather than transmit a high rate stream of data in serial with a single carrier, OFDM transmitter makes use of a large number of closely spaced orthogonal subcarriers that are transmitted in parallel. The term orthogonal frequencydivision is due to the fact that OFDM subcarriers are mutually orthogonal over the specific time interval. The sub-carriers are combined to produce data rates similar to conventional single-carrier modulation schemes in the same bandwidth. Each of the subcarriers modulated with one of the modulation schemes like Quadrature Phase-Shift Keying (QPSK), 16 QAM, 64 QAM (Quadrature amplitude modulation) at lower symbol rate and each of them experiences frequency flat fading channel where the coherence bandwidth of the channel is larger than the bandwidth of the signal. Therefore, all frequency components of the signal will experience the same magnitude fading. In contrast with flat fading channel, in frequency selective channel as the coherence bandwidth of the channel is smaller than the bandwidth of the signal, different frequency components of the signal experience different uncorrelated fading. Thus, an appropriately designed OFDM system converts a frequency selective fading channel into a set of parallel narrowband flat fading channels across the subcarriers. For 3GPP LTE the basic subcarrier spacing equals 15 khz. The number of subcarriers depends on the transmission bandwidth [10]. Figure 2.1 is the representation of key features of OFDM signal in the joint time-frequency domain. Each of the symbols are modulated with adjacent subcarrier independently. The source of Inter Symbol Interference (ISI) is delay spread which can be interpreted as the difference between the time of arrival of the earliest significant multipath component (typically the Line-Of-Sight (LOS) component) and the time of arrival of the latest multipath component [11]. In order to avoid ISI caused by multi-path delay spread guard intervals 6

19 2.2 Multiple access technology in the downlink: OFDM and OFDMA are inserted between the symbols. The guard time must be chosen to be larger than the expected delay spread, such that multi-path components from one symbol cannot interfere with the next symbol. Figure 2.1: OFDM signal represented in frequency and time [2] OFDM Transmitter and Receiver An illustrative description of a basic OFDM modulator is provided in Figure 2.2. It consists of a bank of N complex modulators, where each modulator corresponds to one OFDM subcarrier. Mapping specifies the modulation scheme (Binary phase-shift keying or BPSK, QPSK, 16-QAM...). The resulting information is the amplitude and phase of each sub-carrier as a function of frequency. By passing it through an Inverse Fast Fourier Transform (IFFT), we can compute the in-phase and quadrature components of the corresponding time-domain waveform. Figure 2.2: OFDM transmitter block diagram LTE uses a slightly more complex technique, known as Cyclic Prefix (CP) insertion. LTE 7

20 2 Long Term Evolution (LTE) transmitter starts by inserting a guard period before each symbol, as before. However, it then copies data from the end of the symbol following, so as to fill up the guard period. In fact the CP is a copy of the end of a symbol inserted at the beginning. If the guard period is longer than the delay spread in the radio channel, and if each OFDM symbol is cyclically extended into the guard period (by copying the end of the symbol to the start to create the cyclic prefix), then the inter-symbol interference can be completely eliminated. In the end the information can be mixed up to radio frequency for transmission [10]. In the receiver the reverse operations are done to extract the originally sent message. But there should be some extra blocks to make more sufficient reception such as synchronization and equalization. Synchronization is done to overcome frequency and time offset. Figure 2.3: OFDM receiver block diagram Comparison of OFDM and Code Division Multiple Access (CDMA) As the FFT operations at the heart of OFDM technique was too expensive and demanding in the past, 3GPP couldn t consider that. In 1998, They chose an alternative technology called CDMA. Today the cost of digital signal processing has been greatly reduced and OFDM is now considered as a commercially feasible method of wireless transmission for the handset. Both OFDM and CDMA have significant benefits. However compared to CDMA upon which UMTS is based, OFDM offers a number of distinct advantages [2]: OFDM can be scaled up to wide channels that are more resistant to fading. OFDM channel equalizers are much simpler to implement than are CDMA equalizers, 8

21 2.2 Multiple access technology in the downlink: OFDM and OFDMA as the OFDM signal is represented in the frequency domain rather than the time domain. OFDM can be made completely resistant to multi-path delay spread. This is possible because the long symbols used for OFDM can be separated by CP. OFDM is better suited to MIMO. The frequency domain representation of the signal enables easy precoding to match the signal to the frequency and phase characteristics of the multi-path radio channel. OFDM disadvantages: In OFDM the subcarriers are closely spaced making OFDM sensitive to frequency errors and phase noise. Therefore, OFDM is also sensitive to Doppler shift, which causes interference between the subcarriers. OFDM has no feature to combat with Inter-Cell Interference(ICI) at the cell edge whereas CDMA uses scrambling codes to provide protection against that. Therefore, some form of frequency planning at the cell edges will be required. Figure 2.4 shows a plan for frequency re-use to avoid inter-cell interference at the cell edges. It shows an example that every cell is sharing the same frequency band and every base station is controlling one cell. Within that band, each cell transmits to nearby mobiles using the same set of sub-carriers, denoted f 0. This idea works well because for the users close to base station the received signals are strong enough to overwhelm any interference. Distant users are easily damaged by interference because the received signals are weaker. One approach is neighbouring cells can transmit to those users using different sets of sub-carriers. Half the frequency band is reserved for nearby users, while the remainder is divided into three sets, denoted f 1, f 2 and f 3, for use by distant users. 9

22 2 Long Term Evolution (LTE) Figure 2.4: Example implementation of fractional frequency re-use when using OFDMA. (a) Use of the frequency domain. (b) Resulting network plan [8] OFDMA Technique Orthogonal Frequency Division Multiple Access (OFDMA) is a variant of OFDM, which is commonly used in wireless systems. Since the very narrow band UE-specific transmissions can suffer from narrowband fading and interference in standard OFDM, 3GPP chose OFDMA, which incorporates some elements of Time Division Multiple Access (TDMA). OFDMA allows subsets of the subcarriers to be allocated dynamically among the different users on the channel, as shown in Figure 2.5. The base station can respond to frequency dependent fading, by allocating subcarriers on which the mobile is receiving a strong signal. Figure 2.5: OFDM and OFDMA subcarrier allocation [2] 10

23 2.3 Multiple access technology in the uplink: SC-FDMA By allocating sub-carriers in response to changes in the fading patterns, an OFDMA transmitter can greatly reduce the impact of time- and frequency-dependent fading. 2.3 Multiple access technology in the uplink: SC-FDMA For the uplink communication a different concept for the access technique is used. SC- FDMA used in the uplink where lower Peak-to-Average Power Ratio (PAPR) greatly benefits the mobile terminal in terms of transmit power efficiency and reduced cost of the power amplifier [22]. SC-FDMA is not appropriate for the downlink, because lower PARR is valuable in the User Equipment (UE), but less so in the base stations which is carrying mutiple users and multiple signals. Another issue is that SC-FDMA need extra computation at the receiver. If the receiver is base station this is not a big issue, but for UE as receiver this is costly, because it requires more power and more DSP. Therefore, SC-FDMA has been adopted as the uplink multiple access scheme in 3GPP Long Term Evolution (LTE), or Evolved UTRA (E-UTRA). Although it is still using a form of OFDMA technology, the implementation is called SC-FDMA. In order to understand the differences between OFDMA and SC-FDMA, the Figure 2.6 provides the intuitively graphical comparison of these modulation schemes. Apparently, in this example, QPSK modulation scheme is applied for generation of OFDMA and SC-FDMA symbol. Each QPSK data symbol includes four subcarriers with the payload data. The generation of OF- DMA signal is shown on the left side of Figure 2.6, four adjacent subcarriers at spacing of 15kHz are already mapped into desired place in overall channel bandwidth. Because only the phase of each subcarrier is modulated, the subcarrier power keeps constant between symbols in QPSK modulation scheme. Four contiguous symbols are propagated in parallel and then after one OFDMA symbol period has elapsed, CP is inserted between symbols. For visual clarity CP is shown as a gap on the left side of Figure 2.6. However, it is actually a copy of the end of next symbol. It implies that the transmission power is continuous but the phase is discontinuous at the symbol boundary. Finally, the transmitted signal is created by the performance of IFFT on each subcarrier. IFFT converts the QPSK data symbols in frequency domain into time domain signals. Sum of the final time-domain waveform is referred as the transmitted signal. 11

24 2 Long Term Evolution (LTE) Figure 2.6: Comparison of OFDMA and SC-FDMA transmitting a series of QPSK data symbols [2]. Generation of SC-FDMA signal begins with a special precoding process but then continues in a manner similar to OFDMA. To create a time-domain waveform of the QPSK data sub-symbols, the four color-coded QPSK data symbols form Figure 2.6 are used and the process is to compute the trajectory traced by changing of one QPSK data symbol to that of the next one to achieve one SC-FDMA symbol in the time domain. With the rate of four times the rate of the SC-FDMA symbol, one SC-FDMA symbol contains four consecutive QPSK data symbols. Next step for generating SC-FDMA signal, Discrete Fourier Transform (DFT) process is performed to those symbols from frequency domain into time domain waveforms. The DFT sampling frequency is chosen such that the time-domain waveform of one SC-FDMA symbol is completely presented by four (or M) DFT points (or bins) spacing 15kHz to each other and each bin presents one subcarrier. The most obvious difference between the two schemes is described visually on right hand side of Figure 2.6. The difference is distinguishable in terms of that OFDMA transmits the four QPSK data symbols in parallel with one per subcarrier, whereas SC-FDMA transmits the four QPSK data symbols in series at four times symbol rate, with each data symbol occupying M 15kHz bandwidth. Obviously, OFDMA signal is clearly a multi-carrier with one data symbol per subcarrier, while SC-FDMA signal displays to be similar to a single-carrier (therefore SC in the SC-FDMA name) with each data symbol being envisioned one wide bandwidth signal. 12

25 2.4 Multiple antenna techniques Remarkably, the SC-FDMA symbol consists of sub-symbols (or M = 4 sub-symbols). These multi-symbols have propagated in parallel, which creates the undesirable different high Peak to Average Ratio (PAR) of OFDMA. By transmission of M data symbols in series at M times the rate, the SC-FDMA occupied bandwidth is the same as multi-carrier OFDMA but essentially, the PAR is the same as that used for the original data symbol. By transmitting together many narrow-band QPSK waveforms in OFDMA, the power peak is higher and it would be seen in the wider-bandwidth, single-carrier QPSK waveform of SC-FDMA [2]. 2.4 Multiple antenna techniques MIMO (Multiple Input Multiple Output) is another LTE major innovation, with the ability to further improve LTE s data throughput and spectral efficiency above that obtained by the use of OFDM. Although MIMO adds complexity to the system in terms of processing and the number of antennas required, it enables far higher data rates to be achieved along with much improved spectral efficiency. As a result, MIMO has been included as an integral part of LTE. Essentially MIMO employs multiple antennas on the receiver and transmitter to utilize the multi-path effects that always exist to transmit additional data, rather than causing interference. The schemes employed in LTE again vary slightly between the uplink and downlink. The reason for this is to keep the terminal cost low as there are far more terminals than base stations [26]. For the downlink, a configuration of two transmit antennas at the base station and two receive antennas on the mobile terminal is used as baseline, although configurations with four antennas are also being considered. For the uplink from the mobile terminal to the base station, a scheme called Multi-User MIMO (MU-MIMO) is to be employed. Using this, even though the base station requires multiple antennas, the mobiles only have one transmit antenna and this considerably reduces the cost of the mobile. In operation, multiple mobile terminals may transmit simultaneously on the same channel or channels, but they do not cause interference to each other because mutually orthogonal pilot patterns are used. This technique is also referred to as Spatial Domain Multiple Access (SDMA). 2.5 Radio access modes The LTE air interface supports the Multimedia Broadcast and Multicast Service (MBMS), a technology for broadcasting content such as digital TV to User Equipment (UE) using point-to-multi-point connections. MBMS was first introduced for UMTS in release 6 it is now an attractive option for operators who finally have enough bandwidth to cope with the demand. LTE provided an evolved MBMS service compared to UMTS Release 6 with the target to reach cell-edge spectral efficiency in an urban or sub-urban environ- 13

26 2 Long Term Evolution (LTE) Table 2.1: Transmission bandwidth configuration Channel bandwidth (MHz) Transmission bandwidth configuration (MHz) Transmission bandwidth configuration (NRB UL OR NRB DL ) (RB) ment of 1 bps/hz which is equivalent to support of at least 16 Mobile TV channels at around 300 Kbps per channel in a 5 MHz carrier. Multicast-Broadcast Single-Frequency Network (MBSFN) is the service which operates over a multicast/broadcast over single frequency network using a time-synchronized common waveform that can be transmitted from multiple cells for a given duration Transmission bandwidths In order to support international and regional spectrum regulations LTE s specifications include variable channel bandwidths selectable from 1.4 to 20 MHz with subcarrier spacing of 15 khz. For the new LTE embms, a subcarrier spacing of 7.5 khz is also possible. Subcarrier spacing is constant regardless of the channel bandwidth. LTE air interface is bandwidth agnostic which allows the air interface to adapt to different channel bandwidths with minimal impact on system operation. A Resource Block (RB) is the smallest amount of resource that can be allocated in the uplink or downlink. An RB is 180 khz wide and lasts for one 0.5 ms timeslot. For standard LTE, an RB comprises 12 subcarriers at a 15 khz spacing with 1 ms each symbol, and for embms with the optional 7.5 khz subcarrier spacing an RB comprises 24 subcarriers for 0.5 ms each symbol. The maximum number of RBs supported by each transmission bandwidth is given in the Table FDD and TDD LTE frequency bands The LTE air interface supports both FDD and TDD modes, with different frame structure. Half-duplex FDD allows the sharing of hardware between uplink and downlink since the uplink and downlink are never used simultaneously. This technique, while halving potential data rates, has uses in some frequency bands and also offers a cost saving. FDD spectrum requires pair bands, one of the uplink and one for the downlink. They are paired in order to allow simultaneous transmission on two frequencies. The bands also have a sufficient separation to enable the transmitted signals not to unduly impair the receiver performance. TDD requires a single unpaired frequency band. Therefore, the uplink and downlink share the same frequency, being time multiplexed. As a result, there are different LTE band allocations for TDD and FDD. In some cases these bands may overlap 14

27 2.5 Radio access modes but this does not necessarily simplify designs since there can be band-specific performance requirements based on regional needs. UEs that roam may encounter both types on the same band. Therefore, UEs need to detect what type of transmission is being made on that particular LTE band in its current location. Peak single user data rates In 3GPP standard Release 99, Wideband Code Division Multiple Access (WCDMA) had a peak data rate of 2 Mbps on the downlink and 1 Mbps on the uplink. The introduction of High Speed Downlink Packet Access (HSDPA) in 3GPP Release 5 the peak downlink data rate increased to 14.4 Mbps, by the use of a faster coding rate and a new modulation scheme, 16-QAM. There was a similar increase for the uplink in Release 6, through the introduction of high speed uplink packet access. Later releases have increased the peak data rate further, through the introduction of 64-QAM and spatial multiplexing, and the use of multiple carriers. The peak data rate in LTE Release 8 is 300 Mbps in the downlink and 75 Mbps in the uplink. Figure 2.7 shows the peak data rate of LTE which has increased since its introduction in 3GPP Release 8 and compares it with the peak data rate of WCDMA from Release 99. The data are taken from the most powerful UE capabilities available in FDD mode at each release. Figure 2.7: Evolution of the peak data rates of WCDMA and LTE in FDD mode [8] 15

28 2 Long Term Evolution (LTE) 2.6 LTE Air Interface Protocol Aspects The architecture of E-UTRA radio interface protocol around the physical layer is shown in Figure 2.8 [1]. The physical layer provides data transport services to the higher layers. These services are accessed through transport channels via the Media Access Control (MAC) sub-layer. The physical layer provides transport channels to the Layer 2 MAC sub-layer, and the MAC sub-layer provides logical channels to the Layer 2 Radio Link Control (RLC) sub-layer. Transport channels are characterized by how the information is transferred over the radio interface, whereas logical channels are characterized by the type of information transferred. In the following diagram, the circles between different layers or sub-layers indicate Service Access Points (SAPs) [1]. Figure 2.8: Radio interface protocol architecture around the physical layer [1]. Layer 1 - Physical Layer: To provision the data transport service to the higher layers, the physical layer performs the fundamental functions as following: Error detection on the transport channels, Forward Error Correction (FEC) encoding/decoding of the transport channels, Hybrid Automatic Repeat Request (HARQ) soft-combining, rate matching and mapping of coded transport channels to physical channels, modulation and demodulation of physical channels, frequency and time synchronization, radio characteristics measurements, MIMO antenna processing, transmit diversity, beam forming, RF processing, etc. The physical layer specifications are divided into four major sections: Physical channels and modulation, multiplexing and channel coding, physical layer procedures and physical layer measurements. 16

29 2.6 LTE Air Interface Protocol Aspects Layer 2 - MAC and RLC sub-layers: The MAC layer performs the mapping between the transport channels and logical channels, scheduling of UEs and their services based on their priorities, selecting the proper transport format. The RLC supplies sequenced delivery of Service Data Units (SDUs) to higher layers and eliminate the copy of SDUs. Layer 3 - Radio Resource Control (RRC) layer: RRC is responsible for setting up and managing radio resource blocks. RRC makes decisions of handover or handover based on the neighbouring cell measurement. The requirements for Radio Resource Management (RRM) detailed in covers the procedures and performance requirements for the efficient utilization of radio resources Physical channels and modulation LTE air interface includes physical channels and physical signals defined in [2]. System synchronization, cell identifications and radio channel estimation are the uses of physical signals which are generated in Layer 1. Physical channels carry data from higher layers consisting of control, scheduling and user payload. There are reference signals in both uplink and downlink which are known as pilot signals in other standards. The receiver uses these signals to estimate the phase and amplitude flatness of received signal. In particular at high modulation depth such as 16 QAM or 64 QAM, where small errors in phase or amplitude cause demodulation errors, using the reference signal is crucial Frame structure As described before, the physical layer supports OFDMA on the downlink and SC-FDMA on the uplink. In addition, both paired and unpaired spectrum are supported using frequency division duplexing (FDD) and time division duplexing (TDD), respectively. Although the LTE downlink and uplink use different multiple access schemes, they share a common frame structure. The frame structure defines the frame, slot, and symbol in the time domain. Two radio frame structures are defined for LTE and shown in Figures 2.9 and 2.10 [2]. Frame structure type 1 is defined for FDD mode. Each radio frame is 10 ms long and consists of 10 subframes. Each subframe contains two slots. In FDD, both the uplink and the downlink have the same frame structure though they use different spectra. 17

30 2 Long Term Evolution (LTE) Figure 2.9: LTE frame structure type 1 [2] Frame structure type 2 shown in figure 2.10 is defined for TDD mode. This example is for 5 ms switch-point periodicity and consists of two 5 ms half-frames for a total duration of 10 ms. Subframes consist of either an uplink or downlink transmission or a special subframe containing the Downlink and Uplink Pilot Time Slots (DwPTS and UpPTS) separated by a transmission gap guard period (GP). The allocation of the subframes for the uplink, downlink, and special subframes is determined by one of seven different configurations. Subframes 0 and 5 are always downlink transmissions, subframe 1 is always a special subframe, and subframe 2 is always an uplink transmission. The composition of the other subframes varies depending on the frame configuration. Figure 2.10: LTE frame structure type 2 for 5 ms switch-point periodicity [2] Resource element and resource block A resource element is the smallest unit which occupies one subcarrier in frequency domain and OFDMA or SC-FDMA symbol in time domain. Figure 2.11 shows resource elements 18

31 2.7 System Architecture Evolution (SAE) for downlink. Figure 2.11: Resource grid for downlink [2] The smallest unit which can be scheduled for transmission is resource block which occupies 180 khz in frequency domain and 1 slot (0.5 ms) in time domain. The number of subcarriers per resource block and the number of symbols per RB vary as a function of the cyclic prefix length and subcarrier spacing. For example 7.5 khz subcarrier spacing leads to longer symbols and consequently longer CP which is used to combat the higher delay spread in the multicast /Broadcast applications. 2.7 System Architecture Evolution (SAE) Along with 3G LTE that applies more to the radio access technology of the cellular telecommunications system, there is also an evolution of the core network known as SAE. This new architecture has been developed to provide a considerably higher level of performance that is in line with the requirements of LTE. The new SAE has also been developed so that it is fully compatible with LTE Advanced, the new 4G technology. Therefore, when 19

32 2 Long Term Evolution (LTE) LTE Advanced is introduced, the network will be able to handle the further data increases with little change. 2.8 Self Organizing Network (SON) In order to meet the requirements for increased data capacity and reduced latency, along with the move to an all-ip network, it is necessary to adopt a new approach to the network structure. For 3G UMTS / WCDMA the UTRAN comprising the Node B s or base stations and Radio Network Controllers (RNCs) employed low levels of autonomy [26]. The Node Bs were connected in a star formation to the RNCs which carried out the majority of the management of the radio resource. In turn the RNCs connected to the core network and connect in turn to the core network. To provide the required functionality within LTE System Architecture Evolution (SAE), the basic system architecture sees the removal of a layer of management. The RNC is removed and the radio resource management is devolved to the base-stations. The new style base-stations are called enodebs or enbs. The enbs are connected directly to the core network gateway via a newly defined S1 interface. In addition to this, the new enbs also connect to adjacent enbs in a mesh via an X2 interface. This provides a much greater level of direct interconnectivity. It also enables many calls to be routed very directly as a large number of calls and connections are to other mobiles in the same or adjacent cells. The new structure allows many calls to be routed far more directly and with only minimum interaction with the core network. In addition to the new Layer 1 and Layer 2 functionality, enbs handle several other functions. This includes the radio resource control including admission control, load balancing and radio mobility control including handover decisions for the mobile or UE. The additional levels of flexibility and functionality given to the new enbs mean that they are more complex than the UMTS and previous generations of base-station. However the new 3G LTE SAE network structure enables far higher levels of performance. In addition to this, their flexibility enables them to be updated to handle new upgrades to the system including the transition from 3G LTE to 4G LTE Advanced [26]. With LTE requiring smaller cell sizes to enable the much greater levels of data traffic to be handled, networks have become considerably more complicated and trying to plan and manage the network centrally is not as viable. The term SON came into frequent use after the term was adopted by the Next Generation Mobile Networks (NGMN) alliance. The idea came about as result of the need within LTE to be able to deploy many more cells. Femtocells and other microcells are an integral part of the LTE deployment strategy and 3GPP created the standards for SON. Although LTE SON self-optimizing networks is one of the major drivers for the generic SON technology, the basic requirements remain the same whatever the technology to which it will be applied. The main elements of LTE SON : Self configuration: The aim for the self configuration aspects of LTE SON is to enable new base stations to become essentially Plug and Play items. They should 20

33 2.8 Self Organizing Network (SON) need as little manual intervention in the configuration process as possible. Self optimization: Once the system has been set up, LTE SON capabilities will enable the base station to optimize the operational characteristics to best meet the needs of the overall network. Self-healing: Another major feature of LTE SON is to enable the network to self-heal. It will do this by changing the characteristics of the network to mask the problem until it is fixed. For example, the boundaries of adjacent cells can be increased by changing antenna directions and increasing power levels. 21

34

35 3 Literature review of radio resource allocation in OFDMA networks In this chapter we focus on resource allocation in wireless cellular networks employing OFDMA as a promising air-interface for wireless systems due, primarily, to its inherent resistance to frequency-selective multipath fading and the flexibility it offers in radio resource allocations. As in OFDMA the total bandwidth is divided into a large number of subchannels, throughout the network decisions based on frequency, power allocation can be made such that the resource utilization be maximized. In this chapter we present an overview of previous work in the context of radio resource allocation in OFDMA networks. We provide optimization frameworks based on dynamic and adaptive resource allocation schemes studied in the literature for power, bandwidth and rate allocation in the downlink. Resource allocation problems from different aspects such as centralized and distributed scenarios with single- or multi-cell topologies will be considered. Throughout this chapter, it is assumed that the network employs an initial channel estimation phase so that the frequency selective channel gain between any arbitrary transmitter and receiver through the network is known. It is further assumed that the channel estimation is perfect. 3.1 Single cell multi-user system model A system model for downlink OFDMA-based wireless network, is as follows [27]: Assume K u = {1, 2,..., K} and N u = {1, 2,..., N} are the sets of users and subcarriers, respectively. The data rate of the k-th user R k is given by R k = B N N a k,n log 2 (1 + γ k,n ) n=1 where B is the total bandwidth of the system and a k,n is the subcarrier assignment index indicating wether the k-th user occupies the n-th subcarrier or not; i.e., a k,n = 1 only if subcarrier n is allocated to user k; otherwise it is zero. The bandwidth of each subchannel 23

36 3 Literature review of radio resource allocation in OFDMA networks is B/N = 1/T where T is the OFDM symbol duration. γ k,n is the Signal-to-Noise Ratio (SNR) of the n-th subcarrier for the k-th user and is given by γ k,n = p k,n H k,n = p k,nh 2 k,n N 0 B/N where p k,n is the power allocated for user k in subchannel n, and h k,n, and H k,n denote the channel gain and channel-to-noise ratio for user k in subchannel n, respectively. Thus, The total data rate R T of this system is given by R T = B K N a k,n log N 2 (1 + γ k,n ) k=1 n=1 γ k,n is the effective SNR known by the modulation scheme adjusted to meet Bit Error Rate (BER) requirements which is a measure of received bits of a data stream that have been altered due to noise, interference, distortion or bit synchronization errors. The general form of the subcarrier and power allocation problem is given as or subject to max a k,n,p k,n R T = B N min a k,n,p k,n P T = B N K N a k,n log 2 (1 + γ k,n ) k=1 n=1 K N a k,n p k,n, k=1 n=1 C1: a k,n {0, 1}, k, n, C2: K a k,n 1, n, k=1 C3: p k,n 0, k, n, C4: K N a k,n p k,n P total, k=1 n=1 C5: user rate requirement. The first two constraints are on subcarrier allocation to ensure that each subchannel is assigned to only one user. C3 guarantees that the power values are non-negative. C4 is only effective in problems where there is a power constraint P total on the total transmit power of the system P T. C5 determines the fixed or variable rate requirements of the users. As the optimal solution for this problem is computationally complex, they may not be practical in real time applications [9]. Therefore, suboptimal algorithms have been developed which 24

37 3.2 Rate Adaptive (RA) Radio Resource Allocation (RRA) differ mostly in the approach they choose to split the procedure into several (preferably independent) steps to make the problem tractable and, in their simplifying assumptions to reduce the complexity of the allocation process. The performance of each algorithm greatly depends on the formulation of the problem and the validity of these simplifying assumptions. In the next sections we consider two main classes of dynamic resource allocation schemes which have been reported in the literature as Margin Adaptive(MA) [35, 34, 40] and Rate Adaptive(RA) schemes [30, 29, 32]. 3.2 Rate Adaptive (RA) Radio Resource Allocation (RRA) The optimization problem in RA scheme is to maximize the total data rate of the system with constrains on the total transmit power. Different scenarios of centralized or distributed algorithms, along with single-cell and multicell network topologies have been considered in the literature for the RA RRA problem Schemes for single-cell OFDMA systems A network with single cell and a single fixed Relay Station (RS) was considered in [14]. Such a network with multiple fixed RSs was studied in [23]. The algorithm proposed in [14] performs subcarrier allocation with uniform power distribution (power level of both BS and RS were equal) which was predetermined. The second algorithm achieves an optimal joint power and subcarrier allocation. It has been shown in the literature, e.g., [13] that optimization can be achieved when a subcarrier is assigned to only one user who has the best channel gain for that subcarrier, and also that equal power allocation among subcarriers has almost the same performance as water filling transmit power adaptation but with less complexity. Utility function in multi user OFDM systems The concept of utility functions was used by [31] to formulate the Rate Adaptive (RA) problem in multiuser OFDM systems. Utility is a mapping from network resources used by a user to a real number and it s a function of user s data rate as the most important factor to determine satisfaction of users.therefore, the utility function should be a nondecreasing function of r which is a positive value. The dynamic resource allocation with utility based formulation is the following: K max U k (R k ), a k,n,p k,n k=1 25

38 3 Literature review of radio resource allocation in OFDMA networks subject to: C1: S i S j =, i, j K u, i j C2: k S k {1, 2,..., N}, C3: p k,n 0, k, n, C4: K k=1 n S k p k,n P total, where U k (R k ) is a utility function for the k-th user. R k is the sum of the data rate of user k over all the subcarriers. S k is the set of subcarriers assigned to user k for which a k,n = 1. From the objective function and constrains it is obvious that fairness among the users was not considered. In [31] the extreme case of an infinite number of orthogonal subcarriers each with an infinitesimal bandwidth was investigated by introducing two theorems: theorem I gave the optimal subcarrier allocation assuming a fixed power allocation on all the subcarriers and theorem II gave the optimal power allocation given a fixed subcarrier allocation. Combining the results of the two theorems, the optimal frequency set and the power allocation for the extreme case were obtained. To tackle the mixed integer and continuous variable optimization problem with U k (R k ) = R k the authors of [24] proposed a greedy subcarrier and power allocation in relay networks. Another interesting approach to deal with mixed integer optimization problem was proposed in [20]. Their approach was to transform the integer optimization problem into a linear distribution problem in a directed graph to allow the use of the linear optimal distribution algorithms available in the literature. Centralized RA RRA schemes in multicell OFDMA networks A centralized downlink OFDMA scenario in a multicellular network enhanced with six fixed Relay Stations (RS) per cell was considered in [17, 18]. Relay stations are used to aid the direct communication between the BS and users in order to cope with path-loss characteristic of wireless channel. The RS forwards the received signal to the BS (uplink) or user (downlink) by employing either the Amplify-and-Forward (AF) or the Decodeand-Forward (DF) strategy on the same subcarrier. The problem was formulated as the minimum number of subcarriers required to satisfy a user s Quality of Service (QoS). By considering latency, overhead, and computational complexity, it may be seen that centralized RRA schemes are not the best option for future wireless networks [9]. This has led to the importance of distributed schemes being recognized. 26

39 3.3 Margin adaptive (MA) radio resource allocation Distributed RA RRA schemes in OFDMA relay networks A semi-distributed downlink OFDMA scheme in the form of two algorithms of Separate and Sequential Allocation (SSA) and Separate and Reuse Allocation (SRA), in a single cell enhanced by M half-duplex fixed relays was considered in [15]. It means that the transmission and reception at that relay do not occur simultaneously. The starting point of both the two-step SSA and the SRA algorithm allocation schemes is basically the same. The Subscriber Station (SS) attached to the BS and relays are referred to as the BS SS and RS SS clusters, respectively. In the first step, each RS, along with its SS cluster, is treated as a large SS with a required minimum rate equal to the sum of all the minimum required rates of the SSs in its cluster. The BS allocates the resources among its own SSs and these virtual large SSs. In the second step, the RS allocates resources to the SSs in its cluster based on one of two allocation schemes: Resources assigned to that BS-RS link in the first step are allocated among the connected SSs (SSA). The RS reallocates all N subcarriers to its connected SSs regardless of the BS assignments (SRA). Simulation results for a single cell with one relay showed that the semidistributed scheme, SSA in particular, has a comparable capacity and probability of dropping a user to the centralized scheme [12]. The SSA algorithm showed significant performance stability over the SRA. Since both RS and BS may assign the same subcarriers to their respective SSs in the SRA algorithm, intracell interference may occur, which results in a considerable increase in outage probability. In general, the proposed semi-distributed schemes reduce the amount of overhead required to feedback the CSI and minimum rates to the BS. However, in the case of SRA, there is no need to communicate such information to the BS. These schemes fail to exploit the interference avoidance and traffic diversity gains. 3.3 Margin adaptive (MA) radio resource allocation The MA optimization problem assumes a set of user data rates with fixed QoS requirements. This can be formulated as [27]: 27

40 3 Literature review of radio resource allocation in OFDMA networks subject to: K N min P T = a k,n p k,n, a k,n,p k,n k=1 n=1 C1: a k,n {0, 1}, k, n, C2: K a k,n = 1, n, k=1 C3: p k,n 0, k, n, C4: K N a k,n p k,n P total, k=1 n=1 C5: R k R k,min, k = 1, 2,..., K With help of constraint relaxation and to make the problem tractable, the authors of [34] introduced a new parameter to the cost function, taking values within the interval [0,1], which can be interpreted as the sharing factor for each subcarrier. Then it is shown that the optimization problem can be reformulated as convex minimization problem over a convex set and be solved by computing the Lagrangian of the problem. Lagrange multipliers which satisfy the individual data rate constraints can be found using an iterative search algorithm. However, the iterative computation and search for this algorithm make it prohibitively expensive with complexity of O(NK 3 ) where N is the number of subcarriers and K is the number of MA users. One solution to simplify the algorithm is to assume that the channel is flat for a certain number of subcarriers, as in [36]. 3.4 Adaptive resource allocation in cellular networks A wireless cellular network comprises base stations serving users. The assignment of users to the base stations depends on the strength of receiving signal. As a mobile device can usually receive signals from several base stations, it is typically assigned to the base station with the strongest received signal. Signals from other base stations are known as intercell interference which may cause a low Signal to Interference plus Noise Ratio (SINR). Consequently this decreases the transmission data rate of the users. In order to avoid excessive intercell interference traditional cellular networks employ a fixed frequency reuse pattern so that neighboring base stations do not share the same frequency. As a result the users in cell-edge, which are suffering from interference from 28

41 3.5 Network Optimization neighboring cells, do not interfere with each other. The traditional fixed frequency reuse schemes are effective in minimizing intercell interference, but are also resource intensive in the sense that each cell requires a substantial amount of nonoverlapping bandwidth, so that only a fraction of the total bandwidth can be made available for each cell. Consequently, the standardization processes for future wireless systems have increasingly targeted at maximal frequency reuse, where all cells use the same frequency everywhere [9]. Wireless channels are fundamentally impaired by fading, propagation loss, and interference. Two types of cooperative network that specifically address the issues of intercell interference and path-loss: Base station cooperation: While in traditional networks base stations were operating independently, this type of cooperative network explores the possibility of coordinating multiple base stations. In these networks the transmission strategies among the multiple BS are designed jointly. In particular, the base stations may cooperate in their power, frequency, and rate allocations in order to jointly mitigate the effect of intercell interference for users at the cell edge. Relay cooperation: This type of cooperative network explores the use of relays to aid the direct communication between the base station and the remote subscribers in order to combat against the path-loss characteristic of wireless channel. In both types of cooperative networks, resource allocation is expected to be a crucial issue. However, in this thesis we will focus on BS cooperation and address the different resource allocation strategies regarding that. 3.5 Network Optimization In this part we consider multiple base station cooperative network which an OFDMA scheme is employed within each cell, and no two links in each cell can use the same subcarrier at the same time. So, there is no intracell interference. Given a fixed frequency and power allocation for all transmitters, the network optimization problem is that of coordinating the subcarrier assignment Single user water-filling A single link in OFDM transmit power optimization problem has a classic solution known as water-filling. For a single link problem where the noise and interference are assumed 29

42 3 Literature review of radio resource allocation in OFDMA networks to be fixed, the optimization of achievable rate subject to a total power constraint can be formulated as ( ) N max log 1 + h(n) 2 p n Γσ n=1 2 N s.t. : p n p total, n=1 0 p n p max, where the optimization is over p n, the transmit power on the frequency tone n. h(n) 2 is the channel path-loss and the combined noise and interference σ(n) 2 are assumed to be fixed. The Lagrangian dual of this optimization problem is ( ) ( N L(p n, λ) = log 1 + h(n) 2 p N ) n Γσ(n) n=1 2 λ p n p total. n=1 The constrained problem is now reduced to an unconstrained one in which λ can be interpreted as power price. Setting the derivative of the above Lagrangian yields : p n = [ ] pmax 1 λ Γσ2 (n) h(n) 2, 0 where [.] b a denotes a limiting operation with lower bound a and upper bound b. The fundamental reasons that an (almost) analytic and exact solution exists for this problem are that the objective function of the optimization problem is a concave function of the optimization variables and the constraints are linear. Therefore, convex optimization techniques such as Lagrangian dual optimization can be applied. In practice the exact values of power is not important because the water-filling relation operates on a linear scale on p n, while the rate expression is a logarithmic function, which is not sensitive to the exact value of p n when it is large Multi-user water-filling In a cellular setting, whenever a particular cell implements power adaptation, it changes its interference pattern on its neighbours. Therefore, when every cell implements water- 30

43 3.5 Network Optimization filling at the same time, the entire network effectively reaches a simultaneous water-filling solution, where the optimal power allocation in each cell is the water-filling solution against the combined noise and interference from all other cells. Such a simultaneous water-filling solution can typically be reached via an iterative water-filling algorithm in a system-level simulation where the water-filling operation is performed on a per-cell basis iteratively [38] which has been observed to converge Rate region maximization One of the useful approaches to tackle with the nonlinear utility function of links is to reduce the objective function to weighted sum rate optimization which is essentially linearization of the objective function and decouples that on a per-tone basis. This technique simplifies the problem significantly. A key technique for achieving decoupling is to employ the Lagrangian duality theory. For example, consider the following weighted rate sum maximization problem subject to the power constrains K max ω k R k k=1 N s.t. p k (n) p k,total, n=1 where ( N h kk (n) 2 ) p kn R k = log 1 + Γ( n=1 l k h lk 2 p ln + σ(n) 2 ) p kn denotes the transmit power of user k in subcarrier n, and h kl (n) is the complex channel gain from the transmitter of user k to the receiver of user l. This problem can be solved by dualizing with respect to the power constraint. This results in a dual function g(λ k ) defined as { g(λ k ) = max p kn ( N )} ω k R k λ k p kn p k,total n=1 Now, the optimization problem reduces to N per-tone problems: max p kn { ω k log(1 + h kk (n) 2 } p kn Γ( l k h lk(n) 2 p ln + σ(n) 2 ) λ kp kn. 31

44 3 Literature review of radio resource allocation in OFDMA networks Just as in single-user water-filling, where the solution to a convex optimization problem reduces to solving the problem for each λ, then finding the optimal λ, similar algorithms based on λ search can be applied here. The reduction to an N per-tone optimization problem ensures that the computational complexity for each step of this optimization problem with fixed λ k is linear in the number of subcarriers. The maximum value of the original objective is equal to the minimum of the dual optimization problem which means K max ω k R k = min g(λ k) λ k 0 k=1 The optimum λ k can be found using search methods such as the ellipsoid method (which is a generalization of bisection search to higher dimensions) or the subgradient method [39]. An interesting fact is that this duality technique remains applicable even when the functional form of the rate expression is nonconvex as is the case above, as long as the OFDM system has a large number of dimensions in the frequency domain, which allows an effective convexification of the achievable rate region as a function of the power allocation [39, 21]. 32

45 4 Resource allocation in downlink multi-cell OFDMA networks In this chapter we intend to formulate resource allocation problem in the context of downlink multi-cell OFDMA network closely aligned with 3GPP LTE networks. The formulated problem is known to be NP (Non-deterministic Polynomial) hard and there is no computationally efficient (polynomial time) algorithm known for that [16]. Therefore, we suggest an iterative heuristic approach for user allocation and specifically subcarrier assignment with low computational cost and acceptable performance loss. we primarily intend to formulate a general problem with the following four assumptions. Assumption 1: Each subcarrier at each cell can be used by maximum one user served by that cell, i.e., there is no intra-cell interference. Assumption 2: Each user can be served by one cell at maximum. Assumption 3: Neighboring cells might use the same sub-carriers. Assumption 4: Perfect channel state information at both receiver and transmitter is available during transmission. In addition to these, we will make two assumptions to cast the problem, we want to deal, from the general problem. Throughout this chapter, we concentrate on a subcarrier assignment procedure with the aim of maximizing data throughput of the system. The main focus of this chapter will be on providing mechanisms for adding or removing subcarriers and keeping the interference under control. 4.1 Problem formulation and system model We investigate a solution for the multi user Rate Adaptive (RA) resource allocation problem in multi cell scenario. Assume that we have an OFDMA cellular network in which the set of cells is denoted by I = {1, 2,..., I}, the set of users K = {1, 2,..., K} and the set of subcarriers by N = {1, 2,..., N}. We have K users distributed in I cells with N number of subacarriers available for each cell. The objective is to maximize sum over all 33

46 4 Resource allocation in downlink multi-cell OFDMA networks data rates r ikn which is the rate of user k in cell i given subcarrier n, if it is assigned, with some constrains. The general problem formulation is: max I K N r ikn i=1 k=1 n=1 subject to: C1: a ikn {0, 1}, i, k, n, C2: C3: I N r ikn R k, k, i=1 n=1 N p in P i, i, n=1 C4: p in 0, i, n, C5: I a ikn a iln = 0, k, l = 1,..., K, k l, n i=1 C6: K a ikn a jkm = 0, i j, m, n = 1,..., N, k=1 ( ) p in g ikn a ikn where r ikn = log 2 j i p jng jkn ( K l=1 a jln ) + σin 2 + 1, i = 1,..., I, k = 1,..., K, n = 1,..., N, The rate expression is obtained by the Additive White Gaussian Noise (AWGN) Shannon channel capacity. p in is the power allocated to subcarrier n from the cell i. g ikn is the channel gain from cell i on user k and subcarrier n, and σ 2 in is the noise power. a ikn is an indicator which is one if the user k from cell i is using subcarrier n, and zero otherwise (C1). The received SINR of user k of cell i on subcarrier n can be expressed by: SINR ikn = p in g ikn j i p jng jkn ( K k=1 a jkn ) + σin 2 To maintain fairness, the achieved rate of each user is individually lower bounded in C2. C3 denotes that the power over all the subcarriers in each cell is limited to a certain value. C4 is to guarantee that the transmission power values are non-negative. C5 expresses that each subcarrier has to be assigned to one user at maximum (Assumption 1). C6 denotes that each user can be assigned to one cell at maximum (Assumption 2). 34

47 4.1 Problem formulation and system model With no loss of generality we introduce three 2-dimensional variables instead of a 3- dimensional variable a ikn to reduce the arithmetic complexity. In this line, a ik and a kn denote the association of user k to cell i and association of subcarrier n to user k respectively. Therefore, by definition a ikn = a ik a kn. We also define b in K a ikn. k=1 By substituting the defined variables, the new formulation is given as: ( I K N p in g ikn a ik a kn max log 2 i=1 k=1 n=1 j i p jng jkn b jn + σin ) subject to: C1: a ik, a kn, b in {0, 1}, i, k, n, C2: C3: I N r ikn R k, k, i=1 n=1 N p in P i, i, n=1 C4: p in 0, i, n, C5: C6: K a ik a kn = b in, i, n k=1 I a ik 1, k, i=1 However in the new formulation C5 and C6 are changed, they have the same functionality as before. Assumption 5: In order to simplify the problem we assume that the power is distributed equally among the subcarriers which necessarily means that there is no power control. There is an option for transmit power either to take real positive value, if it s allocated to a subcarrier, or zero when no power is allocated to that subcarrier, essentially means the subcarrier is off. Thus, p in = P i N b in 0, so that we can skip C4. 35

48 4 Resource allocation in downlink multi-cell OFDMA networks Assumption 6: We will also assume that the channel is not frequency selective and we consider it as frequency flat channel which necessarily means g ikn = g ikm n, m. Therefore, p in g ikn = P i N g ik. If we reformulate the problem with new assumptions while substituting c ik = P i N g ik we obtain: ( ) I K N c ik a ik a kn max log 2 j i c jkb jn + σin (4.1.1) subject to: i=1 k=1 n=1 a ik, a kn, b in {0, 1}, i, k, n, I N r ikn R k, k, i=1 n=1 K a ik a kn = b in, i, n k=1 I a ik 1, k, i=1 The complexity of the resource allocation problem is very high as the underlying problem of subcarrier allocation to users is of combinatorial nature and it is even more complex for the case of multi cell scenario, when the objective function is nonlinear due to the interference term. It is called nonlinear integer optimization problem and referred as a NP-hard (Nondeterministic Polynomial hard) combinatorial optimization problem with no standard computationlly efficient optimization algorithm to obtain the optimal solution [16]. Some solution strategies such cutting plane and branch-and-bound algorithms have been proposed to cope with the certain classes of nonlinear integer programming problems in [25]. However, it is hard to guarantee that these algorithms achieve good performance over large scale problems with large number of constraints and variables. In this thesis we propose a heuristic approach solving the issue of subcarrier assignment and user allocation with low computational cost and acceptable performance loss System model Looking precisely at problem (4.1.1) a ik and a kn are independent variables (b in is dependent) which are nothing but user allocation to cells and subcarrier assignment to users 36

49 4.1 Problem formulation and system model respectively. Regarding the user allocation we assign users to cells with strongest receiving signal. The main problem in our context is subcarrier assignment to users. Figure 4.1 illustrates the main building blocks of subcarrier assignment and peripheral blocks for enhancements. Cell Assignment & Initialization Alg. 1 Subcarrier Assignment Alg. 2 Decrease Pre- Assigned Subcarriers Alg. 3 Increase Pre- Assigned Subcarriers Alg. 5 (or 6,7) Compute Gains Cost-Benefit Analysis Alg. 4 Compute Achievable Rates Compute Current Rates Compute Potential Rates Figure 4.1: Block diagram of the whole process We propose two general strategies in serving users which makes the overall procedure different. First strategy is aimed at considering the subcarrier assignments in previous iterations only as side information to promote assignments in upcoming iterations. Second strategy is aimed at continuing the subcarrier assignment in each iteration while keeping those assigned in previous iterations. There are two differences in the implementation of these strategies. One lies on initialization of subcarrier assignment block and the other in the necessity of implementing release 37

50 4 Resource allocation in downlink multi-cell OFDMA networks subcarrier algorithm. Nevertheless, both strategies have the building blocks in Figure 4.1 in common. Throughout the next sections we describe the functionality of each block in detail. 4.2 Cell assignment and initialization In this section we address the user allocation to base station as a primary step for resource allocation. From now on, we use the BS or cell terms interchangeably. Initially we assign each user to a cell by using a simple greedy algorithm. In this algorithm we assign cell i with best SINR to user k. The best cell refers to the cell which can provide maximum SINR for the user in the situation that all the subcarriers are occupied by all the cells. SINR ik = c ik j i c jk + σ 2 i As a remark in SINR formulation, the full interference case, where all the subcarriers are being used by all the cells, is considered. Consequently b jn = 1 n, j. The procedure of the cell assignment is represented in Algorithm 1. Algorithm 1 Cell Assignment (SINR) Input: c ik, σi 2 i, k for all users k do i k arg max SINR ik = i c ik j i c jk + σ 2 i a ik 1 end for return (i k ) k K later on, we use i k as cell of user k if a ik = Frequency Reuse Schemes In OFDMA systems, if two neighbouring cells use the same subcarrier, transmission in one cell interferes with that in the other cell which is known as Inter Cell Interference (ICI). To overcome ICI problem, the system can employ techniques like spectrum spreading [3, 19] and multiple-receiver based interference suppression [7]. Frequency Reuse Factor (FRF) is a widely accepted approach to design channel allocation such that the two neighboring cells allocate subcarriers exclusively. In other words, the neighboring cells will not share same frequencies. 38

51 4.4 Subcarrier Assignment For OFDMA systems FRF 1, which essentially means all the cells use the same frequency band, would be the best choice. However, it has the problem that some users near cell edge cannot get service due to the SINR degradation caused by ICI [33, 37]. In [5] several FRF like conventional integer numbers i.e., 1, 3, 4, or 7 are considered and the possibility of using some non-integer FRFs such as 7/4 and 7/3 is investigated. In order to improve the interference situation for the cell edge users we propose FRF 3 as the first step of subcarrier assignment. In this line, we divide the available bandwidth into three subbands, each will be assigned to one cell. Subcarriers in each subband are assumed to be pre-assigned. The main advantage of considering the so called pre-assigned subcarriers is that the base station can know the interference situation over the subcarriers ahead of the transmission. 4.4 Subcarrier Assignment In [6] it is suggested that in a single user water-filling solution, the total data rate of a zero margin system is close to capacity even with flat transmit Power Spectral Density (PSD) as long as the energy is poured only into subchannels with good channel gains. This important result completely eliminates the major step of power allocation concentrate mainly on subcarrier allocation. The reason is that in multiuser OFDM systems, a flat transmit power might also perform well is that it is assumed due to multiuser diversity, only subchannels with good channel gains are assigned to each user [27]. Nevertheless, in this thesis we neglect frequency-selective fading of channel and assume all the channel gains equal with respect to the subcarriers. Herein the issue is selecting subcarriers with minimum receiving interference from the neighbouring cells so as to provide maximum data rate. Along this goal we will propose an algorithm aiming at assigning the subcarriers which are less affected by the interference of the neighboring cells. Before entering that we explain briefly the frequency reuse schemes and it s advantages in mitigating the interference. The subcarrier assignment is based on pre-assigned subcarriers. we define matrix A with size I N as pre-assigned subcarriers to the cells with the following characteristic. { 1 if subcarrier n is pre-assigned to cell i α in = 0 otherwise Initially we set the values in matrix A according to the frequency reuse scheme 3, which necessarily means that one third of whole subcarriers may be assigned to one of three neighboring cells in a way that no two neighboring cells share the same subcarriers. The distribution of subcarriers is shown in Figure 4.2. This way, the number of cells using a special subcarrier is divided by three. Thus, the maximum spectral efficiency will be 39

52 4 Resource allocation in downlink multi-cell OFDMA networks Figure 4.2: Frequency reuse scheme 3 [5] divided by three primarily. The idea behind pre-assignment consideration is to keep the interference situation of system under control, so that the subcarriers can be assigned without having concerns about the immediate changes in the interference. The main characteristic which distinguishes the first strategy from the second is the following initialization in subcarrier assignment phase. { akn = 0 k, n b in = 0 i, n Setting these variables to zero means that in each iteration we start assigning subcarriers from scratch. In fact, the pattern of pre-assigned subcarriers is the thing which really matters to us. Now, we concentrate on subcarrier assignment regardless of the strategy we chose cause they don t differ. In the subcarrier assignment algorithm we start with computing the SINR for all users and subcarriers. Notice that the SINR values are computed based on the pre-assigned subcarriers which is given already. Thus, during the subcarrier assignment the interference situation of the system doesn t change. By using sinr2rate() function we obtain the achievable rate for each user over each subcarrier. sinr2rate() function maps SINR values to the rate values using Table 4.1. According to LTE system specification, 16 Channel Quality Indicators (CQI) are distinguished. Each CQI corresponds to a supported modulation scheme and code rate for downlink transmis- 40

53 4.4 Subcarrier Assignment sion. The system link budget specification defines what SINR level and what Receiver Reference Sensitivity (RS) level is required to support minimum DL throughput of 95% of the maximum possible at different CQIs. Table 4.1 shows the requirements assuming 10 MHz transmission bandwidth [28] for RS level specification. RS considers thermal noise (-174 dbm/hz) multiplied by transmission bandwidth, receiver noise figure (9dB), an implementation margin (2.5 db for QPSK, 3 db for 16-QAM and 4 db for 64-QAM), and diversity gain (-3 db). CQI modulation code rate efficiency [ ] bps Hz 0 out of range SINR[dB] RS level [dbm] 1 QPSK 1/ QPSK 1/ QPSK 1/ QPSK 1/ QPSK 1/ QPSK 2/ QPSK 3/ QPSK 4/ QAM 1/ QAM 2/ QAM 3/ QAM 4/ QAM 2/ QAM 3/ QAM 4/ Table 4.1: SINR and RS requirements for 10 MHz transmission bandwidth. After finding corresponding rates, we sort the users based on their rate demand so that in each cell the user with minimum rate demand will be served first. The idea of subcarrier pre-assignment enables us carry out subcarrier assignment in cells in parallel. The idea behind that is while assigning subcarriers to a user in a cell, the other users in the same cell will not be affected because there is no inter-cell interference. This eliminate the need of recalculating SINR values for a single assignment which is computationally expensive. The approach is users will either be served completely such that their rate demand be fully satisfied, or they will not be served. In our fixed interference scenario, the possibility that a user could be served or not, is determinable beforehand. This is because we have the 41

54 4 Resource allocation in downlink multi-cell OFDMA networks interference over any subcarrier with respect to pre-assigned subcarriers which is already fixed. This essentially means the SINR of any subcarrier is known a prior and it doesn t change during the assignments. SINR kn = c ik kα ik n j i k c jk α jn + σ 2 i k k, n Therefore we simply have the rate each subcarrier can potentially achieve by sinr2rate() function. We ensure the possibility of serving user l if the following inequality holds. N AchievableRate(l, n) R l n=1 Then among users that inequality holds true, a subcarrier which provides maximum rate will be assigned to the one with minimum rate demand. Once a subcarrier is assigned to a user, the rest of users in that cell can not occupy that subcarrier as stated in assumption 1. We repeat this procedure for all users to fulfill their rate demand if possible. The algorithm of subcarrier assignment is represented in Algorithm 2 with more details. 42

55 4.4 Subcarrier Assignment Algorithm 2 Subcarrier Assignment Input: c ik, α in, a ik i, k, n c ik kα ik n 1: SINR kn j i k c jk α jn + σi 2 k, n Compute the SINR based on pre-assigned k subcarriers (α) 2: AchievableRate = sinr2rate(sin R) 3: for all cells i do 4: Find the users in cell i 5: Sort the users based on their rate demands in ascending order. 6: for all users l in cell i do 7: Rate(l) 0 Rate(l) is current rate of user l 8: if AchievableRate(l, n) R l then n 9: while Rate(l) < R l do 10: (n, maxrate) max AchievableRate(l, n) n 11: b in 1 Indicates subcarrier n is assigned to cell i 12: Rate(l) Rate(l) + maxrate 13: a ln 1 Indicates subcarrier n is assigned to user l 14: for all users l in Cell i do 15: AchievableRate(l, n ) 0 Subcarrier n should no longer be available for the users in cell i 16: end for 17: end while 18: else Add i to OverLoadedCells 19: end if 20: end for 21: end for return a kn, b in, OverLoadedCells 43

56 4 Resource allocation in downlink multi-cell OFDMA networks 4.5 Modifications in pre-assigned subcarriers The outcome of subcarrier assignment shows that the available resources is not fully utilized and there are some users whose rate demands are not satisfied. Although we have controlled the interference issue by employing frequency reuse scheme 3, the ideal FRF for wireless OFDMA systems is 1. In this line we carry out modifications to the pre-assigned subcarriers. The fact that users with heterogeneous traffics are not uniformly distributed among the cells makes the cells overloaded with excessive rate demand. Thus the need to change the pre-assigned subcarriers dynamically according to the user s rate demand seems to be essential. Strategies to cope with lack of resources in overloaded cells will be proposed in the next sections Decrease pre-assigned subcarriers There are some cells in the system which are able to serve all their assigned users an have extra pre-assigned subcarriers. The idea is to release some of their pre-assigned subcarriers in order to eliminate their interference in the further subcarrier assignment. We remind that the SINR of a subcarrier is computed based on pre-assigned subcarriers. It can be easily seen that real SINR obtained by really assigned subcarriers is lower bounded by the SINR computed based on pre-assigned subcarriers. Because, i, n b in α in c ik k j i k c jk α jn + σ 2 i k c ik k j i k c jk b jn + σ 2 i k In order to mitigate the interference we release a portion of the pre-assigned subcarriers which are not assigned (α in = 1, b in = 0) from cells serving all their users. A decision on releasing a subcarrier will be made by considering the induced interference of a subcarrier to the users using that subcarrier in the neighboring cells. A subcarrier with maximum inducing interference (in case it s assigned to a user) is a release candidate. The proposed algorithm for decreasing pre-assigned subcarrier is shown in Algorithm 3. Here we introduced a parameter β [0, 1] to regulate the number of subcarriers going to be released Cost benefit analysis of adding subcarriers Since there are users which do not have enough assigned subcarrier, we extend pre-assigned subcarriers in cells that are heavily loaded. In fact, the decision of adding subcarriers should be taken cleverly. Because adding a subcarrier without considering imposed interference on the other cells might cause the system increasingly unstable. To avoid the 44

57 4.5 Modifications in pre-assigned subcarriers Algorithm 3 Decrease pre-assigned subcarriers Input: a kn, α in, b in, ServingCells, η 1: for all Serving Cells i do 2: for all subcarriers n do K 3: InducedInterf erence(n) = c ik a kn (α in b in ) 4: end for 5: for t = 1 : 6: n argmax η n k=1 ( N ) α in b in do n=1 InducedInterf erence(n) 7: α in 0 release subcarrier inducing maximum interference 8: InducedInterference(n ) 0 9: end for 10: end for return α in consequences of system instability, we perform a cost benefit analysis for each subcarrier in order to ensure a specific subcarrier is beneficial in terms of providing data rate. So far, not-served users and overloaded cells are identified. The Algorithm 4 investigates the influence of adding a subcarrier on the whole system. The cost benefit analyzer algorithm (Algorithm 4) includes the following steps: STEP 1 The purpose is to derive the minimum rate each subcarrier can achieve among all the unserved users. To pursue this aim we assume not-pre-assigned subcarriers (1 α in, i, n) as pre-assigned subcarriers and compute SINR values based on them. As a result the achievable rate of subcarriers will be obtained by using sinr2rate() function. Next step is to find the minimum value of achievable rate over each subcarrier by unserved users in the same cell. In other words, this gives us the minimum benefit we gain in terms of data rate if we assume that subcarrier as a pre-assigned one. STEP 2 In this step the current rate of users over all subcarriers is computed. This is simply done by computing SINR of any user over any subcarrier based on current preassigned subcarriers. STEP 3 We compute the expected rates of subarriers under the condition that all subcarriers in an unserved cell become pre-assigned subcarrier (FRF 1). There are some remarks regarding the above two steps. Remark 1 fixing a specific user from overloaded cell, the rates correspond to newly preassigned subcarriers (α in = 0 α in = 1) in step 2 and 3 are zero cause those subcarriers weren t occupied by the user. 45

58 4 Resource allocation in downlink multi-cell OFDMA networks Algorithm 4 Cost benefit analyzer Input: c ik, a kn, α in, i, k, n 1: for all overloaded cells i do 2: for all subcarriers n do 3: for all unserved users k do c ik (1 α in ) 4: ŜINR kn = j i c jkα jn + σi 2 5: end for 6: AchievableRates = sinr2rate(ŝin R) 7: M inrate(n) min(achievablerates(k, n)) 8: end for 9: for all users l do 10: SINR ln = k c il a ln α il n j i l c jl α jn + σi 2 l 11: end for 12: CurrentRates = sinr2rate(sin R) 13: α α 14: α in 1, n 15: for all users l do 16: SINR ln = c il a ln α il n j i l c jl α jn + σi 2 l 17: end for 18: P otentialrates = sinr2rate( SIN R) 19: r CurrentRates P otentialrates 20: MinGain(n) MinRate(n) k r kn 21: end for return AchievableRates(k, n), M ingain(n), r(n) k, n 46

59 4.5 Modifications in pre-assigned subcarriers l i, n if (α in = 0 α in = 1) CurrentRate(l, n) = P otentialrate(l, n) = 0 Remark 2 fixing a specific user from overloaded cell, the rates correspond to already preassigend subcarriers (α in = 1 α in = 1) in step 2 and 3 are the same cause there is no change in interference situation for those subcarriers. l i, n if (α in = 1 α in = 1) CurrentRate(l, n) = P otentialrate(l, n) Remark 3 the rates obtained in P otentialrates for users out of aforementioned cell are less or equal to rates in CurrentRate. Because, the induced interference of newly added subcarriers affects their rates negatively. l i, n if P otentialrate(l, n) CurrentRate(l, n) The equality holds for the case user l doesn t occupy subcarrier n (a ln = 0) then both rates are zero. STEP 4 the last step is to obtain the difference between CurrentRates and P otentialrates which is always non-negative for all subcarriers due to aforementioned remarks. These values ( r) are losses in user s data rate that system suffers per subcarrier if a subcarrier becomes pre-assigend. In step 1 we obtained the minimum rate that system can achieve per subcarrier. Therefore, we subtract the loss in rates from the minimum achievable rate in step 1 so as to obtain the real (net) gain. Based on this criterion we decide if a subcarrier is beneficial to be added or not. By performing experiments we find a threshold for assessing subcarriers and filter the subcarriers with gain lower than the threshold Increase pre-assigned subcarriers In previous section we established a criterion to eliminate the subcarriers that inflict significant loss in data rates of users. In order to increase pre-assigned subcarriers we select the required number of subcarriers from the remaining ones which are beneficial for the users altogether. For this phase we propose three approaches which have the number of subcarriers to be added in common. The first approach is to take advantage of the criterion obtained in previous section. The procedure of this approach is represented in Algorithm 5. As we have not pre-assigned sufficient number of subcarriers in an overloaded cell we try to add minimum number of subcarriers to fulfill the rate demand of unserved users. To estimate a lower bound of required number of subcarriers we consider the case that there is no interference on the subcarrier. Consequently, we compute SNR for each subcarrier and 47

60 4 Resource allocation in downlink multi-cell OFDMA networks obtain its data rate using sinr2rate() function. As we do not have frequency selectivity, the SNR value for the subcarriers are equal. SNR k = c i k k σi 2 k Thus, the minimum number of subcarriers required for each user k is: R k LowBoundDemand(k) = sinr2rate(snr k ) The tuning parameter α is introduced to regulate the required number of subcarriers. In the next step, we make use of the criterion enables us to assess the subcarriers. For each unserved user we select N umof ReqSc from subcarriers with maximum Gain and add them to pre-assigned ones. Algorithm 5 Increase pre-assigned subcarriers based on cost-benefit analysis Input: µ, c ik, α ik, AchievableRates(k, n), r(n) i, k, n 1: for all overloaded cells i do 2: for all unserved users k in cell i do 3: SNR k c ik σi 2 4: r k = sinr2rate(snr k ) r k is reference data rate for all subcarriers Rk 5: LowBoundDemand(k) 6: Gain(n) AchievableRates(k, n) r k K r ln, 7: SortedGain = Sort(Gain) in descending order 8: N umof ReqSc µ.lowbounddemand(k) µ is tuning parameter 9: m, counter 1 10: while (counter NumofReqSc m N) do 11: if (SortedGain(m) T hreshold α im = 0) then 12: α im 1 pre-assign subcarrier m 13: counter counter : end if 15: m m : end while 17: end for 18: end for return Gain(n), r(n) n The second approach takes into account the minimum received interference over subcarriers. This approach simply selects the subcarriers which are less interfered by neighboring cells and add them to pre-assigned subcarriers. The receiving interference that user k tolerates over subcarrier n is: I ReceivingInterf erence(k, n) = c jk α jn j i k l=1 n 48

61 4.5 Modifications in pre-assigned subcarriers The algorithm which increase pre-assigned subcarriers based on minimum received interference criterion is shown in Algorithm 6. Algorithm 6 Increase pre-assigned subcarriers based on minimum received interference (2nd approach) Input: c ik, MinGain(n) i, k, n 1: for all overloaded cells i do 2: for all unserved users k in cell i do 3: SNR(k) c ik σi 2 4: r k = sinr2rate(snr k ) r k is reference data rate for all subcarriers Rk 5: LowBoundDemand(k) r k 6: I ReceivingInterf erence(k, n) c jk α jn j i k n 7: [SortedRI, Index] = Sort(ReceivingInterf erence) in ascending order 8: NumofReqSc k µ.lowbounddemand(k) µ is tuning parameter 9: idx, counter 1 10: while (counter NumofReqSc k idx N) do 11: m Index(idx) 12: if (MinGain(m) T hreshold α im = 0) then 13: α im 1 pre-assign subcarrier m 14: counter counter : end if 16: idx idx : end while 18: end for 19: end for return α in We propose the third approach based on minimum inducing interference. This approach simply selects the subcarriers inducing less interference to neighboring cells (if they become pre-assigned) and add them to pre-assigned subcarriers. The interference that cell i may induce to users in neighboring cells over subcarrier n is: K InducingInterf erence(n) = c il a ln, l=1 i, n The algorithm which increases pre-assigned subcarriers based on minimum inducing interference criterion is shown in Algorithm 7. So far, we have introduced the fundamental blocks required for subcarrier assignment and the proposed enhancements. 49

62 4 Resource allocation in downlink multi-cell OFDMA networks Algorithm 7 Increase pre-assigned subcarriers based on minimum inducing interference (3rd approach) Input: c ik, MinGain(n) i, k, n 1: for all overloaded cells i do 2: Find unserved users k in cell i 3: SNR k c ik σi 2 4: r k = sinr2rate(snr k ) r k is reference data rate for all subcarriers 5: NumofReqSc [µ. R k ] α is tuning parameter r k k K 6: InducingInterf erence(n) c il a ln n l=1 7: [SortedII, Index] = Sort(InducingInterf erence) in ascending order 8: idx, counter 1 9: while (counter NumofReqSc idx N) do 10: m Index(idx) 11: if (Gain(m) T hreshold α im = 0) then 12: α im 1 pre-assign subcarrier m 13: counter counter : end if 15: idx idx : end while 17: end for return α in 50

63 4.6 Release subcarriers 4.6 Release subcarriers Regarding the second strategy introduced in the beginning of this chapter, the idea is to take advantage of previously assigned subcarriers to decrease computational complexities in the current round of assignment. An important issue raises here is while changing the interference situation by modifying pre-assigned subcarriers in each iteration the subcarriers which were assigned in previous iterations might become excessive or insufficient for rate demand fulfillment. This reflects the need for releasing some of already assigned subcarriers for further assignments. Thus, we propose release subcarrier algorithm for users with over fulfilled rate demand. The purpose is to release maximum possible number of subcarriers from over fulfilled users without causing rate reduction below the user s minimum rate demand. Algorithm 8 shows subcarrier release procedure in details. We finish this Algorithm 8 Release subcarrier Input: c ik, b in, a kn, α in, R k, Rate(k) i, k, n 1: Find users l such that 0 < Rate(l) < R l 2: for all users l do 3: Find subcarriers m such that a lm = 1 4: a lm 0 5: b il l 0 6: Rate(l) 0 7: end for 8: for all cells i do c ik α in a kn 9: SINR kn j i c jkα jn + σi 2 N 10: Rate(k) sinr2rate(sinr kn ), n=1 11: for all users k do 12: (MinRate, n ) min Rates(k, n) {n b in 1} 13: while Rate(k) MinRate R k do 14: Rate(k) Rate(k) M inrate 15: b in 0 16: a kn 0 17: (MinRate, n ) min kn, Rate, a kn ) {n b in 1} 18: end while 19: end for 20: end for return b in, Rate(k), a kn i, k, n k chapter with illustration of block diagram of second strategy in Figure 4.3. Notice that, the release subcarrier block is placed before subcarrier assignment to turn off unnecessary assigned subcarriers either from over-fulfilled or not fully served users and consequently 51

64 4 Resource allocation in downlink multi-cell OFDMA networks the next block (subcarrier assignment) would be able to function effectively. Cell Assignment & Initialization Alg. 1 Release Subcarrier Alg. 8 Subcarrier Assignment Alg. 2 Decrease Pre- Assigned Subcarriers Alg. 3 Increase Pre- Assigned Subcarriers Alg. 5 (or 6,7) Compute Gains Cost-Benefit Analysis Alg. 4 Compute Achievable Rates Compute Current Rates Compute Potential Rates Figure 4.3: Block diagram of second strategy with subcarrier release phase 52

65 5 Performance evaluation 5.1 Simulation Framework In this chapter, the simulation setup is described before presenting and discussing the simulation results. In order to evaluate the algorithms proposed in Chapter 4, we take advantage of a simulated OFDMA network which is properly aligned with LTE 3GPP network. The basic platform for simulation has been developed at the Institute of Theoretical Information Technology within the joint research and development project Self Organization for 4G Multi-Tier Networks with Huawei Technologies. Our contribution is the incorporation of radio resource management. This network consists of 21 hexagonal macro cells (7 sites) and 500 users with heterogeneous traffics distributed randomly. An exemplary network layout is shown in Figure 5.1. Figure 5.1: Simulated network layout In order to improve the interference situation for the cell edge users we will use FRF 3 53

66 5 Performance evaluation as first step of subcarrier assignment. In this line, we divide the available bandwidth into three subbands, each will be assigned to one cell. As explained in subsection the basic unit in downlink transmission of LTE is the Physical Resource Block (PRB), which comprises 12 subcarriers with 15 khz bandwidth each. The effective system bandwidth is 9 GHz. Thus, there are 600 subcarriers available in each cell. System characteristics are shown in Table 5.1. System parameter Total number of users 500 Number of hexagonal cells 21 Carrier frequency Effective transmission bandwidth Number of enbs (macrocells) 12 Min.-Max. Tx power enb Propagation model enb enb antenna gain enb noise figure UE antenna gain UE Noise figure UE initial distribution Simulation area Setting 2 GHz 9 MHz (50 PRBs) 40 dbm - 46 dbm (optionally: switched off) CORLA (raytracing), omnidirectional 14 dbi 5 db 0 dbi 9 db random, uniformly distributed 2.5 km 3.5 km, 5 m resolution Table 5.1: System parameters for urban evaluation environment User Profile For UE simulation, we consider two components that model user behavior: First, the UE traffic profile describes the requested traffic type, i.e., type of service, data rate demand to meet QoS requirements, and the priority level. Table 5.2 shows the considered services and their proportions on overall traffic. We modified the traffic profile with respect to the desired evaluation scenario. We consider priority one for all services. Therefore, the number of active UEs in the system equals the sum of their priorities. 54

67 5.2 Numerical results Service Data rate [kbps] Proportions [%] VoIP Web Data Table 5.2: UE traffic profile considered for evaluation. 5.2 Numerical results Two main strategies proceeded in chapter 4 will be evaluated from different point of views. First strategy was aimed at considering the subcarrier assignments in previous iterations only as side information to promote assignments in upcoming iterations. Second strategy was aimed at continuing the subcarrier assignment in each iteration while keeping those assigned in previous iterations. Simulations are made in both strategies to assess performance of proposed algorithms in terms of indexes such as percentage of served users, spectral efficiency and number of assigned subcarriers. percentage of served users= κ 100 where κ is size of set κ = {k Rate(k) K R k, k K} and K is the size of K (set of all users). I K N spectral efficiency= r ikn /BW i=1 k=1 n=1 I N number of pre-assigned subcarriers= α in In chapter 4 we defined parameters i=1 n=1 µ, η, T hreshold which enables us to investigate dependency of system indexes to pre-assigned subcarrier changes and improve system performance. η (in Algorithm 3) is tuning parameter which regulates the number of pre-assigned subcarriers not being used anymore. µ (in Algorithm 5) is tuning parameter which regulates the number of subcarriers going to be pre-assigned. T hreshold (Algorithm 5) subcarriers with gain above this Threshold will be the candidates for pre-assigned subcarriers. Experiments are performed to investigate percentage of served users with respect to variations of the parameters. We have observed that there are fluctuates in our indexes due to consecutive or nonconsecutive increase and decrease in pre-assigned subcarriers 55

68 5 Performance evaluation of particular cells. In order to keep the system stable, we adopted a policy to limit the times that a particular cell could decrease its pre-assigned subcarriers. Thus, we stop decreasing pre-assigned subcarriers of cells where we had both decrease and increase in previous iterations. Based on several realizations we observed, after some iterations the number of pre-assigned subcarriers will not change anymore which leaves system indexes unchanged. One can keep doing iterations for further assurance. Performance of first strategy According to the first strategy in each iteration we start assigning subcarriers from scratch. The key aim of iterating the procedure is refining pre-assigned subcarriers. We run the algorithms for different values of µ and η to find intervals where the system indexes remain almost steady. In this experiment we suppose 30 iterations would be enough for system to become reasonably stable. Parameters T hreshold, µ and η vary between { 1.5, 1, 0.5, 0, 0.5, 1}, [0.1, 1.5] and [0.05, 1] respectively. Figure 5.1 represents system performance in terms of percentage of served users (1st index) after 30 iterations. It shows that more than 99% of users were served for certain µ, η intervals. Secondly, the results show that for small negative and close to zero threshold the probability of serving more users is higher. Figure 5.2: Percentage of served users with respect to T hreshold, µ and η. 56

69 5.2 Numerical results Performance of second strategy According to the second strategy we keep the subcarriers assigned in the previous iterations while extending pre-assigned subcarriers in overloaded cells and contracting pre-assigned subcarriers in the serving cells. Furthermore, subcarriers assigned to users where rate demands are either not fully satisfied or over fulfilled, will be released. The idea behind this strategy is to take advantage of subcarriers previously assigned, so that in each iteration there exists some users with satisfied rate demand and consequently less time is required for serving the remaining ones. In order to evaluate performance of this strategy we will perform similar experiment with the same parameters. Parameters T hreshold, µ and η vary between { 1.5, 1, 0.5, 0, 0.5, 1}, [0.1, 1.5] and [0.05, 1] respectively. Figure 5.2 represents system performance in terms of percentage of served users (1st index) after 30 iterations. It shows that more than 99% of users were served for certain µ, η intervals. Secondly, the results show that for small negative and close to zero threshold the probability of serving more users is higher. Figure 5.3: percentage of served users with respect to T hreshold, µ and η. Comparing Figures 5.1 and 5.2 we can observe that first strategy outperforms the second one for T hreshold = 0 and negative values close to zero. In order to evaluate performance of first and second strategy in terms of spectral efficiency and number of pre-assigned subcarriers we perform certain number iterations for specific µ and η. The result is shown 57