DU LAYOUT_Layout 1 4/19/13 3:36 PM Page 44

NEXT GENERATION COGNITIVE CELLULAR NETWORKS

COGNITIVE FEMTOCELL NETWORKS: AN OPPORTUNISTIC SPECTRUM ACCESS FOR FUTURE INDOOR COVERAGE

LI HUANG AND GUANGXI ZHU, HUAZHONG UNIVERSITY OF SCIENCE AND TECHNOLOGY XIAOJIANG DU, TEMPLE UNIVERSITY

ABSTRACT Advanced) standard has been developed to sup- port higher throughput and better user experi- Femtocells have emerged as a promising solu- ence. Moreover, it is predicted that in the near tion to provide wireless access cover- future a large amount of traffic (about 60 percent age in cellular dead zones and indoor of voice traffic and 90 percent of data traffic) will environments. Compared with other techniques originate from indoor environments (e.g., resi- for indoor coverage, femtocells achieve better dential home and office) [1]. However, mobile user experience with less capital expenditure and cellular networks have a reputation for poor maintenance cost. However, co-channel deploy- indoor coverage due to the penetration losses of ments of closed subscriber group femtocells cause walls. The attenuation is more prominent at coverage holes in due to co-channel higher frequency in LTE-Advanced. The tradi- interference. To address this problem, cognitive tional cell splitting with denser (BS) radio technology has been integrated with femto- deployments cannot improve network capacity cells. CR-enabled femtocells can actively sense and coverage much. This is because the cell split- their environment and exploit the network side ting gains are severely reduced by high intercell information obtained from sensing to adaptively interference. Furthermore, dense macro BS mitigate interference. We investigate three CR- deployments cause high capital expenditure and enabled interference mitigation techniques, operational cost. Hence, it is important to find an including opportunistic interference avoidance, alternative strategy to improve indoor coverage. interference cancellation, and interference align- Femtocells offer a promising solution for ment. activities can be obtained with- indoor communications. A femtocell access point out significant overhead in femtocells. In this (FAP) is a low-power, low-cost, short-range, article, we present a joint opportunistic interfer- plug-and-play cellular BS deployed in a residen- ence avoidance scheme with Gale-Shapley spec- tial area or small office. FAPs enhance system trum sharing (GSOIA) based on the interweave capacity and coverage via improving link quality paradigm to mitigate both tier interferences in and enabling spectral reuse. In femtocells, higher macro/femto heterogeneous networks. In this data rates and better coverage are achieved by scheme, cognitive femtocells opportunistically shorter communication distance and less penetra- communicate over available spectrum with mini- tion loss. Furthermore, high cellular capacity is mal interference to macrocells; different femto- reached by efficient spatial spectrum reuse in a cells are assigned orthogonal spectrum resources smaller cell size. Femtocells not only provide bet- with a one-to-one matching policy to avoid intra- ter (QoS) to indoor users, but tier interference. Our simulations show consider- also diminish site expenditure, maintenance cost, able performance improvement of the GSOIA and power consumption of base stations [2]. scheme and validate the potential benefits of CR- Nowadays femtocells have attracted a lot of enabled femtocells for in-home coverage. attention from both academia and industry. The channel deployment of femtocells in a NTRODUCTION two-tier has three I options: dedicated-channel deployment, partial- Mobile service has spurred exponential channel-sharing deployment, and co-channel growth in usage, such as video deployment. Co-channel deployment is more sharing, game playing, and movie downloading. attractive to operators due to low cost and back- The explosion of data traffic volume has been ward compatibility. However, co-channel deploy- further fueled by and various ments of closed subscriber group (CSG) portable devices. To enhance mobile broadband femtocells create coverage holes in macrocells. access, the Third Generation Partnership Project In order to solve this problem, cognitive femto- (3GPP) Long Term Evolution Advanced (LTE- cells are developed to sense their surroundings

44 1536-1284/13/$25.00 © 2013 IEEE IEEE Wireless Communications • April 2013 DU LAYOUT_Layout 1 4/19/13 3:36 PM Page 45

Specifications Femtocell DAS Relay Wi-Fi

Outdoor: 250 Outdoor: Typical power 10–100 mW mW–2W, indoor: 250 mW–2W, 100–200 mW < 100 mW indoor: < 100 mW

Coverage Coverage extension Coverage extension of 20–50 m 150 m 100–200 m range of macro macro

Real-time voice and Real-time voice Real-time voice and Real-time voice and Primarily data and Services data and data data data VOIP

Deployment Hot spot/office/ In-home/office Hot spot/office Indoor extension In-home/hot spot scenarios tunnel, high-speed train

Access mode Closed/open/hybrid Open access Open access Open access Closed/open access

DSL/cable/optical or RF Wireless in-band or DSL/cable/optical Backhaul X2 interface fiber links to macrocell out-of-band fiber

Peak data rate LTE-Advanced ( 3GPP R10): 1G b/s (DL) 300M b/s(UL) 600 Mb/s (802.11n)

Table 1. Comparison of various indoor wireless access technologies.

and flexibly adapt their operations to minimize tive alternatives to macrocells, they are still too interference. In the cognitive paradigm, interfer- expensive to be deployed in a house or small ence mitigation approaches can be classified as office. As an energy-saving, cost-efficient small opportunistic interference avoidance, interfer- cellular base station, FAP is suitable for a resi- ence cancellation, and interference alignment. dential home that connects with the service The goal of this article is to study how cogni- provider’s network via home broadband. tive radio (CR)-enabled femtocells mitigate both Wi-Fi is another popular wireless broadband inter-tier and intra-tier interference to improve access technology. A Wi-Fi router is also a plug- indoor coverage in a two-tier heterogeneous net- and-play access point without time-consuming work. We first compare femtocells with other network planning. Wireless LAN (WLAN) main- existing technologies for indoor coverage. Next, ly provides data service and voice over IP (VoIP), we investigate the problem of macrocell coverage while a femtocell realizes any call in real time. holes caused by co-channel CSG femtocells. Then Although the carrier sense multiple access with we study three CR-enabled interference mitiga- collision avoidance (CSMA/CA) mechanism in tion approaches in co-channel deployment. A WLAN is a simple distributed coexistence solu- joint opportunistic interference avoidance scheme tion, it is not robust due to the static resource with Gale-Shapley spectrum sharing (GSOIA) is allocation nature. Because of the selfish coexis- presented to mitigate both intra- and inter-tier tence strategy, their mutual interference acutely interference. Finally, simulations are conducted to increases with widespread deployments of evaluate performance improvement of the pro- WLAN. However, low-cost WLAN is still very posed scheme on interference mitigation. attractive for free spectrum license on the indus- trial, scientific, and medical (ISM) band. In order to compete with Wi-Fi technology, FAP should INDOOR COVERAGE: improve performance at a reasonable price. EXISTING TECHNOLOGIES Various technologies for indoor coverage are FEMTOCELL DEPLOYMENT AND summarized in Table 1. In LTE-Advanced, fem- TECHNICAL CHALLENGES tocells, , relays, and distributed antenna systems (DASs) are low-power nodes in a het- FEMTOCELL DEPLOYMENT SCENARIOS erogeneous network [3]. Picocells are located inside hotspots like airports, shopping malls, and Femtocells are overlaid within macrocells in a stadiums. Picocells work as simplified macrocells two-tier heterogeneous network. Allocating with low power and reduced cost. DASs com- spectrum resources between femtocells and prise many separate antenna elements (AEs) macrocells is a very important issue. There are connecting to macro BSs via dedicated fiber three possible strategies for femtocell resource cables or radoi frequency (RF) links to extend deployment: dedicated-channel deployment, par- macro coverage. A set of low-power distributed tial-channel-sharing deployment, and co-channel AEs replace high-power centralized antennas to deployment [3]. cover the same cell area. Relays are installed in 1) Dedicated-channel deployment: Femtocells cellular dead zones without wired backhaul to are allocated a dedicated carrier frequency dif- compensate for the attenuation loss of barriers, ferent from those of macrocells. This deploy- such as buildings, tunnels, or high-speed trains. ment is a simple solution to avoid mutual Although picocells/DASs/ relays are cost-effec- interference between the two tiers, but the spec-

IEEE Wireless Communications • April 2013 45 DU LAYOUT_Layout 1 4/19/13 3:36 PM Page 46

Inter-tier interference (or cross-tier interfer- 1# 6# CFAP E# CFAP C# L1 MBS ence) and intra-tier L6 5# interference coexist in this network. For 3# inter-tier interference, L2 L10 2# L5 CFAP D# the aggressor and L7 the victim of interfer- L3 L9 ence belong to FeUE L4 L11 L8 different tiers, while MacUE 4# CFAP A# CFAP B# they belong to the MBS

same tier in intra-tier CFAP Internet

interference. Inter-tier interference (L3,L8) CFAP uplink (L4) MBS uplink (L1,L2) Intra-tier interference (L11)

CFAP downlink (L9) MBS downlink (L6,L7) Negligible interference (L5,L10)

Figure 1. Interference scenarios related to femtocells in co-channel CSG deployment.

tral usage is inefficient due to seg- interference scenarios in a heterogeneous net- mentation. work. When a macro base station (MBS) is 2) Partial-channel-sharing deployment: The transmitting to macro user equipment (MacUE) overall bandwidth is segmented into two parts. 4# and 6# in the macro downlink (DL), the One part is exclusively assigned to macro users, downlink of unregistered MacUE 4# (L7) is and the other part is shared by macrocells and interfered by nearby closed CFAP B# (L8). femtocells. Macro users benefit from ubiquitous Strong interference (L8) possibly causes com- coverage on exclusive carrier frequency and par- munication outage of MacUE 4#. On the other tial coverage (outside the femtocell coverage) on hand, interference (L10) from CFAP B# to sharing carrier frequency. This deployment is MacUE 6# is negligible due to the long dis- efficient without causing much bandwidth loss tance between them. Hence, DL resource and mutual interference, but a portion of spec- blocks (RBs or channels) of MacUE 6# can be trum and high-cost carrier-aggregation-capable reused by CFAP B#. Similarly, interference terminals are required. from MacUE 2# to CFAP A# (L3) is strong, 3) Co-channel deployment: In this deploy- while interference from MacUE 1# to CFAP ment, spectral usage is high because femtocells A# (L5) is weak in the macro uplink (UL). are deployed in the same carrier frequency as Since interference between remote macro users macrocells without bandwidth segmentation. This and CFAP is negligible, RBs of macro users low-cost and backward-compatible strategy does can be reused by a distant CFAP. More spec- not rely on high spectrum availability and carrier trum holes can be explored by sensing locations aggregation support at terminals. Co-channel of macro users. deployment is especially attractive for operators Unique features of femtocells pose challenges in crowded spectrum. Nevertheless, co-channel to interference management in a two-tier hetero- deployments of CSG femtocells create coverage geneous network. The ad hoc nature of femto- holes in macrocells due to the inter-tier interfer- cells makes centralized network planning ence. The details are given below. infeasible. Moreover, intercell interference coor- dination (ICIC) cannot be implemented due to MACRO LAYER COVERAGE HOLES the lack of a direct X2 interface in LTE- Interference in macro/femto heterogeneous net- Advanced. The delay of information exchange works is more complicated than that of macro- via wired backhaul (S1 interface) is too long to cells. Inter-tier interference (or cross-tier allow for efficient coordination. A static opera- interference) and intra-tier interference coexist tion, administration, and maintenance (OAM) in this network. For inter-tier interference, the based solution in co-channel CSG deployment is aggressor and the victim of interference belong just “time-of-day” adaptation, which is unable to to different tiers, while they belong to the same adapt to the instantaneous characteristics of tier in intra-tier interference. actual network traffic. As an intelligent wireless Coverage holes in macrocells are caused by technology, cognitive radio provides an adaptive co-channel interference (CCI) in CSG deploy- solution to mitigate interference in dynamic ments of femtocells. Figure 1 illustrates various environments.

46 IEEE Wireless Communications • April 2013 DU LAYOUT_Layout 1 4/19/13 3:36 PM Page 47

INTERFERENCE MITIGATION IN COGNITIVE Self-configuration FEMTOCELL Cognitive module module RRC COGNITIVE FEMTOCELL High layer Traffic Spectrum layer (RLC/PDCP) estimation mobility (QoS requirement) (handover) Cognitive radio technology holds great hope to Cognitive engine break spectrum gridlock through advanced radio techniques and novel coexistence protocols. In order MAC Spectrum to meet stringent requirements of the LTE- MAC statistics state Spectrum sharing database (interference Advanced standard, cognitive radio is integrated layer (collision management) with femtocells to eliminate interference. Cognitive probability) (learning) capability and self-configured capability are the main characteristics of cognitive femtocells [4]. Awareness of the radio scene is essential for CFAPs Spectrum Spectrum configuration to obtain available spectrum opportunities in macro- PHY sensing layer (AMC/power/ cell networks. Spectrum opportunities exist in vari- (spectrum holes) RB/antenna) ous domains: frequency, time, geographical space, code, antenna spatiality [5]. Self-configuration capa- bility helps CFAPs to coexist in a two-tier network without impact on existing cellular networks. Figure 2. Cognitive radio technology in femtocells. Figure 2 shows cognitive radio technology in femtocells from the cross-layer perspective. The cognitive and self-configuration modules are syn- messages to mitigate interference in a two-tier thesized by a cognitive engine. The cognitive mod- cellular network [6]. In the cognitive paradigm, ule first senses the environment and collects specific interference mitigation techniques include information of all layers (e.g., spectrum holes, colli- opportunistic interference avoidance, interfer- sion probability, and QoS requirement). Then the ence cancellation, and interference alignment. cognitive engine analyzes spectrum characteristics Table 2 compares the three techniques. and estimates available resources. Finally, the self- configuration module exploits cognitive information Opportunistic Interference Avoidance — Since femto- in a spectrum state database to optimize parame- cells are overlaid within macrocell networks, ters of all layers in dynamic surroundings. opportunistic spectral access can be exploited for The self-configuration module includes three the hierarchal overlay system to avoid inter-tier components: spectrum mobility, spectrum sharing, interference. After sensing macrocell activities and spectrum configuration. Spectrum mobility that acquire available spectral resources, CFAPs management enables femtocells to smoothly hand exploit unoccupied spectrum holes or white over, which minimizes their performance degra- spaces to minimally disrupt macrocell communi- dation as much as possible when macro channels cations. Accurate macrocell activities are are unavailable or current channel conditions required for orthogonal resource allocation to become worse. Spectrum sharing management avoid inter-tier interference. Opportunistic avoids co-channel collisions of multiple femto- access improves overall spectral efficiency from cells, like the medium access control (MAC) a perspective of temporal, spectral and spatial mechanism in many networks. The spectrum con- reuse of RBs between macrocell and femtocells. figuration module ensures that femtocells work in the best available channels. Configurable parame- Interference Cancellation — This scheme is suitable ters in the physical layer include time-frequency for strong interference that is demodulated or resource blocks of orthogonal frequency-division decoded along with channel estimation to cancel (OFDM), modulation and coding received interference from desired signals. Inter- schemes, transmission power, and antenna angle. ference cancellation requires knowledge of The ognitive module provides femtocells with macrocell messages and codebooks. Dirty paper radio awareness capability, while the self-configura- coding (DPC) or sphere decoding techniques tion module enables femtocells to adapt to a dynam- may be used to cancel interference signals from ic environment. Any change sensed by the cognitive the desired signal [7]. module automatically triggers an adjustment. With learning and reasoning capabilities, cognitive femto- Interference Alignment — In this scheme, all inter- cells keep track of variations and estimate spectrum ference is restricted into approximately half of state from historical data in a dynamic environment. the signal space at each receiver, leaving the Since CFAPs have potential to adaptively allocate other half interference-free space for the desired radio resources and efficiently mitigate co-channel signal. The enabling premise for perfect interfer- interference (CCI) in a two-tier network, it is ence alignment is that cognitive femtocells have promising to apply CR-enabled femtocells to break knowledge of global channel state information. through the spatial reuse barrier of cellular systems. Although interference alignment achieves more degrees of freedom under a time-varying chan- COGNITIVE INTERFERENCE nel, global channel knowledge must be learned before its benefits translate into practice [8]. MITIGATION OVERVIEW The three approaches differ in the nature of CR-enabled femtocells can exploit side informa- the side information obtained from sensing. tion about their environment, including macro- Opportunistic interference avoidance requires cell activities, channel gains, codebooks, and femtocells to communicate using spectral holes,

IEEE Wireless Communications • April 2013 47 DU LAYOUT_Layout 1 4/19/13 3:36 PM Page 48

Opportunistic interference Interference cancellation Interference alignment avoidance

Activities of macrocells in multiple Side information Channel gains, codebooks, and mes- domains: time-frequency, Global channel state information requirement sages of macrocells space/location, antenna spatiality

Align all interference in half of the Interference Exploit spectrum holes to avoid Dirty paper coding (DPC) or sphere received signal space, and leave the mitigation interference in the interweave decoding in the overlay paradigm other half interference-free for the approach paradigm. desired signal.

CT CR ^ ^ W1 H11 W 1 W1 ^ Message W 1 T1 R1 T1 R1 W 1 T1 R1 H21 W1 H21H31 H12H13 H31 CT12 CR ^ Interference CT CR CT CR ^ W2 H W 2 W2 ^ Message W 2 T2 R2 channel model T2 R2 W 2 T2 R2 H22 W2 H12H32 H32 H21H23 Activity of T1 W1 W1 CTH13 CR ^ W3 H23 W 3 Signal link Interference link T3 R3 Signal link Interference link H33 H13H23 H31H32

Table 2. Comparison of CR-enabled interference mitigation techniques.

so ideally they cause no interference. In contrast, cally monitor resource occupancy state and exploit interference cancellation exploits strong interfer- spectrum holes to communicate. Once macrocell ence through sophisticated coding strategies to users are present in these resources, CFAPs facilitate communications for others. Interfer- vacate spectrum holes immediately. ence alignment orthogonalizes all interfering sig- nals and the desired signal to avoid interference. OPPORTUNISTIC INTERFERENCE AVOIDANCE With different network side information, the The challenge of opportunistic interference avoid- three techniques achieve different network ance is how to reliably detect spectrum holes in degrees of freedom. Because macrocell messages macrocell networks. It is noted that spectrum holes or channel knowledge must be acquired with are not only blank macrocell RBs, but also interfer- more communication overhead in the interfer- ence-free RBs allocated to faraway macro users from ence cancellation and interference alignment femtocells. Instead of using a central database of approaches, an opportunistic interference avoid- spectrum usage lists in TV white space bands, CFAPs ance scheme is desirable to mitigate both types detect macrocells’ activities via two methods: listening of interference with less complexity. to scheduled information of a macro BS (MBS) or actively sensing their environment. A hybrid scheme THE GSOIA SOLUTION combining two approaches is summarized below: • A CFAP listens to scheduled information from In this article we present GSOIA to mitigate MBS over the air. As MBS broadcasts sched- both tiers’ interferences. We propose this inter- uled information in a signaling channel with weave-paradigm-based scheme because macro- relatively high transmission power, any CFAP cell activities can be obtained without significant in the cell can get knowledge of resource overhead in femtocells. The goal is to oppor- maps. The “eavesdropping” mechanism is rea- tunistically exploit spectrum holes to improve sonable in practice because MBSs and CFAPs network spectral utilization. The key ideas are: belong to the same operators. • With the opportunistic interference avoidance • CFAP periodically senses spectrum to identify approach, CFAPs exploit orthogonal spectrum nearby MacUE. Various techniques of spec- resources (idle time-frequency RBs of a trum sensing are applied to identify macrocell macrocell) obtained from sensing to avoid activities, such as energy detection, cyclosta- inter-tier interference. tionarity feature detection, waveform-based • GSOIA: A one-to-one policy to avoid multi- sensing, matched filtering, and radio identifica- user collisions is utilized to mitigate intra-tier tion. A simple method is to measure the inter- interference. ference power of each RB. In LTE-Advanced, In the GSOIA scheme, macrocells and femto- the received interference power in DL/UL is cells are modeled as primary users (PUs) and sec- adopted as a mandatory sensing quantity in ondary users (SUs) in a CR network, respectively. BSs. Hence, our approach can be applied to Figure 3 shows the procedure of the GSOIA LTE-Advanced without hardware modification. scheme. First, CFAPs independently sense radio • CFAP compares sensing outcomes with sched- surroundings to explore spectrum holes. Then uled information to identify spectrum oppor- spectrum sharing based on Gale-Shapley (G-S) tunities. The detailed signal processing theory provides a one-to-one matching policy to procedure is discussed in [9]. CFAP updates a allocate available spectrum holes among multiple pool of unoccupied spectrum resources and CFAPs in a distributed fashion. CFAPs periodi- only selects idle channels to communicate.

48 IEEE Wireless Communications • April 2013 DU LAYOUT_Layout 1 4/19/13 3:36 PM Page 49

The hybrid sensing scheme improves reliability Because macrocell of spectrum detection in which scheduled infor- Opportunistic interference avoidance mation increases the accuracy of spectrum sens- messages or channel ing, and spectrum sensing obtains more spectrum Listen to scheduled information opportunities by identifying nearby macro users. knowledge must be After acquiring a pool of available spectrum resources, the subsequent challenge is how multi- Identify nearby MacUE by sensing acquired with more ple CFAPs share spectral resources. Collisions happen when the same channel is simultaneously communication over- selected by neighboring CFAPs (collocated Jointly analyze spectrum opportunities head in the interfer- CFAPs). The Gale-Shapley spectrum sharing scheme is propounded to resolve this issue. Set backoff timers according to ence cancellation GALE-SHAPLEY SPECTRUM SHARING utilities of spectrum opportunities and interference Intuitively, a one-to-one matching policy between CFAPs and channels is a solution to multi-user Wait for backoff timer expiration alignment approach- collisions. The Gale-Shapley theorem can es, an opportunistic achieve a stable one-to-one matching by an iter- N ative procedure. Hence, it is a feasible scheme to No busy tone interference avoid- avoid interference among collocated CFAPs. In the context of cognitive spectral allocation, Y ance scheme is desir- the preference of a CFAP for a channel is denot- Access the corresponding channel ed by channel utility in the Gale-Shapley theorem. able to mitigate both Channel utility R is denoted by the normalized Gale-Shapley spectrum sharing transmission rate of CFAP on a channel. interferences with Figure 3. The procedure of the GSOIA scheme. less complexity. ⎡TT− s ⎤ R = ⎢ ⎥ ××IEKC() × ⎣ T ⎦ (1)

CH1 CH2 CH3 CH4 CH5 where Ts is the channel sensing and waiting time, I is the idle indication of the channel, E(K) indicates multi-user collisions and is inversely User1 5(0.95) 11(0.89) 40(0.60) 22(0.78) 13(0.87) proportional to the number K of competitive users, and C is the channel capability. A one-to- User2 25(0.75) 30(0.70) 35(0.65) 14(0.86) 29(0.71) one matching function between the set of CFAPs and the set of available channels can be defined. User3 12(0.88) 28(0.72) 21(0.79) 4(0.96) 17(0.83) The essence of Gale-Shapley spectrum shar- ing is multichannel opportunistic sensing along Table 3. Utility matrix. with stable channel access control. It has been proven that a stable matching always exists in cognitive access control [10]. Lacking central spectrum sharing achieves a stable one-to-one control, backoff timers are triggered to imple- channel selection in a distributed fashion. ment distributed allocation. The Gale-Shapley spectrum sharing method is described as follows: PERFORMANCE EVALUATIONS • Each CFAP first calculates the utility of every channel in a spectrum resource pool. Then the We conduct simulation experiments to evaluate CFAP sets a backoff timer for every channel the performance improvement of the GSOIA that is inversely proportional to channel utility. scheme, and the simulation results are shown in • When a backoff timer expires, the CFAP detects Fig. 5 and Fig. 6. Femtocells are randomly the busy tone of the corresponding channel. If deployed in the coverage area of a macrocell, there is no busy tone over this channel, CFAP and macro users are dispersed at random loca- immediately selects it for communications. Oth- tions. Users are assumed to have continuous erwise, CFAP abandons this channel and waits data flows to transmit. A group of collocated for next backoff timer expiration. femtocells are considered instead of standalone • The process continues until each CFAP cap- femtocells that have no intra-tier interference. tures one channel or all channels are allocated. Perfect sensing and synchronization of all CFAPs To clearly explain this scheme, we give an are assumed for the GSOIA scheme. If multiple example of the Gale-Shapley spectrum sharing CFAPs access the same channel, only one trans- method. In Table 3, there is a 3 ¥ 5 utility matrix mits successfully. The achievable throughput of a for three CFAPs and five channels. Each ele- femtocell in each available channel is normal- ment is the utility of one user-channel pair, and ized. A random access scheme is chosen as the the value in parentheses is the backoff time. benchmark. Figure 5 shows the throughput gain Backoff time is inversely proportional to channel of femtocells for different macro-user locations, utility, which is calculated by measurement or by which is used to evaluate the benefit of oppor- estimation from historical data. Table 4 shows tunistic access. In Fig. 6, we compare the overall the ping process. Figure 4 depicts channel allo- femtocells throughput of the GSOIA scheme cation results in which each channel is selected with that of random access under different fem- by one CFAP without collisions according to tocell numbers to validate the performance of channel utility order. It shows that Gale-Shapley Gale-Shapley spectrum sharing.

IEEE Wireless Communications • April 2013 49 DU LAYOUT_Layout 1 4/19/13 3:36 PM Page 50

with sensing capability have potential to explore Time User1 User2 User3 more spectrum holes, and the Gale-Shapley spectrum sharing approach greatly improves t=0 CH3 CH3 CH2 overall femtocells throughput. Figure 6 shows performance gains of the t=0.60 CH3 CH3 CH2 GSOIA scheme over the random access scheme under different numbers of collocated femto- t=0.65 CH2 CH2 cells. Twenty channels (RBs) are available to collocated femtocells, and 10 percent of macro t=0.70 CH2 CH3 users are located within the coverage areas of femtocells. Note that when the number of fem- t=0.72 CH3 tocells is more than 20 (the number of available RBs), the performance of the random access t=0.79 CH5 scheme gradually decreases due to more colli- sions with increasing femtocells. However, the t=0.83 CH5 performance of the GSOIA scheme is not impaired as it is a one-to-one matching policy. Table 4. Gale-Shapley ping process. For example, when the number of femtocells and available RBs is the same (20), the GSOIA scheme can provide 170 percent throughput Figure 5 plots the overall femtocells through- improvement over the random access scheme. put with and without sensing capability under The simulation result shows that the GSOIA different ratios of macrocell resource vacancies. scheme considerably outperforms the random Ten femtocells are deployed in the coverage access approach, especially in dense femtocell area of a macrocell. The no-sensing scheme deployments. means that spectrum holes are obtained only by listening to scheduled information of MBS, while CONCLUSION the sensing scheme denotes a hybrid scheme combining sensing with listening to scheduled In LTE-Advanced, femtocells are developed to information. The overall throughput improve- support higher throughput and better user expe- ment of femtocells by the sensing scheme is rience in homes and offices. Because of better more significant than that of the no-sensing link quality and efficient spectral reuse, femto- scheme. The reason is that femtocells can cells not only improve indoor coverage and net- acquire additional interference-free spectrum work capability, but also have less capital opportunities of remote macro users. Spectrum expenditure and maintenance cost than other opportunities are closely related to macrocell low-power nodes in macrocell networks. With resource vacancies as well as the value of a. The widespread deployments of femtocells, interfer- parameter a denotes the percentage of macro ence management is a challenge due to their users in the vicinity of femtocells. It is observed unique features such as ad hoc mode, co-channel that as a increases from 50 to 75 percent, femto- interference, and no direct coordination inter- cells suffer higher inter-tier interference from face. To address the problem of macrocell cover- ambient macro users and have fewer spectrum age holes caused by co-channel interference, opportunities. When the macrocell resource CR-enabled femtocells are studied to mitigate vacancy is 30 percent, sensing capability can complicated interference in the two-tier hetero- offer about 53 percent more gain than a no-sens- geneous network. In the cognitive paradigm, ing scheme (a = 50 percent), and the GSOIA interference mitigation techniques include scheme can provide an additional threefold per- opportunistic interference avoidance, interfer- formance improvement over a random access ence cancellation, and interference alignment. scheme in terms of overall femtocells through- Different side information is required in these put. Figure 5 shows that CR-enabled femtocells techniques to mitigate interference and increase

Frequency

CH5

CH4

CH3

CH2

CH1

Time MBS CFAP1# CFAP2# CFAP3#

Figure 4. Gale-Shapley spectrum sharing.

50 IEEE Wireless Communications • April 2013 DU LAYOUT_Layout 1 4/19/13 3:36 PM Page 51

the network degrees of freedom. Based on the fact that the side information of macrocell activi- Served femtocells throughput vs. macrocell resource vacancy ties is available without significant overhead in 10 femtocells, we propose a joint opportunistic 9 interference avoidance scheme with Gale-Shap- Random, (no sensing) ley spectrum sharing to mitigate both inter-tier 8 and intra-tier interferences. With the oppor- GSOIA, (no sensing) Random, a = 75% (sensing) tunistic interference avoidance approach, CFAPs 7 GSOIA, a = 75% (sensing) interweave their signals with those of macrocells Random, a = 50% (sensing) without significantly impacting their communica- 6 GSOIA, a = 50% (sensing) tions. Gale-Shapley spectrum sharing provides a one-to-one matching policy to avoid collisions 5 among multiple femtocells. Simulation results demonstrated considerable performance 4 improvement of our scheme and showed that CR-enabled femtocells have good potential to 3 break the spatial reuse barrier of cellular sys- 2 tems. Overall served femtocells throughput ACKNOWLEDGMENT 1 This research is supported by the Key Project of 0 the National Natural Science Foundation of 0 20 40 60 80 100 (Grant No. 61231007), and by the U.S. Macrocell resource vacancy (%) National Science Foundation under grants CNS- 0963578, CNS-1022552 and CNS-1065444. Figure 5. Comparison of GSOIA vs. random spectrum sharing, scenario 1: dif- REFERENCES ferent percentages of macrocell resource vacancies and macro users inside 10 [1] D. Lopez-Perez et al., “Interference Avoidance and Dynamic co-channel CSG femtocells coverage. Frequency Planning for WiMAX Femtocells Networks,” Proc. IEEE Int’l. Conf. Commun. Sys., Nov. 2008. [2] K. Zheng, Y. Wang, W. Wang, “Energy-Efficient Wireless In- Home: the Need for Interference-Controlled Femtocells,” Served femtocells throughput vs. number of femtocells IEEE Wireless Commun., vol. 18, no. 6, 2011, pp. 36–44. [3] A. Damnjanovic et al., “A Survey on 3GPP Heterogeous 20 Networks,” IEEE Wireless Commun., vol. 18, no. 3, 18 Random spectrum sharing 2011, pp. 10–21. GSOIA spectrum sharing [4] I. F. Akyildiz et al., “Next Generation/Dynamic Spectrum Access/Cognitive Radio Wireless Networks: A Survey,” Com- 16 puter Network J., vol. 50, no. 13, 2006, pp. 2127–59. [5] T. Yucek and H. Arslan, “A Survey of Spectrum Sensing 14 Algorithms for Cognitive Radio Applications,” IEEE Com- mun. Surveys & Tutorials, vol. 11, no. 1, 2009, pp. 116–30. [6] A. Goldsmith et al., “Breaking Spectrum Gridlock With 12 Cognitive Radios: an Information Theoretic Perspec- tive,” IEEE Proc., vol. 97, no. 5, 2009, pp. 894–14. 10 [7] G. Boudreau et al., “Interference Coordination and Can- cellation for Networks,” LTE Part II: 3GPP Release 8, IEEE Commun. Mag., vol. 47, no. 4, 2009, pp. 74–81. 8 [8] V. R. Cadambe and S. A. Jafar, “Interference Alignment and Degrees of Freedom of the K-User Interference 6 Channel,” IEEE Trans. Info. Theory, vol. 54, no. 8, 2008, pp. 3425–41. 4 [9] M. E. Sahin et al., “Handling CCI and ICI in OFDMA Femtocell Networks through Frequency Scheduling,” Overall served femtocells throughput IEEE Trans. Consum. Electronics, vol. 55, no. 4, 2009, 2 pp. 1936–44. [10] A. Leshem, E. Zehavi, and Y. Yaffe, “Multichannel 0 Opportunistic Carrier Sensing for Stable Channel Access 0 5 10 15 20 25 30 35 40 Control in Cognitive Radio Systems,” IEEE JSAC, vol. 30, Number of femtocells no. 1, 2012, pp. 82–95. BIOGRAPHIES Figure 6. Comparison of GSOIA vs. random spectrum sharing, scenario 2: dif- LI HUANG received her B.S. degree in electronic engineering ferent numbers of collocated femtocells share the same amount of spectrum in 1996 and her M.S. degree in power electronic engineer- resources (RBs = 20). ing in 1999, both from Wuhan University, China. She has worked on system design and evaluation in telecommuni- cation corporations for nearly 10 years. Now she is work- XIAOJIANG DU [SM] is currently an associate professor in the ing toward her Ph.D. degree in electronics and information Department of Computer and Information Sciences at Tem- engineering at Huazhong University of Science and Tech- ple University. He received his B.E. degree from Tsinghua nology (HUST), China. Her research interests include cogni- University, China, in 1996, and his M.S. and Ph.D. degrees tive radio technology and spectrum resource allocation. from the University of Maryland, College Park in 2002 and 2003, respectively, all in electrical engineering. His research GUANGXI ZHU received his B.S. in radio engineering from interests are security, systems, wireless networks, and com- Huazhong Institute of Technology, China, in 1969. Since puter networks. He has published over 100 journal and 1974 he has held a variety of academic positions at HUST. conference papers in these areas, and has been awarded He is a professor and chairman of the academic board of $3 million in research grants from the U.S. National Sci- the Department of Electronics and Information Engineer- ence Foundation and Army Research Office. He is a Life ing, HUST. His research interests include wireless communi- Member of ACM. cation and media signal processing, and he has published over 300 research papers.

IEEE Wireless Communications • April 2013 51