Spectrum allocation with beamforming antenna in heterogeneous overlaying networks

Transcription

1 2st Annual IEEE International Symposium on Personal, Indoor and Mobile Radio Communications Spectrum allocation with beamorming antenna in heterogeneous overlaying networks Sunheui Ryoo, Changhee Joo and Saewoong Bahk INMC, School o EECS, Seoul National University, Korea. Dept. o Inormation and Communication Eng., KUT, Korea. Abstract Two-tier overlay networks that consist o a conventional macrocell network and emtocell hotspots oer an economical solution or high user capacity and extended coverage. However, wireless intererence across tiers causes signiicant perormance degradation and restricts spectrum reuse. In this paper, we explore schemes to mitigate cross-tier intererence with beamorming antennas or overlay networks. In our model, emtocells can operate with requency spectrum that is either shared with or separated rom the macrocell. The enhanced SIR rom beamorming contributes to the population o emtocells with the shared spectrum, anus improve the spectrum eiciency. Given a required SIR level, we show that which emtocells can use the shared spectrum and how much spectrum can be shared to maximize total utility. We show through a numerical perormance evaluation that proposed schemes improve spectrum utilization or two-tier overlay networks. Index Terms Femtocell, cross-tier intererence, beamorming antenna, spectrum allocation. I. INTRODUCTION Recently, a personal-use indoor base station, emtocell, has gained considerable attention or extended in-home coverage and high speed wireless broadband services where usual macrocell system can only provide degraded service indoors or provide no coverage at all. The cells with dierent sizes can be deployed in a hierarchical cell structure to provide multitier network connectivity [], [2]. The beneits or this arrangement are capacity gain, better coverage, and reduced battery consumption o handset [3]. The cross-tier intererence caused by active users can lead to unacceptable perormance degradation over heterogeneous overlay network links. To avoid intererence, the limited spectrum is divided into separate bands and each tier operates in a dedicated spectrum portion. With this partitioned spectrum usage, the cross-tier intererence can be lowered, but multiplexing gain would reduce. On the other hand, by sharing all the spectrum, each radio tier has more available amount o spectrum while suering rom higher cross-tier intererence. The hybrid spectrum usage can be used to take advantage o the two dierent spectrum usages. The idea has been proposed in [4] where the authors calculates the intererence-limited coverage area o a co-channel emtocell base station to determine the inner region or shared spectrum This work was supported by the IT R&D program o MKE/KEIT. [KI89, Intelligent Wireless Communication Systems in 3 Dimensional Environment] Fig.. th d Overlay network with two dierent types o user equipment. and outer region or partitioned spectrum. In [5], a systemdriven criterion to determine the inner and outer emtocells or hybrid spectrum usage has been studied. In this paper, we investigate the impact o beamorming antennas on the perormance o overlaying networks and evaluate the perormance improvement over schemes without beamorming antennas. Tradeos o bandwidth sharing in overlay networks are clariied. For given a required SIR (signal-tointererence ratio), we ine optimal distance and identiy the maximum number o emtocells or bandwidth sharing. Bandwidth allocation method is optimized to maximize the total user utility in the two-tier network. The rest o this paper is organized as ollows. The heterogeneous overlay network system and beamorming antenna models are described in Section II. In Section III, the SIR with and without beamorming antenna are derived. Hybrid spectrum sharing strategy maximizing the number o shared emtocells are presented. In Section IV, the optimal bandwidth allocation or the cellcite utility is analyzed. Numerical results is presented in Section V, ollowed by concluding remarks in Section VI. II. SYSTEM MODEL We consider a two-tier overlay network with macrocell and emtocells, which extends in-home coverage and provides high speed wireless broadband services using a beamorming antenna. Fig. illustrates such a network system with two emtocells. Case A shows that HeNB communicates with a HUE, which, however, does not interere with a nearby MUE owing to beamorming antenna. Thus co-channel operation between macrocell and emtocell is available. On the other hand, case B shows that HeNB transmits signals using omnidirectional antenna, which cause signiicant intererence to MUE within the transmission coverage. In our settings, the macrocell network consists o a single base station (BS) and cellular users that communicate with the BS in the cellsite o radius R m. In the coverage o //$26. 2 IEEE 49

2 2 the macrocell BS, multiple emtocell networks have been deployed with emtocell BSs. Each emtocell hotspot includes a uniormly distributed population o users in a circular coverage area o radius R < R m. We assume that there is no inter-macrocell intererence. In 3GPP LTE [6], users (User Equipment; UE) are classiied into two types based on their access to the overlaying network: MUE accesses the macrocell BS and is denoted by enodeb (enb), and HUE accesses to a emtocell BS and is denoted by Home enodeb (HeNB). Since a emtocell user closer to the macrocell BS is likely to cause intererence in general, we partition the emtocell into two regions based on distance to the macrocell BS. We let the emtocells distance within a threshold distance rom the macrocell BS use a separate spectrum with the macrocell, and denote the set o such emtocells by K p. The emtocells out o distance share the spectrum with the macrocell. Let K s denote the set o the emtocells that use the shared spectrum. Also let K t = K s K p. We model wireless channel using simpliied power propagation model, which computes received signal strength based on distance as P R (d) = P T A, () dα where P T denotes the transmission power o the transmitter, d denotes the distance between the transmitter ane receiver, α denote the path loss parameter, and A is a constant that represents antenna gain. We adopt the beamorming antenna model o [7], where signal propagation area is sectored into two parts: main lobe with gain g m and side lobe with g s as shown in Fig. 2. The beamwidth o the main lobe is denoted by angle θ m, anat o the sidelobe is (2π θ m ). III. ENHANCED SIR WITH BEAMFORMING In this section, we investigate spatial thinning eect o beamorming and derive achievable SIR gain o beamorming antennas over omnidirectional antennas. We consider a overlay network, where a macrocell user and several emtocell users receives packets rom the macrocell and emtocell BSs, respectively. We assume that beamorming antennas are used or all the communications. We ocus on the achievable SIR gain o beamorming antennas over omnidirectional antennas. Fig. 3 depicts the cross-tier intererence between macrocell and emtocell in downlink. The solid lines illustrate two desired transmission links: S m or enb-mue and S or HeNB-HUE. The doted lines indicate intererence across tier: I m or enb-hue and I or HeNB-MUE. In our notation, we use the subscription m and parameters or a macrocell and emtocell networks, respectively, ane superscript B and O to dierentiate beamorming and omnidirectional antenna. We irst obtain the probability that the desired signal beam S and interered signal beam I are overlapping within a single sector o total sectors. P c = P r(θ S < θ m ) P r (θ I < θ m ) P r (θ m ) (2) Fig. 2. Beamorming antenna model S m I m g s I m S g m Fig. 3. An illustration on cross-tier intererence between macrocell and emtocell in downlink. where θ S and θ I denote the antenna alignment angle o desired receiver and intererer rom reerence angle, respectively. For instance, rom the view o the macrocell BS, the desired receiver and intererer are MUE and HUE. These angles depend on the MUE and HUE position, which is independently randomly distributed. Let θ m denote the angle 2π o a single sector, i.e.. We consider or a given sector, θ S and θ I are independent ane probability that the antenna alignment angle exist in an arbitrary sector is. Thereore, P c = () N 2 sec that is. Then, γm B denote the SIR o MUE with beamorming antennas. Using the channel model o () and beam gain, the SIR o MUE can be obtained as γ B m= g mp R m I B i. (3) Note that the intererence imposed on MUE includes cross-tier intererence I i rom emtocell network i with shared spectrum. We assume that cross-tier intererence rom emtocell networks with partitioned spectrum is relatively negligible. The macrocell BS transmits the main beam to the direction o the desired receiver, MUE. Among total K s HUEs, P c portion o HUE will interere with enb. Thus, γm B can be rewritten as γm B g m P R = m (g s ( ) + g m ) Ii O. (4) Let Ψ denote the SIR gain o beamorming antenna over P omnidirectional antennas. Dividing (4) by R m, we obtain i Ks IO i g m Ψ =. (5) ( ) g s + g m Similarly, the SIR gain o HUE is same as given in (5). Overall beamorming antennas suggest gains in two dierent aspects: i) enhanced strength o the desired signal, ii) reduction in the signal intererence. Considering the probability 5

3 3 s Fig. 4. p Bandwidth spectrum allocation or an overlay network. Fig. 5. O B Xi Yi z Y j X j Shared/partitioned spectrum usage based on the criterion distance. o the beam collision and beamorming gain, we can observe the gain o beamorming antennas in two-tier overlay network. IV. BANDWIDTH PARTITIONING Although the beamorming antennas provide signiicant gains in the SIR, there can be still severe cross-tier co-channel intererence i a number o emtocells share the bandwidth spectrum. In such case, we can control the intererence by partitioning the bandwidth spectrum and allocating a separate spectrum to dierent networks. In this section, we consider lexible spectrum allocation or the overlay networks. We consider that a part o spectrum is shared between the two tier networks to improve eiciency, ane rest are partitioned and dedicated to emtocell networks to control wireless intererence as shown in Fig. 4. In this scheme, each emtocell operates with either shared spectrum or partitioned spectrum. Spectrum selection o emtocell is dynamic and depends on its location and traic loads. We study the impact o active emtocell populations on network perormance, and provide a bandwidth allocation scheme that maximizes the network capacity. We irst note the tradeos o bandwidth sharing between eiciency and intererence. I two emtocell and macrocell links do not interere with each other, they can transmit simultaneously and improve the spectrum eiciency. However, as the number o links that use the same requency channel increases, the level o intererence also increases due to the intererence accumulation at the receivers. Since the signal strength depends on distance between two nodes, we can design bandwidth allocation schemes that the macrocell network shares bandwidth spectrum with emtocell networks that are ar away rom the macrocell BS. As the distance rom the macrocell BS increases, the probability that a emtocell network causes intererence to the macrocell network will decrease, anus the possibility o co-channel operations would increase. Motivated by this, we develop distance-based resource allocation schemes, and determine which emtocell networks can share the bandwidth with the macrocell network, and how much bandwidth should be shared to maximize the overall system capacity. A. Distance-Based Bandwidth Sharing Let d O th (or db th ) denote the distance such that emtocell networks within the circular area rom the macrocell BS are allocated to the partitioned bandwidth when omnidirectional antennas (or beamorming antennas) are used. Outside o the distance, emtocell networks share the bandwidth with the macrocell networks. Since beamorming antennas cause less intererence than omnidirectional antennas, we have d B th do th as shown in Fig. 5. We assume the macrocell BS, i.e. enb, locates at the center o the the cellcite (an origin). The position vector o MUE is denoted by z. Let X i and Y i denote the position vector o emtocell BS i and its corresponding receiver, respectively. We assume that each communication link requires to satisy a certain level o SIR. Let γ th and γth m denote the required SIR level or the emtocell and macrocell networks, respectively. Since the SIR at a emtocell user should be larger or equal to γ th. γ B = g mp R I B m = ΨP R I O m γ th, (6) where I O m is the intererence rom the active macrocell link to the emto receiver when omnidirectional antennas are used. Note that the intererence rom macrocell can be written as I O m = P T m Y i α, and rom the transmission power control, we have that P T m = P R m z α. Also, by deinition, we have Y i = when γb = γth. Then, the average threshold distance rom a FUE to the macrocell BS or the spared spectrum usage can be obtained by = ( ) γ th Pm R α ΨP R z. (7) This threshold distance or the shared spectrum usage is reduced comparing to the omnidirectional antenna. It implies that the occupation o the shared spectrum usage will be increased. Similarly, the SIR o the macrocell receiver with beamorming antennas needs to be higher than the required SIR γm th. Hence, we have γ B m = g mp R m I B i = ΨP m R Ii O γ th m, (8) where intererence Ii O rom emtocell i with omnidirectional antenna with power control can be written as Ii O = P T i X i z α α = P R X i Y i X i z. (9) Note that the intererence is normalized due to the power control, i.e. P T i = P R X i Y i α, or each HeNB i, where P R denotes the received power o FUE. Eq. (9) shows that the locations o users and BSs are important to analyze the network perormance. We assume 5

4 4 a randomized model, where users and emtocell BSs are randomly distributed with emtocell density λ = K t πr. Let m 2 m denote the threshold distance such that emtocell networks should be located beyone distance or the macrocell user to satisy the SIR threshold, and let K s denote the maximum number o emtocells that share the bandwidth spectrum with the macrocell. For simpliication, we denote the distance between HeNB i and its communicating FUE as p = X i Y i, which is limited to the transmission range o emtocell R. The intererer HeNBs diststance in shared spectrum q = X i z that range is included in [ m, R m ]. We assume the MUE locate relatively near place rom the enb than HeNB, i.e. X i z. Then the average intererence is E [ I O i ] = P R R = 4P RRα+2 RmR 2 2 (α2 4) p α 2p Rm α 2q R 2 dp q d R 2 th m m dq ( m α+2 Rm α+2 ). () The emtocell networks that share the bandwidth with the macrocell network are located at distance more than, and the number o such emtocell networks K s can be obtained as K s = λ 2π = K t R 2 m Rm m rdrdθ ( R 2 m m 2 ). () Combining (8), (), and (), we can obtain the threshold distance m, beyond which a emtocell network can use the shared bandwidth spectrum without interering the macrocell communication: (R m ) α+2( ) 2 ( ) m d th α+4 ( ) m + R 2 m d th α+2 m = Ψ ( α 2 4 ) P R mr 4 m 4 K t γ th m P R Rα + (R m ) α+4. (2) Since both macrocell and emtocell networks should satisy their SIR threshold, respectively, we have = max ( m, ). (3) B. Bandwidth Allocation We have identiied a set o emtocell networks that can share the bandwidth with the macrocell network using a distance-based approach. Now, the amount o bandwidth that are allocated or either sharing and partitioning to maximize the total utility o the network. We consider both the spectrum eiciency and airness in resource allocation using the ollowing objective unction. Let U denote the utility unction o logarithm sum o capacity or each tiered network. U = log C m + log C i + log C j, (4) j K p Number o shared emtocells, K s Required SIR th [db] = = 6 = 8 = 8 = 36 Fig. 6. The average number o emtocell K s or the shared spectrum usage varying beamorming antenna sector. where C m denotes the capacity o the macrocell network, C i denots the capacity o emtocell networks that use the shared bandwidth spectrum, and C j denotes the capacity o emtocell networks that use the partitioned spectrum. Note that each capacity can be written as C = ω log 2 ( + γ). where ω s and ω p are the shared and partitioned raction. Let γ th be the required SIR threshold at a required macrocell receiver. We also denote the minimum required SIRs at the emtocell receiver with shared spectrum and partitioned spectrum by γ s and γp, respectively. The above equation can be extended as U= log ω s + K s log ω s + K p log ω ( p ) + log (log ( + γ m )) + log log ( + γ i ) i K s + log ( ) + γj. (5) j K p log Since (5) is a concave unction, we can ine optimal ω s that maximizes the total utility using the KKT conditions as ω s = K s + K t + (6) Hence, the optimal allocation is determined by the raction o the emtocell networks that share the bandwidth with the macrocell network. V. NUMERICAL RESULTS We consider a single macrocell site that cooperates with multiple emtocell networks. There are 5 emtocell BSs, i.e. K t = 5. We assume that the macrocell BS is located at (, ) and has a transmission range with radius R m = 5 m. Femtocell BSs are randomly located within the macrocell site. We set the emtocell transmission range R to 2 m ane path loss exponent parameter α to 4. We assume that all the 52

5 5 5 TABLE I OPTIMAL RATIO OF SHARED SPECTRUM, ω s Fig. 7. Aggregated two-tier cell Utility = = 6 = 8 = 8 = s Utility unction with various ω s, when target SIR γ th = 6 db. macrocell and emtocell BSs and UEs have a beamorming antenna with sectors. Fig. 6 shows the increased number K s o emtocell networks sharing the bandwidth. As the number o beamorming sector increases, the normalizereshold distance decreases due to mitigated intererence, which in turn, allows a larger number o emtocell networks to share the bandwidth spectrum with the macrocell network. Under the same network settings, the total network capacity is varying with respect to the raction o bandwidth share ω s. I the number o emtocell in the shared spectrum K s is larger than that o partitioned spectrum users K p, i.e. K s > K t /2, with large portion o ω s, the cellcite aggregated capacity increase. The hal number o emtocell user marks as doted line Fig. 6. In general, the spectrum eiciency improves as the shared bandwidth increases except when K s < K t /2. However, this does not take into account the airness between dierent users. Fig. 7 shows the total utility o (5) at γ th = 6 db. Since the utility maximization accounts or both o capacity and airness simultaneously. There is an optimal value ωs that results in the best perormance. Comparing to the equal bandwidth allocation, i.e. ω s =.5, the optimal bandwidth allocation achieves about to 5% gain. We summarize the optimal sharing spectrum portion, ωs, or the required SIR, γ th ane number o sector in Table I. It is seen that the optimal ω s is reduced with increase o required SIR γ th. With the omnidirectional antenna, i.e. =, the optimal bandwidth allocation, ωs sharply decreases rom.69 to.2. However, i the number o beam sector is large enough, or instance, = 8, 36, the value o ωs is relatively stable. That is because the enhanced SIR gain with beamorming allows the large number o sharing emtocell BSs. Hence, with no regard to the required γ th, the spectrum allocation or the large number o beam sector is recommended to select the sharing spectrum usage. Required SIR γ th [db] utilization or marcocell and emtocell overlaying networks. The SIR enhancement with beamorming antenna in twotier overlay networks is derived anis SIR gain enables a part o spectrum to be shared across tiers or cross-tier to improve eiciency. The emtocell can select its mode in spectrum usage: uses the shared spectrum or eiciency or a separate partitioned spectrum or less intererence. In the cellcite o a macrocell networks, there can exist emtocell networks o a dierent mode. We ormulate the criterion to determine the usage mode based on the equivalent distance rom the macrocell BS and also derivee maximum number o shared emtocells under the homogenous emtocell BSs and UEs distribution assumption. We consider both the spectrum eiciency and airness in resource allocation using a logarithm sum o capacity or each tiered network. For the required SIR ane given number o beamorming sectors, we obtain the optimal spectrum allocation that maximizes the cellcite utility which is proportional to the ratio o shared emtocells. The utility gain o the proposed spectrum allocation over the equal spectrum allocation has been evaluaterough numerical results. REFERENCES [] R. Coombs and R. Steele, Introducing microcells into macrocellular networks: A case study, IEEE Transactions on Communications, vol. 47, no. 4, pp , 999. [2] L.Wang, G. Stuber, and C. Lea, Architecture design, requency planning, and perormance analysis or a microcell/macrocell overlaying system, IEEE Transactions on Vehicular Technology, vol. 46, no. 4, pp , 997. [3] V. Chandrasekhar, J. Andrews, and A. Gatherer, Femtocell networks: a survey, IEEE Communications Magazine, vol. 46, no. 9, pp , 28. [4] I. Guvenc, M. Jeong, F. Watanabe, and H. Inamura, A hybrid requency assignment or emtocells and coverage area analysis or co-channel operation, IEEE Communications Letters, vol. 2, no. 2, pp , 28. [5] L. C. Yong Bai, Juejia Zhou, Hybrid Spectrum Usage or Overlaying LTE Macrocell and Femtocell, IEEE Globecom29, 29. [6] 3rd Generation Partnership Project, Architecture aspects o Home Node B (HNB)/Home enhanced Node B (HeNB), 3GPP TR , 29. [7] Ramanathan, R., On the Perormance o Ad Hoc Networks with Beamorming Antennas, Proceedings o the 2nd ACM international symposium on Mobile ad hoc networking & computing, pp. 95 5, 2. VI. CONCLUSION This paper proposes intererence mitigation with beamorming antenna and hybrid spectrum usage to improve spectrum 53