EMBEDDING femtocells in the current cellular system

Transcription

1 2194 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 61, NO. 5, JUNE 2012 Design and Analysis o Downlink Spectrum Sharing in Two-Tier Cognitive Femto Networks Shin-Ming Cheng, Member, IEEE, Weng Chon Ao, Fan-Min Tseng, and Kwang-Cheng Chen, Fellow, IEEE Abstract In two-tier networks consisting o a macrocell overlaid with emtocells in cochannel deployment and closed-access policy, spatial reuse is achieved at the price o severe intratier and cross-tier intererence rom concurrent transmissions. The intererence causes signiicant perormance degradation, particularly when coordination among base stations BSs is ineasible. Cognitive radio CR is a promising technique or intererence mitigation, where emto-bss with cognitive inormation accomplish concurrent transmissions while meeting a per-tier outage constraint. This paper studies the role o inormation sensed at emto-bss on the transmission capacity. By exploiting dierent cognitive inormation, we propose spectrum-sharing schemes between macrocell and emtocell, as well as among emtocells, to improve spatial reuse gain. Bounds on the maximum intensity o simultaneously transmitting emtocells that satisy a given per-tier outage constraint in these schemes are theoretically derived via a stochastic geometry model. We conduct simulations to evaluate the perormance o the proposed schemes in terms o transmission capacity. The results conirm that, when emto-bss acquire the knowledge o user channel statistics or user location inormation, signiicant spatial reuse gain can be achieved by exploiting the avoidance region and multiuser diversity. Index Terms Cognitive radio CR, downlink capacity, emtocell, spectrum sharing, stochastic geometry. I. INTRODUCTION EMBEDDING emtocells in the current cellular system consisting o planned macro base stations macro-bss has emerged as a promising approach to improve coverage and capacity gains. As shown in Fig. 1, emto-bss acting as shortrange data access points to serve indoor users clearly increase in-building coverage. Additionally, emtocells acilitate the accommodation o a large number o concurrent transmissions in the network known as spatial reuse, thereore yielding enhanced wireless capacity. Macrocells and emtocells typically operate in a common spectrum known as cochannel deployment to eectively utilize scarce resources. Since emto-bss are mostly installed by Manuscript received June 15, 2011; revised November 11, 2011; accepted January 8, Date o publication February 11, 2012; date o current version June 12, This work was supported by the National Science Council NSC under Contract NSC E MY3 and Contract NSC I This paper was presented in part at the IEEE International Symposium on Personal, Indoor, and Mobile Radio Communications, Istanbul, Turkey, September 26 29, The review o this paper was coordinated by Dr. H. Jiang. The authors are with National Taiwan University, Taipei 106, Taiwan smcheng@cc.ee.ntu.edu.tw; r @ntu.edu.tw; armingtseng@gmail. com; chenkc@cc.ee.ntu.edu.tw. Color versions o one or more o the igures in this paper are available online at Digital Object Identiier /TVT Fig. 1. Macro/emto networks. customers, their positions are determined by their owners, as opposed to macro-bss deployed according to detailed network planning. The unplanned positions o emto-bss introduce the occurrence o cross-tier intererence, where macrocells and emtocells play the role o either the aggressor or the victim. Since emto-bss are paid or and maintained by customers or residential and private uses, only mobile subscribers o the emto-bs abbreviated as emto-mss are allowed to establish connections with the emto-bs known as closed-access policy. In this case, unauthorized users can only connect to the macro-bs abbreviated as macro-ms, even i there exists a emto-bs in their vicinity; thus, the macro-mss suer heavy cross-tier intererence. Furthermore, emto-mss might be disturbed by nearby emto-bss, which is known as intratier intererence. A multitude o studies have been recently proposed to study the downlink spectrum-sharing problems between macrocells and emtocells considering cross-tier intererence control 2] 9], among emtocells considering intratier intererence mitigation 10] 14], and both 15] 18]. Spectrum sharing among heterogeneous nodes is typically achieved via cooperation or coexistence 19]. In two-tier emto networks, the delay o connecting via wired backhaul is too long to allow any cooperation between emto-bss and the macro-bs 20]; thus, collecting global inormation or centralized intererence control is not possible. In contrast, emto-bss distributively mitigate intererence in a coexistence ashion by using local inormation without explicit signaling that might be more appropriated. In this paper, we apply a recent innovation called cognitive radio CR to resolve such a coexistence challenge. A CR-enabled emto-bs can automatically sense the environment, interpret the received signaling rom the macro-bs and the surrounding emto-bss, and intelligently allocate resources without /$ IEEE

2 CHENG et al.: DESIGN OF DOWNLINK SPECTRUM SHARING IN TWO-TIER COGNITIVE FEMTO NETWORK 2195 introducing excessive cross-tier and intratier intererence 15], 18]. Given CR capability, how to eectively exploit spectrum resources becomes an urgent issue on spectrum sharing or emto-bss with the macro-bs and other competing emto-bss. Assuming a slotted Aloha channel access protocol, based on dierent availabilities o sensing inormation, we propose several corresponding downlink spectrum allocation and sharing schemes between emto-bss and macro-bss, as well as among emto-bss with consideration or per-tier intererence control. With cognitive inormation rom the macrocell, emto- BSs can share spectrum with the macro-bs in an interweave or underlay manner while guaranteeing a per-tier outage constraint. By utilizing knowledge rom the surrounding emto- BSs, new variants o slotted Aloha, such as Opportunistic Aloha 21], Distance Sense Multiple Access DSMA, and Opportunistic DSMA, are designed to allow more simultaneous and successul emto transmissions. To better understand the eects o inormation sensed at emto-bss on the downlink perormance in the proposed sharing schemes, we develop general models using stochastic geometry, which is widely adopted to investigate the undamental perormance o wireless networks and, thus, or the downlink 3], 22] or uplink capacity 23] in cellular networks. By assuming that the node locations are a realization o a homogeneous Poisson point process PPP, we statistically study the transmission capacity 24] o each proposed scheme, which is deined as the maximum density o successul transmissions, given a per-tier outage constraint 25]. Comparing with the existing analytical results on spectrum sharing in twotier emto networks 3], 22], 23], 26], the proposed model exploits CR to acquire user channel statistics or user location inormation to realize avoidance region and multiuser diversity, so that the transmission capacity is improved. The accuracy o the developed models or downlink capacity in the proposed sharing schemes is validated by simulations. Numerical results are provided to clariy the relationship between local inormation acquired at the emto-bs and the downlink transmission capacity, which acilitates the deployment and development o CR in two-tier emto networks. The remainder o this paper is organized as ollows: In Section II, we outline the existing literature on distributed spectrum-sharing schemes, emphasizing on inormation acquisition technologies. This section also describes the preliminaries related to the proposed taxonomy. Section III presents the system model and briely introduces the proposed spectrumsharing schemes. We analyze the transmission capacities or macro-emto sharing schemes in Section IV and or interemtocell sharing scheme in Section V. Numerical and simulation results are provided in Section VI. Section VII concludes this paper. II. PRELIMINARIES AND RELATED WORKS This section surveys existing spectrum-sharing schemes in two-tier emtocell networks according to the classiication on methods o inormation acquisitions 18]. Typically, orthogonal requency-division multiple access OFDMA is applied in two-tier emtocells networks, and the OFDMA-based systems divide the system bandwidth into N basic time requency units o resource blocks RBs. Exchanging at the BS side. A emto-bs explicitly exchanges inormation between the macro-bs and other emto-bss about their allocation usage, connection behavior, and resource demands via wired backhaul 5], 7], 13] or wireless communication 4], 17]. Perect intererence mitigation can be achieved because the emto-bs is aware o the present actions and uture intentions o the macro- BS and surrounding emto-bss. Such cooperative sharing needs a common control channel and related protocol, and thus, extra communication overheads and delay should be considered. However, the delay o connection via wired backhaul is too long to permit any cooperation between emto-bss and the macro-bs 20]. Measuring at the UE side. In this case, a emto-ms periodically takes its local spectrum measurement and sends back the results to its serving emto-bs 10], 16]. By analyzing the report, the emto-bs can acquire activity and channel condition inormation about the immediate environment o each emto-ms, which acilitates the intererence mitigation. Moreover, the emto-bs can utilize Automatic Repeat-reQuest eedback, indicating the success and ailure o transmission to assist the intererence mitigation 11], 12], 27]. However, perorming these measurements may consume nontrivial amounts o power; thus, it may not be easible or emto-mss that are typically power limited. Measuring at the BS side. By building CR into the emto-bs itsel, detection o the macro-bs and surrounding emto- BS signals can be achieved without any coordination 8], 9], 15], 16]. The CR-enabled solution can serve urgent needs in intererence mitigation while yielding a limited complexity and imposing no impact on the state-o-theart emtocell architecture. The ollowing section discusses the details o inormation acquisition by CR according to the classiication rom 28], including the practicality o obtaining such inormation. Activity inormation: A emto-bs knows which resources are unoccupied by other BSs. For example, i the received intererence power o a channel, e.g., the reerence signal received power in a Long-Term Evolution LTE network exceeds a certain threshold, the emto-bs identiies that the channel is allocated or a macro-ms 8]. Channel inormation: A emto-bs knows the channel statistics or channel gains between some BSs and some MSs, which can be obtained by recent innovations such as CR network tomography 29] without heavy overheads. Location inormation: A emto-bs knows the distance between itsel and an MS/BS and thus knows the strength o the channel to the MS/BS. This implies that knowledge related to a macro-ms, such as the speciic macro-ms allocated to a channel or the location in which the assigned macro-ms will use a channel, is acquired. Typically, such scheduling or zone allocation inormation 4] is encapsulated in

3 2196 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 61, NO. 5, JUNE 2012 PDCCH LTE or DLMAP WiMax by encoding with identities o served MSs. We can simply assign emto-bss a special user identity or decoding such inormation broadcast channel. Moreover, emto-bs could overhear the eedback inormation rom MS to determine the location o MS and, thus, could adjust its own access parameters accordingly 30]. Codebook/message inormation: A emto-bs knows the codebooks o the other BSs. The codebook inormation can be obtained rom the periodic broadcasting inormation, and the message inormation might be obtained ater decoding. While this is impractical or an initial transmission 28], the assumption holds or a message retransmission, where the emto-bs hears the irst transmission and decodes it 27]. III. DOWNLINK SYSTEM MODEL A. Two-Tier Network Architecture The downlink o an OFDMA system with basic time requency units o RBs in two-tier macro- and emtocells is considered, as shown in Fig. 1. Denote the RBs allocated to the macro-ms as macro RB. The cochannel deployment and the closed-access policy are assumed. Macro- and emto-bss transmit to only one user at any given RB at ull power P m and P, respectively, which implies that the transmission power is maintained constant across the RBs. Perect synchronization in time and requency is assumed. Denote H R 2 as the interior o a reerence macrocell with serving area radius R m, which consists o multiple macro- MSs and a macro-bs located at the center o H. The spatial distribution o macro-mss is assumed to ollow a homogeneous PPP with density μ m, and the locations o the macro-mss are denoted as Φ m {X i }. The macrocell is overlaid with emto-bss with radius R, which are randomly distributed on R 2 according to a homogeneous PPP with intensity λ. Such deployment is similar to that in existing works analyzing intererence using stochastic geometry 3], 22], 23], 26]. We let Φ {Y i } denote the locations o the emto-bss. Each emto-bs is coupled with relevant CR capabilities, such as spectrum sensing, intererence management, and eicient spectrum allocation and sharing. Note that we concentrate on the case where intramacrocell intererence is dominant, and thus, intererence originating rom other macrocells is not considered. The eects o intermacrocell intererence in the context o spectrum-sharing schemes proposed herein will be subject to uture research. B. Channel Model We consider the eects o path-loss attenuation, Rayleigh ading with unit average power G, penetration loss due to walls β, and background noise power per RB N 0 in our channel model. The path-loss exponent o transmission is denoted by α. Since emtocells are designed or indoor transmissions, the penetration loss due to inner walls and outer walls shall be considered 31]. The successul reception o a transmission Fig. 2. Allocation results or macro-emto sharing schemes. at an MS depends on whether the signal-to-intererence-plusnoise ratio SINR observed by the MS is larger than an SINR threshold denoted by η. To guarantee the transmission quality, there is an outage constraint at the MS with maximum outage probability ɛ. C. Sharing Schemes The sharing schemes proposed in our paper address mitigating both cross- and intratier intererence by leveraging inormation acquired rom the macro-bs and surrounding emto-bss. Please note that these schemes are designed on the scope o a single RB. Given here are the schemes addressing macro-emto sharing and cross-tier intererence mitigation. Interweave. With activity inormation, a emto-bs acting as a secondary user can passively seek temporary requency voids o RBs or opportunistic allocation. Consequently, emto-mss and macro-mss are operating on orthogonal RBs, and no cross-tier intererence is introduced. However, spectrum interweave restricts the transmission capacity because no spatial reuse is adopted. Underlay. With channel inormation, emto-bss can aggressively exploit RBs occupied by macro-mss to increase spatial reuse as long as the resulting aggregated crossintererence is constrained. However, since the emto-bs is unable to distinguish whether the occupied RBs are assigned to the macro-mss in its vicinity, we must assume the worst case. In other words, the emto transmission should ensure the outage constraints o all macro-mss that may occupy the RB. Thus, there is no appreciable gain rom spatial reuse due to overestimation. Fig. 2 shows the examples o RB allocations or interweave and underlay sharing schemes. Controlled underlay. I a emto-bs is able to leverage location inormation rom scheduling, it is able to determine the set o RBs that can be reused without causing intererence to macro-mss in its vicinity. By deactivating the emto- BSs on the RBs dedicated to a neighboring macro-ms, more concurrent emto transmissions are allowed, and the transmission capacity is increased. Fig. 3 shows a snapshot o spatial distributions o active emto-bss in underlay and controlled-underlay schemes. Note that, i a emto-bs also knows codebook and message inormation, it can assist and improve macrotransmissions.

4 CHENG et al.: DESIGN OF DOWNLINK SPECTRUM SHARING IN TWO-TIER COGNITIVE FEMTO NETWORK 2197 Fig. 3. Spatial distributions o active emto-bss in the let underlay and right controlled-underlay schemes. In the controlled-underlay scheme, there are no active emto-bss within the avoiding region o the macro-ms; thereore, a larger density o active emto-bs can be sustained, given the outage constraint o macro-ms. However, such an overlay scheme 28] only works during periods o retransmissions and, thus, is not discussed. Given here are the schemes dealing with interemtocell sharing and intratier intererence mitigation. Slotted Aloha. Without any inormation, in slotted Aloha, every emto-bs independently tosses a coin in each RB with probability p and transmits i it gets heads. In this case, the active subset o emto-bss is realized by independent thinning and is denoted as Φ p {Y i : B i p 1} with node density pλ, where B i p are independent and identically distributed. Bernoulli random variables with parameter p. Opportunistic Aloha. When the channel inormation o other emto-bss is known, combining the random selection o emto-bss with the occurrence o good channel conditions 21] is a straightorward method to improve transmission capacity. Each emto-bs distributively computes a certain threshold, and only the emto-bss with channel gains larger than the threshold are activated in a reerence RB. The idea is similar to allocating resources to the best link in multiuser networks to retrieve multiuser diversity gain. DSMA. Although slotted Aloha is simple and eective, it cannot completely prevent simultaneous neighboring transmissions rom occurring. The proposed DSMA employs a suitable guard zone 32] o a given radius around each active emto-bs; in other words, the presence o intererers in the guard zone is avoided. It is achieved by allowing each emto-bs to randomly generate a real number between 0, 1] and exchange this inormation with other emto-bss within its guard zone to determine that with the smallest value, implying that the emto-bs must know the location o surrounding emto-bss. The resulting active emto-bss can be modeled by a hard-core point process 33], where each point o the process is the center o a disc that contains no other points, except itsel. We can easily observe the eects o guard zone by comparing the snapshots o spatial distributions o SA and DSMA in Fig. 4. Opportunistic DSMA. Combining the advantages o Opportunistic Aloha and DSMA, Opportunistic DSMA selects the emto-bs with the highest channel gain in a guard zone as the active transmitter. This requires both channel on inormation rom the surrounding emto-bss. Note that those proposed paradigms can be reely combined to mitigate both cross-tier and intertier intererence. In the succeeding sections, we ocus on perormance analysis o the combining schemes, which is summarized in Table I. D. Perormance Metric The metric considered as a measure o spectral eiciency on each RB is transmission capacity, which is deined as the maximum density o successul transmissions satisying a per-tier outage constraint on a reerence RB 24], 25]. Mathematically, the transmission capacity with a per-tier outage probability ɛ is C 1 ɛλ ɛ 1 where λ ɛ is the spatial density o attempted transmissions; hence, λ ɛ 1 ɛ is the spatial density o successul transmissions. Note that, to compute the transmission capacity or emto-bs, we need to consider outage constraints at both emto-mss and macro-ms, rom which two transmission capacities or emto-bss can be respectively derived. The smaller one should be chosen to simultaneously satisy outage constraints at both tiers. IV. MACRO-FEMTO SPECTRUM SHARING With the λ RB allocation activity o the overlaid macrocell, emto-bs can identiy the macro RBs and empty RBs available or allocation. To address the rationale o dierent macro-emto sharing schemes, we adopt the simplest slotted Aloha adopted as the interemtocell sharing scheme. A. Interweave Scheme Based on the activity inormation, macro- and emto-bss orthogonally allocate RBs or their MSs in this scheme; thus,

5 2198 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 61, NO. 5, JUNE 2012 Fig. 4. Spatial distributions o active emto-bss in the let SA and right DSMA schemes. In the DSMA scheme, the active emto-bss are separated with a certain distance. TABLE I PROPOSED SPECTRUM-SHARING SCHEMES we, respectively, discuss the intererence received in a macro- MS at a macro RB and in a emto-ms at an unoccupied RB. The received SINR observed by a macro-ms at a macro RB amounts to γ m G m P m rm α /N 0, where G m is the channel gain or the macro-ms and is supposed to be exponentially distributed with unit mean and r m is the distance between the macro-bs and the macro-ms. Under the near ar eect, to guarantee quality o service QoS or all macro-mss, γ m received at the macro-ms located at the edge o the serving area i.e., r m R m shall satisy Gm P m R α m P η 1 ɛ. 2 N 0 The equality holds when the serving area exactly matches the cell coverage. Here, the serving area is smaller than the cell coverage to account or extra intererence tolerance. Regarding emto-ms, the received SINR γ at an unoccupied RB is γ G P r α /N 0 + I,, where G accounts or the channel gain between a emto-ms and its serving emto-bs and is supposed to be exponentially distributed with unit mean, r is the distance between a emto-ms and its serving emto-bs, and I, is the intratier intererence deined later. Lemma 1: To avoid intererence rom emto-bss violating the outage constraint at a emto-ms, the maximum active emto-bs density utilizing slotted Aloha shall decrease rom λ to λ ln1 ɛ ηn 0 /P R α /R2 ηδ β, δ K α, and the media access probability is p λ /λ. Proo: The SINR γ o a emto-ms located at the cell boundary o its serving emto-bs i.e., r R shall guarantee P G P R α N 0 + I, η 1 ɛ 3 where I, Y j Φ ˆp \{Y 0} G Y j P β, Y j α is the intratier intererence rom surrounding emto-bss to a typical reerence emto-ms located at the origin where slotted Aloha is adopted, and β, is the penetration loss due to inner walls between apartments. G Yj and Y j are the channel gain and the distance between the emto-bs at Y j and a typical emto-ms at the origin, respectively. Note that the emto-ms is located at the cell boundary i.e., R away rom a emto-bs with an arbitrary direction. The spatial distribution o emto-mss at the cell boundary also orms a homogeneous PPP with the same density λ correlated with that o the emto-bss. By the stationary characteristic o homogeneous PPP 34], the intererence statistics measured by a typical emto-ms is representative o that seen by all other emto-mss. The emto- BS at Y 0 is the serving BS or the reerence emto-ms. From 34], the moment-generating unction o I, is u E exp si, ] exp 2π λ du 1 + uα sp 0 β, exp λ P δ β,s δ δ K α 4

6 CHENG et al.: DESIGN OF DOWNLINK SPECTRUM SHARING IN TWO-TIER COGNITIVE FEMTO NETWORK 2199 where K α 2π 2 /α sin2π/α, and δ 2/α. To derive λ, the let-hand side o 3 can be evaluated as ] η P G P R α N 0 + I, exp ηn 0 P R α exp λ R 2 η δ β,k δ α. 5 We thus have λ ln1 ɛ ηn 0 P R α R 2ηδ β, δ K. 6 α The transmission capacity at an unoccupied RB is C e λ 1 ɛ. B. Underlay Scheme In this scheme, emto-bss try to exploit macro RBs to achieve spatial reuse according to channel inormation. We control the number o interering sources i.e., active emto- BSs to prevent violating outage constraints at both the macro- MS and emto-ms. From the perspective o the macro-ms, we have the ollowing lemma. Lemma 2: To avoid intererence rom emto-bss violating the outage constraint at a macro-ms in the macrocell with serving area radius R m, the active emto-bs density must be less than λ ln1 ɛ ηn 0 /P m Rm α ]/ RmηP 2 β,m /P m δ K α ]. Proo: Considering the macro-ms at the serving area edge o the macro-bs, the ollowing outage constraint must be maintained: Gm P m R α m P η 1 ɛ 7 N 0 + I,m where I,m G Y Φ p Yj P β,m Y j α is the cross-tier intererence contributed rom surrounding emto-bss to the j macro-ms, and β,m is the penetration loss due to outer walls. Compared to 2, the cross-tier intererence here makes the macrocell coverage match its serving area. The let side o 7 can be evaluated as ] η P G m P m Rm α N 0 + I,m exp We thus have ηn 0 P m R α m exp λ Rm 2 δ ηp β,m K α. 8 P m λ ln1 ɛ ηn 0 R 2 m ηp β,m P m Rm α P m δ Kα. 9 Corollary 1: The maximum active emto-bs density λ is a monotonically decreasing unction o R m when QoS requirements o all macro-mss are satisied. Another constraint or the active emto-bs density can be derived rom the respective emto-ms. When a emto-ms reuses the macro RB to receive data, the received SINR highly depends on the distance between itsel and the macro-bs, which is approximated by the distance between its serving emto-bs and the macro-bs denoted as d. Moreover, the received SINR at the emto-ms is also limited by intererence rom surrounding emto-bss, and the ollowing lemma describes the permissible emto-bs density: Lemma 3: To avoid intererence rom emto-bss violating the outage constraint at a emto-ms, the active emto-bs density must be less than λ { ln1 ɛ1 + ηp m β m, d α /P R α ] ηn 0/P R α }/R2 ηδ β, δ K α. Proo: To guarantee the outage constraint, a emto-ms located at the cell boundary o its serving emto-bs with the shortest distance to the macro-bs denoted as d should be considered since it suers the most severe cross-tier intererence. The ollowing equation is satisied at this emto-ms: G P R α P η 1 ɛ 10 N 0 + I m, + I, where I m, G m P m β m, d α is the approximate cross-tier intererence rom the macro-bs to the emto-ms, which is located at the cell boundary o its serving emto-bs with distance d away rom the macro-bs. β m, is the penetration loss due to outer walls. The let side o 10 can be evaluated as ] P G η P R α Thus, we have ln λ N 0 + I m, + I, ηp mβ m, d α P R α 1 ɛ exp ηn 0 P R α exp λ R 2 η δ β δ,k α. 11 ] 1 + ηp mβ m, d α P R α R 2 ηδ β δ, K α ηn 0 P R α. 12 Corollary 2: Satisying the QoS requirements o all emto- MS requires the maximum active emto-bs density to be a monotonically increasing unction o d. Note that d must be irst retrieved to compute λ at each emto-bs in a distributed ashion. d can be easily determined either by exchanging its own d with neighboring emto-bss or with the macro-bs recording d or all emto-bss and periodically broadcasting to every emto-bs. Each emto-bs uses the next lemma to determine the maximum active emto-bs density and, thus, the access probability o a emto-bs on a macro RB to guarantee the QoS o both macro- and emto-mss. Proposition 1: Under the situation where R m and d are known, the maximum active emto-bs density at a macro RB

7 2200 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 61, NO. 5, JUNE 2012 denoted as λ or the two-tier network equals min λ, λ to guarantee the QoS o both macro- and emto-mss Corollary 3: I λ λ, the serving area radius o macro-bs may be extended rom R m to Rm u ln1 ɛ/k α ηp β,m /P m δ λ. Proo: When λ λ, Proposition 1 implies λ λ, and 8 becomes P G m ] η P m Rm u α N 0 + I,m exp exp ηn 0 λ R u m 2 ηp β,m P m P m Rm u α δ K α. 13 In the intererence limited regime, we set N 0 0 and have Rm u ln1 ɛ δ. 14 ηp β K,m α P λ m Obviously, as the active emto-bs density decreases, the corresponding serving area radius o macro-bs increases, ollowing Corollary 1. Corollary 4: I λ λ, the emto-bs coverage radius may be extended rom R to a larger value denoted as R u. Proo: When λ λ, Proposition 1 implies λ λ, and 11 becomes ] η P G P R u α N 0+I m, +I, exp ηn 0 P R u α P R u α P R u α +ηp m β m, d α exp λ R u2 η δ β,k δ α. 15 R u does not generally have a closed orm and must be numerically solved. As seen in 12, when active emto-bs density decreases, the emto-bs coverage radius increases. The improvement o transmission capacity at a macro RB that beneited rom the underlay scheme is Cm UL λ 1 ɛ. C. Controlled-Underlay Scheme When the location inormation o the macro-ms who locates an RB is known by a emto-bs, it is able to determine whether a macro-ms detected within its sensing area with radius r s currently occupies an RB. Then, the emto-bs may deactivate itsel at the corresponding RBs occupied by the detected macro- MSs to eliminate its dominant cross-tier intererence. The overall capacity may thereore increase. This scheme implies that emto-bss located within radius r s o a macro-ms will be deactivated at the RB occupied by the macro-ms; in other words, each macro-ms has an avoidance region with radius r s. From the snapshot o spatial distributions o active emto-bss in underlay and controlled-underlay schemes see Fig. 3, we observe the operation o the avoidance region. The ollowing lemma describes the eect o an avoidance region on active emto-bs density: Lemma 4: When all emto-bss located in the avoidance region with radius r s o a macro-ms are deactivated at the RB occupied by the macro-ms, the active emto-bs density λ is larger than that without introducing an avoidance region, i.e., λ in Lemma 2. Proo: Bv, r is denoted as the circle o radius r centered at node v. The intererence rom surrounding emto-bss to a typical macro-ms located at the origin becomes I cul,m Y j Φ p \B0,r s G Yj P β,m Y j α. 16 The moment-generating unction o I,m cul can be computed as E exp si,m cul ] exp 2π λ u du. 1 + uα sp β,m Substituting s η/p m Rm α and α 4in17,wehave E exp η ] { I,m cul exp λ Rm 2 ηp β,m P m R α m π 2 2 π tan 1 r s P m ηp β,m r 2 s R 2 m P m ]}. 18 Compare the active emto-bs density to that without avoidance region by applying α 4 and K 4 π 2 /2in9.Wehave the ollowing ratio: λ π 2 /2 λ π 2 /2 π tan 1 Pm ηp β,m 1 19 rs 2 Rm 2 shows the increase in active emto-bs density when an avoidance region is introduced. Intuitively, by introducing an avoidance region, we prevent the event that the intererence rom a single dominant emto- BS causes outage reception at a macro-ms. Similar to Proposition 1, the maximum active emto-bs density at a macro RB denoted as λ with an avoidance region is min λ, λ. The transmission capacity at a macro RB improved by the controlled-underlay scheme is Cm cul λ 1 ɛ. D. Implementation Considerations To distributively acquire local and real-time inormation, CR technology is adopted in Section II. For the global and timeinvariant inormation, we could leverage the special eatures o the two-tier emto networks. The macro-bs could exploit the preoperational state or coniguration state o the emto-bs 35] to broadcast such inormation to all emto-bss. As the emto- BSs apply slotted Aloha, the corresponding parameters required to calculate the active probability or intererence mitigation i.e., all parameters in 6 or the interweave paradigm can

8 CHENG et al.: DESIGN OF DOWNLINK SPECTRUM SHARING IN TWO-TIER COGNITIVE FEMTO NETWORK 2201 be easily obtained using the same approach. For the underlay approach, emto-bss with CR capability can be easily aware o the existence o macrotransmissions. Equation 12 indicates that an extra parameter, which is the shortest distance between the emto-bss and the macro-bs i.e., d, is needed to calculate the active probability. It is known by the operator since the positions o all emto-bss are obtained in the preoperational state. For the controlled-underlay approach, by exploiting CR, emto-bss could acquire the location inormation o macro- MS to actively deactivate themselves that are located within the avoidance region o the macro-ms. V. I NTERFEMTOCELL SPECTRUM SHARING This section proposes various interemtocell sharing schemes on the basis o slotted Aloha, with emphasis on their transmission capacities. In each section, we irst analyze the case o interweave macro-emto sharing paradigm to emphasize the rationale o the proposed schemes. Then, we urther consider cross-tier intererence to show how to apply the proposed schemes to underlay macro-emto sharing paradigms. A. Opportunistic Aloha Instead o independently tossing a coin with probability p to access the empty RB in slotted Aloha, the emto-bs whose channel gain between itsel and its serving emto-ms is larger than a threshold g th is selected as the transmitter in Opportunistic Aloha. The set o active emto-bss with Opportunistic Aloha at an empty RB can be described by Π {Y i Φ : g Yi >g th } 20 where g Yi is the realization o the channel gain between emto- BS Y i and its serving emto-ms at cell boundary. Note that g Yi is drawn rom G, which is an exponential random variable with unit mean i.e., Rayleigh ading. Proposition 2: Given the same active emto-bs density, the outage probability at emto-ms or Opportunistic Aloha is lower than that or slotted Aloha. Proo: Since G is exponentially distributed with unit mean, its complementary cumulative distribution unction CCDF is F G g P G g e g. 21 I the density o active emto-bss in Opportunistic Aloha is equal to that in slotted Aloha i.e., λ, g th should be chosen to satisy the ollowing equation: PG g th e g th p λ. 22 λ The channel gain o the active emto-bss in Opportunistic Aloha is distributed as G OA, and its CCDF is F G OA g P G OA g P G g G g th { e g /e g th, or g g th 1, or g<g th. 23 Obviously, G OA is stochastically larger than G, i.e., F G OAg F G g or all g 0. The desired signal quality becomes better, whereas the statistics o the interering signals is still the same. Given the same active emto-bs density λ, the outage probability in Opportunistic Aloha is smaller than that in slotted Aloha, due to better channel quality. The aorementioned proposition compares the outage probability o Opportunistic Aloha with that o slotted Aloha based on the same active emto-bs density. To quantitatively evaluate the transmission capacity o Opportunistic Aloha denoted as and C e, we ix the outage probability o Opportunistic Aloha to that o slotted Aloha i.e., ɛ, so that we can derive the perormance improvement in terms o the active emto-bs density. Denote the access probability in Opportunistic Aloha C OA e λ OA λ OA as p OA /λ, where is the active emto-bs density in Opportunistic Aloha. In this case, g th should be chosen to satisy the ollowing equation: PG g th e g th λ OA p OA 24 λ and thus, g th lnλ / λ OA. In the intererence-limited regime, ollowing 24], we set N 0 0 and have ] 1 ɛ P G OA exp Thereore λ OA λ OA η P R α N 0 + I, πr 2 η δ β δ,e G δ πr 2 ηδ β δ, E G δ ] G E OA δ ]. 25 ln1 ɛ ] ]. 26 E G OA δ With 23, EG OA δ ] can be computed as G OA δ ] E e g th Γ1 δ, g th λ λ OA Γ 1 δ, ln λ λ OA 27 where Γs, x x ts 1 e t dt is the upper incomplete gamma unction. In addition, it can be observed that EG δ ]Γ1 + δ, 0. By applying 27 into 26, we have Γ 1 δ, ln λ ln1 ɛ λ OA πr 2 ηδ β, δ Γ1 + δ, 0λ 28 OA rom which λ can be solved. The transmission capacity at an unoccupied RB is Ce OA OA λ 1 ɛ. By incorporating the cross-tier intererence rom the macro- BS i.e., I m, in 25, we could derive the transmission capacity on a macro RB when applying Opportunistic Aloha to underlay-series schemes. Regarding the controlled-underlay scheme, only the emto-bss located outside o the avoidance region o the macro-ms would perorm Opportunistic Aloha.

9 2202 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 61, NO. 5, JUNE 2012 B. DSMA By introducing the concept o a guard zone 32], the presence o simultaneous interering transmissions in proximity is prohibited in DSMA. The set o active emto-bss elected by the DSMA protocol is denoted as Ψ, which can be regarded as a nonindependent Matern thinning o Φ 33]. Deine Ψ {Y i Φ : my i <my j or all Y j Φ BY i,r z \{Y i }} 29 where my i Uniorm0, 1 is a mark or each emto-bs Y i Φ.Aemto-BSY i Φ is selected in the Matern thinning i its mark is smaller than that o any other emto-bs Y j Φ in the guard zone BY i,r z centered at Y i o radius r z. Since each point Y i Φ is independently marked by my i, it is retained with probability p DSMA a 1 0 e λ tπr 2 z dt 1 e λ πr 2 z λ πr 2 z 30 where a ollows that, given that the mark o Y i is t, the number o points in BY i,r z whose marks are less than t is Poisson distributed with mean λ tπrz, 2 and the void probability is e λ tπrz 2. The density o Ψ is observe that λ DSMA 1 e λ πr2 z πr 2 z λ DSMA p DSMA λ, and we λ 1 πrz Proposition 3: Giventhesameactiveemto-BSdensity,the outage probability or DSMA is lower than that or slotted Aloha. Proo: I there exists an r z such that 1 e λ πrz 2 /πr 2 z λ,wehave λ DSMA λ. In this case, the desired signal quality is the same, whereas the intererence is weaker since intererers are with the same density but located with distance r z R + away, where x + Δ maxx, 0. With α 4, we have ] 1 ɛ P G η P R α N 0 +I, exp ηn 0 P R α { exp DSMA λ πr 2 ηβ, 1 2 ]} π 1 r z R +2 2 tan 1 ηβ, R 2 exp ηn 0 P R α { exp 1 e λ πrz 2 πr 2 ηβ, 1 2 πr 2 z π 2 tan 1 ]} 1 r z R +2 ηβ, R λ DSMA From 32, r z and, thus, can be obtained. The transmission capacity o DSMA Ce DSMA at an unoccupied RB is Ce DSMA DSMA λ 1 ɛ. The transmission capacity on a macro RB or combining DSMA with underlay can be derived by incorporating crosstier intererence rom the macro-bs i.e., I m, in 32. For the controlled-underlay scheme, the additional considerations, comparing with the underlay scheme, is that the emto-bss located in the avoidance region o the macro-ms are deactivated, and only the remaining emto-bss execute DSMA. C. Opportunistic DSMA Opportunistic DSMA Instead o marking each point with m Uniorm0, 1 in DSMA, Opportunistic DSMA uses the channel gain between a emto-bs and its serving emto-ms as the mark o the emto- BS. In other words, the advantages o Opportunistic Aloha and DSMA are combined. A point Y i Φ is elected as active when it has the largest channel gain among all the points in BY i,r z. Proposition 4: Given the same active emto-bs density, the outage probability or Opportunistic DSMA is lower than that or DSMA. Proo: Let hg denote the retention probability o a point Y i Φ with mark/channel gain g that no other points Y j Φ with channel gain greater than g exist in BY i,r z. hg can be computed as hg exp λ PG >gπr 2 z exp λ e g πr 2 z. 33 The retention probability o a node is deined as p p 0 0 hg G gdg exp λ e g πr 2 z e g dg 1 e λ πrz 2 λ πrz The density o the retained points is λ p λ. DSMA It is observed that λ is the same as λ. The channel gains o the retained points are distributed as F G g F G g P G <g G > all other nodes channel gains in B0,r z e λ πrz 2 λ πr 2 z k k+1 P max k! G Y i <g i1 k0 k0 e λ πr 2 z λ πrz 2 k 1 e g k+1 k! 1 e g exp λ πr 2 ze g. 35

10 CHENG et al.: DESIGN OF DOWNLINK SPECTRUM SHARING IN TWO-TIER COGNITIVE FEMTO NETWORK 2203 Since exp λ πrze 2 g in 35 is smaller than 1 and the channel gain o a retained point in DSMA is exponentially distributed with unit mean, F G g 1 1 e g e λ πrze 2 g 1 1 e g e g F G DSMAg, or all g 0. Thus, G is stochastically larger than G DSMA, and thus, the outage probability in Opportunistic DSMA is smaller than that in DSMA due to the better channel quality. Rather than comparing outage probability under the same active emto-bs density, in the ollowing, we derive the perormance improvement in terms o active emto-bs density under the same outage probability. In the intererence-limited regime, we set N 0 0 and have 1 ɛ P G η P R α I, ] E G P I, G R αη P. 36 Let q G Δ P /R αη, and deine the set Ξ {Y i Φ : G P Y i α >q, Y i r z R + }. The number o points in Ξ is Poisson distributed with mean R m λ r z R + P G P r α >q 2πrdr λ r z R + exp qrα P 2πrdr. 37 When α 4, with a change o variable t qr 4 /P, 37 becomes λ π P e t t 1 2 d 2 q q P r z R +4 λ π 2 As a result, 36 becomes P 1 q Γ 2, q r z R P 1 ɛ E G PI, q] a < E G PΞ φ] π P E G exp λ 2 q ] 1 Γ 2, q r z R + 4 P π R 4 ηβ, E G exp λ 2 G ] 1 Γ 2, G R 4ηβ r z R + 4, 39 where a ollows that the event I, q implies that there are no node in Ξ. From 39 and 34, we have 1 ɛ<e G exp Γ 1 1 e λ πr 2 z πr 2 z π R 4 ηβ, 2 G ] 2, G R 4ηβ r z R + 4, λ 40 rom where r z and the upper bound o can be obtained. The upper bound o transmission capacity Ce at an unoccupied RB is calculated by multiplying the upper λ bound o with 1 ɛ. The Opportunistic DSMA and underlay-series schemes can be similarly combined, as discussed in Opportunistic Aloha and DSMA. D. Implementation Considerations In Opportunistic Aloha, each emto-bs distributively computes a certain threshold, and only the emto-bss with channel gains larger than the threshold are activated. All inormation required to compute the threshold can be acquired by leveraging the existing CR technologies. In DSMA, a emto-bs needs to acquire the marking inormation rom the neighboring emto- BSs to guarantee that only one emto-bs will be active in a guard zone. To achieve that, each emto-bs just broadcasts the random number selected by itsel. Opportunistic DSMA can be implemented by reerring both that or Opportunistic Aloha and that or DSMA since it is the combination o them. VI. NUMERICAL RESULTS This section uses simulation results to validate the correctness o the proposed analytical models or the sharing schemes. We urther investigate the perormance o the proposed sharing schemes based on the analytical and simulation results. The simulation experiments are built on the Matlab platorm, and the parameter setup ollows the Third-Generation Partnership Project dual-strip model 31]; details are shown in Table II. We deploy the two-tier emto networks, ollowing the network model described in Section III. At each reerence RB, every emto-bs is active/nonactive according to the macro-emto sharing scheme and interemtocells sharing schemes. For each active emto-bs MS pair, we determine i the transmission is successul by checking i the received SINR is higher than the threshold. I it is, the transmission is successul; otherwise, the transmission ailed. Finally, we accumulate the total number o emto-bss with successul transmissions, which is equal to the transmission capacity. We average the results o 100 dierent random deployments at 100 dierent RBs. Fig. 5 shows the eects o R m on Cm UL and Cm cul with slotted Aloha. We observe that the simulation results closely match our analytical models; limited discrepancy exists mainly due to the act that the cross-tier intererence received at the emto-ms rom the macro-bs is approximated by that received

11 2204 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 61, NO. 5, JUNE 2012 TABLE II PARAMETER SETUP Fig. 6. Eects o P on the transmission capacities o the underlay and controlled-underlay schemes with slotted Aloha in a reerence macro RB. The system parameters are set as λ H 500, P m 49 dbm, R m 500 m, R 20 m, N dbm, α 4, β, 5 db,β,m β m, 20 db, η 4, ɛ 0.1, and r s 80 m. Fig. 5. Eects o R m on the transmission capacities o the underlay and controlled-underlay schemes with slotted Aloha in a reerence macro RB. The system parameters are set as λ H 500, P m 49 dbm, P 20 dbm, R 20 m, N dbm, α 4, β, 5 db,β,m β m, 20 db, η 4, ɛ 0.1, and r s 80 m. at the emto-bs. This igure shows that, as R m increases, both Cm UL and Cm cul decrease. As R m increases, the reerence macro-ms located at the boundary o the macrocell receives a weaker signal rom the macro-bs, and the density o active emto-bss decreases to guarantee the outage constraint at the reerence macro-ms. Fig. 6 investigates the eects o P on Cm UL and Cm cul with slotted Aloha, where capacity degrades when P becomes large. This is due to the act that, when P becomes large, the density o active emto-bss becomes small to guarantee a per-tier outage constraint, leading to a small transmission capacity. Both igures show that the transmission capacity in the controlled-underlay approach outperorms that in the underlay approach, which agrees with the results o Lemma 4. This result demonstrates that, in the controlled-underlay scheme with avoidance region, by deactivating the emto-bss on the RB allocated to a neighboring macro-ms, more concurrent emto transmissions are allowed. Fig. 7 shows the eects o ɛ on C e, Ce OA, Ce DSMA, and Ce with the interweave paradigm. The discrepancies existing in Opportunistic Aloha and Opportunistic DSMA are Fig. 7. Eects o ɛ on the transmission capacities o the SA, OA, DSMA, and sharing schemes with the interweave paradigm in a reerence empty RB. The system parameters are set as λ H 500, P m 49 dbm, P 20 dbm, R m 500 m, R 20 m, N dbm, α 4, β, 5dB,β,m β m, 20 db, η 4, and r z 80 m. due to the approximations in 25 and 40, respectively. The results show that the approximated transmission capacities or these sharing schemes are reasonably accurate or small values o ɛ. Regarding the DSMA series schemes, the errors are mainly caused by the assumption that retained nodes away rom r z are modeled as PPP. This igure also shows that the transmission capacities ordered rom high to low are given as ollows: Ce, Ce OA, Ce DSMA, and C e, which it the results o Propositions 2 4. For all schemes, when ɛ increases i.e., the outage constraint releases, the increase in the density o allowable active emto-bss dominates the decrease in 1 ɛ, and thus, transmission capacity increases.

12 CHENG et al.: DESIGN OF DOWNLINK SPECTRUM SHARING IN TWO-TIER COGNITIVE FEMTO NETWORK 2205 Fig. 8. Eects o the density o emto-bss on the transmission capacities o the SA, OA, DSMA, and sharing schemes with the interweave paradigm in a reerence empty RB. The system parameters are set as P m 49 dbm, P 20 dbm, R m 500 m, R 20 m, N dbm, α 4, β, 5dB,β,m β m, 20 db, η 4, ɛ 0.1, and r z 80 m. Fig. 9. Eects o R on the transmission capacities o the SA, OA, DSMA, and sharing schemes with the interweave paradigm in a reerence empty RB. The system parameters are set as λ H 500, P m 49 dbm, P 20 dbm, R m 500 m, N dbm, α 4, β, 5 db, β,m β m, 20 db, η 4, ɛ 0.1, and r z 80 m. Fig. 8 shows the impacts o the density o emto-bss on C e, Ce OA, Ce DSMA, and Ce with the interweave paradigm. We observe that the transmission capacities or the SA and DSMA schemes remain the same since the active emto-bs density is ixed. However, when the density o emto-bss increases, emto BS-MS pairs with higher channel gains can be selected due to multiuser diversity, and thus, the transmission capacity increases. Fig. 9 urther shows the eects o R on C e, Ce OA, Ce DSMA, and Ce with the interweave paradigm. We observe that, when R increases, the signal received at the emto-ms located at the boundary o the emtocell becomes weaker, and thus, the amount o intererence that the emto- MS can suer becomes smaller. In this case, the density o active emto-bss decreases, and the transmission capacity thus decreases. Fig. 10 shows the eects o R on the transmission capacity o slotted Aloha, Opportunistic Aloha, DSMA, and Opportunistic DSMA with the controlled-underlay paradigm in a reerence macro BS. Due to the same reason in the interweave paradigm, when R increases, the transmission capacity thus decreases. Comparing with that or the interweave paradigm, the transmission capacity o the underlay paradigm is reduced due to the existence cross-tier intererence rom macro-bs. When CR-enabled emto-bss are capable o acquiring extra inormation, the SA with poor perormance may not be a desirable choice or interemtocell sharing. Comparing the perormance o the proposed Opportunistic Aloha and DSMA schemes in all cases with the chosen system parameters, we observe that channel inormation is more eective than location inormation or spectrum sharing among emto-bss, which suggests that the emto-bs should be designed to acquire channel inormation or more gains. I both channel and location inormation are acquired, more simultaneous and successul transmissions are allowed. Fig. 10. Eects o R on the transmission capacities o the SA, OA, DSMA, and sharing schemes with the controlled-underlay paradigm in a reerence empty RB. The system parameters are set as λ H 500, P m 49 dbm, P 20 dbm, R m 500 m, N dbm, α 4, β, 5dB,β,m β m, 20 db, η 4, ɛ 0.1, and r z 80 m. VII. CONCLUSION In two-tier emto networks, with the aid o CR, emto- BSs could acquire activity, channel, and location inormation rom macrocell and surrounding emtocells. Based on dierent availabilities o sensing inormation in emto-bss, this work has proposed several corresponding downlink spectrum-sharing schemes between emto-bss and macro-bss, as well as among emto-bss, and derived their theoretical transmission capacities. The numerical results have shown that, compared with the typical interweave and slotted Aloha schemes only utilizing activity inormation to achieve intererence mitigation, the proposed opportunistic-series schemes signiicantly improve

13 2206 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 61, NO. 5, JUNE 2012 spatial reuse gain by urther leveraging channel inormation and, thus, multiuser diversity. Moreover, by additionally exploiting location inormation and the avoidance region, the proposed DSMA-series and controlled-underlay schemes allow many more simultaneous and successul transmissions than that in interweave and slotted Aloha schemes. With implementable eature and intractable analytical results, the proposed schemes acilitate the deployment and development o CR in the two-tier emto networks. REFERENCES 1] S.-M. Cheng, W. C. Ao, and K.-C. Chen, Downlink capacity o two-tier cognitive emto networks, in Proc. IEEE PIMRC, Sep. 2010, pp ] D. Lopez-Perez, A. Valcarce, G. de la Roche, and J. Zhang, OFDMA emtocells: A roadmap on intererence avoidance, IEEE Commun. Mag., vol. 47, no. 9, pp , Sep ] V. Chandrasekhar and J. G. Andrews, Spectrum allocation in tiered cellular networks, IEEE Trans. Commun., vol. 57, no. 10, pp , Oct ] K. Sundaresan and S. Rangarajan, Eicient resource management in OFDMA Femto cells, in Proc. ACM MobiHoc, May 2009, pp ] Z. Bharucha, H. Haas, G. Auer, and I. Cosovic, Femto-cell resource partitioning, in Proc. IEEE Globecom Workshops, Nov. 2009, pp ] Y.-Y. Li, M. Macuha, E. S. Sousa, T. Sato, and M. Nanri, Cognitive intererence management in 3G emtocells, in Proc. IEEE PIMRC, Sep. 2009, pp ] R. Madan, J. Borran, A. Sampath, N. Bhushan, A. Khandekar, and T. Ji, Cell association and intererence coordination in heterogeneous LTE-A cellular networks, IEEE J. Sel. Areas Commun.,vol.28,no.9,pp , Dec ] S.-Y. Lien, C.-C. Tseng, K.-C. Chen, and C.-W. Su, Cognitive radio resource management or QoS guarantees in autonomous emtocell networks, in Proc. IEEE ICC, May 2010, pp ] S. M. Almalouh and G. L. Stüber, Intererence-aware radio resource allocation in OFDMA-based cognitive radio networks, IEEE Trans. Veh. Technol., vol. 60, no. 4, pp , May ] D. Lopez-Perez, A. Ladanyi, A. Juttner, and J. Zhang, OFDMA emtocells: A sel-organizing approach or requency assignment, in Proc. IEEE PIMRC, Sep. 2009, pp ] S. Brahma and M. Chatterjee, Mitigating sel-intererence among IEEE networks: A game theoretic perspective, in Proc. IEEE GLOBE- COM, Dec. 2009, pp ] V. Gardellin, S. K. Das, and L. Lenzini, A ully distributed game theoretic approach to guarantee sel-coexistence among WRANs, in Proc. IEEE INFOCOM, Mar. 2010, pp ] J.-H. Yun and K. G. Shin, CTRL: A sel-organizing emtocell management architecture or co-channel deployment, in Proc. ACM MobiCom, Sep. 2010, pp ] Q. Chen, Y.-C. Liang, M. Motani, and W.-C. L. Wong, A two-level mac protocol strategy or opportunistic spectrum access in cognitive radio networks, IEEE Trans. Veh. Technol., vol. 60, no. 5, pp , Jun ] G. Gur, S. Bayhan, and F. Alagoz, Cognitive emtocell networks: An overlay architecture or localized dynamic spectrum access, IEEE Wireless Commun. Mag., vol. 17, no. 4, pp , Aug ] J. Xiang, Y. Zhang, T. Skeie, and L. Xie, Downlink spectrum sharing or cognitive radio emtocell networks, IEEE Syst. J., vol. 4, no. 4, pp , Dec ] C.-H. Ko and H.-Y. Wei, On-demand resource-sharing mechanism design in two-tier OFDMA emtocell networks, IEEE Trans. Veh. Technol., vol. 60, no. 3, pp , Mar ] S.-M. Cheng, S.-Y. Lien, F.-S. Chu, and K.-C. Chen, On exploiting cognitive radio to mitigate intererence in macro/emto heterogeneous networks, IEEE Wireless Commun. Mag., vol. 18, no. 3, pp , Jun ] J. M. Peha, Sharing spectrum through spectrum policy reorm and cognitive radio, Proc. IEEE, vol. 97, no. 4, pp , Apr ] Third-Generation Partnership Project, 3GPP Report o TSG RAN WG1 Meeting, Oct #62, v ] F. Baccelli, B. Blaszczyszyn, and P. Miihlethaler, Stochastic analysis o spatial and opportunistic Aloha, IEEE J. Sel. Areas Commun., vol. 27, no. 7, pp , Sep ] J. G. Andrews, F. Baccelli, and R. K. Ganti, A tractable approach to coverage and rate in cellular networks, IEEE Trans. Commun., vol. 59, no. 11, pp , Nov ] V. Chandrasekhar and J. G. Andrews, Uplink capacity and intererence avoidance or two-tier emtocell networks, IEEE Trans. Wireless Commun., vol. 8, no. 7, pp , Jul ] S. Weber, J. G. Andrews, and N. Jindal, An overview o the transmission capacity o wireless networks, IEEE Trans. Commun., vol. 58, no. 12, pp , Dec ] J. Lee, S. Lim, J. G. Andrews, and D. Hong, Achievable transmission capacity o secondary system in cognitive radio networks, in Proc. IEEE ICC, May 2010, pp ] H. S. Dhillon, R. K. Ganti, F. Baccelli, and J. G. Andrews, Modeling and analysis o k-tier downlink heterogeneous cellular networks, IEEE J. Sel. Areas Commun., vol. 30, no. 3, pp , Apr ] S.-M. Cheng, W. C. Ao, and K.-C. Chen, Eiciency o a cognitive radio link with opportunistic intererence mitigation, IEEE Trans. Wireless Commun., vol. 10, no. 6, pp , Jun ] A. Goldsmith, S. A. Jaar, I. Mari, and S. Srinivasa, Breaking spectrum gridlock with cognitive radios: An inormation theoretic perspective, Proc. IEEE, vol. 97, no. 5, pp , May ] C.-K. Yu, K.-C. Chen, and S.-M. Cheng, Cognitive radio network tomography, IEEE Trans. Veh. Technol., vol. 59, no. 4, pp , May ] S. Huang, X. Liu, and Z. Ding, Distributed power control or cognitive user access based on primary link control eedback, in Proc. IEEE INFOCOM, Mar. 2010, pp ] Third-Generation Partnership Project, E-UTRA: Further Advancements or E-UTRA Physical layer aspects, Mar GPP TR v ] A. Hasan and J. G. Andrews, The guard zone in wireless ad hoc networks, IEEE Trans. Wireless Commun., vol. 6, no. 3, pp , Mar ] H. Q. Nguyen, F. Baccelli, and D. Koman, A stochastic geometry analysis o dense IEEE networks, in Proc. IEEE INFOCOM, May 2007, pp ] F. Baccelli, B. Blaszczyszyn, and P. Muhlethaler, An Aloha protocol or multihop mobile wireless networks, IEEE Trans. In. Theory, vol. 52, no. 2, pp , Feb ] Third-Generation Partnership Project, Evolved Universal Terrestrial Radio Access E-UTRA and Evolved Universal Terrestrial Radio Access E-UTRAN; Overall description; Stage 2, Sep GPP TS Shin-Ming Cheng S 05 M 07 received the B.S. and Ph.D. degrees in computer science and inormation engineering rom National Taiwan University NTU, Taipei, Taiwan, in 2000 and 2007, respectively. He joined the Graduate Institute o Communication Engineering, National Taiwan University, Taipei, as a Postdoctoral Research Fellow in His research interests include network security, cognitive radio networks, and network science. Weng Chon Ao was born in Macau. He received the B.S. degrees in computer science and physics and the M.S. degree in communications engineering rom National Taiwan University, Taipei, Taiwan, in 2008 and 2010, respectively. Since 2010, he has been a Research Assistant with Intel-NTU Laboratory, National Taiwan University. His research interests include stochastic modeling and control o communication networks.

14 CHENG et al.: DESIGN OF DOWNLINK SPECTRUM SHARING IN TWO-TIER COGNITIVE FEMTO NETWORK 2207 Fan-Min Tseng received the B.S. and M.S. degrees rom National Taiwan University, Taipei, Taiwan, in 2008 and 2011, respectively. He is currently with National Taiwan University. His research interests include network coding, cognitive radio networks, and inormation theory. Kwang-Cheng Chen M 89 SM 93 F 07 received the B.S. degree rom National Taiwan University, Taipei, Taiwan, in 1983 and the M.S. and Ph.D. degrees rom the University o Maryland, College Park, in 1987 and 1989, respectively, all in electrical engineering. From 1987 to 1998, he was with SSE; COMSAT; the IBM Thomas J. Watson Research Center, Yorktown Heights, NY; and National Tsing Hua University, Hsinchu, Taiwan, working on mobile communications and networks. He is currently a Distinguished Proessor and the Director with the Graduate Institute o Communication Engineering and the Communication Research Center, National Taiwan University. His research interests include wireless communications and network science. Dr. Chen has received a number o awards and honors, including the 2011 IEEE ComSoc WTC Recognition Award. He is a coauthor o three IEEE papers that received the 2001 ISI Classic Citation Award, the IEEE ICC 2010 Best Paper Award, and IEEE GLOBECOM 2010 Gold Best Paper.