Open, Closed, and Shared Access Femtocells in the Downlink
Abstract
A fundamental choice in femtocell deployments is the set of users which are allowed to access each femtocell. Closed access restricts the set to specifically registered users, while open access allows any mobile subscriber to use any femtocell. Which one is preferable depends strongly on the distance between the macrocell base station (MBS) and femtocell. The main results of the paper are lemmas which provide expressions for the SINR distribution for various zones within a cell as a function of this MBSfemto distance. The average sum throughput (or any other SINRbased metric) of home users and cellular users under open and closed access can be readily determined from these expressions. We show that unlike in the uplink, the interests of home and cellular users are in conflict, with home users preferring closed access and cellular users preferring open access. The conflict is most pronounced for femtocells near the cell edge, when there are many cellular users and fewer femtocells. To mitigate this conflict, we propose a middle way which we term shared access in which femtocells allocate an adjustable number of timeslots between home and cellular users such that a specified minimum rate for each can be achieved. The optimal such sharing fraction is derived. Analysis shows that shared access achieves at least the overall throughput of open access while also satisfying rate requirements, while closed access fails for cellular users and open access fails for the home user.
I Introduction
Femtocells are small formfactor base stations that can be installed within an existing cellular network. They can be installed either by an enduser or by the service provider and are distinguished from pico or microcells by their low cost and power and use of basic IP backhaul, and from WiFi by their use of cellular standards and licensed spectrum. Femtocells are a very promising and scalable method for meeting the everincreasing demands for capacity and highrate coverage. Since femtocells share spectrum with macrocell networks, managing crosstier interference between femto and macrocells is essential [1][3]. Furthermore, a basic question, particularly for enduser installed femtocells, is which users in the network should be allowed to use a given femtocell.
Ia Motivation and Related Work
Crosstier interference is highly dependent on this femtocell access decision. Closed access, where only specified registered home users can communicate with the femtocell access point (FAP), appears attractive to the home user but can result in severe crosstier interference from nearby cellular users in the uplink (see Fig. 1) or to nearby cellular users in the downlink. To reduce this interference in closed access, previous studies have considered power control [4][8], frequency assignment [9][11], and a spectrum sensing approach [12, 13]. An alternative is to simply hand over cellular users that cause or experience strong interference to the femtocell. This is known as open access. Intuitively, this should increase the overall network capacity [14] at the possible expense of a given femtocell owner, who must now share his femtocell resources (time/frequency slots and backhaul) with an unpredictable number of cellular users.
The uplink performance of femtocell access schemes has been investigated in [15, 16]. The interrelationship between the traffic type, access policy, and performance of highspeed packet access (HSPA) was examined in a simulationbased study [15]. In [16], an analytical framework was presented from open vs. closed access. Both studies suggested a hybrid access approach with an upper limit to the number of unregistered users to access the femtocell. We term this approach shared access since the femtocell is shared with cellular users, but within limits and hence not fully open. With respect to the uplink throughput of registered home users, open (or shared) access reduces interference by handing over the loud neighbor at the expense of FAP resource sharing. The tradeoff is such that open access is generally preferred for both home users and cellular users, since the interference reduction is so important [16]. Does the same tradeoff hold in the downlink?
It would seem that the tradeoff is different in the downlink since here the FAP is the loud neighbor (see Fig. 1). Therefore, serving unregistered users with the FAP benefits them at the cost of FAP resources. Meanwhile, there is at best a very small decrease in downlink interference to the home user. The downlink capacity of open vs. closed access has been studied using simulations for HSPA femtocells [15, 17] and OFDMA femtocells [18][20]. These studies propose and analyze shared access methods with limits on the number of unregistered users [17] and frequency subchannels for them [18, 19]. Indeed, these works find that cellular user performance is improved with open (or shared) access at the cost of reduced home user performance. All these downlink simulations are for very specific scenarios, for example the throughput is averaged over all femtocells and a fixed number of femtocells and outdoor cellular users are considered.
IB Contributions and Main Insights
Clearly, a more general and analytical approach is desirable. It should include key system parameters such as the distance between FAP and MBS, cell sizes, and the density of femtocells and cellular users. It would be more realistic if the femtocell and cellular user positions were not fixed, but rather were modeled as a spatial random process (see [22, 23] and references therein). Ideally, a general statistical distribution of the SINR could be found for both closed and open access. Since metrics such as outage probability, error probability, and throughput follow directly from SINR, once the SINR distribution is known these metrics can be computed quite quickly and easily [23]. Deriving such an SIR distribution (we neglect both thermal noise and interference from other cells) is the main contribution of the paper, and is used to draw a few conclusions about access strategies in the downlink.
First, we see that unlike the uplink [16], the preferred access schemes for home and cellular users are incompatible, with home user preferring closed access. For femtocells with coverage area extending outside the home, i.e. far from the MBS, closed access provides higher sum throughput for home user and lower sum throughput for neighboring cellular users, when compared to open access. For example, of a cell edge femtocell, open access causes at least 20% throughput loss to home user compared to closed access, while the neighboring cellular user experiences outage for typical data service (less than 15 kbps for 5 MHz bandwidth) in closed access. When the cellular user density is high (and/or femtocell density is low), the performance difference between open and closed access increases, since cellular users are increasingly impinged upon by the femtocells and vice versa. Furthermore, we observe that the open access femtocells far from MBS reduce the macrocell load, thereby open access rather than closed access offers higher throughput for a few home users (in its femtocell coverage area smaller than home area) located near and connected to the MBS. Nevertheless, most home users in cell site accessing FAPs are still reluctant to use their femtocells in open access.
Since neither open nor closed access can completely satisfy the need of both user groups, we consider a shared access approach where the femtocell has a timeslot ratio between the home and cellular users it serves, where is closed access. An optimal value of is found to maximize network throughput subject to QoS requirements on the minimum throughput per home and cellular user. For example, given a cell edge femtocell with minimum throughput of 50 kbps/cellular user and 500 kbps/home user, this shared access approach achieves about 80% higher network throughput than open access. When the QoS requirements increase in favor of significantly higher throughput of cellular user, shared access provides the lower network throughput than open access.
Ii System Model
Denote as the circular interior of a macrocell with radius and area centered at a MBS. Since FAPs are installed by the end customer, they are distributed with randomness rather than regular pattern. FAPs are thus assumed to be distributed according to a homogeneous PPP with intensity , denoted , where each is the location of the th FAP. The mean number of femtocells per cell cite is given as . Cellular users are assumed to be uniformly distributed inside . Home users are uniformly deployed in indoor (home) area, a disc of radius centered at their FAP (see Fig. 2). A summary of notation is given in Table I.
Iia Channel Model and MultiLevel Modulation
The downlink channel experiences path loss, Rayleigh fading with unit average power, and wall penetration loss . The path loss exponents are denoted by (outdoor and outdoortoindoor or indoortooutdoor) and (indoorindoor). As in [8],[11], and [25], the downlink femtocell networks are assumed to be interferencelimited and thermal noise at the receiver is ignored for simplicity. The MBS and FAP use fixed transmit powers of and , respectively. We assume that orthogonal multiple access is used (TDMA or OFDMA on a per subband basis), thus no intracell interference is considered. Interference from neighboring macrocell BSs is ignored for analytical tractability^{1}^{1}1Fig. 3 suggests that the assumption is not a significant omission of interference effects for dense femtocell systems. See Section IIID for more discussion. We consider multilevel Mary modulation single carrier transmission that is adapted to the received SIR , thus each user is assumed to estimate its SIR and provide perfect SIR feedback to their MBS (or FAP). Define SIR regions as , where is the minimum SIR providing the lowest discrete rate and . Then, the instantaneous transmission rate (in bps/Hz) is
(1) 
where is the Shannon Gap for multilevel Mary modulation (and may assume some level of coding). Assuming round robin (RR) scheduling with equal time slots, the average throughput is
(2) 
IiB Femtocell Coverage and Access
We assume that all users are served by the station (MBS or FAP) from which they receive the strongest longterm average power as their serving stations. Therefore, a femtocell coverage area as the area with a border at which the longterm average received power from a central MBS and the FAP is the same. The coverage is given by the following lemma.
Lemma 1.
For a FAP at distance from a central MBS located at the origin, the border of femtocell coverage is a circle centered at and the radius is
(3) 
where .
Proof.
Consider a central MBS located at and an FAP at distance from the MBS. Without loss of generality, the FAP is then assumed to be located at . The received power at the position with distances and from MBS and FAP are respectively given as and , where is the path loss at a reference distance . The contour with (zero dB SIR) satisfies , where . For the equation is rewritten as , which is the equation of circle and completes the proof. ∎
The assumption in Lemma 1 is valid for real scenario because where and . Since because , we assume that the center of the cell coverage is equal to the FAP location. For example, for the values in Table I. Lemma 1 states that the femtocell coverage extends towards the cell edge. This further indicates that for an FAP close to the MBS, femtocell coverage can be smaller than indoor area, whereas for an FAP far from the MBS, the coverage leaks into outdoor area.
When a femtocell operates in closed access, only registered users (termed home users) can communicate with the femtocell, whereas in open access, unregistered users within the femtocell coverage (termed neighboring cellular users) as well as home users may connect to the femtocell. Assuming that neighboring cellular users are outdoors, we partition the macrocell into two regions, inner region and outer region, with the threshold distance at which the femtocell coverage area is exactly equal to the indoor (home) area (see Fig. 2). In the inner region, so some “home users” actually communicate with the MBS, while in the outer region, neighboring cellular users would like to connect to the FAP. Home and neighboring cellular users have different signal and interference models due to wall penetration loss. These mean that the SIR of all users needs to be modeled differently for the location of users (indoor or outdoor), the type of base station (MBS or FAP), and femtocell access strategy (open or closed). In order to analyze downlink performance for the femtocell access scheme, we thus define four geographic zones, whereby users in the same zone have the same signal and interference model. Refer Fig. 2.

: the indoor area (a disc with the radius ) covered by the FAP at .

: the outdoor area (a circular annulus with inner radius and outer radius ) covered by the FAP at in open access or covered by the MBS in closed access.

: the indoor area (a disc with the radius ) covered by the FAP at .

: the indoor area (a circular annulus with inner radius and outer radius with respect to the FAP at ) covered by the MBS.
The zone and have the property of and . The zone and have the property of and . Although the cell coverage model based on multiple geographic zones i.e. with multiple SIR model, is more intricate than conventional model [8], [11] with single SIR model for closed access only, it is essential for a comparative study of open, closed, and shared access.
Iii PerZone Average Throughput
Consider a reference FAP at distance from a central MBS , and its home users (or neighboring cellular users) at distance and from and , respectively. As shown in IIB, according to SIR model, all the users divide into four groups located in the zone , , , and . We analyze the throughput for the each zone, and then, based on it, derive pertier throughput in next section. We assume small sized femtocell resulting .
Iiia Cellular User In Zone
Since the neighboring cellular users want to hand off to the FAP, they communicate with (open access) or (closed access), which results in different SIR according to the access scheme as follows:
Closed Access  (4)  
Open Access  (5) 
where is the exponentially distributed channel power (with unit mean) from . is the exponentially distributed channel power (with unit mean) from the interfering FAP . denotes the distance between the user and . The following lemma quantifies the user SIR for the zone .
Lemma 2.
The CDF of spatially averaged SIR over , a circular annulus with outer radius and inner radius , is given as follows:
1) Closed access:
(6)  
(7) 
where , , and is the Gauss hypergeometric function.
2) Open access:
(8) 
If , then
(9) 
(10) 
where and represent the real and imaginary parts of , respectively, is the Exponential integral function.
Proof.
See Appendix A. ∎
IiiB Home User SIR In Zone or
The home user SIR for the zone is given as
(11) 
This is the same as the cellular user SIR for open access given in (5) except for the distinction that this home user is indoors and so the propagation terms are adjusted accordingly. The SIR distribution of the home user in is given in the following lemma.
Lemma 3.
The CDF of spatially averaged SIR over the zone , a disc with radius , is given as
(12)  
(13) 
where . If and , then
(14) 
(15) 
Or, if , then
(16) 
where , , and for calculating are given in (15).
Proof.
See Appendix B. ∎
Note that the user SIR in the zone is given in (11), since the zone is the indoor area covered by a FAP at (in the inner region). The SIR distribution of the home users in is given in the following corollary.
IiiC Home User SIR In Zone
The zone exists for the femtocells with (in the inner region). The user SIR is
(18) 
The SIR is the same as the cellular user SIR for closed access given in (4) except for the distinction that this home user is indoors and so the propagation terms are adjusted accordingly. Thus, the SIR distribution for is given in similar form to (6) as shown in the following lemma.
Lemma 4.
The CDF of spatially averaged SIR over the area , a circular annulus with outer radius and inner radius , is given as
(19)  
(20) 
where .
Proof.
See Appendix C. ∎
IiiD Numerical Results
Fig. 3 shows the spatially averaged throughput for zone , , , and versus FAP of distance from the MBS for a different number of femtocells. Here the system parameters in Table I are used. For the zone , we show two results (closed access) and (open access). The analytic curves given from (21) are very close to the simulated curves. Furthermore, there is not a considerable difference in throughput between the intermacrocell interference (marked with “” and “”) and no intermacrocell interference (marked with “” and “”), which validates our assumption that neglecting intermacrocell interference is acceptable in dense deploy femto.
The higher interference from neighboring femtocells caused by higher decreases the averaged throughput of all the zones. On the other hand, has different effects in different zones as follows. First, since increasing decreases the signal power from , the average throughput ( and ) served by the decreases with . Second, the area of zones and increases with , which enlarges the area with low SIR within the zone, i.e. it has a negative effect on SIR. However, for the users in and (open access), the interference power from decreases with , which enhances SIR. As a result, decreases with increasing indicating that increasing the area of counteracts the effects of decreasing crosstier interference from . Moreover, increases for small , but decreases for large , which indicates that a negative effect of increasing the area of on becomes dominant at higher . Third, for the zone , no negative effect on SIR is observed since the zone area is fixed (independent of ). Thus, monotonically increases with .
Iv PerTier User Throughput: Closed Access vs. Open Access
In this section, we analyze the pertier user throughput of closed and open access based on the number of users in the zone as well as the perzone throughput obtained in preceding section. Denote , , , and is the number of users in the zone , , , and , respectively. Let and denote the number of outdoor cellular users and the number of home users per femtocell, respectively. When is assumed to be fixed for femtocells, total average number of user in a cell cite is then given by
(22) 
where and is respectively the average number of femtocells in the inner region () and the outer region (). Furthermore, (a) follows from that .
Iva Closed Access
Consider a reference FAP in closed access at distance from a central MBS. For the inner region, home users in connect to the FAP, while the neighboring cellular users of the FAP (implying users in ) are served by the MBS. For the outer region, regardless of femtocell access scheme, the home users in connect to the FAP, but the remaining home users in communicate to the MBS. The home users in share the same frequency channel with cellular users by using different time slots. Based on the femtocell/macrocell access scenario of the users, the following theorem quantifies pertier user throughput in closed access.
Theorem 1.
In closed access, the average sum throughput of home users and neighboring cellular users with respect to a FAP at distance from a central MBS is given as
(23)  
(24) 
where the perzone throughput , , , and is given from (21). and is the fraction of timeslot dedicated to the home users in and the cellular users in , respectively, among all users supported by the MBS, which is given as
(25)  
(26) 
where .
Proof.
See Appendix D. ∎
Remark 1.
From (25) and Fig. 3, increasing enhances but reduces , , and . Therefore, in (23) the home user throughput in closed access decreases with in the inner region () but increases with in the outer region (). Intuitively, the signal from the MBS is interference to home users in the outer region, but it is the desired signal to some home users connecting to the MBS in the inner region. This results in throughput degradation (inner region) or improvement (outer region) by increasing .
Remark 2.
From (25), increasing reduces , and does not effect on , , and . Intuitively, many cellular users increase the MBS load and thereby the amount of radio resources allocated home user served by the MBS is decreased. This indicates that for femtocells in the inner region, in (24) is higher for a lower cellular user density, while it is independent of cellular user density in the outer region. Since increases with in (26) and does not effect on , in (24) is high at a high cellular user density.
IvB Open Access
In the outer region, the reference FAP in open access provides service to neighboring cellular users in as well as home users in . Thus, the home users share the downlink radio resource of the FAP with the cellular users in time division manner. On the other hand, in the inner region, the femtocell/macrocell access scenario of the home users in open access is the same as that in closed access. The following theorem quantifies pertier user throughput in open access.
Theorem 2.
In open access, the average sum throughput of home users, , and neighboring cellular users, , with respect to a FAP at distance from a central MBS is given as
(27)  
(28) 
where is given from (21). is the fraction of timeslot dedicated to the home users in among home and cellular users supported from the MBS, and is the fraction of timeslot dedicated to the home users in among home and cellular users supported from the FAP. They are given as
(29)  
(30) 
Proof.
See Appendix E. ∎
Remark 3.
First, in (27), decreases with (), since increasing lowers in (29), and and decrease with from Fig. 3. From (30), increasing reduces . In Fig. 3 an increment in decreases with . Intuitively, is upper limited by the highest rate of Mary modulation at large , while is lower limited by zero. This indicates that begins to decrease at sufficiently large . Second, from (30), increases with . Thus, in (28) is enhanced at a higher cellular user density.
The throughput comparison of both the access schemes is given in the remarks below.
Remark 4.
First, closed access rather than open access increases the number of users supported by the MBS and thereby . This is followed by for the inner region. Intuitively, open access femtocells in the outer region admit neighboring cellular users which reduces the macrocell load. This effectively increases the throughput of home users in inner region, which are supported by the MBS. On the other hand, for the outer region because from (30). Note also from (30) that . Since , comparing and (65) yields . Moreover, is obviously larger than , thus . This indicates that home and cellular users prefer opposite access schemes.
IvC Numerical Results
The throughput results in this section are obtained with the system parameters in Table I. Fig. 4 shows the home user throughput analytically obtained using (23) and (27) versus FAPMBS distance for different numbers of femtocells and cellular users . Since at , we obtain the distance of 130m by substituting into (3). In closed access, home user throughput decreases with , while it increases with per Remark 1. For the home user throughput in open access first increases then decreases with increasing . Additionally, Fig. 4 shows that the turning point moves into the cell interior with increasing . This is because increasing and increases the number of neighboring cellular users, and thus, the time resource allocated to home user in femtocell downlink is reduced. In Fig. 4, the throughput for both open and closed access is degraded, since the aggregated interference from other femtocells increases with . We observe that unlike the case of , open access outperforms closed access for . However, the throughput loss of home users at dominates the home user throughput. Thus, closed access is better for home users.
Fig. 5 plots the sum throughput of neighboring cellular users of a reference femtocell using (24) and (28). The throughput is high at a low femtocell density and a high cellular user density, which agrees with the prediction in Remark 2 and 3. The throughput ( bps/Hz) in closed access is too low to offer typical services (0.003 bps/Hz is equivalent to 15 kbps for 5 MHz bandwidth). Thus, open access is much better for neighboring cellular users, in contrast to the result for home users. Table II summarizes these results of closed vs. open access.
Fig. 6 plots the network throughput in open access and closed access, sum of home user and neighboring cellular user throughput, i.e., and . Note that with respect to the network throughput for , open access is inferior to closed access. The reason is from the inequality given by . Intuitively, since from Fig. 3, the decrement of home user throughput due to time resource sharing with cellular users in open access prevails against the increment of cellular user throughput by substituting for . In a different point of view, this implies that a slight increase in the time fraction provides a high increase at the cost of a slight drop in , i.e., an increase in network throughput . This, as well as the extremely low throughput in closed access motivates the shard access femtocellls using time slot allocation, which will be in the next section.
V Shared Access: Timeslot Allocation
We consider the hybrid access where a FAP allocates fraction of timeslots to home users and the remaining fraction of timeslots to cellular users. Unlike the timeslot allocation in open access, where the time fraction is dependent on the number of home users and cellular users, the timeslot allocation in the shared access optimizes to maximize the network throughput while satisfying QoS requirement. The network throughput is defined as
(31) 
We define the QoS requirement as follows: 1) The average user throughput (cellular user) and (home user) is larger than the required minimum throughput (cellular user) and (home user), respectively, and 2) The average user throughput is at least w.r.t the . Satisfying the QoS, the timeslot allocation problem to maximize the network throughput is formulated as
(32)  
(33) 
where and .
Proposition 1.
The optimal value of the timeslot allocation in (32) is given as
(34) 
The solution is feasible when it is equal or larger than .
Proof.
Denote , , and as a set of satisfying the QoS requirement (33) in the order of description, respectively. we then obtain intersection of the three sets as for . Define a function of as . Since , monotonically increases with . Thus, is the maximum , which yields (34). Moreover, since for , is feasible for . ∎
Remark 5.
Shared access with and is closed and open access, respectively. The QoS parameter determines the priority of home users relative to cellular users with ensuring identical throughput to home and cellular users. In (34), increasing reduces and allocates more timeslots to cellular users. This indicates that shared access provides lower network throughput than open access when is set high, e.g. such that .
Fig. 7 compares the network throughput for different femtocell access schemes, where the system parameters in Table I are adopted. We assume the QoS requirement and respectively corresponding to 50 kbps and 500 kbps for 5 MHz bandwidth. Since at , we obtain the distance of 130m by substituting into (3). For , the throughput of shared access increases with decreasing . Considering lower results in higher , this indicates that increasing home user throughput counteracts the effects of decreasing cellular user throughput . Moreover, this implies that the shared access with higher (more timeslot allocation to cellular users) provides lower throughput than open access as shown in the result for . For , closed access always provides higher throughput than shared access because shared access with , which does not satisfy the QoS requirement, is the same as closed access. For , shared access obtains the same throughput as open access regardless of . The reason is that like open access, the shared access with timeslot allocation allows access from all neighboring cellular users located in the zone . Note that the shared access with appropriate value of achieves higher (at ) or equal (at ) network throughput than open access. We summarize these observations in Table II.
Vi Conclusion
The overall contribution of this paper is a new analytical framework for evaluating throughput tradeoffs regarding femtocell access schemes in downlink twotier femtocell networks. The framework quantifies femtocellsitespecific “loud neighbor” effects and can be used to compare other techniques e.g. power control, spectrum allocation, and MIMO. Our results show that unlike the uplink results in [16], the preferred access scheme for home and cellular users is incompatible. In particular, closed access provides higher throughput for home users and lower throughput for neighboring cellular users; vice versa with open access. As a compromise, we suggest shared access where femtocells choose a timeslot ratio for their home and neighboring cellular users to maximize the network throughput subject to a networkwide QoS requirement. For femtocells within the outer area, shared access achieves higher network throughput than open access while satisfying the QoS of both home and cellular users. These results motivate shared access  i.e. open access, but with limits  in femtocellenhanced cellular networks with universal frequency reuse. []
a Proof of Lemma 2
In closed access the user SIR in (4) is rewritten as , where and and . Then, the complementary cumulative distribution function (CCDF) of the user SIR at distance from the FAP is given as
(35) 
where follows because the CCDF of exponential with unit mean is given as , and is given from [24, Lemma 3.1]. Here is the Laplace transform of (exponential random variable scaled by ), which is given as
(36) 
Moreover, is the Laplace transform of the Poisson shotnoise process . For exponential with unit mean, is given by [24]
(37) 
where . Thus, (35) is simplifies to
(38) 
The cellular users are uniformly located at the zone that a circular annulus with outer radius and inner radius . Then, probability density function (PDF) of the distance is . The spatially averaged SIR distribution over is given as
(39) 
Desired result (6) is obtained by further calculating (39) with the following integration formula [27]
(40) 
Next, in open access, the user SIR in (5) is rewritten as , where and . Using the same approach as in (35), CCDF of the user SIR at distance from the FAP is given as
(41) 
As is the exponential random variable scaled by , from (36) we obtain . As the Poisson shotnoise process is equal to , we get . Thus, we get
(42) 
For open access, from (39) and (42) the spatially averaged SIR distribution over is given as
(43)  