Traffic Assignment Over Licensed and Unlicensed Bands for Dual-Band Femtocells

Transcription

1 Traic Assignment Over Licensed and Unlicensed Bands or Dual-Band Femtocells Feilu Liu, Erdem Bala, Elza Erkip and Rui Yang ECE Department, Polytechnic Institute o NYU, Brooklyn, NY InterDigital Communications, 2 Huntington Quadrangle 4th Floor, South Wing, Melville, NY Abstract More and more broadband user equipments (UE) are equipped with multiple radios to access both licensed and unlicensed bands (via Bluetooth, WiFi and etc.). Cellular operators are exploiting this trend by oloading traic rom their licensed bands to the unlicensed band through a large number o WiFi hotspots. In addition, emtocells are also being used to increase cellular capacity. However, in any o these approaches, a UE can only utilize one requency band when communicating with a emto/macro base stations or WiFi access point, resulting in low spectrum eiciency. In this paper, we extend our prior work on a emtocell simultaneously accessing the licensed and unlicensed bands via a single cellular air interace by studying the non-ully utilized case and proposing an algorithm or a emto base station to jointly assign its traic to the licensed and unlicensed bands. Results demonstrate that the proposed ramework substantially improves the perormance o both macrocells and emtocells. Additionally, the perormance degradation caused by the emtocells to the other unlicensed band users is small and similar to that caused in the current approach o building WiFi hotspots to oload cellular traic. Index Terms emtocell, heterogenous network, small cell. I. INTRODUCTION Small cells have been deployed by cellular operators to boost their network capacities in the past years. Two types o small cells are widely used. One is emtocell, which is ormed by a small-size and low-power base station that is typically installed by cellular users [1], [2]. The other type o small cells are WiFi hotspots that are built by cellular operators to oload traic rom their licensed bands to the unlicensed band [3], [4]. AT&T, or instance, currently owns more than 23,000 WiFi hotspots [4]. However, these two approaches does not use the overall spectrum eiciently. Although both bands are available, they only use one band, either licensed or unlicensed. Moreover, emtocells share the same licensed spectrum with macrocells, which may result in severe cross-tier intererence between them [1], [2]. On the the hand, WiFi hotspots suer rom low MAC layer eiciency, mainly because the collision-avoidance nature o WiFi MAC layer prevents concurrent transmissions in a large geographical area. Besides, WiFi PHY layer is Part o this work was done during Feilu Liu s internship in InterDigital Communications LLC. in the summer o This work is supported by the New York State Center or Advanced Technology in Telecommunications (CATT) and the Wireless Internet Center or Advanced Technology (WICAT), an NSF Industry/University Cooperative Research Center at Polytechnic Institute o NYU. designed or low-intererence local area networks and does not perorm well in intererence-limited environments [5]. In an eort to address the problems in existing smallcell approaches and urther improve cellular network capacity, we proposed a ramework where emtocells simultaneously utilize both licensed and unlicensed bands using LTE-A air interace in an earlier work [6]. Such an approach requires addressing the ollowing challenges. First, a mechanism or the emtocell to access the unlicensed band must be devised and the impact o the coexistence o cellular and non-cellular unlicensed users has to be modeled/analyzed. Another challenge is the assignment o emtocell traic into both bands in order to exploit the beneits o both licensed and unlicensed bands. In [6], we proposed a radio access technology (RAT) or emtocells in the unlicensed band; we also studied the coexistence o a emtocell with WiFi users in a ully utilized unlicensed band. In this paper, we irst analytically study the coexistence o a emtocell with other users in a non-ully utilized unlicensed band, thereore completely addressing the the coexistence issue. Based on the coexistence analysis, we then mathematically ormulate a emtocell traic assignment strategy over the licensed and unlicensed bands in order to minimize the emto-to-macro cross-tier intererence while causing a minimal disruption to other unlicensed band users. The proposed ramework, including the proposed traic assignment algorithm, is evaluated through simulations under a practical deployment scenario. The results show that the perormance o both macrocell and emtocells are improved substantially by the proposed ramework. Under a median emtocell load, our traic assignment algorithm improves the macrocell throughput by over 200% compared with other reerence approaches and existing practices, thanks to a signiicant reduction o the intererence rom small cells to the macrocell. In addition, the total throughput o all cellular and non-cellular users is also improved. Finally, the perormance degradation o other unlicensed band users caused by our emtocell ramework is shown to be small and very similar to that caused by current WiFi hotspot small cell approach. Our work is related to a ew recent papers [7], [8], where emto BSs use cognitive technology to identiy and use the radio resources that are not utilized by the macrocell. In these papers, macrocells are the primary users in the licensed cellular band while emtocells are considered as secondary users. However, these emtocells only operate in cellular licensed

2 channel and use it or a ixed duration, T celltx ; otherwise, the BS will wait or the next access opportunity. As we will see in the ollowing sections, an BS can adjust its usage o the unlicensed spectrum by tuning the parameters T attempt and T celltx. In practice, T sensing should be very small (in the order o 10 microseconds) and is determined by the hardware thus not as tunable as T attempt and T celltx. In addition, an BS is not allowed to access the channel immediately ater a channel use, in order to leave at least T attempt seconds between two consecutive transmissions or other unlicensed band users. Ater obtaining the channel, the emtocell will ollow LTE- A air interace in the subsequent transmissions. Fig. 1. Femtocell channel access mechanism in the unlicensed band. spectrum; while in our work, emtocells jointly utilizes both licensed and unlicensed bands. In the rest o this paper, we assume that the obtained unlicensed band is solely used or downlink transmissions. However, the analysis and conclusions can be easily generalized to the uplink and the shared DL/UL cases. The outline o the paper is as ollows. In Section II, we summarize the emtocell radio access technology (RAT) proposed in [6]. In Section III, we analytically study the coexistence o emtocell with other unlicensed band users. In Section IV, we describe the proposed emtocell traic assignment scheme. In Section V, we validate our analysis and evaluate the proposed algorithm through simulations. II. A FEMTOCELL RADIO ACCESS TECHNOLOGY IN THE UNLICENSED BAND Most current channel access schemes used in the unlicensed band are designed or distributed networks, where collisions may happen both among nodes in the same network (e.g., between WiFi nodes) and among nodes in dierent networks (e.g., between WiFi and Bluetooth). However, in emtocell networks, collisions in the same network can be completely avoided since UEs are coordinated by BSs. Thereore, the approaches that are mainly used to address collisions within the same network, such as the WiFi network allocation vector (NAV) and random backo mechanisms [9], are unnecessary or emtocells. In act, those mechanisms either prevent concurrent transmissions (e.g., the WiFi NAV) or waste channel resources (e.g., WiFi random backo), resulting in ineicient channel access. We have proposed a emto BS channel access mechanism in [6], in order to overcome the drawbacks o distributed channel access schemes. The proposed mechanism is illustrated in Fig. 1. The BS attempts to access the channel only at preassigned periodic time instants, called access opportunities. The period o the access opportunities is denoted as T attempt. Upon the arrival o an access opportunity, the BS starts sensing the unlicensed band. I the channel appears idle or a pre-deined time duration, T sensing, the BS will access the III. MANAGING THE COEXISTENCE OF FEMTOCELL WITH OTHER UNLICENSED BAND USERS In order to manage the coexistence o emtocells with other users in the unlicensed band, we develop analytical tools to adjust emtocell channel usage. In [6], we considered the coexistence o emtocells in an unlicensed band that is ully utilized by the other users. In this paper, however, we consider the case where the unlicensed band is not ully utilized by others. In this case, it is possible or a emtocell to avoid reducing the channel usage o other unlicensed band users. Thereore, we deine a riendly coexistence strategy as controlling emtocell channel usage by tuning BS channel access parameters, so that the channel usages o other users are the same as the case without emtocells. A. System Model We deine two parameters to describe the unlicensed band environment o an BS. Denote r o (0 < r o < 1) as the maximum channel utilization (in terms o raction o time) when unlicensed band users attempt to access the channel all the time. Note that most devices access the unlicensed band based on channel sensing, hence, r o can never be 1; otherwise, no user will be able to transmit. Let r (0 < r < 1) be the average probability that the BS detects the unlicensed band busy. In practice, r and r o can be obtained rom long-term channel sensing. We have the ollowing assumptions. Firstly, we do not need to know the RATs o other unlicensed band users, as long as they are based on channel-sensing and the time instants that they access the channel are random due to their random traic arrivals or channel access schemes. Additionally, the BS and other unlicensed band users may or may not sense each other, depending on the path loss and transmit power. Finally, the desired emtocell channel usage denoted as t cellfrac is set no larger than (r o r) so that the other unlicensed band users will still enjoy the same channel usage r as beore. In the ollowing analysis, we will study how to achieve a desired t cellfrac by tuning BS channel access parameters. B. Perormance Analysis Femtocell channel usage denoted by t cellfrac and the BS channel access success probability, denoted by P cellsucc, can

3 be expressed as unctions o each other. In [6], we developed the ollowing relationship or ully-utilized unlicensed bands: η t cellfrac = 1/P cellsucc + η, (1) where η = T celltx /T attempt, (2) and x means the smallest integer that is larger than or equal to x. It is not hard to argue that (1) also holds or non-ully utilized scenarios. We can also express P cellsucc as a unction o t cellfrac as ollows, P cellsucc = 1 t cellfrac r r = 1. (3) 1 t cellfrac 1 t cellfrac The numerator (1 t cellfrac r) is the raction o time that the channel is idle. The denominator (1 t cellfrac ) is the raction o time that the BS senses the channel. Here we use the assumption that the time instants that the coexisting unlicensed band users access the channel are random. We can obtain P cellsucc and t cellfrac by solving the equations (1) and (3). When η 1, we use the approximation η η to simpliy (1). Then the solution to t cellfrac is obtained as: { 2+η r (2 η r) 2 +4rη t cellfrac = 4, i 0 < η < 1; 1 r 1+1/η, otherwise. (4) Through algebraic manipulations, we can easily show that this is the only easible solution. In practice, (4) can be used by an BS to set a proper channel access parameter η to achieve a desired channel usage t cellfrac. The results o our simulations or overlapping emtocells and WiFi WLANs suggest that the relationship derived in (4) is quite accurate. 1 Furthermore, the simulations conirm that WiFi channel usage is not impaired by emtocells, thanks to our riendly coexistence strategy. IV. FEMTOCELL TRAFFIC ASSIGNMENT OVER LICENSED AND UNLICENSED BANDS In this section, we design a emtocell traic assignment scheme based on the analytical coexistence management tools developed or ully-utilized (in [6]) and non-ully-utilized unlicensed bands (in Section III). The goal is to minimize the intererence rom the emtocell to the macrocell, while maintaining a required rate at the emtocell and causing small throughput degradation to the other unlicensed band users (or both ully-utilized and non-ully-utilized unlicensed bands). A. System Model To describe the basic principles behind our traic assignment strategy, we assume there are one macro base station (mbs) and one BS, each o which has only one user. In the licensed cellular band, there are two interering links in the downlink, one rom BS to macro user equipment (mue) and the other rom mbs to emto user equipment (UE). The macrocell is assumed to make no changes in its transmission 1 These results are not included here due to lack o space, and can be ound in the journal version [10]. strategy (e.g., power allocation, scheduling, etc.) to adapt to the presence o emtocells, so the mbs-to-ue intererence cannot be controlled by the BS and will be ignored in the analysis. All intererence is treated as noise in our analysis. B. A Strategy or BS to Use Both Bands In the traic assignment process, an BS tries to minimize its intererence to macro UEs by tuning its transmission parameters including the transmit power P in the licensed band and the channel access parameter η in the unlicensed band. Hence the objective is min I m (P ) = P h 2 m, (5) where h m is the channel gain o the BS-to-mUE interering link. Three constraints must be satisied. First, the throughput degradation caused by the emtocell to the other unlicensed band users must be small. An BS controls its impacts to others using the riendly coexistence management strategies developed in [6] (or ully utilized unlicensed band, i.e., r r o ) and in Section III-A (or non-ully utilized unlicensed band, i.e., r < r o ). The second constraint is that the BS has to achieve a target data rate or itsel: R (P ) + R R, (6) R where the constant is the required total DL data rate or this emtocell, R is the achieved emto rate in the unlicensed band and R (P ) is the emtocell rate unction in the licensed band, which depends on transmit power P, BS-to-UE channel gain, intererence rom mbs as well as the modulation and coding scheme (MCS). Note that R (P ) is an non-decreasing unction o transmit power P. An BS can estimate the average R based on channel and intererence statistics in the unlicensed band and its target channel usage t cellfrac determined by the coexistence management strategies developed in [6] and in Section III-A. As a result, the BS eectively obtains a ixed out-o-band data rate R rom the unlicensed band. The third constraint is an BS peak transmit power limit, 0 P P, (7) where P is the maximum BS transmit power in the licensed band due to battery limitations and restrictions imposed by regulations. In the unlicensed band, however, we assume no power control thereby ixing the transmit power P. The solution to the optimization problem (5)-(7) is as ollows: I R R 0, inequality (6) always holds. The optimal P is 0 and the licensed band is not used. I R R > 0 and P R 1 ( R R ), the optimal P is R 1 ( R R ), where R 1 (R) corresponds to the lowest transmit power to attain the desired rate R. In this case, the BS uses both bands with data rates o ( R R ) in the licensed band and R in the unlicensed band. Otherwise, there is no easible solution thus the desired emtocell rate R cannot be supported.

4 C. An Algorithm to Implement the Traic Assignment Strategy Based on the analytical results in Sections III and IV-B, we propose a traic assignment algorithm that can be implemented in the MAC layer o BSs. An BS will ollow the steps below. 1) Estimate r and r o through long-term channel sensing. 2) Adjust unlicensed band channel access parameters. a) I r < r o, the unlicensed band is not ully utilized, we set the desired emto channel usage t cellfrac to (r o r) using the coexistence strategy in Section III; then we set BS channel access parameter η using (4) in order to achieve the t cellfrac. b) I r r o, the unlicensed band has already been ully utilized, we will use the coexistence strategy in [6]: irst, set a desired t cellfrac which is air to other unlicensed band users; second, set η using the equation (1) in order to achieve the t cellfrac. The P cellsucc used in (1) can be estimated rom historical data, since it does not vary with emtocell channel usage in a ully utilized unlicensed band. 3) Estimate the unlicensed band data rate R based on t cellfrac and SINR statistics.2 4) Compute the data rate required in the licensed: R = ( R R )+, where R is the average emto DL load obtained rom historical data. 5) Compute BS transmit power P in the licensed band using the results at the end o Section IV-B. 6) With the above operations, an BS obtains data rates o ( R R )+ and R in the licensed and unlicensed bands, respectively. Then the BS proportionally assigns (1 R / R ) + and min(1, R / R ) ractions o its traic to these two bands, respectively. V. PERFORMANCE EVALUATION In this section, we evaluate the proposed traic assignment algorithm o Section IV in practical deployment scenarios. The CoopMAC simulation platorm [11] [12], a customized eventdriven IEEE network simulator built in C language, is used in our simulations. The simulator models random packet arrivals at the IP layer, captures most details o channel access schemes (or both emtocell and WiFi) at the MAC layer, and includes intererence computation and SINR-tothroughput mappings at the PHY layer. The parameters used in the simulations and in our analytical models are summarized in Table I where the path loss models are adopted rom [2], [13]. For simplicity, ading and mobility are not considered. The ollowing topology is considered in the simulations. A mbs is placed at the center o the macrocell with radius 300m. 40 WiFi APs, 40 emto BSs and 50 macro UEs are randomly dropped in the macrocell. Two UEs (or WiFi stations) are dropped within each BS (or AP) s coverage area with a radius o 20m. LTE-A is adopted as the cellular air interace while 2 We can map any given SINR to a corresponding data rate; hence, with SINR statistics, we can compute the expected rate. Thereore, R is just a t cellfrac raction o the expected rate. TABLE I PARAMETERS USED IN SIMULATIONS AND ANALYTICAL MODELS MAC SIFS: 10 µs CWMin: 31 DIFS: 28 µs CWMax: 1023 Slot Time: 9 µs RTS/CTS: Enabled Femto Channel Access r o = 0.9 T sensing = 18 µs IP Packet Size: 1500 Bytes Transmit Power mbs: 40 dbm BS: 15 dbm mue: 15 dbm UE: 15 dbm BW (Lic.): 20MHz BW (Unlic.): 20MHz Ch. Sensing Thresh: -62 dbm (WiFi & Femto) Noise Power: -95 dbm (over 20MHz) Path Loss Models based on [13] PL: path loss in db R: distance in meters L ow: the outer wall loss in db mbs mue PL = log 10 (R) mbs or mue emto node PL = log 10 (R) + L ow, L ow = 10 Femto node emto node in the same emtocell, WiFi node WiFi node in the same WLAN PL = log 10 (R) + 0.7R Femto node emto node in dierent emtocell, WiFi node WiFi node in dierent WLAN, Femto node WiFi node PL = log 10 (R) + L ow, L ow = n with a rame aggregation level o 64K Bytes is used or WiFi. When applicable, cellular attempt interval T attempt is set to 1ms and cellular transmission duration T celltx is set to 20ms (i.e., η = 20). Traic loads o 10 Mbps and 150 Mbps are oered to WiFi WLANs and the macrocell, respectively, which are similar to the practical network situations. Due to the randomness in the topology, some emtocells may be in ully-utilized unlicensed band environments while others in non-ully-utilized environments. I a emtocell is close to many WLANs, it will experience a ully-utilized unlicensed band; however, i a emtocell is surrounded by no WLANs or a small number o WLANs, the channel may not be ully utilized. Five dierent approaches or cellular small cells to use the licensed and unlicensed bands are considered: (WiFi oloading): Cellular WiFi hotspots operate only in the unlicensed band using n air interace. In this paper, cellular WiFi hotspot and WiFi WLAN reer to the WiFi networks accessed by cellular and noncellular users, respectively. (traditional emtocell): Femtocells operate only in licensed bands with LTE air interace. (traditional emtocell + WiFi): Dual-band emtocells operate in both licensed and unlicensed bands with LTE and n air interaces, respectively. At the MAC layer, a simple traic assignment scheme is used to assign IP packets to both bands: packets are assigned to the band which is not in transmission; i no bands is transmitting, packets are randomly assigned to one band. (our proposed emtocell RAT with simple traic assignment): Dual-band emtocells operate in both

5 Throughput Per Cell (Mbps) Throughput Per Cellular Small Cell Throughput Per Cell (Mbps) Throughput Per Macrocell 0 Load o Each Cellular Small Cell (Mbps) 10 Load o Each Cellular Small Cell (Mbps) (a) Small Cells (Femto or WiFi hotspot) (b) Macro Total Throughput (Mbps) Total WiFi WLAN & Cellular Throughput 500 Load o Each Cellular Small Cell (Mbps) Total Throughput (Mbps) Total WiFi WLAN & Cellular Throughput 500 Load o Each Cellular Small Cell (Mbps) (c) Existing WiFi WLANs (d) The Whole System (cellular and non-cellular) Fig. 2. Throughput o small cells, macrocells, WiFi WLANs and the whole system under dierent schemes. licensed and unlicensed bands with LTE air interace. The simple traic assignment scheme in is used. (our proposed emtocell RAT with the proposed traic assignment): Dual-band emtocells operate in both licensed and unlicensed bands with LTE air interace. The proposed traic assignment algorithm (Section IV) is used to assign IP packets to both bands. Fig. 2 shows the throughput o cellular small cells (i.e., emtocells or cellular WiFi hotspots), macrocells, WiFi WLANs and the whole system under dierent schemes. The ollowing observations can be made rom the igure. First, the proposed ramework (Cases 4 and 5) improves the throughput o small cells (Fig. 2(a)) by about 30% under high load, compared with the reerence scheme o. This is mainly due to the lower MAC layer eiciency o the WiFi air interace used by the unlicensed band o. Second, under small-cell traic loads o 40Mbps or lower, the proposed traic assignment algorithm improves macrocell throughput (Fig. 2(b)) by over 200% compared with, 3, 4 which also access the licensed band, due to a substantial reduction o the intererence rom the small cells to the macrocell. Third, WiFi WLAN throughput (Fig. 2(c)) drops by a small raction (<20%) in the proposed schemes ( and 5), compared with which does not access the unlicensed band at all. In act, WiFi WLAN perormance degradation in our proposed schemes is very close to that in the current practice o where cellular operators build WiFi hotspots to oload their traic. Finally, the total throughput o all cellular and non-cellular nodes (Fig. 2(d)) is improved by the proposed schemes (Cases 4 and 5) by over 30% under high small-cell loads. In addition, the user throughput CDF reported in Fig. 3 demonstrates that the proposed traic assignment algorithm almost triples the throughput o every macro user, either at cell center or at the edge. Besides, the packet delay results in Fig. 4 also shows that the delay experienced by macro users are signiicantly reduced by our traic assignment algorithm. 3 3 We note an anomaly that the packet delay o small cells in (Let subigure o Fig. 4) drops with an increasing load. This is due to the MAC layer rame aggregation eature in n. Recall that in, small cells adopt n air interace. With a higher load, a transmitter waits or shorter time to accumulate enough MAC rames or aggregation, resulting in shorter delay.

6 CDF CDF o Small Cell User Throughput Femto User Throughput (Mbps) CDF CDF o Macro User Throughput Macro User Throughput (Mbps) Fig. 3. The CDF o user throughput in two-layered (small cell/macrocell) cellular networks under dierent schemes. Traic loads o small cells are 30Mbps. Delay (ms) Average Packet Delay in Small Cell Load o Each Small Cell (Mbps) Delay (ms) Average Packet Delay in Macrocell 40 Load o Each Small Cell (Mbps) Fig. 4. Average delay o IP packets in two-layered (small cell/macrocell) cellular networks under dierent schemes. VI. CONCLUSIONS In this paper, we urther improve our emtocell ramework o jointly utilizing licensed and unlicensed bands proposed in [6]. We mathematically investigate the coexistence o emtocells with other users in non-ully utilized unlicensed bands. We then propose an algorithm or an BS to assign its traic to the licensed and unlicensed bands in a way that minimizes its intererence to the macrocell while considering the perormance o UEs and other unlicensed band users. With the traic assignment algorithm, our proposed ramework signiicantly improves the throughput o emtocells, the macrocell and the overall system (cellular and non-cellular). In addition, the throughput degradation o other unlicensed band users due to emtocells accessing the unlicensed band is small and very similar to that in the current practice where cellular operators build WiFi hotspots to oload their congested traic. REFERENCES [1] V. Chandrasekhar, J. Andrews, and A. Gatherer, Femtocell networks: a survey, IEEE Communications Magazine, pp , Sep [2] S. Rangan, Femto-macro cellular intererence control with subband scheduling and intererence cancelation, arxiv, [Online] abs/ , [3] D. Darlin, Cellphone carriers are turning to Wi-Fi, too, The New York Times, [Available online: nytimes.com], Sep. 11, [4] AT&T Inc., Third-quarter Wi-Fi connections on AT&T network exceed total connections or 2009, AT&T News Releases, [Available Online]: Dallas, Texas, October 22, [5] V. Sridhara, H. Shin, and S. Bohacek, Perormance o b/g in the intererence limited regime, in Proc. o ICCCN, Aug [6] F. Liu, E. Bala, E. Erkip, and R. Yang, A ramework or emtocells to access both licensed and unlicensed bands, To appear in Proc. o the third Int l Workshop on Indoor and Outdoor Femto Cells (IOFC), Princeton, NJ, USA, [Online] Feilu Liu/emtoWiOpt.pd?ormat=raw. [7] G. Gur, S. Bayhan, and F. Alagoz, Cognitive emtocell networks: an overlay architecture or localized dynamic spectrum access, IEEE Wireless Communications, vol. 17, pp , Aug [8] J. P. M. Torregoza, R. Enkhbat, and W.-J. Hwang, Joint power control, base station assignment, and channel assignment in cognitive emtocell networks, EURASIP Journal on Wireless Communications and Networking, 2010, Article ID [9] IEEE Std n-2009, Part 11: Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Speciications, [10] F. Liu, E. Bala, E. Erkip, and R. Yang, A dual-band emtocell ramework or utilizing both licensed and unlicensed bands, Submitted to IEEE Journal on Selected Areas in Communications (J-SAC). [11] P. Liu, Z. Tao, Z. Lin, E. Erkip, and S. S. Panwar, Cooperative wireless communications: a cross layer approach, IEEE Wireless Communications, vol. 13, pp , Aug [12] P. Liu, Z. Tao, S. Narayanan, T. Korakis, and S. S. Panwar, CoopMAC: a cooperative MAC or wireless LANs, IEEE Journal on Selected Areas in Communications, vol. 25, pp , Feb [13] Femto Forum, Intererence management in OFDMA emtocells, Whitepaper available at Mar