Flexible Spectrum Sharing and Interference Coordination for Low Power Nodes in Heterogeneous Networks
Carlo Galiotto, Nicola Marchetti, Linda Doyle CTVR, Trinity College, Dublin, Ireland email@example.com, firstname.lastname@example.org, email@example.com.
Abstract—Heterogeneous Networks (HetNet) have been proposed as a means of boosting the coverage in areas where users experience weak wireless connectivity. In HetNets, due to the massive deployment of Low Power Nodes (LPN) along side the usual Macro Base Stations (MBS), inter-cell interference becomes a signi?cant issue which, if not properly handled, may degrade the network throughput and the macro-cell edge users performance. In this paper, we focus on the problem of the inter-cell interference to the Macro User Equipments (MUE) generated by the LPNs. We propose an algorithm for ?exible spectrum usage among LPNs that enhances macro cell-edge user throughput. By means of a dynamic redistribution of the frequency resources among MBSs and LPNs, this algorithm reduces the harmful interference generated by the LPNs to the macro cell-edge users, hence increasing the throughput of the latter. In addition to helping the terminals which experience high interference, the proposed solution also limits the overall reduction in network throughput, which is usually one of the side effects of interference coordination in HetNets. The simulation results show that, compared to a full frequency reuse for the LPNs, the proposed algorithm improves the throughput of the macro cell-edge users, which are usually the ones experiencing the lowest performance within the network. Overall, with limited impact in terms of HetNet throughput reduction and with good bene?t to the cell-edge users performance, this algorithm represents an effective solution to the inter-cell interference problem affecting the HetNets. Index Terms—Heterogeneous networks, spectrum sharing, interference coordination, pico-cells, low power nodes, Flexible Spectrum Usage (FSU), Long Term Evolution (LTE).
of this interference, as these UEs are located far from the serving MBS and in proximity of the interfering LPNs. In order to combat the inter-cell interference, frequency planning based techniques (e.g. Hard Frequency Reuse or Soft Frequency Reuse ) for Inter-Cell Interference Coordination (ICIC) are usually adopted in macro-cells networks. However, due to the large number of lower power nodes deployed in HetNets, the frequency planning of massive LPN deployments is excessively complex . Hence, ICIC for LPN in HetNet requires a more ?exible approach for resource sharing compared to the ?xed frequency reuse schemes. In recent years, some solutions for ICIC in HetNets have been proposed. In , the authors propose a Q-learning based approach where each Femto Base Station (FBS) tries to learn a spectrum allocation pattern that maximizes the femtocell throughput and, at the same time, does not interfere with the neighboring MUEs. In , the femto to macro inter-cell interference coordination is achieved by a planned orthogonal subcarrier assignment between MBSs and LPNs. This approach, while being bene?cial in terms of cell-edge user throughput does however considerably limit the network spectral ef?ciency. In most of the existing solutions, the ICIC task for the HetNet is limited to the reduction of transmission power or to the reuse of common bands by adjacent BSs. Unfortunately, any restriction to the BSs’ power/band resources brings with it a cost in terms of network throughput reduction. In this paper we further develop the concept of ICIC I. I NTRODUCTION in HetNets and provide a novel approach for boosting the In order to enhance the coverage in those areas where users performance of the MUEs suffering for high LPN interfermay experience bad wireless connectivity from Macro Base ence. In addition to improving the performance of the cellStations, network operators have recently started deploying edge users, we also strive to contain the network throughput Low Power Nodes . Such a network in which MBSs and reduction when carrying out ICIC for the HetNet. We propose LPNs using the same Radio Access Technology coexist in a distributed Flexible Spectrum Usage Algorithm for HetNets order to improve the wireless coverage is referred to as a (FSUAH) which makes use of a joint ICIC/user scheduling HetNet . technique. FSUAH attempts to provide a target throughput at However, due to the proximity of several Base Stations (BS) the cell-edge UEs ?rst by acting on the MBS user scheduler, sharing the same spectrum, users may experience high inter- which is assumed to be of a Proportional Fair (PF) kind. Then cell interference which can degrade the network performance secondly and only when it is not possible to achieve our , . In this paper we deal with the problem of the high target through the scheduler alone, the inter-cell interference inter-cell interference in the downlink of HetNets and we focus is limited by dynamically reducing the spectrum resources speci?cally on the macro-users. Indeed, in HetNet scenarios, assigned to the adjacent LPNs. Overall, we show that it is the MUEs located at the macro-cell edges are typically victims possible to enhance the HetNet performance by jointly using
978-1-4673-1881-5/12/$31.00 ?2012 IEEE
a PF user scheduler, carried out in an uncoordinated manner among the MBSs, and a ?exible spectrum sharing technique for ICIC purposes. The paper is organized as follows. In Section II we propose the FSUAH algorithm, while in Section III the simulation results are presented and discussed. The conclusions are drawn in Section IV. II. A LGORITHM : FSUAH A. System overview and algorithm framework The FSUAH has been designed for Orthogonal Frequency Division Multiple Access (OFDMA) based systems. Even though in this paper we do not consider any speci?c mobile wireless communication standard, we will refer to Long Term Evolution (LTE) for some of the physical layer implementation details, such as the frequency-time resources organization. For example, in LTE, the smallest frequency and time resource unit we can make use of is the Resource Block (RB) . Although FSUAH is a distributed algorithm which runs in each MBS and in each LPN, this algorithm only controls the spectrum in use by the LPNs. As the frequency planning in a macro cell layer scenario is feasible and of common practice for ICIC purposes , we assume the MBSs use a hard frequency reuse scheme (i.e. reuse 3). Fig. 1 shows an example of a possible scenario, with 3 MBSs coexisting with 3 LPNs. Each MBS sends feedback to the surrounding LPNs in order to signal them whenever MUEs experience high interference. As proposed in , this feedback could be represented by the High Interference Indicator (HII) which reports the indices of the highly interfered RBs to the LPNs. The effect of this feedback is to reduce the frequency resources in use by the adjacent LPNs. The usage of HIIs is the only kind of cooperation of which the FSUAH makes use. No cooperation is required between MBSs or between LPNs. Overall, MBSs and LPNs share the same common frequency resource pool.
Fixed hard frequency reuse
MBS1 f [MHz] MBS2 f [MHz] MBS3 f [MHz] MBS1 MBS3 LPN1 MBS2 Feedback (HII) LPN2
transitions. We de?ne a user operating point as a couple of metrics (i.e. cell-edge user throughput or Rce and average user throughput or Rav ) which will be represented in a 2Dplot. The operating points are a graphical representation of the throughput experienced by the cell UEs and the resource sharing fairness among the UEs themselves. We de?ne the fairness degree as the ratio Rce /Rav ; the higher the fairness, the closer to 1 Rce /Rav will be. In this paper, we assume that UEs are assigned RBs by means of a Proportional Fair (PF) algorithm which allocates α resources to the UEs based on the metric M = d , where d rβ is the current throughput experienced by the UE on a given RB and r is a time averaged measure of the throughput experienced by the UE in the past. The parameters α and β allow the fairness of the algorithm to be speci?ed. By keeping α constant and increasing β we increase the PF fairness, while we favor a less fair throughput increase the other way around. The UE with the highest value of M will then be allocated over a given RB.
Macro cell operating curves 0.25 LPN using the full MBS transmission band LPN partially using MBS transmission band LPN not using MBS transmission band State 2 R < TH av av Rce≥ THce TH 0.1 TH
State 1 R ≥ TH av av R ≥ TH
State 3 R < TH av av R < TH
State 4 R ≥ TH av av Rce< THce 0.3 R
0.35 0.4 [Mbps]
Flexible frequency reuse
LPN1 f [MHz] LPN2 f [MHz] LPN3 f [MHz] LPN3
Fig. 2: MBS operating curves and states. Three different operating curves (blue, red, green) are shown, each of them with 4 operating points. Each operating point corresponds to a given set of PF parameters. As we can see from the example shown in Fig. 2, by means of varying the cell resource share among the same macro cell users (i.e. by varying the PF parameters α and β ) we obtain operating points with different fairness degrees; the lines joining these points are the operating curves. Assuming that: 1) the portion of band used by the LPNs that overlaps with the band used by the MBS remains constant over time; 2) a PF scheduler is used for UEs scheduling; an operating curve is obtained by varying the fairness degree of the PF scheduler. Hence, each operating curve is obtained by keeping the LPN interference conditions constant (i.e. the LPNs transmit over a given portion of band) and by varying the PF scheduler fairness degree or, equivalently, by varying the PF parameters. If we vary the LPNs interference conditions we then obtain a different operating curve.
Fig. 1: MBS and LPN spectrum reuse. B. User operating points In this section we introduce the concept of operating points which will be used as a tool for triggering the algorithm state
state at time t + 1 into the next right-top operating curve (see Fig. 2). Overall, FSUAH will always try to bring and keep the The aim of the MBS is to attempt to meet two user MBS in one of the operating points in state 1, which is the throughput requirements, which are represented by the perstate where the performance targets are met. formance metric thresholds T Hce and T Hav for Rce and Rav The MBS always keeps track of the RBs for which a HII is respectively (refer to Section II-E for further details concerning sent. If the HIIs are sent for all the RBs in use by the MBS, the T Hce and T Hav values setting). In order to achieve this the MBS waits for a given time Treset before beginning again goal, the MBS estimates the operating point (Rce , Rav ) and to respond and take action. In fact, when the HIIs are sent for compares each of these metrics with their respective thresholds all the RBs, there is no other way to reduce the interference of T Hce and T Hav . As shown in Fig. 2, the results of this the LPNs on the MUEs and hence to bring the MBS to state comparison provide the state of the MBS. 1. In this case, after a waiting period Treset has elapsed, some The FSU algorithm running in the MBS always tries to changes within the network may have occurred (i.e. traf?c achieve the required performance ?rst by adjusting its own PF load, users distribution and number of UEs connected to the scheduler parameters rather than requiring a LPN spectrum BSs) and thus it makes sense to reset the algorithm and restart reduction. In Fig. 2, varying the PF parameters corresponds to to evaluate the MBS state. moving the MBS towards different operating points along the same operating curve. Only if the MBS is not able to reach D. Algorithm running in the LPN the expected throughput requirements by means of adjusting As we can see from the ?ow-chart in Fig. 4, after the the PF parameters, a HII is issued to attempt to limit the initialization, the algorithm works with two parallel ?ows. One interference from the neighboring LPNs. If any of the LPNs of these is in charge of increasing the number of the RBs used receiving the HII is the actual cause of the interference, the HII and it is a cyclic operation which runs with period TN EW . has the effect of moving the MBS towards different operating The second ?ow is responsible for limiting the number of curves (see Fig. 2). The algorithm ?ow-chart is shown in Fig. RBs usable by the LPN and is driven by the HIIs sent by the 3. MBSs. C. Algorithm running in the MBS
START Evaluate state
Action No action Decrease PF fairness Send HII Increase PF fairness
Wait for HII Wait for TNEW
If there are still RBs for which an HII has not been sent
Increase number of used RBs
Reduce number of used RBs and update set of available RBs
If HII has been sent for all RBs Reset algorithm
Wait for Treset
Fig. 4: Flow-chart of FSU algorithm running in the LPN. Fig. 3: Flow-chart of FSU algorithm running in the MBS. After initialization, the algorithm periodically evaluates the state in the MBS. Based on this state, an action is taken (see table in Fig. 3), after which the algorithm waits for a certain time Teval . This time is necessary to observe the effect of the action on the operating point. If at time t the MBS state is 2 (or 4), the corresponding action (see table in Fig. 3) will bring the MBS state at time t +1 into the operating point with lower (or higher) degree of fairness along the same operating curve (see Fig. 2). On the other hand, if at time t the MBS state is 3, a HII will be issued (see table in Fig. 3) in order to attempt a reduction of the interference generated by the LPNs. Assuming that the LPNs receiving the HII are the actual cause of the interference on the MUEs, this HII will bring the MBS Whenever the MBSs issue a HII requiring the LPN to release some given RBs (each HII reports the indices of K RBs), the LPN considers these RBs as non-available for some given time. This means that, when the FSU algorithm has to increase the spectrum in use in the LPN, only the available RBs can be chosen. The LPN will select the L RBs where the UEs experience the highest average SINR. Each time the LPN receives a HII from the MBSs indicating that some RBs must be released, the LPN will set these RBs as non-available and will stop using them for a period TL . After a speci?c time has elapsed, the LPN will set these RBs as available again. E. Algorithm parameters Two parameters must be set for the FSUAH running in the MBSs, i.e., the performance targets T Hce and T Hav .
These parameters represent the throughput values which are TABLE I: Summary of the parameters used in the simulations. considered to be satisfactory for the MUEs. In fact, setting System model Spectrum allocation 50 RBs over 10 MHz centered at 2 GHz, T Hce and T Hav to higher values will trigger FSUAH to shared by MBSs and LPNs attempt the achievement of higher cell-edge and average user MBS/LPN TX power 46/30 dBm throughputs for the MBSs. For this reason, T Hce and T Hav MBS antenna Directive antenna, 3GPP model  / 14 dBi pattern/gain will drive the performance achievable by the algorithm. LPN antenna Omnidirectional antenna / 5dBi Since solving a maximization problem for ensuring an pattern/gain optimal setting of the parameters T Hce and T Hav is beyond UE antenna Omnidirectional antenna / 0 dBi pattern/gain the scope of this paper, we follow a different approach for Propagation model choosing T Hce and T Hav . We ?rst measure the average User Channel model 3GPP for HetNet System simulation for outThroughput (UTH), i.e. Rav , and the cell-edge UTH, i.e. Rce , door RRH/Hotzone, model 2, case 1,  in each macro-cell. Then, from the sets of Rav and Rce values Shadow fading 8-10 dB Link-level model obtained from all the macro-cells we compute the Cumulative Link-level abstraction Upper-bounded LTE Shannon’s capacity  Distribution Function (CDF) of Rav and of Rce . Let us refer to Traf?c model Rav CDF and to Rce CDF as Fav (x) and Fce (x) respectively. Traf?c model Full buffer Scenario deployment Finally, the thresholds T Hce and T Hav are set as follows. HetNet BS deployment 3GPP HetNet Spec. : Macro + outdoor First, we choose two percentage values Xav,% and Xce,% of (Macro-cells and pico- RRH/Hotzone, case 1, con?guration 4 for UEs Fav (x) and Fce (x) respectively. Second, we set the thresholds cells) distribution (57 macro-cells, 228 pico-cells, ?1 ?1 3420 users). (Xav,% ) and T Hce = Fce (Xce,% ). With this as T Hav = Fav Number of snapshots 50 setting, FSUAH will attempt to enhance the average UTH of User scheduler the Xav,% % of macro-cells experiencing the worst Rav and Scheduler Proportional Fair the cell-edge UTH of the Xce,% % of macro-cells experiencing (α,β ) - MBS (1, 2), (1, 1.2), (1, 1), (1.2, 1), (1.4, 1), (1.6, 1), (1.8, 1), (2, 1), (4, 1) the worst Rce . In Section III-B we will analyze how different (α,β ) - LPN (1, 1) values of T Hav and T Hce affect the performance achievable FSUAH - MBS by the algorithm. (T Hav , T Hce ) = Set 1: (F ?1 (19%), F ?1 (27%)) = (250, 60) III. S IMULATION R ESULTS A. Simulation parameters and scenario The FSU algorithm has been tested by means of system level simulation, whose parameters are shown in Table I. In relation to the FSU algorithm parameters, only T Hce and T Hav affects the algorithm performance in terms of throughput gain, while K , L and the timers Teval , Treset , TN EW and TL rather have some impact on the convergence speed. The values set for the parameters in Table I guarantee the convergence of the algorithm under the condition of the full buffer traf?c model and static cell-load (i.e. number of UEs attached to the BSs). The analysis of the algorithm under different simulations models will be object of future studies. B. Simulation results The performance metrics which are used for testing the FSU algorithm are the cell-edge user throughput Rce and the average user throughput Rav . The cell-edge user throughput is de?ned as the user throughput value corresponding to the 5th percentile of the user throughput Cumulative Distribution Function (CDF). Rce measures the throughput experienced by the so called "cell-edge users", i.e., the users which suffer from bad radio conditions such as low received power or high interference. The FSUAH makes a trade-off between the average user throughput of the LPNs and the cell-edge user throughput of the MBSs, in favor of the latter. We expect that, by limiting the spectrum used by the LPNs, the FSU attenuates the interference of the LPN to the macro cell-edge UEs. Thus, the effect we observe consists of a reduction of the average
[kbps, kbps] K Teval Treset L TN EW TL
?1 ?1 Set 2: (Fav (40%), Fce (47%)) = (300, 80) ?1 ?1 Set 3: (Fav (56%), Fce (66%)) = (350, 100) 4 0.5 s 4s FSUAH - LPN 5 1.2 s 2s av ce
LPN user throughput on the one hand, and a gain of the macro cell-edge UEs throughput on the other hand. In this section, we compare the user throughput given by the FSUAH algorithm with the one given by a static frequency allocation scheme, namely reuse 3-1 (i.e. R3-1). In R3-1, the LPNs transmit over the whole 10 MHz band while the MBSs adopt the so-called reuse 3 scheme, i.e., each MBS uses 1/3 of the available spectrum which is orthogonally shared among neighboring MBSs. In contrast, in the FSUAH scheme, the MBSs employ reuse 3 scheme while the spectrum in use by the LPNs is dynamically determined by the FSUAH algorithm (see Section II-A). In Fig. 5 we can see the effect of the FSU algorithm in terms of Rce enhancement on the MUEs throughput. In fact, the FSU can improve the MUEs Rce by up to 90% compared to R3-1. Moreover, the MUEs also experience an average user throughput gain, which can be up to 24%. This overall enhancement of the MUEs peformance is due to the joint effect of the PF scheduler and of the ICIC operated by the FSUAH algorithm. While the ICIC (implemented by means of spectrum reduction of the LPN) yields a bene?t in
150 Cell?edge user throughput ? Rce [kbps] FSUAH ? Thresholds set 3 FSUAH ? Thresholds set 2 FSUAH ? Thresholds set 1 Static spectrum allocation ? R3?1 Cell?edge user throughput ? Rce [kbps]
200 180 160 140 120 100 80 60 40 20 0 0 0.5 1 1.5 Average user throughput ? Rav [Mbps] 2 FSUAH ? Thresholds set 3 FSUAH ? Thresholds set 2 FSUAH ? Thresholds set 1 Static spectrum allocation ? R3?1
100 200 300 400 Average user throughput ? Rav [kbps]
Fig. 5: Macro cell user throughput.
Fig. 6: Overall network user throughput.
terms of interference reduction to the MUEs, the PF algorithm makes this bene?t pro?table in terms of cell-edge throughput gain. We observe also that, by changing the parameters T Hce and T Hav , the FSUAH achieves different throughput values. In general, the higher are T Hce and T Hav , the higher is the throughput experienced by the MUEs. However, as aforementioned, the performance gain obtained by the MUEs will be paid in terms of throughput loss for the LPNs users. Hence, increasing the values of T Hce and T Hav will not provide an unlimited bene?t to the overall network throughput. As far as the whole network users are concerned (i.e., the UEs served by the MBSs and by the LPNs), we still observe a gain in terms of cell-edge UTH compared to R3-1 (see Fig. 6). This can be explained if we consider the CDF of the overall HetNet user throughput. In fact, in the scenario assumed for our simulation, it turns out that most the network cell-edge users1 are users served by MBSs. Consequently, improving the MUEs throughput results in a better overall cell-edge user performance. If we consider the HetNet average user throughput (see Fig. 6), FSUAH does not cause any performance reduction. In fact, the Rav gain obtained for the MUEs is suf?cient in order to counterbalance the Rav losses for the LPNs users. Hence, in the scenario we assumed for the simulation results, FSUAH has been proved to be an effective technique for improving the cell-edge user throughput in HetNets, with a gain which can reach up to 88%. IV. C ONCLUSIONS AND F UTURE W ORK In this paper we have tackled the problem of the intercell interference in Heterogeneous Networks (HetNet) and we speci?cally focused on the interference of the Low Power Nodes (LPN) on macro users. We proposed a distributed algorithm, i.e., FSUAH, which allows the LPNs to make ?exible usage of the spectrum, mitigating the harm of the
1 In this case, the term cell-edge user does not refer to the fact that the UE is located at the cell edge, but it means that the throughput experienced by the UE is below the value corresponding to the 5th percentile of the the user throughput CDF.
LPN interference to the macro users and providing, at the same time, a good overall HetNet average user throughput. Moreover, FSUAH only requires limited signaling between Macro Base Stations (MBS) and LPNs, meaning low overhead. By using a joint UE scheduling / Inter-Cell Interference Coordination (ICIC) technique, FSUAH has been shown to increase the macro cell-edge user throughput by up to 90% and to enhance the HetNet cell-edge throughput by up to 88% in the scenario assumed for our simulations. Furthermore, the gain in terms of average user throughput obtained for the MBSs counterbalances the throughput loss of the LPNs, thus containing the overall network reduction in network throughput. The future work will be focused on the problem of the LPN to LPN interference, in order to improve the performance of LPNs layout within the HetNet R EFERENCES
 J. Zhang and G. DeLaRoche, Femtocells: Technology and Development. John Wiley & Sons, 2010.  V. Pauli, J. D. Naranjo, and E. Seidel, “Heterogeneous LTE Networks and Inter-Cell Interference Coordination,” Nomor Research White Paper GmbH, Munich, Germany, December 2010.  A. Damnjanovic, J. Montojo, Y. Wei, T. Ji, T. Luo, M. Vajapeyam, T. Yoo, O. Song, and D. Mallad, “A Survery on 3GPP Heterogeneous Neworks,” IEEE Wireless Communications, vol. 18, no. 3, pp. 10 – 21, 2011.  B. Bennis and D. Niyato, “A Q-learning Based Approach to Interference Avoidance in Self-Organized Femtocell Networks,” IEEE GLOBECOM Workshops (GC Workshops), pp. 706–710, December 2010.  V. Chandrasekhar and J. G. Andrews, “Spectrum Allocation in Tiered Cellular Networks,” IEEE Transaction on Communications, vol. 57, no. 10, pp. 3059 – 3068, October 2009.  H. Holma and A. Toskala, LTE for UMTS - OFDMA and SC-FDMA Based Radio Access. John Wiley and Sons, 2009.  “Winner+ Project - D1.6 Deliverable: Intermediate Report on System Aspects of Flexible Spectrum Use,” November 2009.  3rd Generation Partnership Project (3GPP), “Further Advancements for E-UTRA Physical Layer Aspects (Release 9),” March 2010, 3GPP TR 36.814 V9.0.0 (2010-03).  P. Mogensen, N. Wei, I. Kovacs, F. Frederiksen, A. Pokhariyal, K. Pedersen, T. Kolding, K. Hugl, and M. Kuusela, “LTE Capacity Compared to the Shannon Bound,” Proceedings on the 65th IEEE Vehicular Technology Conference (VTC 2007-Spring), April 2007.