Interference mitigation by distributed beam forming optimization

Transcription

1 English Interferene mitigation by distributed beam forming optimization Matthias Kashub, Christian M. Blankenhorn, Christian M. Mueller and Thomas Werthmann Abstrat Inter-ell interferene is a major issue in OFDMA networks. One approah to redue the amount of interferene is to use beam forming antennas. Further interferene mitigation is ahieved if neighbor base stations oordinate the diretions of their beams. This paper presents a novel distributed interferene oordination algorithm using main-lobe steering beam formers. Our algorithm does not require any expliit signaling between base stations. Also, it does not require any additional hannel measurements to be signaled to the base stations besides those already performed for basi operation. Our evaluations show that signifiant performane gains an be ahieved even with non-greedy traffi. Index Terms Interferene oordination, beam forming antennas, COMP 1. Introdution 1.1 Motivation For ellular radio aess networks, the ever growing demand for apaity makes spetrum a sare resoure. To use the available spetrum as effiiently as possible, these networks use the same frequeny resoures in all base stations (reuse 1) and try to maximize the spetral effiieny. In reuse 1 senarios, the ahievable spetral effiieny is usually limited by the interferene from neighboring base stations, whereas noise does not play a signifiant role. Therefore, the goal of most interferene oordination mehanisms is to redue the interferene from neighboring ell setors on the transmissions in the urrent ell setor. Espeially for users at the ell edge this improves the hannel quality. We onsider the downlink of ellular OFDMA networks suh as IEEE e WiMAX or 3GPP LTE. We assume the base station to be equipped with multiple antennas that are used in ombination with main-lobe steering beamformers. 1.2 Related Work The lassial approah to handle interferene is frequeny reuse. It partitions the spetrum and uses the same partition in spatially disjoint ell setors, only [1]. Frational frequeny reuse is an extension of the lassial sheme. It splits ells into different areas, defines primary and seondary resoures for these areas, and assigns primary and/or seondary resoures to terminals. The so-alled "soft reuse" shemes further redue transmission power for seondary resoures. The partitioning of the resoures is usually done during network planning or in a semi-stati way on a large time sale. Higher gain is ahieved with more dynami shemes, whih adapt to traffi variations and user distribution [2]. Another approah is feasible in ases where the power of an interfering signal is suffiiently high. Then the reeiver an also deode the interfering signal and subtrat it from the desired signal. In addition, modern interferene redution tehniques also use multi-antenna arrays to distinguish the signal soures. When the base stations are equipped with suh antenna arrays, the spatial domain an be utilized to avoid interferene at the transmitter side [3]. The oordination of transmission modes of multiple base stations is alled oordinated beam forming in the 3GPP speifiations [4]. In this paper, we investigate an interferene oordination sheme for the downlink of an OFDMA mobile ommuniation system, operating with main-lobe steering using four antenna-elements at the base station and a single antenna at the mobile station. We assume that the base station always direts the beam to the targeted mobile station (see Fig. 1). Beause eah neighboring base station serves multiple mobile stations simultaneously, the beam diretions differ for resoures assigned to different mobile stations. Therefore, the interferene a mobile station reeives depends on the resoure alloation of its neighbors. Our algorithm strives to minimize the interferene by optimizing this resoure alloation. Fig. 1: Priniple of oordinated beam forming.

2 Previous work has shown that large gains an be ahieved by joint optimization of the sheduling of multiple base stations [5], [6]. However, the algorithms investigated there are not fully distributed and are hard to implement in a real network. Our algorithm works fully distributed and without expliit ommuniation among the base stations. Furthermore, it does not require the mobiles to measure eah omponents of the interferene separately. Instead, the base station must only know the sum of the interferene power and the power of the desired signal measured by the mobile station. Our algorithm was developed for a 3GPP LTE system. An algorithm similar to the presented one is used in [7]. There, interferene in a 60GHz WPAN is mitigated by hanging the set of used resoures if the transmission is jammed by interferene. However, the system presented in that publiation does not use all resoures at the same time. In ontrast, our algorithm is applied in a reuse-1 senario with full transmit power in all ells. In [8], the authors desribe an interferene oordination algorithm for LTE networks. Their approah uses preoding matries restrited by a odebook. Eah mobile station measures the hannel to the neighbor ells and signals whih preodings the neighbor ells shall avoid. In ontrast to our algorithm their algorithm requires additional measurements by the mobile stations and more signaling between mobile stations and base stations as well as between base stations. The remainder of this doument is strutured as follows. The following setion presents the oordination algorithm. Setion 3 desribes the applied evaluation methodology. The oordination algorithm is evaluated as a funtion of several parameters in Setion 4. In addition, the influene of the variability of the traffi is investigated. Setion 5 onludes the paper. 2. Distributed Interferene Coordination Algorithm Fig. 2: Priniple of the oordination algorithm The overall goal of the presented algorithm is to improve the system s apaity by using beam forming and by direting the beams towards the targeted reeivers in a oordinated fashion. Steering the beams provides a degree of freedom, whih we exploit to minimize the overall interferene. Sine the geographial position of the mobiles varies over time, the beams have to be adjusted ontinuously. We assume non-greedy traffi, whih imposes additional boundary onditions. 2.1 Assumptions We assume that all base stations use the same partitioning of the available radio resoures into orthogonal units. We assume that it is not possible to measure the interferene relations between mobile stations diretly. This means that a mobile station an measure the sum of interferene it reeives, but it annot see how muh eah of its neighbors ontributes to this sum. We assume that the system is interferene limited and therefore thermal noise does not have a signifiant influene. We assume that eah mobile station an measure the interferene power as well as the SIR (signal to interferene ratio). Eah oordinated station reports these values to its base station. In typial systems this information an be derived from values that are measured and reported anyway: The CQI (hannel quality indiator) orresponds to the SIR. By ombining the CQI with the RSSI (signal strength indiator) the sum interferene an be obtained. We assume that these values are only measured and reported for the resoures on whih data was transmitted to this mobile station. For simpliity, we further assume that the same number of mobile stations is served in eah setor and eah mobile station gets the same amount of resoures. Therefore, we also have a defined and onstant operation point of eight ative mobile stations per setor. An extension to an arbitrary number of users is straight-forward. Sine the algorithm is designed for low measurement overhead, we assume that frequeny seletive sheduling is not used. We deal with logial hannels that are reated by frequeny diverse permutation of sub-arriers to even out fast fading hannel variations. 2.2 Key idea The algorithm tries to keep the resoure assignment over multiple frames as onstant as possible. In the example in Fig. 2, this is the ase from TTI1 to TTI3 where the resoure assignment does not hange. When the interferene situation beomes too bad, the resoure assignment is hanged. In the example in Fig. 2 this is the ase for the purple (third) mobile station after TTI3. Keeping the resoure assignment as onstant as possible is important to allow other base stations to assess the interferene that their mobile stations have to expet. In ontrast, improving the interferene situation is only possible by hanging the resoure assignment. Obviously, these two goals are onfliting, and an important part of the algorithm is the deision when to hange the resoure assignment. In the following frames, neighboring base stations an adapt to this deision. That way, the entire system strives to a better point of operation.

3 2.3 Algorithm To enable a base station to deide whether it should keep or hange the resoure assignment, eah mobile station measures the urrent hannel situation and reports it to its base station. The base station then evaluates the situation using a ost funtion. Based on the ost and a randomization funtion, the base station deides whether to hange the urrent resoure assignment or not. To hange the resoure assignment, another ative mobile station of the same setor is seleted randomly and the resoure assignment of both mobile stations is swapped. See algorithm 1 for a pseudo ode representation. The following paragraphs desribe the building bloks in detail. Interferene monitoring: For eah of its mobile stations the base station obtains a measurement of the urrent SIR and interferene sum. Sine the resoure assignment of this base station and its neighbors is kept as onstant as possible, this measurement onstitutes a good predition for the SINR and the interferene in the following TTIs. Cost Funtion: The ost funtion aims to desribe how good a given situation is for a mobile station. It is evaluated for eah mobile station and is a funtion of the logarithmi SINR γ (in db) and the interferene sum i (in dbw/hz). A lower value of the ost funtion ( γ, i) represents a better situation. Four different ost funtions have been evaluated and ompared: CONST : IFSUM : SIR : IF SIR : SIR ( γ, i) = 0 ( γ, i) ( γ, i) = i = γ ( γ, i) = i γ CONST IFSUM IF SIR The first option (CONST) is used as a referene, as it does not reat to the atual interferene situation but hanges a resoure assignment by random. Randomized resoure assignment: Instead of deiding whether to hange a resoure assignment by using a deterministi threshold, we apply a randomized threshold to avoid osillating swap sequenes. The higher the value of the ost funtion, the higher the probability to hange the urrent situation. A simple s-shaped urve s x is used to alulate the probability of hanging the resoure assignment. ( ) 0 2 x α δ 2 2δ s( x) = 2 α x + δ 1 2 2δ 1 x α δ α δ < x α α < x α + δ else Based on the alulated probability an unorrelated Bernoulli experiment takes the deision. The threshold value α and the width of the s-urve δ are parameters of the algorithm. Their influene is evaluated in setion 4. Algorithm 1: Interferene mitigation algorithm 1: for all TTI do 2: for all mobile station do 3: Obtain reports of previous TTI (γ and i) 4: Calulate ost funtion: (γ,i) and hange probability p := s() 5: if (random value u(0,1) < p) then 6: Choose pivot partition and perform hange 7: end if 8: Shedule to assigned resoure partition 9: end for 10: end for (1) (2) 3. Evaluation Methodology The algorithm has been evaluated by system level simulation of an LTE-like system. The simulator is based on the IKR SimLib [9]. In this setion, we desribe the simulation model and the methods used to evaluate the algorithms. Our senario onsists of 19 sites and three setors per site, plaed on regular hexagons. To avoid border effets, wrap-around is used. The mobile stations are distributed uniformly over the whole senario and move with a onstant speed of 10km/h. They use a single isotropi antenna without antenna gain. Eah mobile is served by the base station with highest signal strength. Data is transmitted in downlink diretion by a sheduler as desribed in setion 2.3. Uplink is not taken into aount and is assumed to work without delay or paket loss. The hannel model onsists of a pathloss model aording to [10] and of a Gudmundson slow fading model as

4 in [11]. Table 1 gives the parameters of the slow fading model. We do not model fast fading, beause the algorithm is assumed to be used with a frequeny diverse permutation of the systems resoures. Eah setor is equipped with a main-lobe steering beamformer, onsisting of a linear array of 4 setor antennas. These are modeled by a gain pattern derived from measurements of a real setup as desribed in [12], assuming pure line-ofsight for the multi-antenna effets. For eah resoure blok, the main-lobe of the antenna array is pointed at the reeiving mobile station. The auray of the beam is 1 and ideal traking of the mobiles is assumed. Table 1: <tab=1>simulation parameters. To model the variane of the traffi whih influenes the performane of the algorithm, we use an on-off traffi model. During the ON phase, we assume that a mobile station always has data to send. The mean length of the ON phase is set to 0.2s if not stated otherwise. The total number of mobile stations per setor depends on their movement and is 1000/57=17.5 in average. For simplifiation, we assume that eah setor always ontains eight ative mobile stations. Our traffi model is designed suh that the number of ative mobile stations remains onstant. This means, that if the ON phase of one mobile station in a ell ends, one of the previously inative mobile stations will be swithed on. In addition, if there are already eight ative users in a ell and another mobile station handovers to this ell, one of the ative mobile stations is swithed off. Beause of the slow movement of the mobile stations, the influene of the additional swithing aused by handovers is small. For the alulation of the hannel quality (SINR) and the spetral effiieny, the system bandwidth of 10MHz is divided into eight equally sized sub-bands. The sheduler assigns one of these alloation units to eah mobile station. The SINR value is alulated for eah alloation unit aording to the hannel model and the sheduling of the other base stations. Based on the SINR, a transport format is seleted from a set of 27 LTE transport formats, suh that the blok error probability does not exeed 10%. Blok errors are modeled by BLER tables. Retransmissions, as usually implemented by an automati repeat request (ARQ) protool, are not modeled. The overhead of three OFDM symbols for signaling, the ell speifi referene symbols for a single antenna port, and the terminal speifi referene symbols is subtrated from the apaity of eah transmission. For eah ative mobile station, the throughput measurements sum up the apaity of the suessfully deoded transmissions over an interval of 10ms. The 5% quantile of the umulative density funtion (CDF) of these spetral effiieny values is used to evaluate the ell edge throughput. For eah parameterization, 30 independent drops are simulated with 4000 frames per drop. The first half of eah drop is ignored to get rid of transient effets. 4. Performane Evaluation Results 4.1 Influene of different ost funtions Choosing a suitable ost funtion to determine whether to hange the resoure assignment has a high influene on the performane of the oordination algorithm. Fig. 3 shows the spetral effiieny of the ell edge and the entire setor for the ost funtions desribed in setion 2.3. As a referene, the ost funtion CONST shows the performane of pure random behavior. As expeted, there is no oordination gain in this ase. The algorithm ahieves the highest oordination gain when using the IF-SIR ost funtion, whih is about 0.7dB in SINR. Using either the IFSUM or the SIR ost funtion provides a lower oordination gain. Fig. 3: Setor throughput and edge throughput for different ost funtions and α parameter. 4.2 Influene of randomization The presented algorithm employs a randomization of the swap deision to prevent staying in loal minima. The parameters α and δ determine the shape of the s-urve and, thus, the randomization proess. α determines the ost threshold while δ determines the sensitivity to ost value hanges. Fig. 4 shows the average SINR as a funtion of α and δ. If δ is too low, the s-urve has a narrow transition range and therefore the adjustment of α must be very preise, making the algorithm less robust in pratie. A too large δ redues the oordination gain and makes the algorithm slower for varying traffi. We hose δ = 20dB as a good ompromise. One an observe that the oordination algorithm ahieves good results for multiple ombinations of α and δ. All these ombinations have in ommon that the average swap probability is at 2-5%. Fig. 4: SINR for different widths of the s-urve δ. 4.3 Influene of On-Off-traffi Finding a sheduling deision in a deentralized fashion takes time. Thus, this oordination algorithm is sensitive to traffi flutuations. Fig. 5 shows the dependene of the SINR gain to different ON phase durations. One an observe that larger ON times allow for higher oordination gains. The time it takes the oordination algorithm to settle an be estimated from the frame duration of 1ms and the average swap probability of 2-5% at best operation points. One iteration is in the order of magnitude of 20-50ms. Assuming that reahing a good resoure assignment takes several iterations, the algorithm needs several hundred ms to settle. Fig. 6 shows the

5 autoorrelation funtion of the points in time of the swap deision and supports that assumption, sine most energy is loated in the area from 0-1s. In the autoorrelation plot, 1 is the average swap probability. A value bigger than 1 means, that n frames after a swap the hane for an additional swap is higher than average and vie versa. Fig. 5: Fig. 6: SINR for different mean ON times of the ON/OFF traffi model. Effet of the α parameter on onvergene behavior. 5. Conlusion We presented a distributed interferene oordination algorithm whih works without expliit ommuniation among the base stations. The oordination is performed impliitly by the use of hannel measurements by the mobile stations. Nevertheless, our algorithm does not require additional measurements besides those already available in typial systems. The algorithm tries to keep the resoure assignment onstant as long as it is reasonable, whih improves the hannel estimation quality. Interferene oordination approahes that assume to have global knowledge about interferene relations in [13] an ahieve 60% gain in ell throughput and 165% gain in setor edge throughput. Compared to this lass of approahes, the ahievable gains of the presented interferene oordination algorithm is lower (5.7% gain in ell throughput, around 10% gain in edge throughput). Yet, it requires muh less of the knowledge, measurement, and reporting overhead and thus is implementable in real-world systems. Also, the given throughput gains are evaluated with a non-greedy traffi assumption. An additional performane gain ould be ahieved by using ost funtions whih take into aount the maximum possible hannel quality for a mobile station in the given situation. A low amount of signaling between the base stations ould inrease the onvergene speed. Investigating these possibilities will be subjet to further studies. We investigated the algorithm in a senario where the transmissions of all users are oordinated and all available resoures are used. However, the proposed resoure assignment sheme annot be ombined with other sheduling tehniques, suh as frequeny seletive sheduling. The ideal sheme with whih a mobile an be served depends, besides others, on the speed of the mobile and the variability of its traffi. In a real world appliation, only a part of the resoures would be oordinated with the proposed algorithm and other users ould be served on the remaining resoures without oordination. 6. Referenes [1] S. Halpern, Reuse partitioning in ellular systems, in Vehiular Tehnology Conferene, rd IEEE, vol. 33, 1983, pp [2] A. Stolyar and H. Viswanathan, Self-organizing dynami frational frequeny reuse for best-effort traffi through distributed inter-ell oordination, in INFOCOM The 28th Conferene on Computer Communiations. IEEE, 2009, pp [3] G. Boudreau, J. Paniker, N. Guo, R. Chang, N. Wang, and S. Vrzi, Interferene oordination and anellation for 4G networks - [LTE art II: 3GPP release 8], Communiations Magazine, IEEE, vol. 47, no. 4, pp , [4] 3GPP, TR further advanements for E-UTRA physial layer aspets (release 9), v1.0.0, [5] M. Neker, Interferene oordination in ellular OFDMA networks, Network, IEEE, vol. 22, no. 6, pp , [6] M. Neker, A graph-based sheme for distributed interferene oordination in ellular OFDMA networks, in Vehiular Tehnology Conferene, VTC Spring IEEE, 2008, pp [7] M. Park, P. Gopalakrishnan, and R. Roberts, Interferene mitigation tehniques in 60 GHz wireless networks, Communiations Magazine, IEEE, vol. 47, no. 12, pp , [8] L. Liu, J. Zhang, J. Yu, and J. Lee, Inter-ell interferene oordination through limited feedbak, submitted to International Journal of Digital Multimedia Broadasting, [9] Institute of Communiation Networks and Computer Engineering, IKR Simulation Library, Stuttgart, Germany, [10] E. Damosso and L. Correia, COST 231 (Digital mobile radio towards future generation systems), Final report, [11] ETSI, TR , UMTS 30.03, V3.2.0, [12] M. Neker, Towards Frequeny Reuse 1 Cellular FDM/TDM Systems, in Proeedings of the 9th ACM/IEEE International Symposium on Modeling, Analysis and Simulation of Wireless and Mobile Systems (MSWiM 2006), Otober [13] M. Neker, A Novel Algorithm for Distributed Dynami Interferene Coordination in Cellular OFDMA Networks - Communiation Networks and Computer Engineering Report No. 101, Ph.D. dissertation, Universität Stuttgart, First Author Matthias Kashub Universität Stuttgart, Institute of Communiation Networks and Computer Engineering Stuttgart matthias.kashub@ikr.uni-stuttgart.de Other authors Christian M. Blankenhorn, Christian M. Mueller and Thomas Werthmann

6 Universität Stuttgart, Institute of Communiation Networks and Computer Engineering Stuttgart. (Reeived on April 11, 2007) (Revised on June 12, 2007) Table 1: <tab=1>simulation parameters. Frame duration 1ms System bandwidth 10MHz Transmit power 46dBm Inter-site distane 1000m Mobility model Random Diretion Mobile speed 10 km/h Mobile users 1000 Mobile users 8 Pathloss model COST 231 modified Hata [10] Shadowing model time-orrelated Gudmundson (σ=8db, inter site orr.=0.5, orr. dist.=50m) Fig. 1: Fig. 2: Fig. 3: Fig. 4: Fig. 5: Fig. 6: