A Hybrid Clustering Approach in Coordinated Multi-Point Transmission System

Transcription

1 2012 7th International ICST Conference on Communications and Networing in China (CHINACOM) A Hybrid Clustering Approach in Coordinated Multi-Point Transmission System Cui Zeng, Pinyi Ren, Chao Zhang and Gangming Lv School of Electronic and Information Engineering Xi an Jiaotong University, Xi an, China zengcui xjtu@126.com, {pyren, chaozhang and gmlv }@mail.xjtu.edu.cn, Abstract Coordinated Multi-Point transmission technology which can improve cell edge spectral efficiency and meet the demand of high-speed transmission is one of the ey technologies of LTE-Advanced. In this wor, we propose a hybrid cell clustering approach in coordinated multi-point transmission system which is more efficient and practical. First, we divide a cell into three sectors: the intra-cell sector, the intra-cluster sector and the intercluster sector. Users in these three sectors are designed to utilize the non-coordinated approach, the static coordinated approach, and the dynamic coordinated approach, respectively. Second, we propose a method to identify the user s type in order to choose the best coordinated approach. Finally, we give a resource scheduling method to suit the novel clustering approach. Simulations show that, compared with static clustering and dynamic clustering, our proposed hybrid clustering algorithm can achieve the highest cell edge spectral efficiency. Index Terms Coordinated Multi-Point transmission; hybrid cell clustering approach; intra-cell sector; intra-cluster sector; inter-cluster sector. I. INTRODUCTION In order to deal with the conflict between the growing demand and limited frequency resource, full frequency reuse (FFR) has become the consensus of the industry[1]. Based on FFR, the system operates in interference-limited regime. The interference received by mobile terminal mainly comes from the neighboring cells, especially the users located in the edge of the cell. The signal to interference-plus-noise ratio (SINR) of the cell edge users is far lower than that of the central users, which reduces the cell s average spectrum efficiency. In view of this problem, Coordinated Multi-Point (CoMP) transmission technology was proposed[2]. CoMP can improve the cell edge users spectral efficiency and meet the demand for high speed wireless data communications. The core of CoMP technology is that independent cells share data and relevant information with each other. By joint processing or coordinated scheduling, the interference can be reduced or even turned into beneficial factors [3]. Based on the information sharing model and the coupling degree of cells, CoMP can be divided into two categories, namely joint processing and coordinated scheduling[3]. Joint processing requires data sharing among base stations, and channel information interaction between users and base station, which brings a lot of feedbac consumption and greatly increase system overhead [4,5]. Ideal base station coordination is proposed in[6,7]. Full coordination is very difficult to realize, especially for large-scale networs. Therefore, in practical systems, limited number of base stations forming clustering coordination networ should be considered [8]. Popular clustering methods include static clustering, semistatic clustering and dynamic clustering[9,10,11]. Static clustering coordination generally selects adjacent base stations as coordination base stations and does not change with time, so scheduling is low complexity. Users do not need feedbac realtime CSI to base station to cluster. Since it constrains fixed base stations to cooperate, fully distributed macro-diversity cannot be achieved [12]. For dynamic clustering, users adapt to their own environments and need to choose coordinated base station during each time slot [13]. Dynamic clustering meets the real-time needs of the users, and turns the strongest interference into useful signal, which greatly improves the users throughput. Since clusters change dynamically, the complexity of scheduling is high and the feedbac overhead is large [14]. Forming cluster reduce the feedbac overhead of the system, can reduce the complexity of the code design and also reduce the burden of bachaul. On the other hand, however, clustering maes a part of signals to interference rather than useful signal, what s more, unreasonable clustering mechanisms may lead a strong signal into a strong interference. In addition, clustering may cause performance deterioration for the users located in the edge of the cluster, and lead to the unfairness among users. So we need a good clustering strategy to wea these adverse effects and search for an optimal compromise between the design complexity and performance. Our goal is to propose a compromising cooperation approach that maximizes the capability of CoMP and lower the complexity of the system as well. Based on the analysis of the disadvantages of the above strategies, we proposed a hybrid clustering, which has lower scheduling complexity and smaller feedbac overhead but maximizes the capability. First, we divide a cell into three sectors: the intra-cell sector, the intra-cluster sector and the inter-cluster sector. Users in these three sectors are designed to utilize the non-coordinated approach, the static coordinated approach, and the dynamic coordinated approach, respectively. Simulations show that the proposed hybrid clustering approach can approximately get the highest cell edge spectral efficiency. The rest of the paper is organized as follows: In Section II, we give the system model. Section III presents the hybrid clustering approach. Numerical results are presented in Section IV and conclusions are drawn in Section V /12/$ IEEE

2 II. SYSTEM MODEL Consider a cellular MIMO networ consisting of N base stations and K users. Each BS has N t antennas and each user has N r antennas. Each BS chooses only one user of each time-slot t for downlin transmission. The number of data streams to user is L and the data streams are denoted by L 1 vector s, the all networ transmission data streams are denoted by s = [s T 1,...,,s T,...,sT K ]T. In the cluster C of user, there are B (B<N) BSs. The system model of coordinated multi-point transmission system is illustrated in Fig 1. User feedbac CSI to local BS (white), three BSs service the user at the same time (blue), and they share the CSI and the transmission data through the bachaul. Assume user is located in the BS b, we call it local station, the cluster of user is C, the pre-coding matrix is denoted by N t L matrix W b. The received signal of user is given by y =H b Ws b + H i Ws i + H j Wj s +z (1) i C,i b j c, c C where H b is the N r N t channel matrix from BS b to user, z CN(0, I) is additive white Gaussian noise. Superscript denotes BS and subscript denotes user. The first part of (1) denote signal comes from the local BS, the second part denote signal comes from other BSs in the cluster C, and the third part denote signal comes from the other clusters. When non-coordinated, SINR is given by SINR NC = i C,i =b W i Hi 2 + W b Hb 2 j c, c =C, = W j Hj 2 +N o (2) When coordinated, interference coming from cluster becomes useful signal, SINR is given by: W b H b 2 + i C SINR Co =,i b W i H i 2 j c, c C, W j H j 2 (3) + N o Rate of user can be expressed by R =log 2 (1 + SINR Co ) (4) The optimization goal is to maximize the throughput through the clustering: A. Cell division max C SINR Co (5) III. HYBRID CLUSTERING APPROACH The existing researches show that, the networ is in interference-limited when there is full frequency reuse, especially to the edge users, so moderate reducing interference can significantly improve the edge users performance. They introduced coordinated multi-point transmission in LTE-A. As show in Fig 2, the 3 adjacent cells form a static cluster and coordinated. Therefore the main components of interference Fig. 1. System model of coordinated multi-point transmission systems. can be divided into two types: (1) caused by adjacent cells in the static cluster, (2) caused by adjacent clusters. This paper does different treatment of the two types of interference. Based on the above ideas, we divide cell into three sectors, as show in Fig 2 the intra-cell sector (white part), the intracluster sector (blue part) and the inter-cluster sector (red part). In the intra-cell sector, the user has higher SINR. So the users use the traditional transmission directly serviced by the local cell. In the intra-cluster sector, the user has low SINR, and the interference mainly caused by the adjacent cells in the cluster. So we use the static clustering approach. In the intercluster sector, the interference is mainly caused by the adjacent clusters. The user has low SINR and it is hard to improve performance through static clustering approach. So we use the dynamic clustering approach. B. The judgment of the user s type The users have three types: intra-cell users, intra-cluster users, and inter-cluster users. Before transmitting, the system needs to first determine the type of the user, in order to select the best transmission scheme. The type is determined by the users themselves, and then feedbac to the base station. The power received from the local cell i of the user is P local (i) =p i α 2 (, i) (6) ( The power ) received from the coordinated cell j j C static of the user is (j) =p j α 2 (, j) (7) The power received from the cell m in other clusters of the user is (m) =p m α 2 (, m) (8) where p n (n = i, j, m) is the transmission power, α 2 (, n) (n = i, j, m) is the fading factor. The power received from the static cluster cells (including local cell) is = j C static (j) (9) 628

3 h =[h s1 h s2 h s3 ]. Assume ṽ is the right singular vector corresponding to the largest singular value of h. Then the global pre-coding matrix is: [v T s 1 v T s 2 v T s 3 ] T =ṽ (12) Which vs T 1,vT s 2,vT s 3 are the pre-coding matrix of the cell s 1,s 2,s 3 respectively. 3) If the user is an inter-cluster user, uses the dynamic clustering scheme. Details as follows: Fig. 2. The model of the cell division. The power received from the all cells in other clusters is = j B C static When Non-CoMP, SINR is SINR = P local (i) (j) (10) + + N o (11) where N o is noise power. The first step: determination of the cell center/edge user The user is mainly accorded to the received power of the pilot signal to determine the type. If satisfied SINR < SINR edge, the user is a cell edge user; otherwise, the user is an intra-cell user, which SINR edge is the threshold. The second step: determination of the intra-cluster/intercluster user When the user is determined to a cell edge user, continue to be determined. If P co <Pnon co, the user is an intercluster user; otherwise, is an intra-cluster user. Conclusion: if SINR >SINR edge, user is an intra-cell user; if SINR <SINR edge intra-cluster user; if SINR <SINR edge inter-cluster user. and and >, user is an <, user is an C. Hybrid clustering method We propose a hybrid clustering method. Different types of users use different clustering strategy. As follows: Three adjacent cells compose a static cluster, the entire networ is divided into a number of static clusters. First the user calculates the received power of coordinated cells and non- coordinated cells, P co and respectively. 1) According to the above rules of the division, if the user is an intra-cell user, uses the non-coordinated transmission scheme. The user is served by the local cell. 2) If the user is an intra-cluster user, uses the static clustering scheme. The three adjacent cells compose the static cluster. For static clustering scheme, uses the joint processing based the global pre-coding. The cluster of user is C = {s 1,s 2,s 3 }, which s 1 is the local cell, s 2,s 3 are the other coordinated cells. The three cells and the user form a virtual MIMO channel Compute the received power from all cells P j (j B) and sort. Choose the first Y (m) cells with maximum received power to be alternative cells. j =argmaxp j j (13) Choose two cells from the Y (m) cells to coordinate with the local cell. Literate over all the possible clustering combination (total CY 2 () inds), calculate SINR i,j Co of each ind, SINR i,j Co = W b Hb 2 + W i Hi 2 + W j Hj 2 (14) W j Hj 2 +N o j Λ c, Λ c =c, = Choose the maximum one (i,j ) as coordinated cluster. (i,j ) = arg max i,j Y () SINRi,j Co (15) (i,j ) are just the coordinated cells of the inter-cell user. 4) According to the type of the cluster and the coordinated cells, local cell send coordinated request to the coordinated cells. Mae certain the coordination, local cell share the channel information and transmission data to the coordinated cells. D. Scheduling algorithm 1) There are N RBs. And there is a user set {U i } in any RB i. I s (i) is the resource allocation indicator factor of the cell s in RB i. ifi s (i) =1,indicating that the resource i of cell s has been allocated. And if I s (i) = 0, indicating that the resource i of cell s has not been allocated. The maximum rate of user in channel n is r n (i), the historical rate of user is T. Selecting the highest priority user in the resource bloc i, which is determined by: r n (i) =arg max (16) {U } T After a resource bloc been allocated, need to update the historical throughput T = T + r n (i). In order to ensure orthogonal in the cell, each resource bloc can only be assigned to a user in the same cell. 2) The cluster of user is {s 1,s 2,s 3 }, if i {1,2,3} I s i (i) 1, indicates that in the coordination 629

4 TABLE I SYSTEM-LEVEL SIMULATION PARAMETER SETTINGS Parameter Setting Channel model SCM Simulation scene Urban macro-cell Number of cells 57 Number of MSs per cell 10 bandwidth 10MHz Carrier frequency 2.5GHz Sub-carriers interval 15Hz BS antenna structure N t =4 MS antenna structure N r =2 MIMO mode SU-MIMO ( single stream) Channel estimation error Ideal estimation Receiving mode MMSE Lin to system mapping MIESM Service type Full buffer CDF(%) Static CoMP Dynamic CoMP 10 Performance of Proposed cell edge user Normalized Cell Average Spectrum Efficiency(bps) cluster cells, some cell resource have been allocated. The user scheduling failed, then update the user set{u i } {U i } = {U i } (17) 3) If the scheduling of user is successful, let I si (i) = 1,i {1, 2, 3}, {U i } = {U i } 4) Follow by cycle, until all users are scheduled, that is {U i } =. 5) If {U i } =, update the resource bloc. Followed by cycle, until all resource blocs are scheduled. IV. SIMULATION RESULTS In this section, we give the simulation results of the proposed algorithm. As comparison, two other algorithms are considered: Static-CoMP The local cell and the adjacent two cells compose the coordinated cells. And use the joint processing technology of the global pre-coding. Dynamic-CoMP The coordinated user selects the optimal two cells in all cells to compose the coordinated cells. And use the joint processing technology of the global pre-coding. Simulation settings follow the ITU M.2135 standard. The simulation system was consisted of 57 cells, each cell contains 10 users and users are distributed in the cell with a uniform random distribution. Wraparound technology was used to eliminate the edge effect. The threshold which to determine the cell intra or edge users is determined based on the proportion of the coordinated users. We assumed the proportion of the coordinated user is 20%. The related simulation parameter settings are shown in table 1. A. The performance of different coordinated solutions Fig 3 gives the CDF of normalized cell average spectrum efficiency in urban macro cell scenario. And table 2 gives the performance of different coordinated solutions. It can be seen from the figure and the table that the performance of the proposed scheme is the best, the dynamic CoMP is second Fig. 3. CDF of normalized cell average spectrum efficiency. TABLE II THE PERFORMANCE OF DIFFERENT COORDINATED SOLUTIONS The type of Cell average Gain Cell edge Gain transmission spectral user spectral efficiency efficiency (bps/hz) (bps/hz) Static- CoMP % % Dynamic- CoMP % % Proposed % % and the static CoMP is the worst. The performance of the dynamic CoMP and the proposed scheme are very similar. But regardless of the cell average spectrum efficiency or the cell edge spectral efficiency, the performance of the proposed scheme is better than the dynamic CoMP. Figure 4 and figure 5 show the cell average spectral efficiency histogram and the cell edge user spectral efficiency histogram in the different coordinated scheme, respectively. We can see from that when static CoMP, cell average spectral efficiency is 1.87 bps/hz, and cell edge spectral efficiency is bps/hz. Compared to static CoMP, dynamic CoMP not only improve 8% of the cell average spectral efficiency but also 34.5% of the cell edge spectral efficiency. The performance of the proposed scheme is the best. It gets the cell average spectral efficiency gain of 13% and the cell edge spectral efficiency gain of 35%. The proposed scheme is outperformed from the two other schemes due to users selecting the best transmission scheme. B. The influence of different threshold According the threshold SINR edge, divided users into intracell user and cell edge user. When SINR >SINR edge, the user is an intra-cell user, and uses Non-CoMP scheme. When SINR <SINR edge, the user is a cell edge user, and uses CoMP scheme. Threshold will affect the number of the coordinated users and will also affect the cell performance. 630

5 Fig. 4. Cell average spectral efficiency histogram. Fig. 5. Cell edge user spectrum efficiency histogram. Fig 6 shows the cell average spectrum efficiency of the proposed scheme in different thresholds. The threshold is smaller, the cell average spectrum efficiency is greater. This is because the threshold is larger, more resources are assigned to the coordinated users, less resources are assigned to the non-coordinated user, so the cell average spectrum efficiency will reduce. The threshold is smaller, the cell edge spectrum efficiency is greater. This is because the threshold is larger, coordinated users are more. And the chance to be scheduled is smaller. So increase the threshold will not improve the throughput of the cell edge users. V. CONCLUSION We propose a hybrid clustering approach in coordinated multi-point transmission system. For the proposed approach, the users location information is taen into consideration, and users with different locations utilize different coordinated schemes.three schemes, namely Non-comp, static-comp, and dynamic-comp, are applied to intra-cell users, intra-cluster users, and inter-cell users, respectively. Simulations show that, that our scheme cam improve both the cell average spectral efficiency and the cell edge spectral efficiency. ACKNOWLEDGMENT This wor was supported in part by the National Natural Science Foundation of China ( ),the Nation- CDF(%) %coordinated user 10 20%coordinated user 30%coordinated user Normalized Cell Average Spectrum Efficiency(bps) Fig. 6. Cell average spectrum efficiency in different threshold. al Science and Technology Major Project(2012ZX ), and the National High-Tech Development Program (2011AA01A105). REFERENCES [1] TR v1.0.0, Further Advancement of E-UTRA, Physical layer aspects, 3GPP Technical Report. [2] M. Karaayali, G. Foschini and R. Valenzuela, Networ Coordination for spectrally efficient Communications in Cellular Systems, in proc.of IEEE Wireless Communications, Volume:13, Issue:4, pp.56-61, Aug [3] 3GPP TR , Further Advancements for E-UTRA Physical Layer Aspects(Release 9), v.1.5.2, January Sep [4] G. J. Foschini, K. Karaayali, and R. Valenzuela, Coordinating multiple antenna cellular networs to achieve enormous spectral efficiency, in proc.of IEEE Communications, Vol. 153, Issue:4, pp , August [5] Stefan Bruec, Lu Zhao, Jochen Giese and M. Awais Amin, Centralized Scheduling for Joint Transmission Coordinated Multi-Point in LTE- Advanced, in proc.of IEEE WSA, Bremen,pp , Feb [6] Someh, O. Simeone, Y. Bar Ness, A. M. Haimovich. Distributed multicell zero-forcing beamforming in cellular downlin channels, in proc.of IEEE Global Telecommunications Conference, vol.27, Dec 2006, pp.1-6. [7] Jing Shen, N. C. Tse David, B. Soriaga Joseph, Hou Jilei, E. Smee John, Padovani Roberto. Downlin macro-diversity in cellular networs, in proc.of IEEE International Symposium on Information Theory, pp.1-5, [8] Linan Sun, Zhongzhao Zhang, Xuejun Sha, and Weilin Jiang. Improved static clustering base station coordination, in proc.of CCWMC, shanghai, China, pp , Dec [9] 3GPP TR R , Clustering for CoMP Transmission, Jan. 12-Jan [10] 3GPP TR R , Setuo of CoMP Cooperation areas, January [11] 3GPP TR R , Cell Clustering for CoMP Transmission/Reception, Feb. 9-Feb [12] Fan Huang, Yafeng Wang, Jian Geng, Mei Wu, Dacheng Yang, Clustering Approach in Coordinated Multi point Transmission/Reception System, in proc.of IEEE VTCF, pp.1-5, [13] Shanghui Xiao, Zhongpei Zhang, Qinmin Wang, Zhiping Shi, Coordinated Multi-point Transmission Systems with Dynamical Cell Clustering Strategies, in proc.of IEEE CHINACOM, pp.1-5, [14] Papadogiannis.A,Gesbert.D,Hardouin.E, R D Div, A Dynamic Clustering Approach in Wireless Networs with Multi-Cell Cooperative Processing, in proc.of IEEE ICC, pp ,