Practical Performance Evaluation of Coordinated Multi-Point (CoMP) Networks Application to Saudi Arabia.pdf

Proceedings of the 8th IEEE GCC Conference and Exhibition, Muscat, Oman, 1-4 February, 2015

Practical Performance Evaluation of Coordinated Multi-Point (CoMP) Networks Application to Saudi Arabia Ali H. Muqaibel

Ayham N. Jadallah

Electrical Engineering Department King Fahd University of Petroleum & Minerals (KFUPM) Dhahran 31261, Saudi Arabia

Operation and Maintenance Department Mobily Al-Khobar, Saudi Arabia [email protected]

[email protected] Abstract—In Coordinated Multi-Point (CoMP) networks, Base Stations (BSs) cooperatively process User Equipment (UE) connected to multi-points to eliminate the inter-cell interference. CoMP networks convert the interference signal into useful information by controlling the interfering signals among adjacent cells. This is important for UEs at the cell edge. This paper examines practical deployment of CoMP and investigates its enhancement to the spectral efficiency and Signal-to-Interference and Noise Ratio (SINR) in practical inter-site and intra-site CoMP scenarios. We consider the topography and terrain data of AlKhobar city in Saudi Arabia to evaluate the received power and SINR behavior and estimate the SINR improvement in practical CoMP networks. Keywords—clustering; CoMP; Coordinated Multipoint.

I. INTRODUCTION Demands for mobile communications services and higher data rates are increasing day by day, which is a serious concern for network operators. Research and development departments in telecom operators and vendors have been working on finding higher spectral efficiency network solutions to meet the huge data traffic and required quality of service. Frequency reuse is one of the solutions to satisfy high data rate demands and at the same time reduces the interference, which limits the performance of cellular systems. Even with frequency reuse, other problems persist like the inter-cell interference which limits the cell edge User Equipment (UE) data rates. Coordinated Multi-Point (CoMP) is an advanced wireless mechanism in mobile communication transmission and reception which proposes a better solution to overcome the inter-cell interference and enhance the cell edge data rates [1]. In CoMP technology, Base Stations (BSs) connect the UE to multipoint in a coordinated way to cooperatively eliminate the inter-cell interference. CoMP transmission enhances the network average data rate and increases the spectral efficiency [2]. Coordinated BSs exchange information about the received signals from the served UE. This information is needed to perform multi-cell joint signal processing [3]. The UE can be served cooperatively from more than one BS which enhances The authors acknowledge the support by King Fahd University of Petroleum & Minerals (KFUPM)

978-4799-8422-0/15/$31.00©2015 IEEE

the performance at the cost of additional overhead data bits and huge backhaul infrastructure [1]. The network is enabled to transmit in a cooperating way to eliminate inter-cell interference under ideal conditions of perfect time and frequency synchronization, as well as short delay for exchanging data and Channel State Information (CSI) [4]. Yong Cheng and others proposed an optimal approach to tradeoff between the gain and the overhead of CoMP systems [5]. Each group of coordinated antennas in CoMP forms a cluster which jointly serves a group of UEs. In general, the number of the cluster’s coordinated cells is limited due to synchronization factors, signal overhead and the availability of extended fiber connection between BSs [6]. Clustering can be intra-site or inter-site. An inter-site cluster is formed across sites. Intra-site CoMP is desirable because it does not incur extra backhaul data exchange [7]. If a sector is participating in more than one cluster then we have overlapping clusters otherwise the clustering is non-overlapping. Choosing the coordination cluster antennas depends on many factors such as distance between antennas, transmitted power level, UEs locations and number of coordinated antennas per cluster. The optimum selection of cluster cells is reflected on the cellular systems spectral efficiency improvement. Clustering can also be categorized into static and dynamic clustering. Cells cluster distribution in static clustering is selected one time according to pre-defined deployments that uses the site location and site radio frequency parameters as main characteristics in forming fixed clusters. Static clustering requires little signals overhead and less complex compared to dynamic clustering [8]. A major disadvantage for a given UE with static cell clustering mechanism occurs when the cells that have the strongest link gain may not lie in the same cluster [9]. These cells would then cause strong interference. Dynamic clustering can solve these issues. March et al. in [10] discussed ideal and practical clustering layouts to show how static clustering can yield a performance close to the ideal clustering. SINR is one of the main aspects that should be optimized in any wireless communication system. Micheal Grieger et al.

Proceedings of the 8th IEEE GCC Conference and Exhibition, Muscat, Oman, 1-4 February, 2015 evaluated the system performance in terms of the SINR of the equalized transmit signals [11]. In [12] the authors evaluated the SINR performance in LTE-Advanced system at UEs for two different intra-cell CoMP schemes. Many publications have shown how CoMP mitigates interference and improves both the data rate and the spectral efficiency in ideal networks. Patrick Marsch, Gerhard Fettweis and Michael Grieger considered practical CoMP field trials [1], [2], [8], [11], and [6]. There is definitely a need for more practical research with different topologies. This paper examines clustering in practical deployment of CoMP. It investigates how CoMP communications can enhance the spectral efficiency and SINR in inter-site and intra-site CoMP scenarios. We consider real city topography and terrain data to simulate the received power and SINR behavior and estimate the SINR improvement. This paper is organized as follows. In section II, we present the ideal and the used practical system models. Performance analysis is conducted using SINR as main performance criteria. In Section III, clustering is addressed for the ideal scenario. The impact of the cluster size is examined. In Section IV, Al-Khobar network is evaluated in the presence of inter-site, and intra-site cooperation. The paper is concluded in Section V. II. IDEAL AND PRACTICAL SYSTEM MODELS This part presents two different system models. In the ideal system model, we address some important parameters such as path-loss, antenna pattern and SINR; while we focus on the practical aspects in the real system model such as BSs information, terrain data and different configuration parameters for antennas and UEs. This is the information needed to evaluate the SINR for comparison between coordinated and non-coordinated scenarios. Ideal models are simulated based on the hexagonal setup using Matlab® while the practical network is simulated using ray tracing simulation tool called WinProp ®. Both setups consist of a set of BSs/sites, each one has three directive antennas/sectors and each directive antenna basically serves one cell through the entire network. A. Ideal Network Model In the considered ideal network layout, each BS serves three hexagonal cells using three directive antennas. Each antenna covers one cell area, and the azimuth for each antenna is beaming 120 degrees out of phase from the neighboring antenna in the same BS. We considered three hexagonal tiers network which have 37 BSs, and each one consists of three different sectors making a total number of 111 sectors. Fig. 1 shows the ideal network setup where each unique color refers to one BS that consists of three cells. Following 3rd Generation Partnership Project (3GPP) standards, we fix the Inter Site Distance (ISD) to 500 m as we are considering an urban area with macro BSs. Serving cell selection criteria depends on UE downlink received power. Hence, the UE will be served from the sector which supports this UE with the highest Received Signal Level (RSL).

Fig. 1. Ideal network layout

Path-loss is affected by the antenna gain, atmospheric conditions and multipath effects. We consider a flat-plane pathloss, 𝑃𝐿, model [10]: 𝑃𝐿 = 130.5 + 37.6 log10 (

𝑑 𝑘𝑚

)

(1)

[dB]

where 𝑑 refers to the distance between the UE and the BS. The path-loss exponent in equation (1) equals to 3.76 assuming shadowed urban areas. The antenna pattern depends on the angle, 𝜃, between the antenna azimuth and UE, the front-to-back ratio of the antenna and the maximum possible attenuation (𝐴𝑚), and the 3dB main lobe beam-width, 𝜃3𝑑𝐵 . The antenna pattern equation can be defined as in [10]: 𝐴𝐿(𝜃) = 𝑚𝑖𝑛(12 |

𝜃 𝜃3𝑑𝐵

2

| , 𝐴𝑚)

[dB]

(2)

The number of sectors per BS is affected by the values of 𝜃3𝑑𝐵 and 𝐴𝑚. In our case, each BS consists of 3 sectors with an antenna gain equals to 14 dB𝑖 where 𝜃3𝑑𝐵 and Am are 70 degrees and 20 dB respectively. Each sector has directional antennas with 120-degree beam-width. Cellular communication systems generate various types of interference such as inter-cell and co-channel. CoMP technique improves the UEs SINR values and reduces the interference at cell edge, where appropriate network optimization that care about choosing the coordinated sectors in CoMP systems can enhance the network efficiently. We are considering 𝐽 UEs and 𝑀 cells. The experienced SINR by a UE 𝑗 from the serving cluster 𝑀` is formulated as: ̀

𝑆𝐼𝑁𝑅𝑗𝑀 =

𝑚 𝑃 ∑𝑚∈𝑀 ̀ 𝜆𝑗 𝑚 2 𝑃 ∑𝑚∈{𝑀/𝑀 ̀ } 𝜆𝑗 +𝜎

(3)

Proceedings of the 8th IEEE GCC Conference and Exhibition, Muscat, Oman, 1-4 February, 2015 where 𝜆𝑗𝑚 is the path gain for the UE 𝑗 served from the cell m. 𝑃 is the BS transmitted power and 𝜎 2 is the UE noise variance. The summation in the denominator is over the non-coordinated BSs, 𝑚 ∈ {𝑀/𝑀̀}. By considering that each UE 𝑗 is served by one cluster 𝑀`. The UE j path gain from cell m, 𝜆𝑗𝑚 , is directly related to the received power, Pr𝑗𝑚 . The received power can be calculated as: Pr𝑗𝑚 = 𝑃 − 𝑃𝐿𝑗𝑚 − 𝐴𝐿(𝜃𝑗𝑚 )

[𝑑𝐵]

(4)

where 𝑃𝐿𝑗𝑚 and 𝐴𝐿(𝜃𝑗𝑚 ) are the path-loss and directional loss from cell 𝑚 to UE 𝑗, respectively. B. Real Practical Network The chosen practical network data represents a unique topography from Al-Khobar city in Saudi Arabia that was not used in previous CoMP research. The elevation topography of this coastal city is flat and close to the sea level. We consider 20 BSs and 60 directive sectors. The distribution of these BSs depends on practical network implementation with nonuniform ISD. We simulated Frequency Division Duplex (FDD) Long-Term Evolution (LTE) air interface network with 20 MHz bandwidth using Multiple-Input and Multiple-OutputOrthogonal Frequency-Division Multiple Access (MIMOOFDMA). This technology requires a precise propagation model to estimate the network interference. Choosing UE serving cell in this model depends on the highest received power of all carriers in the network and the minimum required SINR. We fixed two threshold for these two parameters. Fig. 2 shows the satellite view of the real system model using Google Earth®. Each 4-digit number in the figure refers to one real BS as categorized in the network database. Serving areas in real network are unequal in size due to the different locations of antennas, antennas height, environment, interference, BS elevation and transmitted power level. Original Digital Elevation (ODE) generated by Shuttle Radar Topography Mission (SRTM) and a Ray tracing design

Fig. 2. Satellite view of the practical network model of Al-Khobar

TABLE I. PRACTICAL MODEL PARAMETERS

Parameters Layout ISD Carrier Frequency Bandwidth UE Noise Figure Antenna Gain Resolution of prediction results Sampling rate Subcarrier spacing Symbol duration Separation between Uplink and Downlink Max Tx power BS antenna height

Value 20 sites with 3 cells (sectors) 500 m (in average) 2.6 GHz 20 MHz 6 dB 11.6 dBi 81 m 384/250 15 KHz 66.67 usec 120 MHz 40 dBm 17 - 35 m

software are used to simulate the real setup. We simulated 20 BSs using one of the LTE frequencies which is 2.6 GHz and 20 MHz for bandwidth, and the ISD is non-uniform because we are using real BSs locations. Each sector maximum transmitted power is 40 dBm which is a reasonable level of power that can be used in urban or suburban networks. Table I briefs the used parameters. These parameters are feed to the propagation modeling software. III. CLUSTERING IN IDEAL COMP NETWORKS One major concern in the coordination is how group of BSs and UEs can be selected. Choosing appropriate clustering can reduce the experienced interference between the sectors. Assigning the coordinated sectors inside each cluster, choosing clustering type and cluster size are the most important aspects in clustering selection. Clustering selection is easier in ideal CoMP networks compared with practical ones because of the uniformness of BSs and UEs distribution which give more flexibility in clustering design. We applied the two optimization criteria in [10]; maximize mean SINR and outage measure probability. We investigated non-overlapped and overlapped static clustering techniques. The initial clusters are selected according to a quality function passing through a number of steps. The first step is to assign one serving cluster to each UE 𝑗 where each cluster consists of 3 different sectors. We calculate the received power to each UE from all the 111 sectors. Then we find out the sectors that provide the highest three received power values to a specific UE, and those three sectors will form one initial cluster. Hence, the number of clusters basically in this step is equal to the number of UEs 𝑗 and we called those clusters as initial clusters. Some of the initial clusters are common to different UEs. The second step is to rank the initial clusters according to their

Proceedings of the 8th IEEE GCC Conference and Exhibition, Muscat, Oman, 1-4 February, 2015 occurrence, where the most frequent clusters are the most desired ones. The number of occurrences is subject to some constraints such as the number of UEs, UEs location, BSs locations and UEs density distribution. The third step is to cancel the repeated clusters and keep only one copy from each cluster so we have unique clusters. The optimization of maximizing mean SINR or outage measure depends on the initial clusters and the number of simulated UEs 𝐽. In our case, simulating 441 UEs in the nonoverlapped clustering leads to 263 unique clusters. Increasing 𝐽 assures that all network cells are included in the initial clusters. However, increasing 𝐽 linearly, results in an exponential increase in the simulation complexity. We decreased this complexity by simulating specific granularity from the entire network. We used a simulated granularity equals to one quarter of the entire network that uses 400 UEs distributed in the granularity. In this case, we got 89 unique clusters out of the total 400 initial clusters. In optimization outage probability, Fig. 3 demonstrates the SINR CDF differences between clusters with different number of coordinated cells using 1600 UEs in the ideal system model. Cluster size is an important factor to determine the UE experienced SINR level. The experienced SINR level increases when the number of coordinated sectors inside a cluster increases, whilst increasing the cluster size increases the system signal overhead and complexity. The largest SINR improvement is when the cluster size is changed from 2 to 3 cells. The cluster size advantage is less significant, when the cluster size is increased from 4 to 5 coordinated cells. The gain starts to saturate as the cluster size exceeds 6. IV. CLUSTERING IN PRACTICAL COMP NETWROKS We applied different optimization criteria in the ideal model where the system is more flexible to accept changes because sites and UEs in that model are somehow systematic. Applying a standard optimization criterion for real network is more complex due to different conditions such as; elevation, sites location, users distribution and antennas height. We simulated two main types of coordination in the real SINR CDF comparison between different CoMP cluster size

1

Cluster size =1 Cluster size =2 Cluster size =3 Cluster size =4 Cluster size =5 Cluster size =6

0.9 0.8 0.7

CDF

0.6 0.5 0.4 0.3 0.2 0.1 0 -20

-10

0

10

20

30

40

50

60

SINR

Fig. 3. SINR CDF comparison between different CoMP cluster sizes

70

CoMP network using WinProp®. The first type is held between sectors within the same BS which is called intra-site coordination, and the second type is between a set of sectors which belong to different BSs. This type is called inter-site coordination. Both of them are simulated for non-overlapped clustering. We can simulate up to seven carriers and two horizontal and vertical polarizations, thus we have 14 unique RF and polarization air interfaces. Using the same carrier causes different types of interference, and the practical system model has more than 14 clusters in both intra-site and inter-site coordination. The demands here are more than the available resources because there are only 14 unique RF and the number of clusters is more than 14. Reuse of frequency spectrum and reduced cell size have to be implemented. This may lead to an increase in the interference level in the multi-cell networks. Applying frequency reuse should occur between geographically separated clusters to prevent any overlapping between the spectrum resources as the interference and distance between clusters are inversely related. A. Inter-site clusters coordination The real system model consists of 20 sites with 60 cells. We simulated this system model using 23 clusters in the inter-site coordination type. Choosing the 23 clusters in this stage is primary based on sector locations where some of them are in the network borders and the others in the middle of the network. Fig. 4a shows the 23 clusters network coverage, the clusters are of different sizes due to non-uniform ISD and real network distribution. Since there are 23 clusters and 14 unique RF parameters, we reused 9 RF parameters in different clusters, and only 5 clusters have unique RF parameters. Fig. 4b demonstrates how the downlink interference differs from one cluster to another. Less interference exists in the 5 clusters which have unique RF parameters while the interference is higher in the remaining 18 clusters because their RF spectrum has been reused twice. Despite of the benefits of using frequency reuse technique, UEs still suffer some interference between the sectors in the networks especially in short ISD networks. UEs in the middle of the network experience higher interference because they are affected by many undesired signals from neighboring sectors while the UEs in the network edge receives less undesired signals. B. Intra-site clusters coordination The coordination in intra-site clusters takes place on the BS level where all BS cells belong to one cluster. The number of the clusters in intra-site coordination is equal to the number of sites, which equal 20 sites in the practical system model. Fig. 5a shows the 20 intra-site clusters in the network where each color refers to one cluster. Frequency reuse is applied in the intra-site coordination case also, where there are 6 clusters with reused RF parameters. Fig. 5b shows the SINR level distribution in the system with intra-

Proceedings of the 8th IEEE GCC Conference and Exhibition, Muscat, Oman, 1-4 February, 2015

(a) Coordinated clusters distribution

(a) Coordinated clusters distribution

(b) SINR level

(b) SINR level

Fig. 4. Inter-site coordination in real network; cluster distribution and SINR

Fig.5. Intra-site coordination in real network; cluster distribution and SINR

site coordination. There is no direct comparison between intersite and intra-site coordination SINR levels since their sites have different RF plan. Figure 6 shows the SINR CDF for no-CoMP, intra-site CoMP, inter-site CoMP (cluster size is 3) and inter-site (cluster size is 4). This figure quantitatively shows the SINR behavior in the real network system model compared with the ideal case

presented in Fig. 3. This figure illustrates the fact that we can get better SINR enhancement in inter-site CoMP in comparison to intra-site CoMP, but this will be at the cost of the needed backhaul to exchange data between BSs. At CDF equals to 0.4, inter-site CoMP SINR exceed intra-site CoMP by 6 dB. Increasing the cluster size leads to improvement in the SINR values in inter-site coordination case, but the signal overhead to exchange CSI increases. The plots in Figure 6 are based on real data so their behaviors is not very smooth.

Real Network SINR CDF Comparison Between Different CoMP status 1

V. CONCLUSIONS

0.9 0.8 0.7

CDF

0.6 0.5 0.4 0.3 No Coordination

0.2

Intra-Site CoMP Inter-Site CoMP (Cluster size is 3)

0.1

Inter-Site CoMP (Cluster size is 4) 0 -10

0

10

20

30 SINR

40

50

60

70

Fig. 6. Practical network SINR CDF comparison between different CoMP approaches

In this paper, we have investigated the ideal and real CoMP networks in different types of clustering. It was shown that real CoMP can enhance the SINR to noticeable levels as in ideal CoMP networks. We have quantified the SINR improvement and discussed the effect of cluster size on the achieved SINR for practical and ideal networks. Increasing the number of coordinated sectors in small clusters size shows more SINR improvement compared with larger clusters. We addressed clustering and SINR behavior for inter-site and intra-site clustering in Al-Khobar. We believe a comprehensive investigation for applying coordination in noncoastal cities and comparing the gain with coastal ones deserves further work.

Proceedings of the 8th IEEE GCC Conference and Exhibition, Muscat, Oman, 1-4 February, 2015

REFERENCES [1] P. Marsch, M. Grieger and G. Fettweis, "Field Trial Results on Different Uplink Coordinated Multi-Point (CoMP) Concepts in Cellular Systems," in Global Telecommunications Conference (GLOBECOM 2010), 2010 IEEE, 2010. [2] R. Irmer, H. Droste, P. Marsch, M. Grieger, G. Fettweis, S. Brueck, H. P. Mayer, L. Thiele and V. Jungnickel, "Coordinated multipoint: Concepts, performance, and field trial results," Communications Magazine, IEEE, vol. 49, no. 2, pp. 102-111, February 2011. [3] P. Marsch and G. Fettweis, Coordinated Multi-Point in Mobile Communications - From Theory to Practice, Cambridge University Press, 2011. [4] J. Yu, Q. Zhang, P. Chen, B. Cao and Y. Zhang, "Dynamic joint transmission for downlink scheduling scheme in clustered CoMP cellular," in Communications in China (ICCC), 2013 IEEE/CIC International Conference, China, 2013. [5] Y. Cheng, M. Pesavento and A. Philipp, "Joint Network Optimization and Downlink Beamforming for CoMP Transmissions Using Mixed Integer Conic Programming," Signal Processing, vol. 61, no. 16, pp. 3972 - 3987, 2013.

[6] M. Grieger, P. Marsch, Z. Rong and G. Fettweis, "Field trial results for a coordinated multi-point (CoMP) uplink in cellular systems," in Smart Antennas (WSA), 2010 International ITG Workshop on, 2010. [7] Q. Li, R. Hu, Y. Qian and G. Wu, "Intracell Cooperation and Resource Allocation in a Heterogeneous Network With Relays," Vehicular Technology, vol. 62, no. 4, pp. 1770 - 1784, 2013. [8] M. Grieger, G. Fettweis and P. Marsch, "Large Scale Field Trial Results on Uplink CoMP with Multi Antenna Base Stations," in Vehicular Technology Conference (VTC Fall), 2011 IEEE, 2011. [9] S.-h. Xiao, "Discussion on Strategies for Adaptive Dynamical Clustering in Cooperative Multi-point Downlink Transmission Systems," Wireless Personal Communications, vol. 67, no. 3, pp. 525-539, 2012. [10] P. Marsch and G. Fettweis, "Static Clustering for Cooperative MultiPoint (CoMP) in Mobile Communications," in Communications (ICC), 2011 IEEE International Conference on, 2011. [11] M. Grieger, P. Helbing, G. Fettweis and P. Marsch, "Field trial evaluation of compression algorithms for distributed antenna systems," in Sarnoff Symposium, 2011 34th IEEE, 2011. [12] Q. Li, R. Hu, Y. Qian and G. Wu, "Cooperative communications for wireless networks: techniques and applications in LTE-advanced systems," Wireless Communications, IEEE, vol. 19, no. 2, pp. -, April 2012.