Abstract Optimum basestation positioning in the framework
of LTE wireless networks is not an easy task since there are
several factors that need to be taken into account. The ever increasing demand for geographically inhomogeneous capacity, which has become the limiting factor in the network planning process, as well as the presence of relay nodes and mixed cell scenarios, must all now be considered along with the systems constraints and capabilities. In this paper, the optimization problem for an LTE system is described and results are produced with a proposed partially combinatorial optimization algorithm. The optimization analysis is based on coverage, capacity and cost criteria. Example results are provided for a typical case study that give useful planning insights for LTE systems. Index Terms Radio network planning, Basestation location, LTE systems. I. INTRODUCTION RF planning is a multi-step process of a rather iterative nature. In the context of 2G wireless systems this process could be divided in two main unrelated steps, i.e. the coverage and capacity planning stages. This however is not the case with 3G WCDMA and 4G LTE systems, where capacity and coverage planning are strongly interrelated and should be treated simultaneously. Current LTE deployments (Rel. 8/9) offer up to 100Mbps peak rates with traffic that includes voice and data from cell phones and other mobile devices such as tablets. The plethora of data services (web browsing, social networking, audio, video, software downloading) requires a sound estimation of the user throughput requirements which is rather difficult to achieve. Additionally, at the end of 2012 ~10% of the population was under LTE while the projections suggest that this percentage will climb up to 60% at the end of 2018 [1]. Key feature of Release 12 and beyond is the utilization of heterogeneous deployments that comprise both high (macro) and low power nodes [2]. Finding the optimal LTE network deployment amounts to solving an optimization problem of combinatorial nature. As a consequence, system modeling that allows for rapid performance assessment and hence selection among a typically huge number of candidate solutions, is crucial. The cost and complexity of a network is closely related to the number of basestations required to achieve the operator's service objectives. The location of basestations is not an easy task and there are numerous factors that must be taken into account when deciding the optimum position. Planning and performance optimization in cellular networks form a very active line of research with a considerable amount of published work in the open literature especially on UMTS network planning and optimization, ranging from BS location and coverage planning to antenna parameter conguration and cell load balancing [3-5]. The problem of base station (BS) positioning planning becomes complicated in the framework of 4G wireless networks since the presence of relay nodes (RNs) and mixed cell structures must now be included in the planning process. Typical optimization algorithms (combinatorial, genetic, greedy, simulating annealing) have already been used for this type of problems (e.g. [6]). However, the combined problem of optimized base-station and relay-node planning in order to improve network coverage and capacity as well as cost, especially in 4G systems, has received little attention. In this paper, we have developed a rigorous LTE system model that works for a general network topology and explicitly accounts for mixed cell structures and relay nodes as well as non-uniform trafc demand (hotspots). A partially combinatorial optimization algorithm that exploits specific characteristics of the LTE system (CQI) has also been developed in order to automatically identify an optimum solution that meets the operators requirements in terms of coverage and capacity in a given area, with the minimum number of basestations and hence, cost. The remainder of the paper is organized as follows: Section II describes the general BS positioning problem in the context of an LTE network, section III describes the optimization algorithm, section IV presents simulation results for a case study and finally section V concludes the paper. II. THE PROBLEM A number of discrete user supplied points or control nodes (CNs) are used to represent the capacity and coverage requirements in the operational area [7]. The initial number of possible BS locations is user supplied (i.e. locations where planning permission and/or agreements have been made). The initial selection of user supplied sites will determine to a great extent, the final outcome, since no additional locations will be identified during the optimization process, and the user selected positions will not be modified. A typical problem is shown in Figure 1, where a number of control nodes (CNs) have been distributed in an area. A number of possible base station (BS) locations are also Automatic Location of Base-Stations for Optimum Coverage and Capacity Planning of LTE Systems G.E. Athanasiadou, D. Zarbouti, G.V. Tsoulos University of Peloponnese, Department of Informatics and Telecommunications, Tripoli 22100, Greece, E-mail: {gathanas, dzarb, gtsoulos}@uop.gr
displayed in the map. Each macro base station has three sectors, pointing towards 30 o , 150 o and 270 o in azimuth. The antenna pattern of [8] is used for each sector antenna. In the direction of maximum power, each sector can also have an omnidirectional Relay Node (RN) which is activated only if a CN is connected to the basestation through this relay node. In the simulation, the micro basestations are omnidirectional and they dont have relay any nodes. The propagation module is based on the WINNER model [9], which gives different path loss values for macro, micro and relay node scenarios. The LOS distance in the WINNER model is set to 500m and shadowing is not included in this stage, in order to be able to check the validity of the simulation outcome. 0 500 1000 1500 2000 2500 3000 0 500 1000 1500 2000 2500 3000 Test Area (axes in meters)
Figure 1: Test area of the case scenario. Thick red dots are possible locations for sectorized BSs, thick green dots are possible locations for micro BSs, black stars are control nodes and thick magenta stars represent hot spot areas.
The propagation parameters of the simulation are summarized in Table 1.
EIRP of a macro BS (sector) 46dBm + 18dBi = 64dBm EIRP of a micro BS 30dBm + 7dBi = 37dBm EIRP of a Relay Node 30dBm + 5dBi = 35dBm Propagation model used for macro base stations WINNER Path Loss Models: Scenario C2 NLOS with hTx = 25m Propagation model used for micro BS WINNER Path Loss Models: Scenario B1 NLOS with hTx =10m, d LOS = 500m Propagation model used for Relay Nodes WINNER Path Loss Models: Scenario B1 NLOS with hTx = 5m, d LOS = 500m User/micro/RN antenna pattern Omnidirectional Sector antenna pattern 3GPP antenna pattern for 120 o sector [8] Total cable losses 2dB Central frequency 2.12GHz
Table 1: Propagation parameters of the simulation
The capacity module that provides the LTE rate calculation for each user is described with the following flow chart where the Rank Index (RI) and the number of RBs that will be allocated to the user are known. SIR mapping to CQI T a b l e
Figure 2: Rate calculation for each user according to LTE. (Table 7.2.3-1 can be found in [12]). The user demands are represented by two different sets of control nodes. There is a set of CNs which require a relative small data rate (e.g. 1-3Mbps) and there are hot spot areas which require ten times more capacity. All bandwidth resources are available to each sector and microcell (reuse pattern 1/3). Hence, each sector of a macrocell and each microcell have 100 Resource Blocks (RBs) to allocate to users. The number of RBs which are allocated to each CN is a function of the channel Signal-to-Noise-plus-Interference Ratio (SINR) and the user demand for data rate. For the LTE rate calculations the authors use a similar approach as the one presented in [11]. For determining the minimum number of RBs that need to be allocated by the serving node to the CN, the process described in Figure 2 is followed. More specifically, a) the SINR that is derived at the CN is mapped to a single Channel Quality Index (CQI) value using the SINR-to-CQI mapping function given in [10]. b) the CQI is mapped to a proper modulation level (Q m )and code rate (CR). This mapping is derived according to Table 7.2.3-1 of [12]. c) equation 1 provides the required number of RBs (N RB ) when the CN throughput (CN t ) is known. In equation 1 N RE
is the number of Resource Elements (REs) per RB that for the purposes of this work is set to 126 (CFI=3, Transmission Mode 1), T slot is the time duration of the LTE slot, while the 24bits-CRC of LTE has also been considered. CRC CR Q N N RI T CN m RE RB slot t 8 1 8 (eq.1) The RNs are considered to be outband, i.e. the communication of the RNs with the corresponding macro basestation is not using these bandwidth resources. The RBs
needed by the users connected to an RN, are subtracted by the pool of RBs of the corresponding macro basestation. The algorithm must provide the bandwidth resources according to the SINR and the data rate request, as well as coverage to all control nodes using the smallest sub-set of possible eNBs. Among solutions with equal number of base stations, those which provide better overall channel SINR are preferred. III. THE OPTIMIZATION ALGORITHM The proposed optimization algorithm follows a partially combinatorial approach. The combination approach fits perfectly with the idea of pre-selected antenna positions. Similar approaches have been used for automatically locating basestations for 2G systems with coverage and traffic criteria [13-14]. The idea is that if all possible combinations of basestation positions are examined the optimum combination for this set of BSs will be found. However, the number of combinations increases dramatically as the number of possible locations is increased. To avoid this problem this algorithm segments the total number of possible locations into smaller groups, which are randomly selected, and for each smaller group an exhaustive search for solution is performed. The test area is first divided into four equal subareas (e.g. 1.5Kmx1.5Km for the area of Figure 1) where the partially combinatorial algorithm is applied. Half of the positions of each subarea form the set of BS positions to examine for the whole area. All basestations which participate in any of the produced solutions for the subareas form the group of candidate positions for the total area. Hence, the number of possible locations reduces to half (or even more if required), improving dramatically the algorithm speed, making it possible to examine scenarios with a large number of candidate positions. The initial number of BSs for each segment is determined by estimating the number of BSs needed to fulfill the capacity criterion assuming a moderate channel quality in terms of SINR for all links (CQI = 4). If the algorithm after going through the produced combinations cannot find a solution, a lower CQI is assumed (i.e. a higher number of required BSs) until a solution is found or CQI reaches its lowest value (CQI=1). Also, a first estimate for the number of BSs needed to fulfill the coverage criterion is found by applying a simple greedy algorithm. Given a number of BSs and control nodes, the greedy algorithm first selects the BS that covers the most control nodes. The BS and the control nodes are then removed from the area of study and the same procedure is repeated until there are no control nodes left to cover. The larger of these two estimations (one for the coverage and one for the capacity criterion) sets the initial number of locations in the examined combinations. For each combination of possible basestations, the algorithm first checks if all CNs are covered, i.e. the received power is above the threshold of -95dBm, as it is calculated by the LTE link budget. If the CNs are not covered then the proposed solution is invalid and the algorithm proceeds to the next combination of possible BSs. If the solution passes the coverage test, the SINR and the required RBs for all possible links are calculated. The CNs which are covered by only one BS, are the first to be allocated to an eNB. Then, starting from the CNs which require the highest number of RBs, each CN is allocated to the eNB where it has the highest SINR. If this eNB has not enough available RBs, the next eNB is considered, until there are not any eNbs left. If a CN cannot be served by any of the basestations, the solution is invalid. If all CNs are allocated to an eNB, the combination is stored as one possible solution and the algorithm continues the search (for other solutions with equal or smaller number of basestations) until a predefined number of trials (e.g. 100) is reached. At the end of the search, the algorithm chooses the solution with the minimum cost. Given that macrocells are much more expensive to deploy, the cost of each microcell and RN is considered to be 1/10 and 1/20 of the cost of a macrocell, respectively [1]. Channel quality (i.e. required bandwidth resources) as well as 95% coverage of the total area, are also considered as secondary criteria. IV. SIMULATION RESULTS FOR A CASE STUDY In this section a case study is performed for a 3Km x 3Km area (Figure 1) where 80 CNs are randomly placed, 10% of which represent hot spots. The required data rate for the CNs is 1Mbps, and for the hot spots 10Mbps. In the area there are 40 randomly placed possible macrocells with three sectors each, as well as 40 microcells with omnidirectional antennas. The distance between the RN and the corresponding macrocell is 400m. The solution shown in Figure 3.a has four sectorized macro basestations (thick red dots in the figure) and one micro basestation (thick green dot). At all sectors the RNs (black circles) have been activated since they have CNs connected to them (black lines), but there are also connections directly to the sectors (cyan, green and magenta lines for the three different sectors). Note that according to the WINNER model, the channel gain reduces considerably beyond the LOS distance. Hence, the CNs connected to the RNs or the microcells are within a range of ~500m (e.g. in Figure 3.a the CN in the bottom left corner is served directly by the sector since it is out of the microcell range). The algorithm chooses basestation/RN positions close to the hotspots so that the channel quality is high and hence, less bandwidth resources (RBs) are used for the link. Another possible solution (Figure 3.b) is also the final outcome of the algorithm. Note that 3 out of 5 proposed basestations are common for both solutions. However, the second solution is preferred (i.e. it has better fitness function) because the usage of microcells, reduces the cost of the solution considerably, and also improves the overall channel quality of the CNs since the interference from the microcells is much lower. From the same figure can be noticed that one of the basestations covers CN positions
almost 2Km away. This is not surprising, since the coverage of the employed macrocells is ~3Km. The reason why the coverage of the chosen cells in this test scenario appears smaller, is that it is a capacity limited scenario i.e. the cell range remains small in order to have better channel quality and also the eNBs only have enough bandwidth resources (RBs) to serve the CNs which are closer to them.
(b) Figure 3: (a) A possible solution produced by the partially combinatorial algorithm, (b) The final optimum solution. V. CONCLUSIONS The positioning optimization problem of heterogenous cells with relay nodes for an LTE system with non-uniform traffic is described in this paper and results that give useful insights are produced with a partially combinatorial optimization algorithm that exploits LTE characteristics. The optimization analysis is based on coverage, capacity and cost criteria. ACKNOWLEDGEMENTS The work in this paper has been co-financed by the European Union (European Social Fund ESF) and Greek national funds through the Operational Program Education and Lifelong Learning of the National Strategic Reference Framework (NSRF)-Research Funding Program: THALIS NTUA, Novel Transmit and Design Techniques for Broadband Wireless Networks (MIS 379489).The authors would also like to acknowledge discussions with colleagues in COST VISTA IC1102. REFERENCES [1] Ericsson Mobility Report, June 2013, Available Online: http://www.ericsson.com/res/docs/2013/ericsson- mobility-report-june-2013.pdf [2] Astely, D.; Dahlman, E.; Fodor, G.; Parkvall, S.; Sachs, J., "LTE release 12 and beyond" Communications Magazine, IEEE, vol.51, no.7, pp. 154-160, July 2013 [3] Hurley S (2002) Planning effective cellular mobile radio networks. IEEE Transactions on Vehicular Technology 51(2):243253 [4] E.Amaldi, A. Capone, F. Malucelli, Radio planning and coverage optimization of 3G cellular networks, Wireless Netw (2008) 14:435447, DOI 10.1007/s11276-006-0729-3, Springer. [5] Mathar RM, Niessen T (2000) Optimum positioning of base stations for cellular radio networks. Wireless Networks 6:421428 [6] Raisanen L, Whitaker RM (2005) Comparison and evaluation of multiple objective genetic algorithms for the antenna placement problem. Mobile Networks and Applications 10:7988 [7] A.Molina, A.R.Nix, G.EAthanasiadou, A combinatorial algorithm for optimum base station location, IEE Electronics Letters, March 2000, vol. 36, no 7, 30 pp. 668-669. [8] Spatial channel model for multiple input multiple output (MIMO) simulations, 3GPP TR 25.996 v.6.1.0, 2003. [9] WINNER II WP1: Channel models, Deliverable D1.1.2, 30/11/2007. [10] Ikuno, J.C.; Wrulich, M.; Rupp, M., "System Level Simulation of LTE Networks," IEEE VTC 2010-Spring, 2010 IEEE 71st, vol., no., pp.1,5, 16-19 May 2010. [11] D.Zarbouti, G.V.Tsoulos, G.E.Athanasiadou, C.A.Valagiannopoulos, The Effective Radiation Pattern concept for realistic performance estimation of LTE wireless systems, International Journal of Antennas and Propagation, Hindawi, September 2013.. [12] 3GPP TS 136.213: EVOLVED Universal Terrestrial Radio Access (E-UTRA); Physical Layer procedures [13] A.Molina, G.E.Athanasiadou and A.R.Nix, The Automatic Location of Base-Stations for optimised cellular coverage: A new combinatorial Approach, IEEE VTC 1999, pp. 606-610. [14] A.Molina, A.R.Nix, G.EAthanasiadou, Cellular Network Capacity Planning Using the Combination Algorithm for Total Optimisation, IEEE VTC 2000, pp. 2512-2516.