Assessment of Cellular Planning Methods for GSM Pedro Assunção, Rui Estevinho and Luis M. Correia Instituto das Telecomunicações/Instituto Superior Técnico, Technical University of Lisbon Av. Rovisco Pais, 1049-001 Lisboa, Portugal; Ph: +351 218418478; Fax: +351 218418472 rui.rui@megamail.pt, pedro_assuncao@net.sapo.pt, luis.correia@lx.it.pt Abstract The performance of various cellular planning methods for GSM, which decrease co-channel interference without sacrificing traffic capacity, is evaluated. For measuring the advantages of each method, the Carrier-to-Interference Ratio, is calculated. The use of different clusters and the benefits of sectorisation in C/I are studied. An evaluation of the influence of Frequency Hopping on C/I is also done. The Multiple Re-use Patterns (MRP) method is investigated as well, and a comparison is performed with previous results. The results show that changing the size from 4 to 3 leads to a decrease in C/I of 2.6 db. Several real antennas are studied for sectorisation, the best results being obtained for the 90º beamwidth one. The use of sectorisation, together with MRP, allows an increase in network performance, without visible additional costs. I. INTRODUCTION Due to limitations in the available spectrum, and because of the continuous increase of subscribers, mobile operators are forced to introduce advanced cellular planning management strategies in order to increase system capacity, without increasing interference. The concept of cellular network allows the use of cellular clusters, which are re-used until all the service area is covered, and this has been extensively used in GSM.. As a consequence, interference problems arise from using the same frequencies throughout the network. Also, there is a growing demand for new services; hence, it became necessary to expand the current systems without an unmeasured increase of costs. New planning methods are created for reducing interference and increase capacity, like frequency hopping or multiple re-use patterns. One way to study the interference that co-channel cells create to a user, is by calculating the Carrier-to- Interference Ratio, C/I. Different clusters lead to different re-use distances: a small cluster means that the distance between the users and the interfering cells is smaller, therefore, the higher the re-use distance the higher the C/I, but less frequencies will be available per cell and the capacity will be smaller. Another way to increase C/I is by reducing the number of interferers on the network, which can be done by using sectored cells. A simulator was developed to evaluate several planning methods that allow an increase of capacity without decreasing C/I. The next section presents the theoretical models used on the simulator, i.e., sectorisation, frequency hopping and multiple re-use patters. Section III contains a description of the implementation of the simulator. The main results are discussed in Section IV. Finally, conclusions are drawn in Section V. A. Sectorisation II. THEORETICAL MODELS The technique of sectorisation is well known, and consists in dividing the cell into several equal angular sectors, usually 3 or 6 [1]. The coverage of a sector is ensured by a directional antenna, which radiates mostly to the area of that sector and not to the others. By doing this, the number of interfering Base Stations (BSs), is reduced and the co-channel interference decreases. The most common sectorisation is the division into three sectors of 120º each. However, the use of real antennas (i.e., of nonideal radiation patterns) has an impact in the performance of the network, and needs to be evaluated. B. Frequency Hopping With Frequency Hopping (FH), a user hops from one carrier to another within a group of available frequencies [1]. FH is also well known, and allows improving the immunity to interference, being used to reduce the influence of Rayleigh s fading, to increase frequency diversity, and to reduce C/I. When hopping, the Mobile Station (MS) does not stay too long on a fading peak; therefore, the overall fast fading effect is reduced. Moreover, since each frequency has a different level of interference on different locations and at different time instants, when hopping, the interference will be distributed by all channels, and, consequently, the probability of a user experiencing an unacceptable level of interference is reduced [2]-[5]. For these reasons, frequencies can be re-used more intensively, and a higher spectral efficiency is obtained. GSM uses slow frequency hopping with 217 hops per second [6]. C. Multiple Re-use Patterns Multiple Re-use Patterns (MRP) is a method for reaching high capacity, by using a tight frequency re-use and frequency hopping. It is implemented with different clusters, each one with an associated re-use distance, and uses FH to combine these clusters into an average one. This average cluster has more capacity than the one of the original (larger) cluster, and a larger re-use distance than the other (smaller) cluster [7]-[8]. The objective with MRP is to achieve a balance between increasing capacity (in order to satisfy a growing number of users) and maintaining the necessary quality (i.e., keeping a good C/I). Another advantage of using MRP is that it can be applied only to the areas where there is a lack of capacity, i.e., it does not have to be applied to the entire network.
The available frequencies are divided and assigned to different clusters. However, it is important to assign the BCCH (broadcast) frequencies to the higher cluster that has the longer re-use distance, so that these frequencies are completely protected from interference. Protecting BCCH frequencies is very important, because an interference problem in these frequencies would lead to wrong choices in cells, access, paging, etc. III. SIMULATOR IMPLEMENTATION A. Overview A simulator was built with the purpose of studying the interference problem in a network, and to evaluate some of the available methods that implement the models mentioned above. The output produced by the simulator is the C/I. The simulator was used to measure the interference in the downlink, since usually this one is more limited than the uplink, where antenna diversity can be used. Since the purpose is to compare different cellular planning techniques, a constant traffic situation was considered, with all calls being generated at the same time; no drop calls and no handovers are considered. The structure of the simulator is presented in Fig. 1, where only the key blocks are shown. less; this model considers a typical urban environment with antennas on the top of the buildings and a distance decay coefficient of 3.8. For distances higher than 5 km, the Okumura-Hata propagation model was used [10]. The path loss is calculated, and from there, the carrier and interfering received powers. Only the first ring of interfering BSs was considered for calculating the total interference (maximum of six interferes): (C/I) [db] = C [dbm] - n= N act 1 I n (1) [ dbm] N act = 0, 1,..., 6, being the number of active co-channel BSs. Fig. 2 shows an example of a homogenous network, for a cluster size of N cs = 4, with all BSs having the same characteristics, and placed at the same distance from each other. The reference BS, to which the MS is linked to, is placed in the middle of the network and the surrounding rings of co-channel BS are presented in different shades. Input Data BS and MS Distribution Choose Antenna Type Channel Allocation Path Loss and Received Power Calculation Determination of Interferers C/I Result Fig. 1 - Simulator structure. A typical urban environment was chosen for simulation. After loading the input parameters, the simulator starts to distribute uniformly the cells by the service area, and also to distribute uniformly the users in the cells. After that, the antenna type is chosen from an operator s database. Simulations with several types of omnidirectional and directional antennas were performed. Ideal (theoretical) and real radiation patterns were used, so that the real situation and the theoretical one could be compared. A user gets a random frequency and a random time-slot, with a uniform distribution. For the propagation analysis, the COST231-Walfisch Ikegami [9] propagation model was used for distances between the BS and MS of 5 km or Fig. 2 - Hexagon cells used in network planning, with a regular cluster of 4. B. Sectorisation Sectorisation was implemented by dividing into 3 sectors. For antennas with an ideal radiation pattern, there is a maximum of three interferers (the axes of the sectors oriented to 60º, 180º and 300º). When using real antennas, the radiation pattern is not confined to the 120º sector, and an overlap occurs between adjacent sectors; hence, there will be a maximum of six interferers, although not all of them reaching the receiver with the same interfering power. Some simulations were made with this cell configuration. The use of 60º, 90º and 120º antennas was studied, as well as their influence on the calculation of C/I, being also compared with the ideal case. Fig. 3 shows the radiation patterns for ideal and real 120º.
(a) ideal 120º (b) real 120º C. Frequency Hopping Fig. 3 Radiation patterns. Using the model described before, a random FH system was implemented. This allows the MS to change randomly the frequency in each new frame, preserving the time-slot in each call. An average call duration of 90 s was considered, which leads to a total of 19 530 hops, at a rate of 217 hop/s. The co-channel interference will vary from hop to hop, for the duration of the call, and the total interference will be lower than the non hopping case, because the probability of two users having the same frequency and time-slot is smaller. The possibility of reducing the cluster size from 4 down to 3 cells, and then to 1 cell, was also taken into account. When a cluster of 3 is used, all available carriers are evenly distributed among the cells, and all available frequencies in each cell are used for hopping. For the 1- cell cluster situation, each sector uses all available frequencies for hopping, resulting in co-channel interference among different sectors; hence, for this simulation, two rings of interference must be considered. D. Multiple Re-use Patterns The MRP technique allows having the benefits from FH and some more. In each cell, there will be two clusters: N cs = 3 and N cs = 4. FH combines these two clusters, and an intermediate cluster will exist. It is intended, with MRP, to increase the number of frequency channels without sacrificing the C/I quality. A total of 40 carriers was considered (as explained later). For MRP simulations, these 40 carriers were distributed between the two clusters, as shown in Table I. Table I Carrier distribution N cs = 4 and N cs =3. Case Nº of carriers for the 4-cell cluster 0.0-10.0-20.0-30.0-40.0-50.0 Nº of carriers for the 3-cell cluster 1 40 0 2 28 12 3 16 24 4 0 40 Case 1 shows the typical 4-cell cluster, with all frequencies allocated, while Case 4 shows the 3-cell cluster. The two middle cases try to get the advantages from both plans: higher capacity and lower interference. An MS connecting to a cell will randomly get a frequency from one of the two plans. Each time a hop occurs, the new frequency can be from either of the considered clusters. Fig. 4 illustrates the channel distribution for Case 3. In this case, 16 frequencies are allocated to the N cs = 4 cluster, and 24 frequencies are allocated to the N cs = 3 cluster. So, the available frequencies per cell are 16/4 + 24/3 = 12; from these 12 frequencies, 4 are assigned to the 4-cell cluster and 8 are assigned to the 3-cell cluster. When hopping, users can be assigned to any of the clusters, since the frequency channel is randomly selected. The total C/I results for the duration of the call are the average ones, obtained from using only the 4-cell or the 3-cell cluster. Fig. 4 - Example of MRP channel allocation. For the previous cases, the number of frequencies per cell is shown in Table II. Case Table II Carriers per cell calculation. Nº of carriers per cell for N cs = 4 Nº of carriers per cell for N cs = 3 Total of carriers 1 40 / 4 0 10 2 28 / 4 12 / 3 11 3 16 / 4 24 / 3 12 4 0 40 / 3 13 It is obvious that there is an increase in capacity with MRP, since the total of carriers per cell increase when mixing the two plans. IV. RESULTS A total of 40 carriers are available for the entire network, which is the value used by GSM operators in Portugal, for the 900 MHz band. Table III summarises the input parameters used by the simulator. Table III - Summary of the input data in the simulations Cell radius BS height MS height BS output power MS reception gain 1 km 30 m 1.8 m 10 dbw 0 dbi Cluster size 3 or 4 Number of sectors in each cell 3
For simulation purposes, the mobile was considered in two positions: at the border of the cell, and appearing randomly in any position within the cell. The results are presented for several types of antennas (ideal omnidirectional, ideal sectorial and real sectorial with 60º, 90º and 120º beamwidths) and the two above positions (border, random), Table IV. The results are for an average C/I over 100 simulations, for standard planning, without using FH or MRP. The 4-cell cluster and the tighter 3-cell one were considered. It can be seen that the results for the N cs = 3 plan are worst at the border, because the distance between the co-channel cells and the reference cell is shorter. Cells with ideal sectorisation have only three possible interferes, against six from omnidirectional cells, and so the C/I is higher. For the random case, results are not as conclusive. Table IV - Average C/I results for standard planning. Average C/I N cs = 4 N cs = 3 [db] Border Random Border Random Ideal Omni 16.65 30.98 14.05 30.84 Ideal Sect 120º 22.45 37.98 18.54 34.45 Real Sect 120º 24.15 36.54 18.56 37.03 Real Sect 90º 26.29 39.18 18.73 40.83 Real Sect 60º 26.56 42.55 17.92 41.40 It would be expected that the best results would occur for ideal antennas, but that does not happen. The case is that ideal antennas have a uniform directional gain in all the coverage area, hence, the co-channel BS interferes with maximum power. For real antennas, the directional gain is not uniform, it being lower in the directions near the angular borders of the sectors, therefore, there is less interference for a user at these angular borders of the sector. It can also be seen from Table IV that the best results for the 4-cell cluster were obtained for the 60º antenna. This happens because the beamwidth is smaller than the in other cases and the interference that reaches the user is lower. However, for this kind of antennas, there is a coverage problem for users at the sector s border, due to the low value of the directional gain of the antenna. The 90º antennas have similar results at the limit of the cell, when compared to the 60º antennas, but they have the advantage of better coverage between sectors. For the 3-cell cluster, the results indicate that a 60º antenna is not so good as in the previous case. The problem is that the power received from the BS and the power received from the co-channel BS does not vary in a linear way. When using FH, all the available frequencies are used for hopping, except for the BCCH carrier, which contains the control and signalling channels and should not interfere in any way with the traffic channels. Because of the high processing time, the average C/I when using random FH, Table V, is the average result over only 10 simulations. Table V - Average C/I results for FH, at the border. Average C/I [db] N cs = 4 N cs = 3 N cs = 1 Omni ideal 16.41 15.13 11.93 Sect ideal 120º 22.35 17.87 17.93 Sect real 120º 23.68 18.01 18.13 The results for FH should be better than the standard plan, but since fast fading was not considered in the links, one could expect no major differences. Besides that, the benefits of FH are dependent on the number of frequencies used for hopping. For the 4-cell cluster, there is no improvement in the results, because there are not enough frequencies for hopping, considering the actual initial conditions of the simulator. For the 3-cell cluster, with more frequencies for hopping, there is an improvement in the results for omni antennas, but still no gain for sectored antennas. For the 1-cell cluster, one can achieve a large increase in capacity, because all available frequencies are used for hopping. From the theoretical point of view, it is possible to reduce the cluster from 4 cells to 1: the average results for the latter are above the 9 db margin for GSM; however, it is impossible to implement this cluster, since the minimum results obtained for C/I could reach 3 db. To make the most of FH, the system should be planned with a tighter frequency re-use. Other studies [2] show that when channel occupancy is lower, there are more benefits in using FH. In this simulator, only the interference averaging aspect was studied. Table VI shows the average C/I results for MRP, when using also FH. In the MRP simulations, the available frequencies were divided into two different bands, which are used by the two clusters (N cs = 4 and N cs = 3). Table VI - Average C/I results for MRP. Frequency distribution N cs =4 N cs =3 Average C/I [db] for Omni Ideal Average C/I [db] for Sect Ideal 40 0 16.74 22.23 28 12 15.93 21.92 16 24 15.42 21.57 0 40 14.97 21.05 The MRP results in Table VI show that C/I decreases when gradually going from the N cs = 4 cluster to the N cs = 3 one. This happens because the re-use distance becomes smaller and co-channel cells get nearer to each other. On the other hand, more frequencies are available per cell, and so capacity increases. The C/I results obtained for the two intermediate cases are between the results obtained for the two extreme cluster sizes, as expected. With MRP, an increase in capacity is achieved,
without excessively compromising the quality of the network. Fig. 5 and Fig. 6 illustrate the average C/I results for each method, and the variation between the minimum and maximum values. When using MRP, a tighter cluster size can be used (N cs = 3), with increased capacity, and still maintaining C/I at a good level. VI. ACKNOWLEDGEMENTS The authors would like to thank Telecel Vodafone Network Development Department for their support on technical questions about the implemented models. VII. REFERENCES Fig. 5 Graphical results for omnidirectional cells. Fig. 6 Graphical results for sectored cells. As it can be seen by the obtained results, C/I is higher when sectored cells are used, because fewer interferers exist. More results are provided in [11]. V. CONCLUSIONS After studying the actual conditions of the networks and the existing cellular planning methods, there is no doubt that it is very important to optimise the existing methods. When using ideal antennas, and the mobile placed on the border of the cell, the results show that changing the cluster size from 4 to 3 leads to a decrease of 2.6 db in C/I. Using sectorisation in the N cs =3 cluster allows to achieve a 4.5 db gain, while for the 4-cell cluster the gain is of 6 db. From the several types of real antennas studied (60º, 90º and 120º beamwidths), the 90º beamwidth one presented the best results. When using random FH a better performance can be achieved without degrading C/I. A tighter cluster size can be used, leading to a higher capacity, and the global interference is reduced. The improvement in network performance when using FH would be more visible if fast fading would have been considered in the simulator. [1] M.D. Yacoub, Foundations of Mobile Radio Engineering, CRC Press, Boca Raton, FL, USA, 1993 [2] G.W. Tunnicliffe, A. Sathyendram and A.R. Murch, Performance Improvement in GSM Networks Due to Slow Frequency Hopping, in Proc. of VTC 97 47th IEEE Vehicular Technology Conference, Phoenix, Arizona, USA, May 1997 [3] H. Olofsson, J. Naslund, J. and Skold, Interference Diversity Gain in Frequency Hopping in GSM, in Proc of VTC 95-45th IEEE Vehicular Technology Conference, Chicago, Illinois, USA, Apr. 1995 [4] A. Mohammed and S. Sali, Co-channel Interference Management in Cellular Networks, in Proc. of ACTS Mobile Communications Summit 97, Aalborg, Denmark, Oct. 1997 [5] J. Wigard, T.T. Nielsen, P.H. Michaelsen, and P. Mogensen, Improved Intelligent Underlay-Overlay Combined with Frequency Hopping in GSM, in Proc. of PIMRC 97 8th IEEE Personal Indoor and Mobile Radio Communications, Helsinki, Finland, Sep. 1997 [6] M. Mouly and M.B. Paulet, The GSM System for Mobile Communications, Mouly et Paulet, Palaiseau, France, 1992 [7] F.A. Cruz-Péres, D. Lara-Rodrigues and M. Lara, Multiple Reuse Patterns in Urban Microcellular Environments, in Proc. of VTC 99 49th IEEE Vehicular Technology Conference, Houston, Texas, USA, May 1999 [8] S. Engstrom, T. Johansson, F. Kronestedt, M. Larsson, S. Lidbrink and H. Olofsson, Multiple Reuse Patterns for Frequency Planning in GSM Networks, in Proc of VTC 98-48th IEEE Vehicular Technology Conference, Ottawa, Canada, Apr. 1998 [9] E. Damosso and L.M. Correia (eds.), Digital Mobile Radio towards future generation systems, COST 231 Final Report, COST Secretariat, Brussels, Belgium, 1999 [10] J.D. Parsons, The Mobile Radio Propagation Channel, Pentech Press, London, UK, 1992 [11] P.L. Assunção and R. Estevinho, Assessment of Cellular Planning Methods for GSM (in Portuguese), Graduation Thesis, IST, Lisbon, Portugal, Nov. 2000