Small cell network topology comparison

INFORMATION AND COMMUNICATION TECHNOLOGIES AND SERVICES VOLUME: 11 |NUMBER: 5 |2013 |SPECIAL ISSUE Small Cell Network Topology Comparison Jan OPPOLZER, Robert BESTAK Department of Telecommunication Engineering, Faculty of Electrical Engineering, Czech Technical University in Prague, Technicka 2, 166 27 Prague, Czech Republic opp[email protected], robert.b[email protected] Abstract. One of the essential problems in a mobile network with small cells is that there is only a limited number of Physical Cell Identifiers (PCIs) available. Due to this fact, operators face the inevitable need for reusing PCIs. In our contribution, we are dealing with a PCI assignment to Femtocell Access Points (FAPs) in three different topologies. The first model places FAPs randomly within the network while respecting overlapping defined. The second model places FAPs in a grid without other restrictions. The third model forms a grid as well, although buildings and roads are taken into account and FAPs are always inside buildings. The proposed models are compared and a conclusion is made based on simulation results. Keywords Collision, confusion, femtocell, physical cell identifier, small cell, topology. 1. Introduction Femtocells, also known as Femtocell Access Points (FAPs), are here for a few years yet. They are small, low-power and mainly low-cost personal or enterprise Base Stations (BSs) deployed by customers [1], not by operators as in case of macrocells, etc. Although FAPs are small, they are a big market worth $2,7 billion by 2017 [2]. Originally, FAPs were intended mainly for improving indoor coverage because poor coverage affects up to 30 % of businesses and 45 % of households [3]. Further reasons were enhancing Quality of Service (QoS) and network capacity as well as offering new services to customers and raise customer retention [1], [3]. Nowadays, FAPs and metrocells collectively called as ”small cells”, are used even to improve outdoor signal coverage in city centres and busy streets. For example, in Newcastle and Bristol, small cells are trail deployed and during the testing data transmission was three times faster compared to 3G network [4]. By 2016, small cells are expected to make up almost 90 % of all base stations [5]. Moreover, by 2016, small cells and Wi-Fi access points will carry up to 60 % of all mobile data traffic [6] which is a huge portion if we take into account the increase in mobile data usage. Although small cells are a huge market, there are still some unresolved challenges. One of them is a Physical Cell Identifier (PCI) assignment mechanism. A PCI is composed of 168 unique groups each containing 3 identities which makes 504 identities in total [7]. Since every cell in the network need an identifier, this number is not sufficient and PCIs have to be reused which brings challenges, namely i) collision events and ii) confusion events. A PCI collision means that neighbouring cells have identical identifier assigned. Such a problem produces interference which creates so-called coverage hole and none User Equipment (UE) is able to connect to any femtocell. A PCI confusion arrives when a cell has more neighbours with the same identifier. In such a situation, when a handover should take place it will fail due to ambiguous destination where to transfer the connection [1], [3]. The aim of this paper is to develop and compare various topologies for small cell network simulations mainly dealing with PCI assignment techniques. Identifiers are assigned automatically, PCI collisions are completely avoided by scanning radio environment for neighbours’ identifiers and PCI confusions are solved whenever occurred. This paper is structured as follows. The second section briefly describes related works in this field of study which is not too wide yet. The third section is focused on three proposed topologies, their description, basic features and characteristics. The fourth section is devoted to simulations, comparing individual topologies and results. The fifth section summarizes the paper and outlines possible future work. c 2013 ADVANCES IN ELECTRICAL AND ELECTRONIC ENGINEERING 303 INFORMATION AND COMMUNICATION TECHNOLOGIES AND SERVICES VOLUME: 11 |NUMBER: 5 |2013 |SPECIAL ISSUE 2. Related Works Nowadays, identifiers are usually assigned either manually by operators using network planning tools or automatically using random selection. However, neither method is efficient. The first method is costly, timeconsuming and prone to human made errors. The second method is not reliable and might produce a confusion event or even worse a confusion event. There is a number of articles focusing on PCI assignment techniques. For example, [8], [9], [10]. However, all those works are concerned about macrocell level only. Authors in [8] are working with real 3G network from Vodafone Germany; however, 3G network with outdoor macrocells is far from 4G network with densely deployed small cells. In [9], handover measurements are utilized just to detect issues related to identifiers in macro and microcells. In another article [10], authors simulated very few BSs not representing future dense deployments. In all honesty, we have really tried to find out any similar study trying to compare various topology models that deal with PCI assignment techniques, but we have not discovered any. 3. Proposed Topologies All the three proposed placements described later have a few common basic characteristics. There is always a single Macrocell Base Station (MBS) with circular coverage area under which all FAPs are deployed. For simplicity, all the FAPs have the same radius. To simulate various FAP densities (for example city centres on one hand and rural areas on the other hand), the number of FAPs generated within a topology is varying. Although the Long Term Evolution (LTE) and LTEAdvanced (LTE-A) standards support up to 504 different PCIs, we have allowed only 480 of them at a maximum to be assigned in our simulation. The first reason behind this upper limit is that we would like to know whether a smaller portion of the identifiers is sufficient. The second reason is that other cells in a real network topology, such as macrocells, picocells, etc., require an identifier, too, so we have reserved at least a tiny PCI portion for those cells. And finally, the lower PCI range limit is required in order for the algorithm to converge and assign PCIs correctly. FAPs need information about their neighbourhood (i.e. PCIs of neighbouring cells) whenever they are choosing an identifier in order to evade i) a PCI collision and ii) a PCI confusion. Since FAPs have limited power, they can ask about neighbourhood only adjacent neighbours. However, this is not a sufficient amount of information when confusion events should be eliminated. To obtain data about unreachable neighbourhood in order to become aware of a greater part of the topology and evade a PCI confusion as mentioned, FAPs can employ neighbours as well simply by asking about their neighbourhood data. This is how a FAP can obtain information about neighbours multiple hops away. We term this as a hop count. In all the three proposed topologies, a simulation works as follows. At first a MBS is generated. Then, depending on the topology selected, a defined number of FAPs is placed within the MBS area i) randomly, ii) in a precise grid, or iii) in a grid where FAPs are allowed to be placed only inside of buildings and forbidden outside. Whenever a FAP is deployed, it scans radio environment for neighbouring cells. When the FAP has no neighbours we call it as a “standalone FAP” and such a FAP can select a PCI randomly. Alternatively, when neighbours are detected, the FAP selects such a PCI in order to avoid producing a PCI collision. After selecting a collision-free identifier, a bidirectional interface is established with detected neighbours for later usage. By establishing this interface a Neighbour Relation (NR) is set up and Neighbour Relation Tables (NRTs) containing lists of neighbouring PCIs are exchanged. Now, a PCI confusion procedure check is launched. When a confusion event is discovered, a FAP that is confused by neighbours initiate a solving procedure. In our previous work, we have designed and implemented two techniques for solving confusion events, we call them ”random method” and ”smart method”. Here, we deploy the mature one – smart method – which outperforms the other technique in terms of overhead introduced to the network. Our ”smart method” works as follows. When a FAP encounters a confusion event, it requests the involved FAPs (i.e. confusion producers) to report how many adjacent cells they have. After acquiring those numbers, the FAP with the fewest neighbours is chosen to reselect its PCI. Before a new PCI is chosen, the FAP scans radio environment (collision avoidance) and exchanges NRTs with neighbours up to 3 hops away (confusion avoidance). If the confusion event is still present, for example, when there are more than two confusion producers, this process is run again with remaining confusion producers. Reselecting a PCI means that new NRs have to be established between neighbours and thus overhead is produced. Firstly, the FAP with a new PCI has to inform all its neighbours about this change. Secondly, c 2013 ADVANCES IN ELECTRICAL AND ELECTRONIC ENGINEERING 304 INFORMATION AND COMMUNICATION TECHNOLOGIES AND SERVICES VOLUME: 11 |NUMBER: 5 |2013 |SPECIAL ISSUE Tab. 1: Common simulation parameters. Parameter Value MBS radius, rMBS 564 m MBS area, AMBS 1 km2 FAP radius, rFAP 15 m FAP count, NFAPs 250–1500 PCI range 80–480 Hop count, Nh3 Tab. 2: Random placement parameters. Parameter Value FAP overlapping <50 % the neighbours have to acknowledge this modification. When a FAP has nneighbours, this will eventually lead to 2nmessages sent over the network. Basic simulation parameters common to all topologies are stated in Tab. 1. 3.1. Random Placement The first model is random placement which does not fully represent real conditions or a real topology even though it might be close enough by tweaking various parameters such as FAPs overlapping, etc. In this model, FAPs are randomly placed within the MBS area using uniformly distributed pseudo-random numbers. Neighbouring FAPs can overlap each other; however, their mutual area is limited so scenarios where multiple FAPs are deployed at the same place (above each other) is eliminated. Overlapping and other parameters are stated in Tab. 2. In Fig. 1, a demonstration of random placement is shown. Red dots represent individual FAPs. Blue lines between dots symbolise so-called NRs which means that those neighbours can communicate mutually and exchange information about neighbours including their PCIs stored in NRTs. Numbers next to red dots are PCIs assigned and numbers next to blue lines are euclidean distance between FAPs in metres. 3.2. Grid Placement #1 The second model in our comparison study is grid placement #1. It is the easiest topology where individual FAPs are deployed in a precise square grid under the area covered by a MBS. Although some spots might be left empty depending on the total number of FAPs deployed in the actual simulation. Grid placement #1 parameters are introduced in Tab. 3. −100 −50 0 50 100 −100 −80 −60 −40 −20 0 20 40 60 80 100 Distance [m] Distance [m] 16.0916.09 9.489.48 11.6211.62 11.8511.85 18.4518.45 17.1717.17 17.91 15.40 10.82 17.91 15.40 10.82 17.5617.56 15.1015.10 16.9616.96 14.9214.92 8.54 16.74 11.72 8.54 16.74 11.72 16.5716.57 9.639.63 14.89 9.80 14.89 9.80 16.6016.60 16.37 18.99 16.37 18.99 15.95 17.03 17.46 10.11 15.95 17.03 17.46 10.11 10.5410.54 19.26 16.17 8.84 19.26 16.17 8.84 15.95 10.10 15.95 10.10 15.8715.87 16.0416.04 13.79 18.58 13.79 18.58 17.11 19.20 17.11 19.20 16.32 10.74 16.32 10.74 12.1312.13 13.2413.24 14.8214.82 12.3612.36 18.1518.15 18.7918.79 18.68 16.42 18.68 16.42 15.95 16.42 11.07 15.95 16.42 11.07 19.18 19.97 19.18 19.97 15.0115.01 16.5113.48 16.5113.48 19.1119.11 11.88 12.48 8.61 11.88 12.48 8.61 18.48 12.72 18.48 12.72 18.70 9.24 10.59 18.70 9.24 10.59 15.9215.92 19.61 12.46 9.44 13.19 19.61 12.46 9.44 13.19 18.8519.64 18.8519.64 15.57 19.23 18.67 17.89 15.57 19.23 18.67 17.89 14.34 9.47 14.34 9.47 17.21 17.41 9.63 18.16 17.21 17.41 9.63 18.16 11.67 15.16 11.67 15.16 17.59 17.77 17.59 17.77 11.62 19.14 13.41 15.12 11.62 19.14 13.41 15.12 19.9819.98 9.189.18 9.27 18.66 19.42 9.27 18.66 19.42 12.09 17.40 13.04 10.62 15.26 12.09 17.40 13.04 10.62 15.26 14.08 10.94 14.08 10.94 11.21 17.71 11.44 17.38 11.21 17.71 11.44 17.38 376 227 344 61 455 352 296 269 109 424 246 201 371 334 474 284 75 63 82 32 180 348 186 17 3 314 361 32 163 238 149 322 387 324 446 306 274 475 26 41 460 194 328 214 477 401 118 199 177 379 15 382 16 9 401 381 449 96 413 123 29 464 31 110 16 61 326 14 408 135 202 313 441 140 463 373 41 86 91 48 Fig. 1: Random placement demonstration. Tab. 3: Grid placement #1 parameters. Parameter Value Vertical side of grid 10 m Horizontal side of grid 10 m Grid placement #1 demonstration is depicted in Fig. 2. The meaning of red dots, blue lines and numbers are the same as in the random placement demonstration. The figure is quite similar to random placement demonstration; however, it can be seen that FAPs are not placed randomly but in a precise grid. Also, it is obvious that in this demonstration there are more standalone FAPs than in random placement. −100 −50 0 50 100 −100 −80 −60 −40 −20 0 20 40 60 80 100 Distance [m] Distance [m] 10.0010.00 14.1414.14 14.1414.14 10.0010.0010.0010.00 14.14 14.14 14.14 14.14 10.0010.00 10.0010.00 10.0010.00 10.0010.00 14.14 10.0010.00 14.14 14.1414.1414.14 10.00 14.14 10.00 14.1414.14 10.0010.00 14.14 10.0010.00 14.14 10.0010.0010.0010.00 10.00 10.00 10.00 10.00 14.14 10.00 14.14 10.00 14.1414.1414.1414.14 10.00 10.00 10.00 10.00 14.1414.14 14.14 14.14 14.14 14.14 10.0010.00 14.14 14.14 14.14 14.14 14.1414.14 14.14 10.00 14.14 10.00 10.00 14.14 10.00 14.14 14.1414.1414.1414.14 14.1414.1410.0014.1414.1410.00 14.1414.1414.1414.14 14.1414.14 10.00 10.0010.00 10.00 10.0010.00 10.0010.00 10.00 14.14 10.00 14.14 10.0010.00 14.14 10.0014.14 14.14 10.0014.14 14.1414.14 14.1414.14 10.00 14.1414.14 10.00 10.00 14.14 10.00 14.14 10.0010.00 14.14 10.00 14.1410.00 14.14 10.00 14.1410.00 10.0010.00 107 203 298 44 71 72 43 325 265 386 192 275 130 291 437 274 472 9 221 241 342 331 475 129 264 28 120 258 234 146 67 164 402 302 118 229 476 348 14 265 306 181 122 267 362 46200 238 411 285 306 28 232 62 322 367 59 344 327 16 445 254 97 273 330 127 208 374 442 189 187 319 457 440 176 270 383 430 443 198 Fig. 2: Grid placement #1 demonstration. c 2013 ADVANCES IN ELECTRICAL AND ELECTRONIC ENGINEERING 305 INFORMATION AND COMMUNICATION TECHNOLOGIES AND SERVICES VOLUME: 11 |NUMBER: 5 |2013 |SPECIAL ISSUE Tab. 4: Grid placement #2 parameters. Parameter Value Vertical side of grid 10 m Horizontal side of grid 10 m Flats in a building 10 Flat distribution 5 ×2 flats/building Flat dimensions 10 ×10 m Road width/height 10 m −100 −50 0 50 100 −100 −80 −60 −40 −20 0 20 40 60 80 100 Distance [m] Distance [m] 10.0010.00 14.1414.14 10.0010.00 10.0010.00 14.1414.14 10.00 14.14 10.0010.00 14.14 10.00 14.1414.14 14.1414.14 14.1414.1414.1414.14 10.00 10.00 10.0010.00 10.00 10.00 10.0010.00 10.00 14.14 10.0010.00 14.14 10.00 14.1414.1414.1414.14 10.0010.0010.00 10.0010.00 10.00 10.0010.00 10.0010.00 10.0010.0010.0010.00 14.1414.14 10.0010.0010.0010.00 10.0010.00 10.0010.0010.0010.00 10.0010.00 10.00 10.0010.00 10.00 10.0010.00 14.1414.14 10.0010.00 10.00 14.14 10.0010.0010.0010.00 14.14 10.0010.0010.00 10.00 10.0010.00 10.00 10.0010.00 10.00 14.1414.14 10.00 14.1414.1410.0014.1414.1414.1410.0014.1414.1414.14 10.0010.00 10.00 10.00 10.00 10.00 10.0010.00 10.0014.1414.1410.0010.0014.1410.0014.1414.1410.0010.0014.14 10.0010.0010.00 14.1414.14 10.00 14.1414.14 14.14 10.0010.00 14.14 10.0010.00 14.1414.14 10.0010.00 10.00 10.0010.00 10.00 10.0010.00 10.0010.0010.0010.00 59 223 19 262 461 88 20 213 190 465 321 350 53 237 443 322 203 153 475 361 156 181 272 351 196 294 466 410 184 131 400 236 225 70 48 145 471 37 126 174 308 46 12 149 457 434 349 314 295 102 339 90 428 28 155 409 58 359 226 145 289 102333 142 92 155 447 320 360 466205 145 190 352307 414 187 476 151 19 Fig. 3: Grid placement #2 demonstration. 3.3. Grid Placement #2 The third model, we have implemented for this comparison study, is grid placement #2 which is a variation on previous grid placement #1. Area covered by a MBS is composed of single-floor rectangular buildings separated by roads. Every single building consists of flats arranged into a rectangular arrangement. Flats’ and road dimensions as well as flats’ distribution in buildings are the same throughout the whole topology. All the parameters are summarized in Tab. 4. And for simplicity, when a FAP is deployed, it might be placed only in the exact middle of a flat. Figure 3 illustrates how this topology looks like. As this model is a variant of the previous one, it can be seen some similarities; however, separation of buildings by roads is very obvious at first sight. The meaning of dots, etc. is the same as in previous demonstrations. 4. Simulation Results In Fig. 4, it can be seen that the absolute number of standalone FAPs (they have no neighbours) is not very varying in individual topologies. It is even more obvious in Fig. 5 where the number of standalone FAPs is expressed in percents of the whole topology. From these figures we can assume that all the proposed topologies are very similar eventually. The only difference seems to be in visual appearance of a particular model. 200 400 600 800 1000 1200 1400 1600 0 20 40 60 80 100 120 140 160 180 200 Number of FAPs [ ] Number of standalone FAPs [ ] random grid#1 grid#2 Fig. 4: Number of standalone FAPs (absolute values). 200 400 600 800 1000 1200 1400 1600 0 10 20 30 40 50 60 Number of FAPs [ ] Number of standalone FAPs [%] random grid#1 grid#2 Fig. 5: Number of standalone FAPs (percents of the topology). Figure 6 shows overhead introduced to the network while solving PCI confusion events. The overhead is counted in messages that have to be sent. This figure indicates that confusions are more common in random placement. Such a discovery is evident because FAPs in this model might be placed almost anywhere if they do not exceed allowed overlapping. However, in both grid topologies, there are more strict rules for placing FAPs. This means they can not be so close and it eventually leads to fewer confusion events. Although grid #2 model topology experiences the fewest confusion events when only short PCI range is applied; however, with greater PCI range the differences are insignificant even when comparing to random placement as shown in Fig. 6. c 2013 ADVANCES IN ELECTRICAL AND ELECTRONIC ENGINEERING 306 INFORMATION AND COMMUNICATION TECHNOLOGIES AND SERVICES VOLUME: 11 |NUMBER: 5 |2013 |SPECIAL ISSUE 0 100 200 300 400 500 0 500 1000 1500 2000 2500 PCI range [ ] Overhead in messages [ ] random, 1500 FAPs random, 1000 FAPs grid#1, 1500 FAPs grid#1, 1000 FAPs grid#2, 1500 FAPs grid#2, 1000 FAPs Fig. 6: Overhead caused by solving confusions events. 5. Conclusion In this paper, we have proposed three different topologies for small cell network simulations dealing mainly with PCI assignment techniques. It has been shown that there are no significant differences among those models, although the complexity of particular topologies are considerable. From our simulation results, we can conclude that the easiest model to implement (random placement in this case) is the most suitable at the same time. In the future, we are going to enhance our models further. For example, by enabling FAPs to have varying radius we get closer to a reality because this option could simulate different attenuation in separated buildings. Also, having more MBSs that partially cover the same area is another way how to get more real simulation results. Acknowledgment This research work has been supported by the Grant Agency of the Czech Technical University in Prague, grant no. SGS13/199/OHK3/3T/13. References [1] CHAMBERS, D. Femtocell primer. 2nd ed. United Kingdom: Lulu Com, 2010. ISBN 978-1445744-346. [2] TERAL, S. and R. WEBB. Small cell market forecast to hit $2.7 billion by 2017. Infonetics Research [online]. 2013. Available at: http://www.infonetics.com/pr/2013/ 2H12-Small-Cell-Equipment-Market-Highlights. asp. [3] ZHANG, J. and G. DE LA ROCHE. Femtocells: Technologies and Deployment. 2nd ed. Chichester: John Wiley, 2010. ISBN 978-0-470-74298-3. [4] CURTIS, S. Virgin Media Business to offer ”small cells as a service”. Techworld [online]. 2012. Available at: http: //news.techworld.com/networking/3400674/ virgin-media-business-to-offer-small\ -cells-as-a-service/. [5] DUFFY, D. Small Cells to Make Up Almost 90% of All Base Stations by 2016. Informa Telecoms &Media [online]. 2012. Available at: http://www.smallcellforum.org/ newsstory-small-cells-to-make-up-almost\ -90-percent-of-all-base-stations-by-2016. [6] CURTIS, S. Small cells and WiFi to carry 60% of mobile traffic by 2016. Techworld [online]. 2012. Available at: http: //news.techworld.com/networking/3362145/ small-cells-and-wifi-to-carry-60-of\ -mobile-traffic-by-2016. [7] 3GPP 36.211. Evolved Universal Terrestrial Radio Access (E-UTRA); Physical channels and modulation. 3GPP, 2013. Available at: http://www. 3gpp.org/ftp/Specs/html-info/36211.htm. [8] BANDH, T., G. CARLE and H. SANNECK. Graph Coloring Based Physical-Cell-ID Assignment for LTE networks. In: Proceedings of the 2009 ACM International Wireless Communications and Mobile Computing Conference, IWCMC. New York: ACM Press, 2009, pp. 116–120. ISBN 978-160558569-7. DOI: 10.1145/1582379.1582406. [9] AMIRIJOO, M., P. FRENGER, F. GUNNARSSON, H. KALLIN, J. MOE and K. ZETTERBERG. Neighbor Cell Relation List and Physical Cell Identity Self-Organization in LTE. In: ICC Workshops ’08. IEEE International Conference on Communications Workshops, 2008. Beijing: IEEE, 2008, pp. 37–41. ISBN 978-1-42442052-0. DOI: 10.1109/ICCW.2008.12. [10] JAESEUBG, S., M. TIEJUN and P. PETER. Towards Automated Verification of Autonomous Networks: A Case Study in Self-Configuration. In: 8th IEEE International Conference on Pervasive Computing and Communications Workshops: PERCOM Workshops 2010. Mannheim: IEEE, 2010, pp. 582–587. ISBN 978-1-4244-66054. DOI: 10.1109/PERCOMW.2010.5470504. c 2013 ADVANCES IN ELECTRICAL AND ELECTRONIC ENGINEERING 307 INFORMATION AND COMMUNICATION TECHNOLOGIES AND SERVICES VOLUME: 11 |NUMBER: 5 |2013 |SPECIAL ISSUE About Authors Jan OPPOLZER received his bachelor’s (2009) and master’s (2011) degrees from the Czech Technical University in Prague, Faculty of Electrical Engineering. In 2011, he has joined the Department of Telecommunication Engineering as a Ph.D. student. His research interests include future mobile networks with small cells, Physical Cell Identifier assignment methods, Self-Organizing Networks and related topics. Robert BESTAK obtained a Ph.D. degree in Computer Science from ENST Paris, France (2003) and a MSc. degree in Telecommunications from Czech Technical University in Prague, CTU (1999). Since 2004, he has been an Assistant Professor at Department of Telecommunication Engineering, Faculty of Electrical Engineering (FEE). His research interests include radio resource management techniques in HSPA/LTE, cognitive networks and femtocells. c 2013 ADVANCES IN ELECTRICAL AND ELECTRONIC ENGINEERING 308