Cost efficient dimensioning of integrated fixed and mobile networks

Transcription

1 Cost efficient dimensioning of integrated fixed and mobile networks Tom Pallini Promotoren: prof. dr. ir. Mario Pickavet, dr. ir. Sofie Verbrugge Begeleiders: dr. ir. Bart Lannoo, dr. ir. Koen Casier Masterproef ingediend tot het behalen van de academische graad van Master in de ingenieurswetenschappen: computerwetenschappen Vakgroep Informatietechnologie Voorzitter: prof. dr. ir. Daniël De Zutter Faculteit Ingenieurswetenschappen en Architectuur Academiejaar

2 Acknowledgments I would like to take this opportunity to thank all the people supporting me in any way during the writing of this dissertation. Without them, and their support and guidance, this dissertation would not have been completed. First of all, I would like to thank my promoters Dr. Ir. Sofie Verbrugge and Prof. Dr. Ir. Mario Pickavet for giving me the opportunity to work on this interesting subject and providing me everything I needed to complete this study. I would also like to express my deepest gratitude to my advisors Dr. Ir. Bart Lannoo and Dr. Ir. Koen Casier for their great suggestions, valuable information and useful feedback. I was truly amazed by the amount of time they spent helping me and I m very grateful to them. Special thanks go out to Ir. Ger Bakker, director of Unet, who showed me around his company and gave me all the information I needed to complete my studies. Last, but by no means least, I thank my friends and family for their support and encouragement. Especially my sister Katia, who made it possible I could fully focus on my dissertation, and my girlfriend Marie for giving me endless support and for her profound understanding.

3 Authorization for loan The author gives permission to make this master dissertation available for consultation and to copy parts of this master dissertation for personal use. In the case of any other use, the limitations of the copyright have to be respected, in particular with regard to the obligation to state expressly the source when quoting results from this master dissertation. Tom Pallini, may 2012

4 Cost efficient dimensioning of integrated fixed and mobile networks by Tom Pallini Masterproef ingediend tot het behalen van de academische graad van Master in de ingenieurswetenschappen: computerwetenschappen Academiejaar Promotoren: Prof. Dr. Ir. M. PICKAVET, Dr. Ir. S. VERBRUGGE Scriptiebegeleiders: Dr. Ir. B. LANNOO, Dr. Ir. K. CASIER Faculteit Ingenieurswetenschappen en Architectuur Universiteit Gent Vakgroep Informatietechnologie Voorzitter: Prof. Dr. Ir. D. DE ZUTTER Summary In this dissertation we build a tool that computes an optimal dimensioning of a mobile network, which is interconnected by a fixed network, based on geographical information. The cost modeling of this dimensioning is also done by the tool. We perform techno-economic analyses with the tool in two distinguishing cases: a rural and an urban case. We discuss and compare two types of wireless technologies: LTE-advanced and Wi-Fi and the influence of their MIMO-configuration, their height, the requested data rate, the adoption and the considered area. The developed dimensioning tool finds a balance between the costs of the wireless network and the costs of the fixed network. These costs play an important role in the result. In both cases LTE-advanced turns out to be the best option although one should be careful with the uncertainty involving the price of the 2.6GHz-band license. A choice between Transmit Diversity and Spatial Multiplexing should be made when using MIMO. We discovered that this choice depends of the chosen wireless technology and the given scenario. Key words Dimensioning, wireless networks, fiber networks, MIMO, economic analyses

5 Cost Efficient Dimensioning of integrated fixed and mobile networks Tom Pallini Supervisors: dr. ir. Sofie Verbrugge, prof. dr. ir. Mario Pickavet, dr. ir. Bart Lannoo, dr. ir. Koen Casier Abstract In this article a comparison of wireless technologies is made by computing the dimensioning of a mobile network, along with the fixed network interconnecting its antennas, in different types of cases and comparing the costs. A dimensioning tool based on GISdata was developed to perform these calculations. Keywords Dimensioning, wireless networks, fiber networks, MIMO, economic analyses I. INTRODUCTION More and more people use their smartphones or tablets to surf on the internet and use bandwidthconsuming applications. That s why the demand of high data rates in mobile networks is increasing rapidly. These high data rates generate a lot of traffic between the base station and its backhauling connection. Therefore, there s a need to connect these base stations to an optical fiber network, which can handle very large data rates. In this paper we combine the optimal dimensioning of a wireless network with the dimensioning of the fixed fiber network which interconnects its base stations. Two different cases have been evaluated, one in an urban environment and the other in a rural area. A. LTE-advanced II. TECHNOLOGIES LTE-advanced is a mobile communication standard developed by the 3 rd Generation Partnership Project (3GPP) [1]. To satisfy the enhanced peak data rates LTE-advanced uses Carrier Aggregation to create a bandwidth of up to 100 MHz. In the downlink OFDMA is used, while in the uplink LTE-advanced uses SC- FDMA to make the user equipment less complex and more power-efficient. LTE-advanced supports MIMO. In this paper 8x2 MIMO is used. Note that a license is required when operating a LTE-advanced network. B. Wi-Fi Wi-Fi is a technology that allows devices to transfer data over a wireless network. Wi-Fi is per definition any WLAN product that is based on the IEEE standards. The n standard was released in 2009 to create a next generation Wi-Fi capable of much higher throughputs than other IEEE standards. By using 4x4 MIMO, allowing channels of up to 40 MHz, using more OFDM subcarriers and improving the coding rates, a maximum data rate of 600 Mbps is possible. Next to the higher data rates, a higher range was achieved with n. In this paper 4x2 MIMO is used as standard configuration. C. Transmit Diversity and Spatial Multiplexing Multiple Input Multiple Output (MIMO) is a technique where more than one antenna at receiver and transmitter side is used to transfer data between them. Two techniques are used in this paper to benefit from MIMO: Spatial Multiplexing (SM) and Transmit Diversity (TD). TD results in a higher range of the base station, while SM increases the capacity. D. Passive Optical Network A Passive Optical Network (PON) is a point-tomultipoint optical network architecture in which all the equipment between two endpoints of the network is passive. PONs are used in this paper to interconnect the base stations of the mobile network. III. THE DIMENSIONING TOOL We created a dimensioning tool based on GIS data that can handle many different types of requests with many different inputs. An important point was to support the gradual roll-out of the networks, which is made possible in the created dimensioning tool. The tool first calculates the required base stations for a target year and it will use this information to calculate the optimal locations for the years between the first and the target year. By using the information of the target year, only base stations and fiber eventually needed in the target year will be installed in earlier years Based on the tool, we worked out some cases, discussed in the next section, to show the power and the possibilities of the tool. A. City of Ghent IV. RESULTS We used the tool to look at the composition of the costs when deploying an LTE-advanced network in Ghent (157 km 2 ) over 5 years. As shown in Figure 1, the biggest cost is the fiber cost, which is responsible for 45% of the total cost. This cost is linearly related to the amount of fiber that is installed, so this amount should be made as small as possible. The fixed network clearly has a big impact on the total cost.

6 Costs (M ) 15% 33% 4% 3% 45% Equipment Fiber Fiber Installation BSs Maintenance BSs License Figure 1 Composition of costs case Ghent In another scenario, the center of the city of Ghent, an area of 7 km 2, is assumed as the input area, where we look at a comparison of LTE-advanced and Wi-Fi. Different scenarios per technology, all with different data rates, were investigated, 13 in total. The downlink data rate varies from 1 to 13 Mbit/s and the uplink bitrate is always one fifth of the downlink bitrate. A comparison of the costs of all these scenarios is given in Figure 2. As we see in the graph, at lower downlink bitrates LTE-advanced is the cheaper solution, but when the data rate is higher than 7 Mbit/s Wi-Fi gives us the cheapest solution. We can state that the cost of both technologies at data rates between 5 and 10 Mbit/s are more or less the same, but with lower or higher rates, the chosen technology has more impact on the total cost. Note that for LTE-advanced, a license cost of 341,818 is taken into account. 2,5 2 1,5 1 0, Bitrate download (Mbit/s) Wi-Fi LTE-advanced Figure 2 Comparison costs LTE-a vs Wi-Fi We also inspected what the achieved coverage in the different experiments was. We discovered that Wi-Fi has many problems reaching the desired coverage at high bitrates. We noticed that at data rates higher than 11 Mbit/s, the coverage of LTE-advanced also gets below the demanded coverage of 99%. We can conclude that the reached coverage of LTE-advanced is always higher or equal than the coverage of Wi-Fi. Only comparing the costs made us doubt between the technologies when the data rates are high, but comparing the reached coverage shows that at the high data rates LTE-advanced performs better. So in any case, LTE-advanced seems to be the better choice to cover the city center of Ghent, although the exact price of the license can have a big impact, but this is discussed in subsection C. B. Rural area In the second case we tested the tool on a more rural area of 35 km 2 in Flevoland. We connected all farmers in a given area to provide a triple-play package, resulting in 25 Mbit/s downlink per house, and compared different types of technologies to make this connection. Coverage (at a lower bitrate) of the mobile network is also provided on the fields of the farmers. We also inspected the influence of SM and TD Table 1 gives an overview of the results of the different scenarios. It is quite clear that LTE-advanced (LTE-a) with SM is the best technical choice. It outperforms all other technologies. Not only the Capex is lower, but also the yearly Opex is better than in the other scenarios. Note that in this case the license cost is estimated at 766. Table 1 Comparison scenarios rural case LTE-a SM LTE-a TD Wi-Fi SM # BSs Km fiber Wi-Fi TD Capex 409, ,576 1,820,789 1,328,672 Opex 21,000 42,000 44,200 30,550 We can also conclude that SM and TD have a big impact on the performance of the wireless networks. A basic rule is that areas with a high demand of data transfer, due to the high density or the demanded data rate, need SM to increase the ranges of the antennas. While in areas where the demand of data is lower, TD seems a good method to increase the ranges. We discovered that the choice between the two depends on the area and the situation, but also on the chosen technology. C. Influence of license cost and adoption We estimated the license fee for the city of Ghent at 341,818. In a pessimistic view, we assume that the license can cost 1,504,000, while a best-case scenario gives us a value of 113,939.The worst-case scenario results in an increase of 15.5% of the total cost, and the best-case scenario is 3% lower than the reference scenario. It is clear that the license cost has a big influence on the final cost. We also discovered that the adoption has a great impact on the resulting cost. The total cost under a lower adoption is 32% lower, and it is 27% higher under a higher adoption. Spending some money in making a good prediction of the adoption is definitely a good choice. V. CONCLUSION This article gives a good view on how important the fiber network is when dimensioning a fixed-wireless network and shows that one should consider different technologies and different settings of them. It also shows the influences of the license fees and the predicted adoption when deploying such a network. REFERENCES [1] ITU, Framework and overall objectives of the future development of imt-2000 and systems beyond imt-2000, ITU Recommendation ITU-R M.1645, [2] P. E. Mogensen et al., Lte-advanced: The path towards gigabit/s in wireless mobile communications, Wireless Communication, pp , 2009.

7 Kostenefficiënte dimensionering van geïntegreerde vaste en mobiele netwerken Tom Pallini Begeleiders: dr. ir. Sofie Verbrugge, prof. dr. ir. Mario Pickavet, dr. ir. Bart Lannoo, dr. ir. Koen Casier Abstract In dit artikel worden er verschillende draadloze technologieën vergeleken door de dimensionering van een mobiel netwerk, waarvan de antennes verbonden zijn door een vast netwerk, te berekenen in verschillende soorten gebieden en hun kosten te vergelijken. Er werd hiervoor een dimenioneringstool gemaakt dat werkt op basis van GIS-data. Trefwoorden Dimensionering, draadloze netwerken, glasvezel netwerken, MIMO, economische analyses I. INTRODUCTIE Meer en meer mensen gebruiken hun smartphones of tablets om op het internet te surfen en belastende applicaties te gebruiken. Daarom is de vraag naar hoge data-rates in mobiele netwerken snel aan het stijgen. Deze hoge data-rates genereren veel traffiek tussen de antennes en hun backhaul-connectie. Daardoor is er een nood om deze antennes te verbinden aan een optisch glasvezel-netwerk dat hele hoge data-rates aankan. In deze paper combineren we de optimale dimensionering van een draadloos netwerk met de dimensionering van een vast netwerk dat diens antennes interconnecteert. A. LTE-advanced II. TECHNOLOGIES LTE-advanced is een mobiele communicatie standaard ontwikkeld door het 3GPP [1]. LTEadvanced gebruikt Carrier Aggregation om een bandbreedte van maximaal 100 MHz te creëeren. In de downlink wordt er OFDMA gebruikt en in de uplink gebruikt LTE-advanced SC-FDMA om de apparatuur van de gebruiker minder complex en energie-zuiniger te maken. LTE-advanced ondersteunt MIMO en in deze paper wordt er een 8x2 MIMO-configuratie gebruikt. Merk op dat er een licentie nodig is wanneer men een LTE-advanced netwerk opereert. B. Wi-Fi Wi-Fi is een technologie dat apparaten toelaat om data te transfereren over een draadloos netwerk. Wi-Fi is per definitie elk WLAN-product dat gebaseerd is op de IEEE standaarden. De n standaard is gelanceerd in 2009 met als doel een next-generation Wi-Fi te maken dat veel hogere data hoeveelheden aankan dan eerdere standaarden. Door 4x4 MIMO te gebruiken, kanalen van maximaal 40MHz toe te laten, meer OFDM subcarriers te gebruiken en de codeer-rates te verbeteren, is er een maximale data-rate van 600 Mbps mogelijk. In deze paper zullen we voor Wi-Fi een 4x2 MIMO-configuratie gebruiken. C. Transmit Diversity en Spatial Multiplexing Multiple Input Multiple Output (MIMO) is een techniek waarbij meer dan één antenne bij de ontvanger en de zender wordt gebruikt om data te transfereren. Twee technieken worden gebruikt in deze paper om voordelen te halen uit MIMO: Spatial Multiplexing (SM) en Transmit Diversity (TD). TD resulteert in een hoger bereik van het basis station en SM verhoogt de capaciteit. D. Passief Optisch Netwerk Een Passief Optisch Netwerk (PON) is een punt-totmultipunt optisch netwerk architectuur waarbij al de apparatuur tussen twee eindpunten van het netwerk passief is. PONs worden in deze paper gebruikt om de antennes van het netwerk met elkaar te verbinden. III. DE DIMENSIONERINGSTOOL We hebben een dimensioneringstool gemaakt, gebaseerd op GIS-data, dat verschillende soorten scenarios kan verwerken. Een belangrijk punt was om een graduele uitrol van het netwerk te ondersteunen. De tool berekent eerst het nodige aantal basis stations voor het doeljaar en gebruikt dit om de optimale locaties te bepalen voor de jaren ervoor. Door de informatie van het doeljaar te gebruiken, zullen enkel basis stations en glasvezel dat uiteindelijk nodig is in het doeljaar, geïnstalleerd worden in de eerdere jaren. M.b.v. deze tool werkten we enkele cases uit om de mogelijkheden van de tool aan te tonen. Deze cases en hun evaluatie zijn weergegeven in de volgende sectie. A. Stad Gent IV. RESULTATEN We gebruikten de tool om de samenstelling van de kosten te bekijken wanneer we een LTE-advanced netwerk in Gent (157 km 2 ) uitrollen over 5 jaar. Zoals te zien op Figuur 1 is de glasvezelkost goed voor 45% van de totale kos, wat veel is. Deze fiber-kost heeft een lineair verband met het aantal km glasvezel dat geïnstalleerd is, dus voor een optimale kost moet deze afstand zou zo klein mogelijk zijn. Het vaste netwerk heeft dus duidelijk een grote invloed op de totale kost.

8 Costs (M ) 15% 33% 4% 3% 45% Equipment Fiber Fiber Installation BSs Maintenance BSs License Figuur 1 Samenstelling van kosten case Gent In een ander scenario nemen we het centrum van Gent (7km 2 ) als input en we vergelijken er twee draadloze technologieën: LTE-advanced en Wi-Fi. We onderzochten 13 verschillende scenarios per technologie, elk met een andere data-rate. De downlink data-rate laten we variëren van 1 tot 13 Mbit/s, waarbij de uplink data-rate steeds een vijfde daarvan bedraagt. Figuur 2 geeft een vergelijking van de kosten van al deze scenarios. We zien in de grafiek dat bij lagere bitrates LTE-advanced de goedkopere oplossing is, maar wanneer de data-rate hoger dan 7 Mbit/s is, dan geeft Wi-Fi ons de goedkoopste oplossing. We kunnen zeggen dat tussen de 5 en de 10 Mbit/s de kosten van beide technologieën ongeveer gelijk zijn, maar bij hogere of lagere data-rates heeft de gekozen technologie meer impact op de totale kost. Merk op dat voor LTEadvanced een licentie-kost van meegerekend werd. 2,5 2 1,5 1 0, Bitrate download (Mbit/s) Wi-Fi LTE-advanced Figuur 2 Vergelijking kosten LTE-a en Wi-Fi We onderzochten ook wat de bereikte dekkingsgraad was in de verschillende experimenten. We ontdekten dat Wi-Fi veel meer problemen heeft om de vereiste dekkingsgraad van 99% te behalen bij hoge data-rates. We merkten dat bij data-rates hoger dan 11 Mbit/s de dekking van LTE-advanced ook onder de gewenste dekkinsgraad gaat. We concluderen wel dat de bereikte dekking van LTE-advanced steeds hoger of gelijk aan de bereikte dekking van Wi-Fi is. Door enkel de kosten te vergelijken hadden we twijfels over de keuze van de technologie bij hoge datarates, maar wanneer we ook de dekkingsgraad in rekening brengen, zien we dat LTE-advanced eigenlijk beter presteert bij die hoge data-rates. We kunnens dus stellen dat LTE-advanced steeds de betere optie lijkt om het centrum van Gent te dekken, al kan de exacte prijs van de licentie hier een grote invloed op hebben, maar dit wordt uitgelegd in sectie C. B. Landelijk gebied In een tweede case hebben we de tool gebruikt om een landelijk gebied in Flevoland uit te werken. We hebben alle boeren in een gegeven gebied geconnecteerd om zo een triple-play pakket aan te bieden, wat resulteert in 25 Mbit/s downlink per huis, en verschillende technologieën vergeleken om deze connectie te maken. Er wordt ook dekking (aan een lagere bitrate) voorzien op de velden van de boeren. We onderzochten ook de invloed van SM en TD. Tabel 1 geeft een overzicht van de resultaten van de verschillende scenarios. Het is duidelijk dat LTEadvanced (LTE-a) met SM de beste optie is. Niet alleen de Capex is lager, maar ook de jaarlijkse Opex ligt lager dan in de andere scenarios. Merk op dat in dit geval de licentie-kost geschat is op 766. Tabel 1 Vergelijking scenarios landelijke case LTE-a SM LTE-a TD Wi-Fi SM Wi-Fi TD # BSs Km fiber 4,5 6 36,8 28,4 Capex Opex We kunnen ook concluderen dat SM en TD een grote impact hebben op prestatie van de draadloze netwerken. Een basisregel is dat gebieden met een hoge vraag aan data-uitwisseling, door de hoge dichtheid of de vereiste data-rate, SM nodig hebben om het bereik van de antennes te vergroten. Terwijl dat wanneer deze vraag lager ligt, TD een goede methode lijkt om het bereik te vergroten. We ontdektden dat de keuze tussen SM en TD afhangt van het gebied en de situatie, maar ook van de gekozen technologie. C. Invloed van de licentie-kost en de adoptie We hebben de licentie-kost voor Gent geschat op In een pessimistische kijk, veronderstellen we dat de licentie kan kosten en in het beste geval kost ze Het slechtste geval resulteert in een stijging van 15,5% van de totale kost en het beste geval ligt 3% lager. Het is duidelijk dat de licentie-kost een grote invloed kan hebben op de totale kost. We ontdekten ook dat de adoptie een grote invloed heeft op de resulterende kosten. De totale kost bij een lagere adoptie ligt 32% lager en ze is 27% hoger bij een hogere adoptie. Betalen voor een goede voorspelling van de adoptie is aan te raden, zo kan men onnodige kosten vermijden, of het netwerk beter aanpassen aan de vraag. V. CONCLUSIES Dit artikel biedt een goede kijk op het belang van glasvezel-netwerken wanneer men een vast-mobiel netwerk dimensioneert en toont dat men zeker ook verschillende technologieën en instellingen moet beschouwen. Het geeft tevens het belang van de licentie-kost en de voorspelde adoptie aan. REFERENTIES [1] ITU, Framework and overall objectives of the future development of imt-2000 and systems beyond imt-2000, ITU Recommendation ITU-R M.1645, [2] P. E. Mogensen et al., Lte-advanced: The path towards gigabit/s in wireless mobile communications, Wireless Communication, pp , 2009.

9 Table of Contents 1. Introduction Background Technology Wireless techniques Wireless technologies Optical Fiber GIS and Shapefiles Dimensioning Technical modeling of the wireless technology Genetic algorithm Minimum spanning tree Economic Prices and costs Adoption Investment analysis The dimensioning tool Wireless Tool User data and GIS data Technical parameters and technical modeling Costs, other inputs and cost modeling Dimensioning TESS framework The area module The costing module The evaluation module Importance to master thesis The wireless-fixed dimensioning tool Input Core of the tool... 31

10 3.3.3 Output Optimizations Speed Splitting the target area Downside of the density coverage Conclusion Integration of fixed and mobile networks Urban case Input City of Ghent Wi-Fi vs. LTE-advanced Conclusion Rural case Input Comparison of technologies Conclusion Refined economic analyses Influence of expected adoption Influence of cost of extra MIMO antennas Influence of license cost Conclusion and future work... 64

11 Table of abbreviations 4G AON ASK BIPT BPSK BS CA Capex CoMP CP DL EDGE FDD FDMA FTTB FTTB GIS GSM IBCN IEEE IMT IPTV IRR ITU LTE LTE-a MIMO MISO MST MU-MIMO NPV ODF OFDMA OLT ONT ONU Opex PAPR PON PSK 4 th generation mobile telecommunications Active Optical Network Amplitude Shift Keying Belgian Institute for Postal Services and Telecommunications Binary Phase Shift Keying Base Station Carrier Aggregation Capital Expenditures Coordinated Multipoint Cyclic Prefix Downlink Enhanced Data rates for GSM Evolution Frequency Division Duplex Frequency Division Multiple Access Fiber To The Business Fiber To The Business Geographic Information System Global System for Mobile Communication Internet Based Communication Networks and Services Institute of Electrical and Electronics Engineers International Mobile Telecommunications Internet Protocol TeleVision Internal Rate of Return International Telecommunication Union Long Term Evolution LTE-advanced Multiple Input Multiple Output Multiple Input Single Output Minimum Spanning Tree Multi User-Multiple Input Multiple Output Net Present Value Optical Distribution Frame Orthogonal Frequency Division Multiple Access Optical Line Terminal Optical Network Terminal Optical Network Unit Operational Expenditures Peak-to-Average Power Ratio Passive Optical Network Phase Shift Keying

12 QAM QPSK SC-FDMA SIMO SISO SM SU-MIMO TD TDD TDM TDMA TESS TESS UE UL UMTS VOIP VPN WDM WLAN XDSL Quadrature Amplitude Modulation Quadrature Phase Shift Keying Single Carrier-Frequency Division Multiple Access Single Input Multiple Output Single Input Single Output Spatial Multiplexing Single User-Multiple Input Multiple Output Transmit Diversity Time Division Duplex Time Division Multiplexing Time Division Multiple Access Transactional Environmental Support System Techno Economic Software Suite User Equipment Uplink Universal Mobile Telecommunication System Voice Over IP Virtual Private Network Wavelength Division Multiplexing Wireless Local Area Network (any type of) Digital Subscriber Line

13 1 1. Introduction Wireless technologies are evolving fast the latest years and next generation wireless networks are being deployed all over the world. More and more people use their smartphones or tablets to surf on the internet and use bandwidth-consuming applications, e.g. to watch video files online. Many people would like to be able to do this anywhere, at any time. That s why the demand of high data rates in mobile networks is increasing rapidly. Nowadays, we already see deployments of fourth generation (4G) networks, which are able to cope with those high demands. In the future even networks of higher generations will be deployed that have higher data rate capacities. New technologies also allow the ranges of the base stations of these wireless networks to be larger than before. This means that more people can be served with just one base station. Giving each of the people in this coverage area a high data rate generates a lot of traffic between the base station and its backhauling connection. Therefore, there s a need to connect these base stations to an optical fiber network, which can handle very large data rates. Fiber networks are also becoming popular because they can deliver a very high bandwidth. Installing a fiber infrastructure to support the base stations has the advantage that it will certainly meet the data rate requirements of the current wireless networks as well as of future generations. A typical approach in network planning starts from an optimal placement of the wireless access points followed by a separate planning in which they will look to connect the base stations to a fixed network. The cost of this fixed connection is often underestimated. In this master thesis we combine the optimal dimensioning of a mobile network with the dimensioning of the fixed fiber network which interconnects the base stations. Because the demands and the usage of a network grow every year, a gradual roll out of the network should be possible. Based on predictions of the amount of customers and the data rate and coverage demands of every year, we aim to find the most cost efficient roll-out of the wireless and fixed networks. A tool was developed that is able to dimension wireless-fixed networks and perform a detailed cost modeling of this dimensioning. We made this tool as flexible and generic as possible, so it can be used in many different scenarios and under many different technologies for both the fixed and wireless equipment. After the development of such a tool some exemplary cases are set out to show the power and the possibilities of the tool. Two different cases have been evaluated, one in an urban environment and the other in a rural area. The tool offers solutions the placement of the antennas and fiber and the costs of the dimensioning of these areas under different technologies and perform some refined economic analyses on these results. The results of our study show that the optimal dimensioning depends on different factors, the most important one being the chosen technology, which has a great impact on the ranges of the antennas, but also on the final cost of the network. The choice of this technology is influenced greatly by the license costs needed when operating in a licensed band. Other factors, such as the MIMOconfiguration and the expected adoption of the network influence the cost of the network and should be studied before deploying a fixed-mobile network.

14 2 2. Background Dimensioning both wireless and fiber networks requires knowledge of both topics. First, an overview of the technology background is given, starting with the information of fourth generation (4G) wireless technologies. As the fixed network is an optical fiber network, background on fiber networks is added. Both networks are dimensioned in a different way and section 2.2 discusses the algorithms and techniques needed to dimension them. Finally, the cost modeling of the dimensioned network is also an important part since cost efficiency is desired when rolling out networks. In section 2.3 we give the required definitions and information of some investment analysis terms used throughout this dissertation. 2.1 Technology The chapter of the technology background consists of four sections. Section discusses the techniques that will be used in the newest wireless technologies and those technologies are given in section The base stations of the wireless networks under these technologies will be interconnected using optical networks. Information about techniques and latest developments of optical networks is given in section The final section takes a quick look at the usage of Geographical Information Systems used in the tool of this master thesis Wireless techniques The main goal of this thesis is to dimension a wireless network. Many different wireless technologies exist and many new ones will arise in the future, each having their own characteristics. The choice of the technology might be crucial when deploying a network. Requirements of such networks grow pretty fast in time and choosing a technology that can cope with these growing demands is important. This section sets out the techniques used in the different wireless technologies that are deployed in the cases discussed later in the dissertation. Duplexing In mobile communication there is an explicit distinction in the two directions of communication: from the base station (BS) to the user, the downlink (DL), and the other way around, which is called the uplink (UL). Providing this two-way communication is called duplexing and the following two techniques can be used for it: Time Division Duplex (TDD) [1] will use the same frequency band in the DL and the UL, but assigns different timeslots to both directions as shown in Figure 2.1. Between the UL and DL timeslots, a guard interval is implemented to prevent UL and DL interference. This is indicated as GP (Guard Period) in the close-up of the bandwidth allocation on the figure.

15 3 Figure 2.1: Time Division Duplex [2] Frequency Division Duplex (FDD) [1] will use two frequency channels to communicate between user and BS. Figure 2.2 clearly shows that one frequency band will be assigned to the DL and one to the UL. A guard band separates both bands to prevent interference between UL and DL. This is indicated in the figure as the duplex frequency distance. Figure 2.2: Frequency Division Duplex [2] Modulation scheme Digital modulation is the process of translating digital data to an analogue signal. This process is necessary if the digital data needs to be transferred over a medium that only allows analogue transmission, e.g. a wireless medium [1]. Phase Shift Keying (PSK) uses different shifts of the phase of the signal to represent the data. Shifting over 180 is called Binary PSK (BPSK), which is shown in Figure 2.3 (a). Only 1 bit is coded per shift. Figure 2.3 (b) shows Quadrature PSK (QPSK), where higher bitrates can be achieved because 2 bits are coded per phase shift, but the more symbols used in the phase domain, the harder it will get to separate them and decoding errors are more likely to occur after the analogue transmission. Another modulation scheme is Quadrature Amplitude Modulation (QAM), which is a multilevel coding that uses both phase shifts and differences in amplitude to code the data [1]. Two examples of QAM are shown in Figure 2.3, where (c) gives 16-QAM using a 4-bit multilevel coding and (d) shows 64-QAM, which implements a 6-bit multilevel coding. Here too, a higher bit-level corresponds with a higher bitrate, but is more sensitive to modifications of the analogue signal due to transmission.

16 4 Figure 2.3: Modulation schemes Multiple Access Different ways exist to share the wireless medium amongst its users. In some current technologies Code Division Multiple Access (CDMA) is used, which uses codes with certain characteristics to separate different users in code space and enable access to a shared medium without interference [1]. In modern technologies following two techniques are used: Orthogonal Frequency Division Multiple Access (OFDMA) creates slots in the time-frequency space and assigns users to a different number of sub-carriers as shown in Figure 2.4. An advantage over CDMA is that it can handle multipath interference with more robustness and less complexity at the user side. If there is less complexity in the receiver, it will become cheaper and will need less power. OFDMA will also achieve a higher MIMO spectral efficiency (see further). Figure 2.4: OFDMA Single Carrier-Frequency Division Multiple Access (SC-FDMA) is an adapted form of OFDMA [3] and their difference is explained in Figure 2.5 [4]. On the left side, N (in this case 4) adjacent 15 khz subcarriers are each modulated the OFDMA symbol period by one QPSK data symbol. After one OFDMA symbols, holding 4 QPSK symbols, a Cyclic Prefix (CP) or guard period is inserted to make sure there s no interference between the symbols and the next four symbols are transmitted in parallel. SC-FDMA signal generation begins with a special linear pre-coding process but then continues as OFDMA. The right side of the figure shows that SC-FDMA transmits the four data symbols in series at 4 times the rate, with each symbol occupying N times 15 khz bandwidth. As the figure shows, SC-FDMA is clearly a single carrier (which explains the SC prefix) and OFDMA is a multi-

17 5 carrier. Note that the SC-FFMA symbol length has the same length as the OFDMA symbols, but it contains N sub-symbols. The complexity of SC-FDMA is comparable to OFDMA, but the peak-to-average power ratio (PAPR) of the transmitted signal in OFDMA is higher due to parallel transmission. This means that in SC-FDMA the transmit power is more efficient, resulting in a longer battery life. SC-FDMA is an attractive alternative to OFDM, especially in uplink communication where lower PAPR greatly benefits the transmit power efficiency and costs of the mobile terminal. Figure 2.5: OFDMA and SC-FDMA [5] Carrier Aggregation Carrier Aggregation (CA) allows to aggregate two or more component carriers of the same or different bandwidth into a carrier with a larger spectrum [6]. In Figure 2.6 we see that 5 carriers are aggregated to create a larger carrier with a higher maximum bandwidth. Subcarriers of different bands or consecutive subcarriers of the same band can be aggregated to form a bigger carrier. This way, a larger bandwidth can be used in UL and DL. This technique is used in modern wireless technologies, e.g. LTE-advanced (see section 2.1.2). Figure 2.6: Carrier Aggregation [7]

18 6 MIMO A key feature in 4G wireless communication is the use of multiple antennas on the transmitter and the receiver. Multiple Input Multiple Output (MIMO) is a technique where more than one antenna at receiver and transmitter side is used to transfer data between them (see Figure 2.7). This benefits the spectral efficiency greatly. Using only one antenna at both sides is called Single Input Single Output (SISO). Other variants like Single Input Multiple Output (SIMO) and Multiple Input and Single Output (MISO) or also possible, but in this dissertation we will focus on MIMO. The notation of a MIMO configuration with Y antennas on the transmitter and Z antennas at the receiver is: YxZ MIMO. The antennas in MIMO can be used in different ways [6]. A distinction has to be made about the amount of users using the antennas at a given time. Single User-MIMO (SU-MIMO) assigns the timefrequency resources to one user only. This user can now reach a maximal spectral efficiency. Multi User-MIMO (MU-MIMO) allocates different users in the same time-frequency resource. This way, more users can transfer data at the same time. Different techniques exist to benefit from MIMO, Spatial Multiplexing (SM) and Transmit Diversity (TD) being two common techniques explained further. Figure 2.7: SISO, SIMO, MISO and MIMO [8] TD is a technique that uses the multiple antennas to send replicas of a given data stream [9]. As shown in Figure 2.8, all antennas will send the same stream, but coded differently across space and time, and the receiver can now make a better decision by combining the received signals. This results in a higher range of the BS. Figure 2.8: Transmit Diversity [10] SM will split a given data stream into pieces and will send each fragment with a different antenna as shown in Figure 2.9. Each data stream is now independent of each other. This increases the capacity of the MIMO-link.

19 7 Figure 2.9: Spatial Multiplexing [10] In Spatial Multiplexing, the maximum data rate of the normal configuration will be multiplied by the number of antennas that can transmit at the same time. This in contrast to Transmit diversity, which doesn t change the data rate but adds positive input gains to the antenna, resulting in a higher range Wireless technologies Nowadays there are many different wireless technologies. We are now at breakthrough of the fourth generation (4G) of telecommunication systems. Figure 2.10 gives an overview of the generation of telecommunication systems that have already been deployed. The first generation (1G) was a collection of analogue systems. Of the second generation (2G) systems the Global System for Mobile Communications (GSM) standard is the most used one. Nowadays many third generation (3G) networks are rolled out. Enhanced Data Rate for GSM Evolution (EDGE) and Universal Mobile Telecommunication System (UMTS) are the most known examples of 3G systems. In this section we will point out the requirements for the 4G systems as defined by IMT-advanced, and discuss LTEadvanced, a 4G standard. We also take a look at the Wi-Fi standard, which matches the 4G data rates. Figure 2.10: Overview of generations of telecommunication systems [11]

20 8 IMT-advanced International Mobile Telecommunications (IMT)-advanced is a global framework made by the International Telecommunication Union (ITU) to set the requirements for a 4G system. Its predecessor is IMT-2000, a framework for 3G systems. Figure 2.11 shows the goals for the successor of IMT IMT-advanced aims to achieve a higher mobility and higher data rate than IMT Figure 2.11: IMT-2000 and IMT-advanced [12] The most important properties of IMT-advanced are [13]: a high degree of commonality of functionality worldwide while retaining the flexibility to support a wide range of services and applications in a cost efficient manner compatibility of services within IMT and with fixed networks capability of interworking with other radio access systems high quality mobile services user equipment suitable for worldwide use user-friendly applications, services and equipment worldwide roaming capability enhanced peak data rates to support advanced services and applications (100 Mbit/s for high and 1 Gbit/s for low mobility were established as targets for research) These features enable IMT-Advanced to address evolving user needs and the capabilities of IMT- Advanced systems are being continuously enhanced in line with user trends and technology developments. LTE-advanced One technique that meets the IMT-advanced requirements is Long Term Evolution (LTE) advanced [14]. LTE-advanced is a mobile communication standard developed by the 3 rd Generation Partnership

21 9 Project (3GPP). An important demand was that LTE-advanced would remain compatible with LTE, a 3G standard that meets the demands of the IMT-2000 framework. To satisfy the enhanced peak data rates LTE-advanced uses Carrier Aggregation to have a bandwidth of up to 100 MHz. Note that using CA increases the complexity of the User Equipment (UE). In the downlink OFDMA is used, while in the uplink LTE-advanced uses SC-FDMA to make the UE less complex and more power-efficient. LTEadvanced supports MIMO of up to 8x8 in the DL and 4x4 in the UL. Coordinated Multipoint (CoMP) transmission and reception techniques are also introduced. CoMP techniques try to decrease the interference by communicating through the backhaul network between the Base Stations, which increases the performance [15]. To enhance the coverage in difficult conditions, such as big buildings or the UE being indoor, LTE-advanced uses relaying. A relay node is connected to the Base Station through a fixed or wireless link, while the user connects wirelessly to the relay node. Figure 2.12 shows a relay situation with a wireless link between the base station and the relay node. Figure 2.12: Relaying [7] Wi-Fi Wi-Fi is a technology that allows devices to transfer data over a wireless network. Wi-Fi is per definition any Wireless Local Area Network (WLAN) product that is based on the IEEE (Institute of Electrical and Electronics Engineers) standards. IEEE is a set of standards for implementing WLAN communication in the 2.4, 3.6 and 5 GHz frequency bands, which are licensefree. The first standard, , dates from 1997 and was able to reach 2 Mbps. Many releases of the standard exist, some much more popular than others. Following releases are the most common ones. The first one that was widely accepted was b, which was released in 1999 along with a. The first operated in the 2.4 GHz band and could achieve 11 Mbps, the latter in the 5GHz band with a maximum net rate of about 20 Mbps. In g was released to reach data rates of up to 54 Mbps n was released in 2009 by IEEE to create a next generation Wi-Fi capable of much higher throughputs than other IEEE standards. By using 4x4 MIMO, allowing channels of up to 40 MHz, using more OFDM subcarriers and improving the coding rates, a maximum data rate of 600 Mbps is possible. Next to the higher data rates, a bigger range was achieved with n. More next generation releases are being developed and will be released in the future.

22 Optical Fiber The goal of this thesis is to interconnect the wireless network with a fiber-optical network. Two types of optical access networks can be distinguished: Active Optical Networks (AONs) and Passive Optical Networks (PONs). An AON uses electrically powered switching equipment to manage signal distribution and direct signals to specific customers. A PON, on the other hand, uses optical splitters to separate and collect the optical signals as they move through the network and does not contain electrically powered equipment. PONs are more efficient than AONs because each fiber optic strand can serve up to 32 users. They also have a low building cost relative to AONs along with lower maintenance costs. That s why we chose PONs over AONs as the backhauling networks in this dissertation. This section will explain how a PON works and will set out the different types of PONs. The equipment needed for a PON will also be listed and discussed in this section. Passive Optical Network A Passive Optical Network (PON) (Figure 2.13) is a point-to-multipoint optical network architecture in which all the equipment between two endpoints of the network are passive. A single optical fiber can serve multiple premises by using unpowered optical splitters. Different types of PONs can be distinguished according to the Division Multiplexing method. A common type is Time Division Multiplexing PON (TDM-PON), which assigns a time-slot to each subscriber. Wavelength Division Multiplexing PON (WDM-PON) (Figure 2.14) is a newer technique where each subscriber is assigned a different wavelength [16]. The advantage of this method is that WDM-PON can be seen as an aggregation of per wavelength point-to-point connections between each subscriber and the central office. Figure 2.13: Passive Optical Network

23 11 Figure 2.14: WDM-PON [17] Figure 2.15 shows the equipment model used in this dissertation. The model shows the central office and all the equipment needed to connect one fiber to this central office with a PON. Assume the X at the splitter in the figure is equal to 32, which is a common configuration for a PON splitter. The most important components of the model are explained underneath. Figure 2.15: Equipment model of a PON

24 12 Optical Line Termination An Optical Line Termination (OLT) is a device which serves as the service provider endpoint of a PON. It has two main functions: It performs conversion between the incoming signals received from the service provider s equipment to the signals used in the optical network. It coordinates the multiplexing between the Optical Network Units (ONUs). Optical Network Unit and Optical Network Terminal An Optical Network Unit (ONU) performs conversion between the incoming optical signal from an optical network and the in-house cabling at the customer s premises. An Optical Network Terminal (ONT) is a special case of an ONU that serves a single subscriber, instead of multiple subscribers. Optical Distribution Frame An Optical Distribution Frame (ODF) is a fiber optic management unit used to organize the fiber optic cable connections. An ODF exists of an ODF rack and several ODF slots GIS and Shapefiles A Geographic Information System (GIS) is a system designed to capture, store, manipulate, analyze, manage and present all types of geographical data. A shapefile, short for Esri 1 shapefile, is a geospatial vector data format for GIS software. Shapefiles store primitive geometrical data types of points, polylines and polygons. These three data types are used to represent many different geographic elements, points for example can represent water wells, while a river can be shown as a line and the polygon could be a lake or an ocean. Each data type can also have a series of attributes that describe the items, e.g. the name. In this thesis, shapefiles are used to represent possible locations of antennas and fiber and geographical data that contain attributes, e.g. demographic data. These shapefiles will be used in the dimensioning tool discussed in chapter 3 and the different files used in this study will be discussed in chapter 4 when introducing the different cases. 2.2 Dimensioning An important step in deploying a network is dimensioning it: making the decision where to place the components of a network. How this works depends on the type of network and the type of its components. In this thesis we have to dimension a wireless network in combination with an optical fiber network. The wireless dimensioning consists of two steps: first a technical modeling is performed to determine the data rates and the ranges of the antennas and then the optimal locations of the base stations have to be found. Section discusses the techniques used for the technical modeling. To find the optimal locations of the base stations this master thesis uses a heuristic algorithm to select an optimal solution from a set of possible locations. This heuristic algorithm is explained in section Esri is a software development and services company providing GIS software and geodatabase management applications. Esri developed and regulates the shapefile format as an open standard for data interoperability.

25 13 The fiber dimensioning in this master thesis depends upon these optimal locations of the antennas. Given these locations, the optimal interconnection is searched. This is the interconnection that has the minimal total cost, which is linearly linked to the distance and can be optimally calculated with a Steiner Tree. In this dissertation, we used an approximation of the Steiner Tree by using the Minimal Spanning Tree. Section explains the concept of a Minimal Spanning Tree and gives an algorithm to calculate it Technical modeling of the wireless technology The technical modeling of the wireless technology consists of two parts: finding the possible data rates and determining the maximum range of the antennas. The possible data rates for every modulation scheme of the given technology are used to determine the number of users that one base station can cover. Finding the maximum range of a base station consists of two steps: first a link budget is calculated and then the maximum range can be found based on this link budget and a path loss model. Bitrate The bitrate depends of the properties of the used technology. The formula differs for TDD and FDD: when using TDD an extra factor the DL:UL ratio has to be added. The formula [18] is given underneath. The code rate refers to the amount of information located in the data stream. A code rate of k/n means that for every k useful bits (holding information), n bits are generated by the coder, resulting in n-k redundant bits. A lower code rate is more resistant to errors due to the higher number of redundant bits and thereby results in higher ranges. M is the number of symbols in the modulation scheme, e.g. for 16-QAM, M is 16, and for QPSK, M is 4. ( ) ( ) With BW the bandwidth in hertz n the sampling factor N data N FTT k/n M G the number of available subcarriers the total number of subcarriers code rate number of symbols in modulation scheme the guard interval Link budget The first step in calculating the maximum range is computing the link budget. In the link budget the gains and losses of the transmitter, through the medium to the receiver are taken into account. The maximum allowed path loss PL MAX that a signal can suffer and will still be detectable by its receiver will be calculated. The path loss is given in the formula underneath. It sets out the relationship between the emitted power and the received power of the signal.

26 14 The input power, gains, losses and margins are input parameters that are known when selecting a certain technology and a certain configuration of it. These parameters are discussed later when their values will be used. The receiver sensitivity is a combination of the thermal noise, receiver noise and receiver implementation loss. The receiver sensitivity is different for each modulation scheme and it s the reason why higher modulation schemes achieve lower ranges. When using Transmit Diversity, the ranges are increased, due to a higher PL MAX. Two extra gains are added to the link budget: a cyclic combining gain of 3dB and a MIMO gain. The MIMO gain depends from the amount of transmitter and receiver antennas and its formula is given underneath. ( ) Path loss model The PL MAX is used to determine the ranges of the antennas by using a path loss model. A path loss model defines the relationship between the range of an antenna and the path loss. This thesis will use the Erceg model [19], which has different types of configurations, each responsible for a different terrain type. The formula for the Erceg model is given underneath [20]. Three different types of terrain categories are defined in this model: A, B and C. Category A is developed for a hilly terrain or moderate-to-heavy tree density, while Erceg C models a mostly flat terrain with light tree densities and category B is somewhere in between these two. The difference in these categories is determined by the three parameters a, b and c and there values for each different category is given in Table 2.1. In this dissertation we only use the Erceg C model, but we use two different values for AF (the correction factor of the environment).when calculating the path loss in an urban area, AF is set to 3dB, while in rural areas AF is equal to -5dB. With h BS the wavelength (in meter) the path loss variable: = (a b x h BS + c/h BS ) the height of the Base Station a, b, c constants that represent the terrain type d the distance between the user and the Base Station (in meter) d 0 f h m AF s = 100m the frequency (in MHz) the height of the Mobile Station (in meter) the correction factor of the environment the shadowing fading component that takes into account the signal variations between different locations on an equal distance of the transmitter Category A Category B Category C a b c Table 2.1: Numerical values of Erceg model parameters [19]

27 Genetic algorithm To dimension the wireless network we need to select a number of locations from a collection of possible antenna locations that will result in a network that meets the demands when installing an antenna on the selected locations. In many cases there are many possible antenna locations and the number of antennas needed to cover a certain area is unknown. Computing every possible solution and comparing the results will therefore be a very time-consuming assignment (e.g. for selecting 10 antenna locations from 20 possible locations already leads to more than possibilities to be checked). A better way to approach this problem is using a heuristic algorithm, which won t run over every possible solution but will try and find an optimal but not necessarily the best solution in a smart way. In this thesis we use the genetic algorithm to find a good solution to our dimensioning problem. A genetic algorithm is a search heuristic that mimics the process of natural evolution. The algorithm is part of the larger class of evolutionary algorithms. The evolution of a genetic algorithm starts from a population, consisting of randomly generated solutions. In each generation, the fitness of every solution in a population is evaluated. In this master thesis this fitness function will be a combination of the total coverage and the cost of a solution. The best solution (based on the fitness) needs to be saved and only overwritten if a better solution is found. In a next step solutions with the lowest fitness values are removed. Multiple solutions are then stochastically selected from the remaining population and modified to form a new population. The population is now used in the next generation of the algorithm. The solutions selected from a previous generation will be called the parents. The evolution in a generation will happen through a genetic operator invoked on the parents: either a mutation or a cross-over. This will result in the children or the offspring, which are the solutions of the current generation. Figure 2.16 shows a graphical explanation of both genetic operators. Consider a solution as a string of bits with a fixed size the amount of possible antenna locations, where each bit is associated with a possible antenna location. When a bit equals 1, an antenna on the corresponding location would be installed if the solution would be implemented. A crossover is the operation where two parents will swap a piece of their string resulting in two children that are a combination of both parents, as shown on the left side of Figure In a mutation, a child is formed from a parent by changing the value of a randomly chosen bit in the string. The algorithm stops when a satisfactory fitness level is reached or when a predefined maximum number of generations has been reached. Note that it is not certain that a satisfactory solution has been found at this maximum number of generations. Figure 2.16: Crossover and mutation

28 Minimum spanning tree In order to calculate the cost of a fiber network, a fast way of estimating the installation length has to be added to the dimensioning tool. For this we work with a minimal spanning tree, which is defined as follows: A spanning tree of a connected, undirected graph is a connected sub graph that connects all the vertices together. If we assign weights to the edges, we can calculate the MST of the graph, which is the spanning tree with the lowest total weight. There are many existing algorithms for calculating such a minimal spanning tree fast. We worked with the Kruskal algorithm which is explained in [21]. Define n as the number of vertices we need to connect. The algorithm follows these steps: Create a set F, which is empty Create a set S containing all the edges of the graph While S is not empty or F is not yet a spanning tree of size n-1 o Remove the edge with minimum weight from S o If F is empty or if it connects to just one or none of the trees of F, add the edge to F o If the edge connects two different trees of F, remove those trees from F and add the new tree to F, which is the combination of the two original trees and the edge o Else: throw the edge away F now contains one element: the MST of the graph As the dimensioning tool is also built to cope with gradual roll-out scenarios, where the parts of the network that were already installed in an earlier year need to be taken into account to form the extended network, this algorithm had to be changed in the following manner: In the first step, add the given sub graph to the set F. In the second step, remove all edges from the sub graph in the set S. All other steps remain the same. 2.3 Economic An important part of this master thesis will handle the cost modeling of the dimensioned networks. To do so, some economic background is required. Section will set out the costs and prices of the equipment and the installation used in this master thesis. Because the discussed wireless technologies are new, people need to be given time to get used to its existence and they will not always use it right away. The amount of people using a certain technology at a given time is called the adoption. The costs and the income will depend on this adoption, which is discussed in section The most frequently used operation for performing an economic analysis based on these costs is the Net Present Value, which is formulated and explained in section

29 Prices and costs This section sums up the prices and costs used in this thesis. First, the costs for the wireless network are given, including the equipment costs and the rent or price for a pole to place the base station on. The other subsection gives the costs for the equipment and installation of the optical network. Wireless The wireless costs depend on the chosen technology. In this master thesis we use two different technologies: Wi-Fi and LTE-advanced, which were discussed in section The equipment of a 2x2 MIMO BS of LTE-advanced comes to a total of 30, In this master thesis we will use 8x2 MIMO LTE-advanced, which implies that for each sector on the BS 6 extra antennas have to be placed. The costs of these extra antennas can be very dependent on the case, the vendor and the type of contract with the vendor. We assume an extra cost of 66.67% per BS to upgrade it to an 8X2 MIMO BS. Later, we take a look at the influence of this extra cost. For Wi-Fi, a 4x2 MIMO BS costs 4, The maintenance costs of the base stations are 10% of the equipment costs and the average lifetime of the wireless equipment is 5 years. To install a BS and connect to the power and to a network an installation cost of 4,000 for LTE-advanced and 2,000 for Wi-Fi is needed. This cost is lower for Wi-Fi because the base station is smaller and easier to install. Also the area of the cases used in this thesis has an influence on some costs. As we will see in chapter 4 we have one urban case in Ghent (Belgium) and one rural case in Flevoland (The Netherlands). A distinction has to be made in the cost of the locations, because in Ghent we can rent locations on poles of the Belgian Institute for Postal services and Telecommunications (BIPT) or, in case of Wi-Fi, even just install the base stations on buildings. In Flevoland we have to install our own poles, either just straight into the ground or a smaller pole on the roof of a building in the area. In the case of Ghent, for a location of BIPT a yearly rental cost of 3,000 for LTE-advanced and 2,700 for Wi-Fi will be asked [18]. A new location on a building has a rental cost of 250, but also an extra installation cost of 2,000 is required to be able to place a BS on a building. On this new location a building permit for the installation is needed, such a permit costs 500. We will not need this permit for a BIPT location. In the rural case (see section 4.1.3) no poles of the government or buildings are used for placing the base stations but new poles have to be installed. There are two options: a 100ft (33m) pole on the ground or a 30ft (11m) pole placed on top of a roof on a building. A 100ft pole costs 10000, consisting of 8000 equipment and 2000 placement costs. The 30ft pole costs 2500, off which 2000 are equipment costs and 500 are paid for installation. In Table 2.2 an overview of all the discussed costs is given. 2 This figure is based on an estimation made on rough data received from Ericsson. 3 This estimated number is based on information received from Zapfi.

30 18 Ghent Flevoland LTE-advanced Wi-Fi Wi-Fi new loc. LTE-advanced Wi-Fi Base station Pole Installation BS Building permit Total Capex Rent location Maintenance Total Opex Table 2.2: Wireless equipment and installation costs License costs When operating in the 2.6 GHz band, so when using the LTE-advanced technology, a license has to be paid to use this frequency band. Determining a fixed cost per MHz is not easy, because these licenses are auctioned in every country and depend greatly from country to country, e.g. the licenses in Sweden were sold at almost 4 times the price as they were sold in Belgium [22]. Since we will handle two cases in two different countries, we need to look up the license fees for both countries. In Belgium, the licenses were auctioned in November The licenses were sold at 4.6 euro cents per MHz-PoP, which is the bandwidth in MHz divided by the population covered [23]. So, to give an idea, a 2x15 MHz band (3 sectors FDD) was bought for 15,040,000 and a 45 MHz band (3 sectors TDD) is worth 22,510,000 in Belgium. The licenses are valid for 15 years. In the Netherlands, the 2.6GHz auction happened in April Licenses were sold much cheaper than in Belgium, averaging only 0.13 euro cents per MHz-PoP, which is derisory compared to the Belgian price [24]. 1 MHz in the paired spectrum for the whole country was sold at an average of 20,200, but prices were different for every bidder [25]. These licenses are valid for 20 years, but the operators were obliged to start deploying their network within 2 years. The exact license costs used in this dissertation depend on the case and are discussed in chapter 4. Fiber costs The cost for installing 1 meter fiber is estimated at 50. These costs include the digging costs and the cost of the material of the fiber. Table 2.3 shows an overview of the costs of equipment used in the central office. Also the average lifetime of the equipment is given. At the end of this period, the equipment needs to be replaced. When no entry is given, the lifetime is considered infinite.

31 Adoption Equipment Price Average lifetime ODF rack 800 ODF slot 20 System rack 600 Shelf with switching fabric years OLT card years Control card 0 5 years Transport card 0 5 years Small pluggable optical port 15 5 years Power supply 700 Layer 2 switch years 1:32 splitter years Table 2.3: Fiber equipment costs and lifetimes [26] User adoption is often the basis of the economical calculations since users are the main revenue driver. The adoption is considered to be given by a general adoption model. The Gompertz model is chosen to estimate the user adoption of Mobile internet through the years [27]. According to [28], this adoption model gives the best fit in case of a telecom business case compared to other models. The formula of the Gompertz model has three parameters: the inflection point (a), the slope (b) and the market size (m). The parameters used for Ghent are shown in Table 2.4, assuming we are now (i.e. 2012) in year 0. We will mainly use the values for the inhabitants. The formula of the Gompertz model is given underneath and Figure 2.17 shows the shape of the Gompertz curve for the parameters of the inhabitants. ( ) ( ) User type a b m (%) Inhabitants Students Tourists Industry Table 2.4: Values of parameters of Gompertz model Figure 2.17: Gompertz curve

32 Investment analysis When deploying a network gradually, every year costs will be made and revenues will be earned resulting in a cash flow. Different calculations and operations can be made on this cash flow. The most important of them all is the Net Present Value (NPV). The NPV of cash flows is defined as the sum of the present values of the individual cash flows of same entity [29]. A present value is the value on a given date of a payment made at other times. If this payment is in the future the payment is discounted to reflect the time value of the money. The discount rate used in this dissertation is 10%. The formula of NPV is given underneath. Other instruments to measure the profitability exist, such as Internal Rate of Return (IRR), which is the rate at which the NPV is 0. The higher the IRR, the more profitable a project is. Another instrument is the payback period, which refers to the period of time required to repay the investment. Shorter payback periods are preferable to longer ones. In this dissertation we only use the NPV as an instrument for the investment analysis. ( ) With t the time of the cash flow i the discount rate R t N the net cash flow (inflow minus outflow) at time t the total number of periods

33 21 3. The dimensioning tool The dimensioning tool is a very important part of the master thesis, because it will do the hard work for the research. If we want to make an estimation of the costs for installing a wireless network backhauled over a fixed fiber network and want to get a better feeling for the different tradeoffs in this, we need to be able to calculate the optimal placement of the wireless antennas and fiber topology. Once the amount and placement of the antennas and the topology of the fiber is available, we can proceed to estimate the costs for such a project. With this ability in place, we are able to link the wireless technologies (see chapter 2) to a market estimation in terms of customer adoption, willingness to pay, expected bandwidth per user, etc. and come up with a set of profitable scenarios (technology, placement, etc.) and their detailed business cases, which will be further explored in chapter 4. Finding the optimal combination of antenna placement and fiber topology on top of geographical information system (GIS) data, is a non-trivial task and clearly an infinite amount of solutions for this exist. For this, we have built upon an existing heuristic tool based on genetic algorithms for the optimal placement of the wireless antennas. This tool is described in more detail in subsection 3.1. To keep the calculation time as low as possible, only a stub calculation is used for the costs and revenues, and a more detailed calculation is required once a good optimal candidate is found. To accomplish this we could use the Techno-Economic Software Suite (TESS), for which we describe the parts we used in subsection 3.2. Both tools wireless dimensioning and TESS are combined and extended to form the core of the fixed mobile dimensioning tool. Clearly here it is important to capture the requirements correctly, to enable us to perform all simulations later on. For this the input and output has to be defined and linked into the existing tools. Especially the code of the wireless tool had to be extended and updated to cope with those new requirements and to enable fixed-wireless instead of wireless-only cost estimation. The structure and details of the different elements in the tool are described in subsection 3.3. A final part of the key research and work on a heuristic tool is focused on reducing the runtime while maintaining (or improving) the outcome. Subsection 3.4 details the most important runtimeoptimizations introduced in the tool. It also concludes this section and will clearly show the input, output and runtime for different exemplary cases. 3.1 Wireless Tool The Wireless Tool is a tool that was developed within IBCN. It was developed to find an optimal placement of antennas in a given area when someone wants to roll out a wireless network that has to meet a given adoption requirement and given capacity requirements in ten years. Note that the tool assumed the network will be rolled out within the coming year, so all antennas will be placed at once. The dimensioning of the wireless network is well handled in this tool. A detailed technical modeling is done, based on path loss models, to determine the range of each antenna, while taking into account the population density. The dimensioning itself is computed with a heuristic algorithm,

34 22 which is a good choice due to the many possible solutions of a wireless dimensioning problem. This is caused by the lack of placement restrictions for an antenna. Since the goal of the dissertation is to determine the roll-out of a wireless network interconnected by a fixed network, this tool is a good starting point. The tool even offers a basic calculation to connect the base stations to given backhauling connection points. Though, in many situations these calculations won t give an optimal solution, so one of the main working points will be to extend this fixed network dimensioning. Another important working point will be the flexibility of the tool, since many parameters in this tool are hardcoded. Finally, to be able to run many experiments with this tool, some runtime-optimizations will be needed. Figure 3.1 shows the structure of this tool. Based on the GIS data of the area, the technical parameters of the desired wireless technology and the costs of the antennas, a dimensioning is made. First the modeling of the wireless technology to find the ranges of the antennas is run, followed by the actual dimensioning placing antennas as good as possible to cover the given areaand eventually a cost modeling, based on the input costs. In the following sections each part of the tool will be discussed and changes that have been made are highlighted. Figure 3.1: Wireless tool [18] User data and GIS data The tool requires several files with GIS data (in a shapefiles format) as input. A shapefile of the target area is needed. The area can be split into sectors, each having a different value for the number of inhabitants, called user data in Figure 3.1. There is also an option to provide another shapefile to determine the area to cover. Besides that, a shapefile containing the possible locations of the base stations is needed. The tool can also process three-dimensional shapefiles to determine the height of buildings in the area and use these buildings as possible base station locations. This will be important

35 23 when using technologies such as Wi-Fi, because the ranges of the antennas won t always be big enough to place them only on the public locations. Finally, the original tool also requires a shapefile with the locations of possible backhaul connections. For every antenna installed, the closest connection point is searched and a fiber connection is made between the antenna and this connection point. All of this GIS data will be used in the updated dimensioning tool, except for the shapefile containing the possible backhauling locations, which will not be used as the fiber dimensioning will be more detailed and requires a different input (see section 3.3.1) Technical parameters and technical modeling The technical parameters for the chosen wireless technology are stored in a spreadsheet. There are many parameters to consider when dealing with wireless technologies. All the values of the parameters are processed by the same calculation sheet. The sheet calculates the maximum uplink and downlink data rates of a base station under given technologic constraints. Based on the maximum data rates, and the demographic data from the shapefile, a range for every possible antenna (based on the given antenna locations) is calculated. These ranges are calculated using a path loss model, discussed in chapter 2. This range translates into a coverage area that has the shape of a circle. So every possible antenna location now also has a given area it would cover if the antenna is installed. This technical modeling information will be used as input for the dimensioning of the wireless network Costs, other inputs and cost modeling In some ways, the Wireless Tool is very limited. As mentioned before, it assumes the roll-out will be performed all at once. Therefore, the tool will look at the demand for network connectivity over ten years. It assumes the adoption of the network usage will be 0.2 and that the DL capacity requirement is 5 Mbit/s, while the UL capacity is 1 Mbit/s. These settings are not easy to edit in the original tool, because they can t be changed in the main class. The fixed period of ten years is also hard to change, because not only are the calculations for the adoption based on it, but also the equipment costs are calculated based on this number. The tool has one cost as input, the antenna cost. This cost is based on the installation cost, and 10 years of maintenance costs, all discounted with 5 percent over these years. This implies that changing the installation or maintenance cost requires some effort too. For the fiber a fixed cost per meter is assumed. This cost includes all equipment costs and installation cost. The latter will be the most expensive of both, because it includes the digging costs, which certainly in the city is not to be underestimated. The cost modeling of the wireless tool is rather simple: multiplying the amount of antennas by this fixed antenna cost and adding the amount of fiber multiplied by the fiber cost. In the dimensioning tool of this master thesis easy changes of these parameters should be possible. First of all, the roll-out should be able to be gradual, so every year antennas can be added, to cope with the capacity and adoption requirements of the given year. This also makes the cost modeling more complex, which for the antennas should be based on an installation cost and a yearly maintenance cost and should calculate for any period what the total cost will be. To be able to have a gradual roll-out, the requirements of capacity and the adoption should also be variable every year instead of fixed.

36 Dimensioning As mentioned before, the tool uses a genetic algorithm to dimension the wireless. This algorithm mimics normal genetics and its selection of the fittest, leading to an optimization over different generations. Each generation, a number of solutions will be created based on solutions of the previous generation. One solution has as its genes the placement of a number of antennas and the total cost, considering the antenna costs and the cost of the fiber connection. In every generation, the tool will calculate the fitness value for every solution and compare it with the current best solution. The fitness value is a combination of the coverage fitness and the cost fitness. The coverage fitness is the ratio of area covered by the current solution to the total area of the target input area. The cost fitness is equal to one, subtracted by the ratio of the cost of the current solution to the worst case cost, and this number is multiplied by 100 to get a percentage. The worst case cost is the total cost, assuming an antenna would be installed and connected to a backhaul connection point on every possible antenna location. To determine the total fitness value, a weight factor is calculated to determine the influence of the cost fitness upon the total fitness. The minimum coverage is the goal coverage, e.g. 99%, subtracted with a fixed value, e.g. 5%. The exact calculations are shown in the formulas below. When the coverage of a solution is situated between the minimum and the goal coverage, a quadratic ascending function decides the influence of the cost fitness. The weight is implemented to achieve a good coverage: the closer the coverage is to the coverage goal, the higher the fitness will become, due to the impact of the cost fitness. If the coverage of the solution is below the minimum coverage, the cost fitness won t have any influence on the total fitness at all. ( ) ( ) { This dimensioning algorithm gives good results, but the result might not always be based on fair assumptions. Considering the cost of 1 meter fiber is set in the tool at 50, the impact of the amount of fiber needed to connect a base station is big. A small example scenario to highlight this impact is given in Figure 3.2. As shown, two base stations (BS1 and BS2) have a common closest connection point.

37 25 Figure 3.2: Example scenario fiber cost impact The Wireless tool will simply connect both base stations directly to the connection point (Figure 3.3). Figure 3.3: Example scenario solution of Wireless Tool This solution is not optimal, because obviously it is cheaper to connect BS1 to BS2 instead of directly to the connection point as shown in Figure 3.4. Figure 3.4: Example scenario good solution The impact of the non-optimal solution is big, because the cost of the fiber has a big influence on the total antenna cost. In this scenario, BS1 actually has a false cost, because in a realistic case, the last solution will always be implemented. Because the cost of BS1 is high, it will have less chance to be selected as an antenna that has to be installed, even though it might be optimally located to get a good coverage. To solve this problem, we have to improve the fiber dimensioning in the tool. Some simple changes can give us a more realistic solution and a correct cost for every antenna. These improvements are explained in section TESS framework The Techno Economic Software Suite (TESS) is a network cost and business modeling framework developed by IBCN. It is designed to be able to perform a detailed cost modeling of different types of fixed networks. The framework is currently being expanded to support the cost modeling of wireless networks. Great effort was made to make this framework as generic as possible. That s why every input and output is based on an object of the class TimeFunction. A time function is a function in time off which the type of value of the function can be chosen freely. Time functions are used for the input and output of the tool and will make the cost modeling easier. Figure 3.5 shows a typical calculation chain of TESS for a fixed network. There are three chained modules: area, cost and evaluation.

38 26 Figure 3.5: The Techno Economic Software Suite [26] The area module The area module is a hierarchical structure that can have several sub-areas. For this dissertation the sub-areas won t be needed. Each area consists of three smaller calculation modules: the adoption module, the dimensioning module and the equipment tree module. The adoption module will forecast the number of subscribers of the network for every coming year. The dimensioning module is responsible of calculating the amount of ducts, fiber and trenching needed to deploy the network. For now this dimensioning is based on analytical models, but IBCN is currently working on a GISbased dimensioning module. The last calculation module of the area module is the equipment tree module. It consists of a tree structure that allows determining the amount of equipment needed to connect the subscribed customers at any given time. Figure 3.6 gives an example of such a hierarchical equipment tree structure. On every branch there is a granularity factor representing the relationship between different levels. The lowest level equipment of the tree is connected to drivers. A driver can be different things, e.g. the amount of customers or the amount of street cabinets. The calculations of the tree will be based on those drivers and performed in a bottom-up manner.

39 27 Figure 3.6: Example of tree structure of equipment model [26] The costing module The costing module is responsible for calculating the costs of the infrastructure and operation costs. The input of the costing module is obtained from the output of the area module. It links the amount for equipment, cable, trenching and customers to the unit costs. Different calculations are done in this module. First of all, the Capital Expenditures (Capex) are calculated, which is the upfront cost of the infrastructure, including installation of the equipment and costs for deploying the network. The re-installation costs are also part of the Capex. These costs occur when equipment has to be replaced at the end of its lifetime. Next to the Capex, the Operational Expenditures (Opex) are computed, which is the sum of the costs of maintenance and repair, connection and service provisioning of customers and daily operational costs like power consumption. Finally, the revenues are calculated in this module, depending on the number of customers the network has connected The evaluation module The evaluation module is a final step in the TESS module chain. It allows automatically combining all the calculation results of the cost module and determining the final results for a specified period of time. The output consists of some predefined economic calculations like NPV, IRR, etc Importance to master thesis The TESS framework will be an easy tool to use in the cost modeling part of the master thesis. The flexibility given by the time functions is a great benefit using this framework and the fact that everything is made generic also contributes to the favor of using this framework. From the three modules in the area module, only the equipment module will be used in this dissertation. For the dimensioning and adoption calculations we rely on the Wireless Tool, because it is much more extended than in TESS. We have to make sure the output of the dimensioning and adoption calculations are time functions, so we can plug those calculation steps into the calculation chain. We connect this output to the area module, which is limited to the equipment tree module. The driver for the equipment module is the number of base stations, which equals the number of

40 28 fiber connections with the central office, since our equipment model assumes the splitter is located in the central office. TESS calculates the yearly Capex and Opex in the calculation module, given a time function with the values of this driver as input. Based on this information the evaluation module can calculate the NPV of the solution. 3.3 The wireless-fixed dimensioning tool The new dimensioning tool has two main goals: add a better fiber dimensioning and allow the choice to gradually expand the network over time in a cost efficient manner. The realistic fiber dimensioning is crucial to give every possible antenna location an equal chance to get selected for the dimensioning. The gradually expansion is also a feature that creates more realistic scenarios of wireless network roll-outs. The tool first calculates the required base stations for a target year and it will use this information to calculate the optimal locations for the years between the first and the target year. By using the information of the target year, only base stations and fiber eventually needed in the target year will be installed in earlier years. Note that the tool is also programmed to determine a roll out over several years without first calculating a target solution. Still, the options for this calculation are not as flexible as required in this dissertation and we are more interested to first calculate the network required in the target year. All these requirements resulted in a tool, of which the structure is shown in Figure 3.7. As said before, the tool is based on the original wireless tool, and expanded in many ways to meet the requirements for the wireless-fixed dimensioning. Section will discuss the input of the tool, while section will take a more detailed view on the upgrades of the core of the tool. At last, section takes a look at what comes out of the tool.

41 29 Figure 3.7: the dimensioning tool Input An overview of the input of the tool is given in Figure 3.8. Compared to the original wireless tool, we added two types of input: the TimeFunctions and the equipment model. This section discusses the changes in input compared to the original Wireless Tool. The wireless technology input is equal to the input of the technical parameters of the original wireless tool. There are many parameters to consider when dealing with wireless technologies. Some parameters will remain unchanged in all the considered scenarios, which are listed in Table 3.1. The gains and losses at the receiver and the transmitter that are needed to calculate the link budget are given, as well as the number of sectors and the average height of the mobile station used later in this dissertation. The fade margin is a margin that takes the temporal fading (e.g. changing weather conditions) into account and is determined by the expected yearly availability of the system. The noise figure is a measure for the degradation of the signal caused by the components in the radiofrequency signal chain. Many other parameters may vary according to the considered technology; those parameters will be discussed in the chapter of the case using the certain technology. Note that the parameters given here can easily be changed when e.g. using a new technology.

42 30 Parameter Value Base station (BS) BS antenna gain 17 db BS feeder loss (Tx/Rx) 0.5 db Noise Figure (Rx) 2 db Implementation loss 2 db Sectors 3 Mobile station (MS) MS height 1.5 m MS feeder loss (Tx/Rx) 0 db Noise Figure (Rx) 7 db Implementation loss 2 db Margins Fade margin 10 db Cell interference margin (DL) 2 db Cell interference margin (UL) 3 db Table 3.1: Fixed technical parameters We combined the GIS data and the user data of the input of the original wireless tool into one block and called it the GIS data. Only the shapefile with the fiber connection points will not be used anymore. Where in some scenarios it is desired to give some connection points as input, it is not mandatory. If one would like to give the location of existing fiber as input, the possible connections points on this fiber network should be given as input. With this info, the tool will be able to build a wireless network upon a given fiber infrastructure. Many input parameters of the tool are time functions. All time functions should have the same length in time. For every year a certain UL data rate, DL data rate, coverage and adoption can be provided. If a parameter should remain the same over time, a constant time function can be given as input. Due to the usage of these time functions, many different scenarios can be evaluated with the tool; from roll-outs that have to be completed by the end of the years to roll-outs of networks that will gradually be built during many years. Figure 2.15 shows the equipment model which was used for the tool. Note that changing this model in the tool is very easy. Every base station that is installed in the target area after dimensioning is seen as an incoming fiber for this model. This input is only used in the cost modeling of the tool, which is the last step. As discussed before, the cost input is now somewhat different: an installation cost and a yearly maintenance cost has to be provided for the base stations. For the fiber, there s also a fixed cost per meter, mainly consisting of the digging cost per meter. Along with this fixed cost, the costs of the equipment in the central office have to be provided as input, together with their average lifetime. Figure 3.8: Input of the dimensioning tool

43 Core of the tool The core of the tool is shown in Figure 3.9. A big difference to the original wireless tool is the separation of the wireless and the fiber dimensioning. The outer loop, which runs a wireless-fixed dimensioning over several years, is also a major modification. This was added to include the possibility to roll out a network gradually. The first iteration is the dimensioning of the target year. This way, we know what the network in the future will have to look like and we can base the dimensioning of the other years upon this solution. After the target year, the tool dimensions all the other years in ascending order, starting from the first year. Note that the tool also works without setting a target year and just gradually adding antennas every year, based on the previous year. Figure 3.9: Core of the dimensioning tool Dimensioning of fiber First of all a more realistic dimensioning of fiber was added. Therefore the Minimum Spanning Tree between the base stations will be calculated using Kruskal s algorithm (see section 2.2.3). Now the distance between the base stations will affect the cost. By adding this, the tool will try to minimize the total length of the MST between the base stations. Compared to the original wireless tool, the dimensioning of fiber is now executed after the wireless dimensioning. In the original wireless tool it happened all at once, just by deciding beforehand which connection point the base station will connect to. Because now the dimensioning is split into two parts, the wireless dimensioning is more correct, due to the fact that each base station has the same price. Different variants of the fiber dimensioning have to be implemented. For the target year, the fiber dimensioning is just the MST between the installed base stations and, if desired, between the fiber connection points given as input. For the other years, we need to keep track of which fiber was installed in the previous years, because this fiber will certainly have to belong to the solution of

44 32 future years. This is done by deleting the existing fiber from the set S and adding it to the forest F in Kruskal s algorithm. Now an MST between the new and old antennas is calculated and connected to the existing fiber network. Besides that, the dimensioning of a non-target year needs an adapted implementation of Kruskal s algorithm, because in this case, only fiber connection that will be installed in the target year should be chosen. So instead of having the complete graph of connections between all selected antennas as the starting set S for Kruskal s algorithm, set S now only contains the fiber connections that will be needed in the future. This means, that a non-optimal amount of fiber may be installed in the earlier years. To make this more clear, we return to the example scenario of Figure 3.2, now assuming that both base stations will be needed in the target year and only BS1 will be needed in the first year to meet the requirements. This means that Figure 3.4 would be our most favorable solution of the fiber dimensioning in the target year, which is the MST between the three points. Figure 3.10 (a) now shows the fiber dimensioning problem of year 1. BS2 is colored gray to emphasize the fact that it is not part of the first year s solution, but it will be part of the solution of the target year. If we calculate the normal MST on this scenario, we would obtain the solution shown in Figure 3.10 (b): BS1 directly connected to the connection point. This would give the cheapest solution in terms of fiber in that year. If now in the target year BS2 is added to the network, the fiber dimensioning will connect BS2 directly to the connection point. This will give the solution shown in Figure 3.3, which is not optimal. To solve this problem, we use the adapted algorithm, where the starting set is restricted to the fiber that will be needed in the future. The solution is shown in Figure 3.10 (c), which is not optimal for the current year, because the amount of fiber could be less if BS1 is connected directly to the connection point. But adding BS2 in the target year now doesn t require any extra fiber to be laid, because Figure 3.4 now is the target year solution and all the fiber in this solution was already installed in year one. Looking at it year by year, the fiber dimensioning might not be optimal, but if we look at the total amount of fiber installed over all the years, using the adapted form of Kruskal s algorithm in the nontarget years gives us an optimal solution. Figure 3.10: (a) top: example scenario year 1; (b) middle: MST example scenario year 1; (c) bottom: adapted MST example scenario year 1

45 33 Density coverage The current formula of coverage fitness (see section 3.1.4) cannot be considered as correct because the loss of coverage in a dense area costs as much as the loss of coverage in a less dense area. This problem is explained with a small example given in Figure The total area exists of both the blue and the red square, which we assume is 2 units big (1 unit equals the area of 1 square). The populations of the areas shown on the Figure 3.11 are given as input in the tool, but only used to calculate the coverage areas of the antennas. This actually explains why on the figure the coverage area of BS2 is smaller than the one of BS1. If we had to choose to install one of both antennas, intuitively we would select BS2, because it covers an area with 100 people living in it instead of 50 in the area covered by base station 1. The figure shows when we install both base stations 1 and 2, the tool would clearly value the coverage fitness at 100. If we would now remove BS1 from the solution, the coverage fitness would become 50%, because only 1 unit of the total area of two units would be covered. Removing BS2 results in the same coverage fitness of 50%, based on the same reasoning. Assuming that the installation and fiber connection costs of both base stations are equal, both solutions only installing BS1 and only installing BS2 will result in the same total fitness value. So running this problem with a coverage demand of at least 50%, this will result in an outcome of one of both solutions, each having an equal chance to be selected. This is not correct, because BS2 covers more people, so it should always be selected by the tool. This problem is solved by adding the density value to the coverage fitness calculations. The new formula for coverage fitness is shown underneath. This results in a more realistic fitness value. Applying this formula to the problem of Figure 3.11, only installing BS1 results in a coverage fitness value of 33% and only installing BS2 a value of 67%. So using this formula in the tool always gives a more correct and logic result of the wireless dimensioning. Figure 3.11: Coverage fitness problem Cost of worst case scenario Another thing that had to be revised is the cost of the worst case scenario that was discussed in section 3.1.4, which is needed for the calculation of the cost fitness. In the original wireless tool this was rather basic, just adding all the costs of all the possible antennas and the cost of their connection to the backhauling. In this case we will also need to consider the costs of all the possible antennas, but we will use a different calculation for the total fiber cost of this worst case scenario. There can be many possible antenna locations, so a simple calculation is preferred to save some calculation time. Since the only requirement of a worst case scenario cost is that it has to be higher than the cost of

46 34 any other possible case, the worst case fiber cost is the addition of the costs of a fiber connection between each antenna and a certain point in the target area. This point can be a fiber connection point or the first antenna location. This creates a star network of the antenna locations, which cannot be shorter in distance than the MST between all the antennas. The total worst case scenario cost is the sum of this fiber cost and the total antenna cost. Even if all possible antennas would be part of the solution, its cost will still be smaller or equal than the worst case scenario cost Output The tool has two different types of output, as we can see in Figure One output type contains the placement of the base stations and the fiber connections for each year of the planning. The other a cost modeling of the considered project is calculated after the iterations over all the years. Both output types are discussed in following sections. Figure 3.12: Output of the dimensioning tool Location of base stations and fiber This type of output consists of shapefiles that contain the location of the base stations and fiber of the best solution of each year. The original wireless tool already had methods to provide a shapefile with the location of the base station and one where the coverage of each base station is displayed as a circle. A new method is provided to create a shapefile where the fiber connections are stored. Showing these shapefiles together with the one of the target area gives a view on the solution of the tool. Cost modeling After running the genetic algorithm for every year the costs of the best solution is modeled using the TESS framework. For each year it calculates which base stations need to be installed and which ones were already there. The latter will have a maintenance cost for every year they are up and running. The same thing has to be calculated for the fiber and the equipment needed for the fiber. A number of mathematical operations for time functions available in the TESS framework are used in these calculations. All the costs are discounted with a rate of 10% and the total cost is calculated. 3.4 Optimizations This section discusses the main optimizations that are made in the core of the wireless tool to increase the performance of the tool Speed Many optimizations that have been implemented to improve the runtime are listed underneath.

47 Time (in minutes) 35 Population The population, which is the number of solutions calculated in each generation, affects the speed of the tool. Figure 3.13 shows a graph that gives the relation between a fitness value and the time when this fitness value is reached for different population sizes calculated over 1000 generations. We see that in case of a smaller population a higher fitness value is reached in a lower generation. Note that these results are affected by the randomness of the genetic algorithm and that the results are used to find a good population size, but this might not be the optimal one. These results were used to estimate around which population size we have to look to find a good size to use in the tool. In the next section we check if there is a better size than 25. population=25 population=50 population=75 population= Fitness value Figure 3.13: impact of population time on runtime of the tool Efficiency By performing some local optimizations on the solutions the tool gives better results. Every 60 generations the tool tries to optimize the solutions of the population. It checks for every active base station if the total fitness would improve when the base station and its fiber connections are removed from the solution. This local optimization removes base stations that don t have a big influence on the coverage fitness. A downside is that executing this optimization requires more calculation time. Figure 3.14 shows a graph, again with the relation of a fitness value and the time it is reached for a population of 25. It shows that implementing the optimization ( full optimization in the graphs legend) improves the result, because the fitness value reaches a higher value, but the execution time is higher. By only performing the local optimization on the best half of the population a better result is achieved. By best half we mean the 12 solutions that have the best fitness values at that time. The green line on the graph shows that this optimization results in a better fitness value and that fitness values are reached much earlier than in the case of no optimization or full local optimization. It is clear that by implementing the half local optimization, less time is needed to reach a certain fitness value, which makes the tool more efficient. Note that when the term local

48 Time (in minutes) Time (in minutes) 36 optimization is used in the remaining of this paper, it refers to the half local optimization as described above. To make sure that the population size is still a good choice with the implementation of the local optimization, we compared the runtimes of the tool again for different populations. Figure 3.15 shows the fitness-time relations for a population of 20, 25 and 30. It is clear that 25 is a good choice of population size, because it reaches a certain fitness value earlier than the other two options and also a higher fitness value after 1000 generations is reached. full optimization no optimization best half optimization Fitness value Figure 3.14: comparison of different configurations of local optimization population=25 population=20 population= Fitness value Figure 3.15: comparison of population sizes with implementation of local optimization Caching A great optimization of speed was to cache the fitness values and only recalculate them when really necessary. Most of the runtime of the tool is spent while calculating fitness values. The runtime improved greatly by caching them smartly. As discussed in the worst case scenario cost is

49 37 needed for the calculation of this fitness value. In the original wireless tool this cost was calculated every generation, but since this cost is equal in every generation, it is enough to calculate it at the start of the genetic algorithm and caching this value to use it in the other generations. Following the same reasoning, there is no need to recalculate the fitness values of solutions that are not changed, so we cache these values too. The runtime is reduced thanks to the caching. Stopping rule Thanks to the optimization in efficiency a good solution is obtained faster than before. A stopping rule was added to speed the tool up even more. If for a number of generations no better fitness is found, the genetic algorithm will stop and the current best solution is selected as the optimal solution. In this dissertation this number of generations is 350 for the dimensioning of the target year and 200 for the dimensioning of the years before the target year. This number is lower in the years before the target year because the number of base stations to choose from is lower and a good solution is achieved faster Splitting the target area If we want to run the dimensioning tool on a big area, the tool has the possibility to split the area into smaller parts. This optimization was already part of the original wireless tool, splitting up the wireless dimensioning of a bigger area in 9 parts, but is expanded to be able to handle splitting the fiber network into parts. The tool can split the area into 9 parts based on the number of antenna locations, so every part contained practically the same amount of possible antenna locations. To be able to use the density coverage optimization and not using too much time with it, each part has to be given the appropriate population information. For every part, a connection point (which is one of the possible antenna locations) that is closest to the center of the target area will be selected to take care of the connection of the fixed networks of the different parts. The genetic algorithm then runs on all the parts and looks for a good partial solution in each part. When these partial solutions are merged, another local optimization, which was added for this master thesis, is performed. This local optimization checks for every active base station what the total fitness would be if we remove it from the solution. Then we remove the base station which gave the greatest improvement on the total fitness. These two steps are iterated until no improvement is made by removing another base station. This local optimization gives better results than the one discussed before, but it takes a lot more time to complete Downside of the density coverage In section we introduced the density coverage to create more correctness in the solutions. A downside to the density coverage is that the runtime of the tool increases by these calculations. To give an idea, an overview of the solution of the urban case discussed in section 4.1 is given in Table 3.2, once with, and once without the density coverage. It shows us that the solutions are almost the same (except for the location of the antennas), but we see that there is a big difference in runtime. When using the geographic coverage instead of the density coverage, the tool only needs 70% of the time. Then again, it doesn t give a correct solution.

50 38 Geographic coverage Density coverage Antennas in year Fiber in year km 5.2 km Total cost 527, ,580 Runtime 12 min 17 min Table 3.2: Geographic coverage versus density coverage 3.5 Conclusion To complete this master thesis a refined tool was needed to dimension a wireless network interconnected by a fixed network. The Wireless Tool seemed a good starting point and expanding it in various ways made it possible to make a generic fixed-wireless dimensioning tool. We also used the TESS framework to make the inputs and outputs more generic and time-sensitive and optimized the final tool in several ways to improve the runtime. This created a relatively fast dimensioning tool that can handle many different types of requests with many different inputs. An important point was to support the gradual roll-out of the networks, which is certainly possible in the created dimensioning tool. The tool is a good starting point to consider different cases describing totally different situations. Based on the tool, we worked out some cases to show the power and the possibilities of the tool. These cases and their evaluation are discussed in the next chapter.

51 39 4. Integration of fixed and mobile networks A powerful tool was developed to dimension an integrated fixed and mobile network. We handled two different cases to determine the strengths of this tool and to see where improvements and changes may be needed in future work. Two very different cases are discussed in this chapter: one case is situated in an urban area, while in the other case the target area is rural. Many other cases could be handled with the tool. The two distinguishing cases are used to show what the tool is capable of. 4.1 Urban case In most cases a broadband network is installed in an urban area. In this master thesis we determined the gradual roll-out of a 4G network in the city of Ghent. Two different scenarios are discussed. The first is the deployment of an LTE-advanced network in the whole city of Ghent, an area of 157 km 2. This case is used to show what the output of the tool looks like. Later in the section, we assume the small center of the city of Ghent as the input area, where we look at a comparison of two possible wireless technologies: LTE-advanced and Wi-Fi. This shows that the tool is capable of making a good comparison between two technologies. Choosing the wireless technology is a key decision when rolling out a wireless network and this tool could help make the decision easier. The chapter starts with giving all the input and assumptions used in both scenarios Input Area Different types of shapefiles are used as input of the tool. First of all, the target area has to be selected. This shapefile also contains the demographic data of the area. This is added as a property to every different feature of the shapefile. The left side of Figure 4.1 shows the shapefile of the city of Ghent, where each area separated by the darker lines represents one feature. The city center is shown in green while the rest of the area is orange. Since the goal was to deploy a network in the urban, the docks of Ghent are removed from this shapefile (they were located at the north-east corner of the given map). The right side of Figure 4.1 shows a close-up of the center of Ghent, which is about 7 km 2. Another shapefile that is required to run the tool is the shapefile containing the possible antenna locations. In Ghent we can rent a location on a pole of the BIPT. The locations of those poles are publically known and can be retrieved on the website of the BIPT. Figure 4.2 shows the shapefile containing all the possible BIPT locations in Ghent for placing an antenna. We also use this shapefile when solving the scenario of the center of Ghent, since all the poles in the center are also shown on this figure. When considering the scenario in the city center, we also deploy a Wi-Fi network to compare it with an LTE-advanced network. With Wi-Fi, we have the possibility to install the base

52 40 stations on the rooftops of buildings in the center of the city. For this, a shapefile containing 3D-data of the center, containing the height of each building, is available. The tool selects every building that is higher than 15m to be a possible antenna location for the Wi-Fi network. The city of Ghent is mainly an urban environment, so we will need to use the corresponding parameters in the Erceg C model, with the urban correction factor, explained in section This is done by setting the path loss model mode to Urban in the tool. Figure 4.1: Shapefiles Ghent Figure 4.2: Antenna locations of BIPT Ghent

53 41 Requirements A case depends on the requirements for a certain area. In this case, we look at a gradual roll-out spread over 5 years. The data rate requirements of the scenario considering the total area of Ghent are different every year and are listed in Table 4.1. These data rates are increasing a lot every year as means of an example to show the influence of the data rates on the ranges of the antennas. When considering only the center of Ghent we would like to see the influence of the required data rate on the solution, so we consider 13 different situations per technology, each with a different data rate requirement. The roll-out also happens in 5 years, but in each of these years the data rate is equal. The considered data rates are given in Table 4.2. Year DL (Mbit/s) UL (Mbit/s) Table 4.1: Yearly data rate requirements city of Ghent Scenario A B C D E F G H I J K L M DL (Mbit/s) UL (Mbit/s) Table 4.2: Data rate requirements of different scenarios center of Ghent (equal each year) Another factor that influences the ranges of the antennas is the adoptions of the technology in the area. To apply this adoption to the case the Gompertz curve given in section is used as input for both scenarios. The required coverage also has to be given as input and is equal in every year: 99% (of the inhabitants) in both scenarios. Wireless technology In the scenario of the whole city of Ghent we selected LTE-advanced as the wireless technology to show the possibilities of the tools. The base station has an input power of 46 dbm. We use the FDD spectrum with 2x5 MHz per sector, because studies in [18] showed this is the best configuration for this case. A MIMO configuration of 8x2 is applied and Spatial Multiplexing is used to increase the data rates. In urban environments SM has much more use than TD, because many people are located on small areas, so high total data rates in one coverage area are needed. The carrier frequency is 2.6 GHz, for which a license is required. All these parameters are listed in Table 4.3. The cyclic combining gain is a gain in the link budget when using Transmit Diversity, because this increases the range. So for Spatial Multiplexing this remains 0. The MIMO gain is also only positive if TD is used and it depends on the number of antennas used at the transmitter and receiver, while this number doesn t impact the cyclic combining gain.

54 42 Parameter Value Input Power 46 dbm Frequency 2.6 GHz Duplexing FDD Bandwidth 2x5 MHz Base station Tx antennas 8 Rx antennas 8 Cyclic combining gain (Tx) 0 db Mobile station Tx antennas 2 Rx antennas 2 Cyclic combining gain (Tx) 0 db MIMO Configuration SM MIMO gain (Tx/Rx) 0 db Table 4.3: Parameters LTE-advanced FDD When considering the scenario of the center of Ghent, we also run the tool with Wi-Fi. The parameters are similar to the ones of LTE-advanced. The input power of the base stations are less though, only 35 dbm instead of 46 dbm. Wi-Fi also operates in the license-free band at 2.4 GHz with a bandwidth of 20 MHz. Wi-Fi uses TDD and has a DL:UL ratio of Wi-Fi only uses 4 MIMO antennas at the BS, so 4x2 MIMO is used. Spatial Multiplexing is also applied to get higher data rates, so no extra gains are added to the link budget. All this information is listed in Table 4.4. Parameter Value Input Power 35 dbm Frequency 2.4 GHz Duplexing TDD Bandwidth 20 MHz Base station Tx antennas 4 Rx antennas 4 Cyclic combining gain (Tx) 0 db Mobile station Tx antennas 2 Rx antennas 2 Cyclic combining gain (Tx) 0 db MIMO Configuration SM MIMO gain (Tx/Rx) 0 db Table 4.4: Parameters Wi-Fi TDD

55 43 Costs Section discusses all the costs needed and used in the cases. Table 2.2 shows the costs of the wireless equipment used in this case. Table 2.3 contains the costs of the equipment of the fiber network; these costs are the same for every case. The price of laying one meter fiber equals 50 per meter. For LTE-advanced a license is needed to operate in the 2.6 GHz band. In this case we use a 2x5 MHz bandwidth in FDD mode, but we have to consider three sectors per base station, so we need a license for 2x15 MHz. This license was auctioned for 15,040,000 for Belgium and is valid for 15 years [23]. Now, we will try and estimate the value of a license for Ghent only. We could divide the total license cost by the area of Belgium and multiply it by the area of Ghent, but this proportion is not really fair since a license is more seen as a permission to cover an amount of people instead of an area. The area covered is more a consequence of covering a certain amount of people. That s why a better estimation of the license cost would be to divide the license by the total population of Belgium about 11 million and multiply it by the population of Ghent, which is around 250 thousand. This results in a license cost of 341,818. To compare with the other method, its calculations result in a cost of 77,348, so there is certainly a big difference in both methods. Note that these calculations are only estimations and could be different in reality, e.g. due to touristic value of the city of Ghent or due to the fact you can t buy or own a license for only Ghent. Nevertheless this estimation seems a fair indication of the license costs City of Ghent We put all the input of the scenario of whole Ghent into our tool and let it run. An overview of the output of the cost modeling is given in Table 4.5. The overview proofs that the higher the data rate and the adoption are, the higher the amount of base stations will be. Deploying a network of 52 antennas, interconnected by 67 km of fiber, over a period of 5 years will cost around 7.5 million euros. We also set out the yearly costs in a graph shown in Figure 4.3. It s clear that the major part of the investment happens in the first year, because the biggest part of the infrastructure 25 antennas and 64.5 km fiber is installed this first year and in later years we also take a discount rate into account. We can see that, apart from the first year, the yearly cost is higher in later years, even though an equal cost is actually lower in a later year due to the discount. This is explained by the fact that every year there are two input values that increase: the data rate and the adoption. So the increase in the amount of antennas needed to cover the region is not linear, but is higher in later years. After year Number of base stations Amount of fiber (km) Cumulative cost 5,306,928 5,687,208 6,177,708 6,664,562 7,468,978 Table 4.5: Overview solution case Ghent

56 Year Figure 4.3: Yearly costs case Ghent The visual output of the tool for each year is shown in Figure 4.4. These figures give a good view on the way the dimensioning algorithm works. First the network for year 5 is installed, that s why it s displayed first. It s clear that the antennas in year 1 to 4 are all antennas that will eventually be placed in year 5. Antennas placed in an earlier year will also remain in the solutions of later years. The same reasoning is valid for the fiber. If we would dimension the fiber between the antennas placed in year 1 without thinking about the coming years, we would determine the MST. As we can see this is certainly not the case, because the fiber network makes some big detours. But as we know, this is done on purpose, because in later years antennas will be placed along these detours. This explains why almost all the fiber is installed in the first year. To reach a base station at an edge of the area, the fiber passes through many future locations before reaching the base station. This is confirmed by the fact that in year 2 only 1 extra fiber is installed and that in year 3 all the fiber needed in year 5 is already placed. That s not the only thing that changed, because if we look closely at some of the base stations that are both in year 1 and year 5, we notice that their ranges are smaller in year 5. This is a consequence of the increase in data rates, which is explained by the fact that the range of an antenna is limited by two things: the input gain of the antenna and the total data rate it has to provide. If the data rate doesn t exceed the maximum allowed bitrate, the range will be as high as the maximum path loss (calculated based on the gains) allows it to be. If at a certain range, the maximum data rate is reached, the antenna will not be able to provide any more coverage beyond this range, because it is fully occupied. The limitation by the path loss is the same for every base station under a similar configuration, but the limitation by the data rate depends on the demands of the area a base station has to cover. In year one for example, all ranges are about the same, because the ranges are limited by the path loss model. In higher years the required data transfers are higher in the center, which decreases the range of that base station due to the need of higher modulation schemes. This is definitely noticeable in year 5, where all the ranges in the city center are really small compared to the ranges in the suburbs because the density in the center is much higher.

57 45 Figure 4.4: Visual output of each year of case Ghent To see what the influences of the different components of the network are, we set out the composition of the total discounted cumulative cost over the 5 years in Figure 4.5. The biggest cost is the fiber cost, which is responsible for 48% of the total cost. We made a distinction between the costs directly dependent of the amount of fiber and the equipment costs at the central office. The latter is only responsible for 3% of the cost. This means the amount of fiber that is installed has a great influence on the total cost, in this case 45% and should be as small as possible. The installation costs of the base stations are good for 33% of the total cost, while maintaining them and their location is good for 15%. For the license we pay 4% of all the money that has to be spent. 15% 4% 3% Equipment Fiber Fiber Installation BSs Maintenance BSs 33% 45% License Figure 4.5: Composition of the total cost of solution

58 Costs Wi-Fi vs. LTE-advanced In this subsection we focus on the center of Ghent only, where we calculate the deployment of two different types of wireless networks: LTE-advanced and Wi-Fi. We would like to see which technology is cheapest to roll it out over a period of 5 years. We have performed a comparison for different values of bitrates, as shown in Table 4.2. It s interesting to see which technology can handle higher data rates, so which one is more future-proof. The results of this experiment are given in Figure , , , , ,00 0, Bitrate download (Mbit/s) Wi-Fi LTE-advanced Figure 4.6: Comparison of LTE-advanced and Wi-Fi costs under given data rates As we see in the graph, at lower DL bitrates LTE-advanced is the cheaper solution, but when the data rate is higher than 7 Mbit/s Wi-Fi gives us the cheapest solution. We can state that the cost of both technologies at data rates between 5 and 10 Mbit/s are more or less the same, but with lower or higher rates, the chosen technology has much more impact on the total cost. Note that for LTEadvanced, the license costs are also taken into account. To give an idea in the difference of number of base stations and km fiber that are installed in every situation, an overview of these figures is given in Table 4.6. Mbit/s LTE- #BS advanced km fiber Wi-Fi #BS km fiber Table 4.6: Number of base stations and km fiber The antennas of the two technologies clearly have a different range. Wi-Fi needs 5 to 10 times more base stations to cover the same area. The fact that there is more fiber needed for Wi-Fi is a direct consequence of the high number of base stations. Table 4.6 shows some strange numbers for Wi-Fi at 12 and 13 Mbit/s. It states we need less base stations for a higher data rate if we compare both numbers. This can be explained by inspecting what the achieved coverage in the different experiments is. We did set the required coverage at 99%, but the fitness function discussed in section doesn t guarantee that this coverage is met. In Figure 4.7 we set out all the coverage percentages reached in the experiment.

59 Fitness value Coverage % 99% 98% 97% 96% 95% 94% Data rate DL (Mbit/s) Wi-Fi LTE-advanced Figure 4.7: Reached coverage percentage in Wi-Fi vs. LTE-advanced experiment We can see that Wi-Fi has many problems reaching the desired coverage at high bitrates, while it actually has many more possible antenna locations. Table 4.6 also learns us it needs many more base stations to cover a same area, so it is quite logic that when even more BSs are needed, some parts of the area might not get covered well enough, due to the lack of antenna locations. We can see that at data rates higher than 11 Mbit/s the coverage of LTE-advanced also gets below 99%. We notice that at 6 and 7 Mbit/s the coverage is above the coverage goal. This is explained by the fact that an extra antenna can result in a better coverage than needed, but without the antenna the coverage goal would not be reached. We can conclude that the reached coverage of LTE-advanced is always higher or equal than the coverage of Wi-Fi. At high data rates there s a difference of up to 2%, which results in an incorrect comparison of the costs. We can add this achieved coverage-level in the comparison by using the fitness formulas introduced in the tools. We have all the data needed to calculate the fitness values, except for the worst case scenario cost, which has to be the same for both technologies. We arbitrary chose the worst case cost as the sum between 1,500,000 and the DL bitrate multiplied with 200,000. We chose to increase this cost with increasing bitrate because it makes sense that more money is spent for deploying networks that can achieve higher data rates. If a fixed worst case cost would be chosen, the fitness values of lower bitrates would be very high and the one of higher bitrates low, while both solutions might have the same cost efficiency. The fitness values are set out on the graph given in Figure Bitrate downlink (Mbit/s) Wi-Fi LTE-advanced Figure 4.8: Fitness values of Wi-Fi vs. LTE-advanced comparison

60 48 From this graph we can conclude that LTE-advanced is always the better choice. Only comparing the costs made us doubt between the technologies when the data rates became higher, but comparing the fitness values shows that at the high data rates it is actually even more favorable to choose LTEadvanced than in the lower data rates. In any case, LTE-advanced seems to be the ideal technology for covering the city center of Ghent. To see the composition of the costs of both technologies we compare the situation with 5 Mbit/s DL and 1 Mbit/s UL, because both the total cost and the fitness value are about the same. Figure 4.9 shows the composition of the costs when using LTE-advanced. We see that 32% of the costs are license fees. 28% of the money is spend on the fiber network, while 27% is dedicated to the installation of the base stations and 13% for maintaining them. Figure 4.10 shows the same diagram for Wi-Fi. Here, a remarkable 69% of the costs are caused by the fiber network, the biggest part to the placement of the fiber, which is 56% of the total costs. No license costs are applicable here. The installation and maintenance costs of the wireless network are responsible for respectively 21% and 10% of the costs. The main factor for the high costs of LTE-advanced is the license fee and also the expensive equipment for the base stations. Although the latter gets compensated due to the fact not many of them are needed. If the license fees would be less in reality than estimated, we should certainly go for LTE-advanced in the choice of technology. On the other hand, if the estimation of the license cost is lower than in reality, we should consider Wi-Fi to cover the city center with a wireless network. The influence of this license cost will be investigated in the next chapter. When using Wi-Fi, the main factor is obviously the fiber network. This makes sense, because many cheap base stations are installed through-out the city-center, and they all have to be connected to the fiber network. Due to the high digging costs, the influence of a couple more meters of fiber is pretty big. If these costs could be suppressed, e.g. by renting dark fiber or when an FTTH network is available, Wi-Fi seems a good option for the network. The actual costs of this network could be higher too, due to the abstractions made in the dimensioning: we didn t take the roads into account. We just took the shortest path between two points, which is straight. Considering the roads would give us a longer trajectory of the fiber, increasing the costs. On the other hand, this abstraction is also made when implementing LTE-advanced, so also in that scenario the costs would go up without this abstraction. 23% Fiber 32% 5% Equipment Fiber Installation BSs Maintenance BSs License 13% 27% Figure 4.9: Composition of costs LTE-advanced

61 49 10% 56% Fiber 21% Equipment Fiber Installation BSs Maintenance BSs 13% Figure 4.10: Composition of costs Wi-Fi Conclusion The first scenario gives a good aspect on what the dimensioning tool is capable of and what type of output the tool produces. It shows how the different years of the gradual roll-out are computed. The second scenario shows the comparison of Wi-Fi and LTE-advanced in the city center of Ghent and teaches us that LTE-advanced seems to be the best option, although the exact price of the license and the equipment has a big impact. The influence of the license will be examined in chapter 5. Also the fixed network has a big influence on the decision, as Wi-Fi will need a much longer net. So if the costs would be significantly lower, it might be a good decision to go with Wi-Fi instead of LTEadvanced. On the other hand we can also conclude that LTE-advanced works better at higher bitrate demands, so it s more future-proof than Wi-Fi.

62 Rural case In the second case we would like to test the tool on a more rural area. Within this dissertation there was a close cooperation with Unet B.V. Unet is a Dutch Fiber To The Business (FTTB) company that provides services over their fiber to small and medium sized companies. They also rent dark fiber to other companies. On their active networks they provide services such as internet, VOIP, XDSL, IPTV, VPN, PIN and surveillance. Unet gave us detailed information to benchmark the tool on a rural case in Flevoland. They would like to connect all farmers in a given area and provide a triple-play package, including internet, telephony and digital television and compare in a study different types of technologies to make this connection. Coverage (at a lower bitrate) of the mobile network should also be provided on the fields of the farmers. Different wireless technologies and configurations of them are tested with the tool to find an optimal technology to meet the demands. We inspect the influence of Spatial Multiplexing and Transmit Diversity, since in rural areas the ranges of the base stations become important because the population is spread over the area Input Area The target area in Flevoland has a surface of about 35 km 2. Figure 4.11 shows a satellite view of the area limited by the road N305 in the north, by the road N301 in the east, by the N704 in the west and by the red line drawn on the map in the south. Figure 4.11: Sattelite view of the target area in Flevoland Unet gave us GIS-data in which the buildings and the fiber that s already available are given. We made a shapefile out of this data, which is shown in Figure We assigned the population manually to the areas. We aim at a bitrate of 5 Mbit/s DL and 1 Mbit/s UL per person. We assign a population of 100 to the total rural area without the areas of the houses assuming not more than 100 farmers will be at work at once. Each house on the shapefile has a population of 5, assuming a 10 Mbit/s channel for HDTV and three channels of 5 Mbit/s for telephony and internet usage. This results in a total of 25 Mbit/s per house.

63 51 Figure 4.12: Fiber available in target area Besides that, Unet also gave as GIS data of a fiber network that they would deploy in order to connect all the farmers in this area. The shapefile with their proposal is shown in Figure Their solution has the benefit that there is basically no limit in terms of data rate, but on the other hand, they only connect farms and there would be no network access on the fields. Additionally this solution leads to much trenching and a high final cost. Figure 4.13: Solution Unet case Flevoland We aim to compare their solution to a couple of solutions computed with our tool. Therefore we need to select the possible antenna locations. We chose to select every house (or their garden) as a possible antenna location. Making this choice, we can compare the scenario where large poles are installed on the ground with the scenario where smaller poles are put upon roofs. This results in a collection of about 75 possible locations shown in Figure Figure 4.14: Possible antenna locations case Flevoland

64 52 A final step to prepare the geographical input is to create fiber connection points. We created (manually chosen) points on the given fiber that crosses the area, making sure each farm near the fiber is not too far away from a connection point. In the data received from Unet there was also an extra connection point given at the north border of the area. This connection point is part of another fiber line running through Flevoland that belongs to Unet. As shown in Figure 4.15 we also added this connection point to the input. Figure 4.15: Fiber connection points case Flevoland Since this area is a rural area, we will need to use the correction factor in the Erceg C model that corresponds with a rural environment. This can be done by setting the path loss model mode on Rural in the tool. Requirements In this case, we are not talking about a gradual roll-out. We just want to know how many base stations of a given technology need to be installed to meet the demands. As stated before, DL data rate of 5 Mbit/s and an UL data rate of 1 Mbit/s is required and this at 99% coverage. The adoption here will be 100%, because every house will pay for the use of their amount of bandwidth. Wireless technology In some scenarios of this case we use LTE-advanced, but this time we also use it in the TDD. The TDDmode in LTE-advanced works better together with the Transmit Diversity technique used in MIMO [18]. The base stations have an input power of 46 dbm. TDD LTE-advanced also operates in the licensed band at 2.6 GHz with a bandwidth of 15 MHz. A MIMO configuration of 8x2 is also used and now Transmit Diversity is used to increase the range of the antennas. Note that not that many people use the base stations at the same time, so increasing the ranges could be better than increasing the data rate. The cyclic combining gain is now 3 db and the MIMO gain is also positive, due to the usage of TD. The MIMO gain depends only from the amount of transmitter and receiver antennas and in case of 8x2 MIMO it equals 12 db. All these parameters are listed in Table 4.7. When using the FDD mode of LTE-advanced we refer to the previous case, where all the technological details are set out in section

65 53 Parameter Value Input Power 46 dbm Frequency 2.6 GHz Duplexing TDD Bandwidth 15 MHz Base station Tx antennas 8 Rx antennas 8 Cyclic combining gain (Tx) 3 db Mobile station Tx antennas 2 Rx antennas 2 Cyclic combining gain (Tx) 3 db MIMO Configuration TD MIMO gain (Tx/Rx) 12 db Table 4.7: Parameters LTE-advanced TDD We use two configurations of Wi-Fi in this case. The Wi-Fi parameters used in one scenario are the same as the ones used in the urban case, so also a 4x2 MIMO configuration, using Spatial Multiplexing in a TDD frequency band of 20MHz. In another scenario we use the same configuration still operating in TDD-mode along with Transmit Diversity instead of Spatial Multiplexing. Similar to LTE-advanced, this adds a cyclic combining gain of 3 db to the BS and MS parameters and a MIMO gain of 9 db (4x2 MIMO). LTE-advanced base stations are placed on poles of 30m high, while Wi-Fi antennas are placed on smaller poles of 10m on rooftops, resulting in a total height of 15m. Costs Table 2.2 shows the costs of the wireless equipment used in this case. Digging in a rural area is much cheaper than digging in the streets of a city. Here, fiber can be laid in a trench next to the street under soft ground. This is why the cost per meter fiber given in is much lower for this case. Unet also gave us some financial information about their solution, which consists only of a fiber network. From that data we derived a price of 23.7 per meter for the installation of fiber in the area. Again, for LTE-advanced a license is needed. In the first scenario we use 15 MHz bandwidth in TDD mode, but again considering three sectors, so we need a license for 45 MHz. In the Netherlands licenses were sold at an average of 20,200 per MHz. So, 909,000 would be sufficient to have the license for the whole country. Again, we are only deploying the network in a small part of the country and this time, the part is not touristic or populated at all. It s an area where about 75 farmers live, and where the interest for a 4G network is not very high. If we take the same approach as in the previous case, estimating the population of the given area at 250 (considering wives and kids) out of a total of 16,736,736 people living in the Netherlands, the license cost would result in This

66 54 price can be neglected and might be unrealistic. If we now try the geographic approach, considering 35 km 2 for the area and 41,528 km 2 for the whole country, the license cost is estimated at 766. This might be a better indication, so we will work with this price. Note that the 3G licenses in the Netherlands will have a much smaller impact than in Belgium. For LTE-advanced operating in a FDD-mode, with the configurations given in the urban case, only 30 MHz is needed. The same calculations as in the TDD-mode result in a license cost of Comparison of technologies LTE-advanced When using LTE-advanced in the TDD-mode and with the TD technique, the output gives us a network with 8 antennas and about 6 km fiber shown in Figure In this figure both the newly installed fiber and the fiber that was already available are shown. The total cost of this solution is 754,576, including the license cost. The total cost includes one year of maintenance and operational costs. Figure 4.16: Solution Flevoland case LTE-advanced TD When we compare it to the scenario using LTE-advanced in FDD-mode, so with Spatial Multiplexing, we observe that only 4 antennas interconnected by about 4.5 km fiber are needed. The result is shown in Figure This solution will also be a lot cheaper, only costing 430,279. We stated before that we thought using Transmit-Diversity in TDD mode would give us better results, but here we discover it s not true. This can be explained by the fact that the required data rates are fairly high. One house, which is just a small area on the map, demands a data rate of 25 Mbit/s. So, 10 houses for example, require 250 Mbit/s (with an adoption of 100%). As explained before, the range of an antenna is limited by both the input gain of the antenna and the total data rate it has to provide. In Figure 4.16 for example, the coverage-circle of the center antenna is much smaller than the one in the top-left corner, because it serves more houses. So it seems the range-increase that TD brings along has no influence in this situation, because the original maximum range is not even reached due to the high rates. On the other hand, SM creates enough bandwidth to cope with these high data rates for many users, so the ranges won t be constrained by the data rate, but by the maximum path loss function. As

67 55 Figure 4.17 shows, the ranges of the base stations are all about the same. This indicates they are limited by the path loss function, instead of the maximum allowed bitrate. So in this scenario, the ranges limited by the path loss are higher than the ones limited by the maximum data rate. Figure 4.17: Solution Flevoland case LTE-advanced SM Wi-Fi A first study for Wi-Fi runs the tool with a configuration that implies Spatial Multiplexing. We discover that when we install an antenna on every possible location given in Figure 4.14, we achieve a coverage-percentage of 75. From the previous case we learned that the ranges for Wi-Fi are not as high as the ones for LTE-advanced. This explains why we can t reach the required coverage when only using these antenna locations for Wi-Fi. Although we prefer to use the previously shown possible locations, we created a shapefile with much more locations covering the area, assuming poles of 15m can also be placed in the fields. The new possible locations are shown in Figure Note that the poles and installation costs are higher on locations that are not on rooftops of buildings, which only require poles of 10m. So we added a cost of 1000 to cope with this problem. Figure 4.18: Possible antenna locations case Flevoland for Wi-Fi Using this shapefile as input and the Wi-Fi configurations implementing SM, results in a network of 68 antennas connected by almost 36.8 km fiber. Although a Wi-Fi BS is much cheaper, the total costs will be 1,864,989, which is much higher than both scenarios using LTE-advanced. Not only is this technology much more expensive, its optimal solution also achieves only 97.7% coverage, instead of the 99% achieved with LTE-advance. The visual output of the solution is shown in Figure 4.19.

68 56 Figure 4.19: Solution Flevoland case Wi-Fi SM We notice that the ranges of the antennas are really small, which means that one antenna doesn t cover many houses. It s clear that the ranges are limited by the path loss, because they are all the same. In this scenario, TD might help increasing these ranges because the required data rate by one base station is not so high and it increases the ranges of the antennas if they are only limited by the path loss function. Figure 4.20 shows the result of the scenario where TD is used. This time, only 47 antennas are needed and about 28.4 km fiber to connect them all. This results in a total cost of 1,359,222, which is less than the SM scenario of Wi-Fi, but still much more than the scenarios using LTE-advanced. In this case, Transmit Diversity does help to increase the ranges and to bring down the total cost. So for this scenario, there are fewer antennas needed when using TD than when using SM. Note that the obtained coverage here is 98.6%. Figure 4.20: Solution Flevoland case Wi-Fi TD Overview of solutions Table 4.8 gives an overview of the results of the scenarios discussed before. It is quite clear that LTEadvanced (LTE-a) with SM is the best technical choice. It outperforms all other technologies. Not only the Capex is lower, but also the yearly Opex is better than in the other scenarios.

69 57 We can also conclude that SM and TD have a big impact on the performance of the wireless networks. A basic rule is that areas with a high demand of data transfer, due to the density or the data rate, need SM to increase the ranges of the antennas. While in areas where the demand of data is lower, TD seems a good method to increase the ranges. We discovered that the choice between the two depends on the area and the situation, but also on the chosen technology. This case clearly showed that in one technology SM performs better, while in the other the choice should definitely be TD. LTE-a SM LTE-a TD Wi-Fi SM Wi-Fi TD # BSs Km fiber Coverage 99.1% 99.2% 97.7% 98.6% Capex 409, ,576 1,820,789 1,328,672 Opex 21,000 42,000 44,200 30,550 Table 4.8: Overview results Flevoland case Influence of height In previous scenarios, LTE-advanced base stations where placed on 30m poles and Wi-Fi antennas on 10m poles on 5m high roofs, resulting in a height of 15m. The difference between the two constructions supporting the equipment is 7,500. This made us wonder what would happen if we put the LTE-advanced antennas on the smaller poles, which cost much less. The ranges of the antennas depend on this height, a higher placed antenna will reach further. We tested if placing (probably more) antennas at 15m would be cheaper than the previous scenarios. We ran the tool with both LTE-advanced settings, so once for SM and once for TD and retrieved the results given in Table 4.9. We see that both scenarios are worse compared to the case when they are placed on a pole of 30m. Again, the fact that the choice between SM and TD depends strongly on the situation is emphasized here. With base stations placed at a height of 15m we notice that it s actually better to implement TD. Both the Capex and Opex are lower and fewer antennas are needed to cover the area. Actually, when comparing the solution of the TD-scenario with its solution at 30m, we see that there s only one antenna difference between both solutions. When we make a similar comparison for SM we notice a difference of 7 antennas. This is explained by the fact that only SM could benefit from the increase of the range of the antennas by putting them higher. TD didn t really need this increase in range, because it didn t get that far anyway due to the limitations caused by the required data rates. LTE-a SM 15m LTE-a TD 15m # BSs 11 9 Km fiber Coverage 99.3% 99.1% Capex 888, ,553 Opex 57,750 47,250 Table 4.9: Solution case Flevoland with BS at 15m

70 Conclusion The results of our tool tell us the best way to provide the farmers a wireless connection is to use LTEadvanced with SM, placed on 30m poles. Deploying this network would cost us around 410k with a yearly cost of 21k. We also learned from the previous case that LTE-advanced can cope better with increasing data rate demands, so if in the future the required data rates might go up, LTE-advanced does seem a good choice of technology. By studying this case we also discovered there s no real strategy in choosing between SM and TD, but both techniques have to be tested before making a decision. We discovered there are different factors that influence their performances: The chosen technology The input power of the BS The height of the BS The density and the placement of the population through-out the area

71 59 5. Refined economic analyses In this chapter, we perform some refined economic analyses. We show how the tool can also be used to perform such analyses. Our reference scenario is the scenario of the whole city of Ghent in the urban case, except when studying the influence of the expected adoption. There we use the center of Ghent as reference scenario. 5.1 Influence of expected adoption Figure 2.17 showed us the adoption curve we used in the cases. It represents the predictions of the amount of people that will use the network. There doesn t exist a prediction that is 100% certain, so also this adoption is just estimation and possesses a certain amount of uncertainty. In this chapter we want to take this uncertainty into account. To do so, we constructed alternative adoption curves that represent this uncertainty. The adoption influences the range of the base stations, so a different adoption as input will give us a different output. We will discuss the changes of the output due to these changes in adoption. For creating the curves, we assume that in year 0 there s 60% chance that the regular adoption curve will be followed, 20% that the curve will be higher than expected and 20% that it will be lower. The differences between the curves are situated in parameter m of the Gompertz model. For the higher curves m is 0.25 and the lower curve has 0.15 as value for m, while in the regular curve m is 0.2. These curves and their chances are given in Figure 5.1. Figure 5.1: Possible adoption curves year 0 This uncertainty is present in every year, so after 1 year, we should also take the uncertainty into account. Two new curves are built for every curve created in year 0 in a similar way. Again, a higher and a lower curve with 20% chance are made. This time, the higher curve is a composition of its reference curve until year 1 and the Gompertz curve with an m that s 0.05 higher but shifted down to be equal as its reference curve in year 1. The lower curve is constructed in the same fashion. The higher and lower curve for the regular curve (m=0.2) are shown in Figure 5.2 along with their chances. Note that these chances have to be multiplied with the chance the reference curve appears,

72 60 which is 60% in this case. The higher curve is displayed as RU, which is the notation for regular-up, and the lower curve as RD, corresponding with regular-down. Figure 5.2: Possible adoption curves year 1 This process is repeated each year for every curve that s created in the year before. The only difference is that every year the difference of parameter m between the higher and lower curve and their reference curve is cut in half. This is done to cope with the fact that every year the uncertainty is a bit less, because the predicted period that is shorter. So in year 2 the difference is as shown in Figure 5.3. This difference equals in year 3 and in year 4. Year 5 is the last year of the gradual roll-out so no more prediction is made in that year. Figure 5.3: Possible adoption curves year 2 Creating all the curves in every year results in a collection of 243 curves that are shown in Figure 5.4. We ran the tool with the input of the urban case, in the scenario of the center of Ghent at 5 Mbit/s DL, once for each of these curves.

73 61 Figure 5.4: All possible adoption curves sensitivity analysis We set out the total costs of the results along with the chance they appear on a graph shown in Figure 5.5. The costs vary from 800,000 to about 1,500,000, which is almost double. Compared to the reference scenario (normal adoption) resulting in 1,180,098 the lowest cost is 32% lower, and the highest cost is 27% higher. Note that a higher adoption brings a higher income, because more customers will pay for the service. So even though the cost is higher, the income is higher. Actually, if we look at year five, the adoption varies between and 0.275, so in case of higher adoptions our income will be about 3.5 times as high. This means a higher adoption would be beneficial when taking an income per person into account. Back to Figure 5.5, we notice that costs near the reference scenario the only one with a 0.08 chance have more chance to appear than scenarios with much lower or higher costs. We calculated the mean, which is the sum of the costs multiplied by the chance they appear. The mean equals 1,170,375, so there s 10,000 difference between the reference scenario and the mean. This means that in reality, there s a little more chance the costs will be lower than expected. Figure 5.5: Influence of expected adoption