Millimetre-Wave Spectrum Sharing in Future Mobile Networks: Techno-Economic Analysis

IN

DEGREE PROJECT ELECTRICAL ENGINEERING, SECOND CYCLE, 30 CREDITS

STOCKHOLM SWEDEN 2016 ,

Millimetre-Wave Spectrum Sharing in Future Mobile Networks

Techno-Economic Analysis EHAB ELSHAER

KTH ROYAL INSTITUTE OF TECHNOLOGY

SCHOOL OF INFORMATION AND COMMUNICATION TECHNOLOGY

SCHOOL OF INFORMATION AND COMMUNICATION TECHNOLOGY

i

Abstract

Mobile operators passed through many phases in the market over the last several years, since the beginning of mobile broadband services. When the first smart phone was introduced in 2007, it caused a huge increase in the traffic and the users demand kept on increasing. This exponential growth has led to a severe shortage in the available capacity which caused that mobile users don’t have their promised quality of service and coverage. Operators began to put different scenarios for the next-generation mobile networks, putting in consideration the expected increase in the number of users along with their high demand.

As a new proposed solution to the scarcity of empty spectrum slots, operating in higher frequency bands (noted as mmwave) emerged as a solution that will provide larger bandwidths with lower prices for the license. Mmwave will provide users with high data rates but on the other hand, has a low penetration rate that can be fixed by increasing the base stations. Another technique for the operators to follow is that they share their own spectrum with each other, by changing the classic way of exploiting the spectrum which proved a low efficiency and high cost, operators can increase their spectrum and coverage with lower cost.

To get a clear understanding of how the operators will decide their future

strategies, a technical analysis of the new strategies will not be enough, a

technical one also will make it clearer and will help the operators in making

the decision. The objective of this thesis is to perform a Techno-Economic

analysis to get a full image of the system performance. Our system will consist

of 2 operators working in mmwave band with antennas equipped with

directional beamforming and the base station transmitters will consist of

small cells serving outdoor users only. The main question we want to answer

is what will be the effect of decreasing the beamwidth on the system

performance and when the operators will need to share their spectrum with

each other. The performance evaluation will be based on measuring the

downlink achievable rate. As we will be performing an economical evaluation,

the number of base stations required in each strategy will be an important

parameter to evaluate its economic feasibility and cost savings. The different

scenarios will include variations of the beamwidth and coordination between

the operators with an objective of seeking the best performance along with

cost savings. The results should give us a clear look on how the operators will

decide for a certain strategy depending on downlink data rate as a KPI and the

number of deployed base stations as a limiting factor.

ii

Acknowledgments

This work would not have been possible to complete without the support of many

people to whom I want to show my gratitude. First of all I would like to dedicate this

thesis to my wife Doaa and my daughter Layla, their support was infinite. I would like

to thank my family for their continuous support. I also want to thank my supervisor

Ashraf Widaa for his guidance and continuous help, our continuous talks helped me

and guided me throughout the thesis as we kept sharing our ideas to come up with a

good formulation. I really appreciate his help and understanding. I would like to

thank my examiner Anders Västberg for his patience and understanding. Finally I

want to thank my brother Hesham for his continuous support before and during the

thesis, his help and motivation had an essential role.

Table of Contents

1 Introduction ... 1

1.1 Background ... 1

1.2 Approaches to gain better spectrum utilization ... 2

1.3 Problem Formulation ... 3

1.3.1 Research gap and questions: ... 4

1.3.2 Benefits, Ethics and Sustainability ... 4

1.4 Basic Methodology ... 5

1.5 Delimitations ... 6

1.6 Outline ... 7

2 Theoretical background & Related Work ... 8

2.1 Mmwave Band ... 8

2.1.1 Poisson Point Process in Stochastic geometry ... 9

2.2 Small cells ... 11

2.3 Beamforming and massive MIMO ... 13

2.4 Spectrum Sharing ... 14

2.5 Breakdown of economic gain ... 15

2.6 Related Work and Contribution ... 16

2.6.1 PPP for Base Stations positioning ...16

2.6.2 Beamforming in small cells ...16

2.6.3 Massive MIMO ...17

2.6.4 Sharing Scenarios ...17

2.6.5 Measuring Cost Savings ...19

2.7 Our Contribution ... 20

3 System Model ... 21

3.1 Methodology ... 21

3.2 Operation in 28 GHZ ... 22

3.3 Locating Serving Base Station ... 22

3.4 Main lobe Beamwidth ... 23

3.5 Measuring SINR & Data Rate ... 24

3.6 Applying Spectrum Sharing ... 24

3.7 Simulation Parameters ... 25

3.8 Economic Model ... 25

4 Analysis and results ... 27

4.1 Simulation Analysis ... 27

4.1.1 Typical user analysis ...27

4.1.2 How the cell edge users are affected by BS density ...28

4.1.3 Interference Measures ...30

4.1.4 Coordination effect on the performance ...31

4.1.5 Setting a downlink data rate as a KPI ...32

4.1.6 How the increased user density will affect the performance ...34

4.1.7 Variation of Allocated Bandwidth ...34

4.2 Economic Model Analysis ... 35

5 Conclusion ... 38

References ... 39

1

1 Introduction

1.1 Background

Wireless and mobile internet usage is facing a high increase in their

penetration rate and the user demand, the increasing rate is expected to be higher in the next few years. The traditional way of spectrum exploitation is not efficient as the spectrum price is increasing and mobile operators need higher BW in different frequency ranges. The expected data rates at 2020 will be in the range of 1Gbps indoor and 100Mbps outdoor [‎1][‎2], the total mobile traffic are expected to reach 40 Ebytes per year as seen in fig.1 where the traffic grows around 10 percent quarter-on-quarter and 60 percent year-on- year [‎2], the graph includes the data traffic only as the voice traffic is stable with no clear increase over the years and small consumption of the bandwidth.

Given the amount of available spectrum allocated to mobile communication systems, no operator will be able to meet the demand for high data-rates on its own, when the exclusive traditional way of spectrum utilization continues.

Figure 1 Mobile traffic per year

As seen in fig.2, the revenues from using mobile broadband currently represent a small share from the total revenues. The graph represents what 1GB will provide as revenues if it’s from mobile broadband and its equivalent in SMS and voice calls [‎3]. Assuming that SMS consumes 160 bytes, so 6554 SMS will consume 1 MB, average price per SMS is 0.13 SEK. Voice consumes 47 kbps including overhead. Voice calls and SMS don’t consume the available Bandwidth compared to mobile broadband usage.

Figure 2 Revenues from different usages

0 10 20 30 40 50

2011 2016 2020

D at a tr affic in EB per year

1 10 100 1000 10000 100000

Mobile

broadband Mobile voice SMS

EU R

Source: OECD Internet traffic exchange, Svensk telemarknad Source: Ericsson Mobility Report, June 2016

2

It was shown that 5 minutes youtube video consumes data equal to 11500 SMSs while the difference in revenues is 1:23 000 SEK assuming regular prices [‎3].

In conclusion, traffic is growing but revenues aren’t expected to increase proportionally with it, which give us a strong motivation to search for new ways for better spectrum utilization and reduce the costs necessary to operate the network. In our thesis, the main problem that we want to tackle is how to make a balance between upgrading the network and keeping the revenues higher that the expenses.

1.2 Approaches to gain better spectrum utilization

As shown in the previous section, the need for more spectrum without the conventional ways is getting more important. The operators had to search for new efficient ways to exploit the existing spectrum in a better way and acquire more spectrum. In the past years, many studies were made to present new thoughts and techniques for spectrum usage. Operating in mmwave band emerged as a solution for the scarcity of empty spectrum slots in the conventional cellular frequencies [‎4]. The large pieces of spectrum available at this band give possibilities of much larger bandwidth and higher data rates.

With the usage of mmwave, many transmittion techniques were proposed to achieve the best technique suitable for mmwave band. A proposed technique that proved to be efficient in previous mmwave models [‎7] is using highly directional antennas transmitting with small beamwidth [‎5]. Noted as beamforming, this technique has the advantage of reducing interference between the neighboring base stations operating in the same frequency bands.

Beamforming requires modifications and some enhancements in the antennas of both transmitter and receiver side to operate correctly.

A companion for the operation in mmwave band is small cells deployment.

Normal macrocells are not suitable in the mmwave band due to their low Revenues for similar

amount of data

*Assumption: YouTube 0.5 Mbit/s, Price 0,05 SEK per MB.

6250 SMS per MB. Price 0,20 SEK per SMS

Source: OECD Internet traffic exchange, Svensk telemarknad

3

penetration rate which requires the usage of cells with lower coverage area.

Small cell usage has been adopted by many operators nowadays to be used alongside macrocells in high density areas for their low power consumption, deployment cost and high spectrum efficiency as they have a maximum cell range of 50m [‎6].

As exclusive licensing of the spectrum proved to be a big waste in terms of spectrum usage, sharing the spectrum among operators is a suitable solution for next-generation mobile networks as it gives an efficient spectrum usage whether the spectrum is licensed or unlicensed. Spectrum sharing has many strategies and scenarios for the operators to apply together in order to get the best performance. All these concepts can be used together to create a more efficient mobile network that meets the user’s increasing high demand.

An effective example of network sharing among operators in the market is the LTE project in Sweden between Telenor and TELE2 [‎41]. They decided to deploy one LTE network and build it from scratch together and share their spectrum. They equally founded a new joined venture called Net4Mobility which will manage the LTE network. They manage to buy license in 900MHz and 2.6 GHz band as they totally shared it along with the infrastructure. This joint venture made cost savings that helped the operators to achieve LTE coverage in all the country and they also managed to make a 50 % extension on GSM network coverage.

1.3 Problem Formulation

The problem that the mobile operators are trying to deal with here can be divided into two parts. First, the increased demand from mobile users that require high quality of service and data rates with less spectral and physical resources to fulfill the demand. Secondly the revenues problem, as mentioned earlier the operators are facing an increase in their expenditures while the revenues expected from new technologies are not increasing proportionally with it. When we try to define our main research questions, some main parameters has to be taken into consideration including data rate, SINR, base stations density.

Our main focus will be on applying beamforming and spectrum sharing concept in a network based on small cells operating in high range frequencies.

The main questions here are will the focusing of the beam be enough for the mobile operators to satisfy their user’s future demand without the need for another solution like spectrum sharing? How the mmwave spectrum can be authorized from the regulator and operator’s side, can it be Exclusive use or shared use? What will be the cost savings resulting from following the different scenarios?

By setting the exclusive use, each operator works in its own licensed spectrum while facing the shortage of empty slots and the high price to buy a license for new slots and gain more bandwidth. In shared use, operators are sharing their bandwidth so that both each operator can gain more capacity. Many problems arise with the shared use:

 Increased Inter-Operator interference resulting from the high number of BSs operating in the same frequency band.

 Expensive dedicated backhaul for real-time signaling in case of

coordination among operators.

4

 Operators will have to change their strategies in the market to a non- selfish one.

As shown in the first section, revenues aren’t expected to increase while the cost of deploying a network is expected to highly increase due to the changes in the base stations density and the complexity of the equipment in the base stations and Backhaul network.

From the operators side, the economic gain is as important as improving the performance. Controlling the network’s deployment cost is a problem the operators are dealing with while improving and modernizing their network.

Part of our motivation is the economical part where we connect our analysis of the network’s performance with the cost savings of the network in study.

1.3.1 Research gap and questions:

Although extensive research efforts were made in mmwave band, the main focus was on applying it in an indoor environment, and without focus on the outdoor users and how they will be served. In [‎7], the analysis was based on outdoor user performance without putting into consideration the economic aspects of each scenario, only the technical performance. Also the effect of beam focusing wasn’t related to the spectrum sharing and putting it as a replacement. What we are trying to do in our thesis is that we study how decreasing the beamwidth can enhance the performance and what is its best performance without the need for the operators to share their licensed spectrum and coordinate with each other. The other gap we want to research it is the economic gain of the analyzed model. A Techno-Economic analysis was made in [‎8], which was mainly about indoor-users and with spectrum sharing scenarios only. We will try to reach a concrete comparison between the mentioned scenarios from both technical and economical points and study the possible trade-offs.

Our main research questions will be as follows:

 How focus the beam should be in order for the operators to share millimeter wave spectrum without cooperation?

 What are the trade-offs between the improvement in the experienced data rate per end-user and modernization cost?

The goal of the thesis is to find a suitable strategy for the operators to work together maximizing their performance while controlling the Total Cost of Ownership (TCO) so that the new strategies can achieve a cost savings that will motivate the operators to follow them. Through this thesis, we will develop a system model that simulate the distribution of base stations of 2 mobile operators using stochastic geometry which will help us deduce the required base station density to achieve the target data rate with reducing the transmitter’s beamwidth and taking into consideration different levels of coordination among the operators.

1.3.2 Benefits, Ethics and Sustainability

In our thesis, the aim of the Techno-Economic Analysis is to put the economic

aspects along with the technical ones to help the operators to reduce the cost

5

of operating the network. Reducing the operators cost will directly reflect on the users as the service cost will decrease for them accordingly. The costs for the infrastructure to operate make the operators pay a large bill which represent 50% of the OPEX [‎9]. The users will pay less and get bigger data bundles in return as the operators will have bigger bandwidth available. The low power consumption of small cells is considered as an advantage for the environment compared to the high power consumption of traditional macro cells [‎10]. As a basic concept in green cellular networks, coordination and collaboration among operators is highly related to construct a network that is friendly to the environment. As a result of reducing the network energy consumption, CO

emissions will be limited along with the energy expenses which attract researchers for this area [‎11].

Another point is regarding sharing spectrum and infrastructure among operators, ethical problems can arise from the fact that each operator has its own economic gain as a priority when setting their marketing strategies. By taking the decision of cooperation instead of competition, the operators agrees together on a common target where they work together to achieve it. On the other hand, at the competition decision each operator aims at maximizing its own profit in a selfish manner [‎8]. For the operators to agree on a common objective is a big challenge where they have to set guidelines for coexisting rules which are sometimes noted as “etiquettes”. It may be considered as an ethical challenge for the operators to follow these etiquettes strictly.

When the concept of operating in mmwaves band was put in research, the idea was to put a sustainable solution that will provide the increased data rates and high demands from the users. The large bandwidth that will be available to each operator compared to traditional systems now is expected to be sufficient for the operators for many years to come which provides a sustainable network [‎12].

1.4 Basic Methodology

The thesis will perform a cost-capacity study of an outdoor ultra-dense network scenario. Thus, the quantitative methodology is considered the most appropriate since the study will depend on numerical data derived from simulations of the examined technologies. The quantitative method will emphasize the numerical analysis of the data collected in an experimental way to establish associations between different variables. In order to answer the research questions, a techno-economic analysis will be made to evaluate the system performance and the economic gain with several scenarios for network simulation that will help in gathering the required data.

The thesis procedure will follow the listed phases:

 Literature study of the possible solutions.

 Deciding the desired design of the network with the required parameters.

 Setting the different scenarios to simulate.

 Simulation of the network using MATLAB.

 Analyzing the network’s performance in terms of SINR and data rate.

 Formulate a dimensioning problem to quantify the required number

of BSs

6

 Setting the economic model.

 Apply the previously obtained results to the economic model for cost saving figures.

 Proposed strategy based on the obtained results.

Figure 3 Thesis Methodology

1.5 Delimitations

In our thesis, our analysis focuses on outdoor users only due to the low penetration rate of mmwaves which make it hard to reach indoor users. Which make us assume that indoor users will be served by separate indoor hotspots so that outdoor small cells serve only outdoor users.

The time frame of our analysis is the time around 2020. Operating in mmwaves band is not yet applied in the market and it’s currently still in the experimental phase. As a result, the revenue parts will not be included in the economic gain as we will consider the costs part only. In addition, getting accurate numbers to use in the cost model wasn’t possible and any assumptions regarding the current costs for small cells wouldn’t give clear results due to the many variables that will affect the process. One way to overcome this issue is by using the base station densities as a main reference to compare the cost savings between different strategies. This can be considered as a neutral element that gives a relative cost comparison where cost is normalized by a reference and without the need of applying numbers.

As we focus on the data rate and the Base Stations density, physical

architectural implementations and design protocols will not be studied. Also

handover techniques between different BS and load balancing will be out of

our scope.

7 1.6 Outline

This thesis is organized as follows. We explain briefly the literature review regarding the needed topics, in chapter 2, we introduce the mmwave band and how it works. Along with it stochastic geometry application used for designing the network with Poisson Point Process (PPP) compared to the classical distribution of base stations while applying the Euclidian Distance concept in our design. The reason of using Small cells in our network is also mentioned and why conventional macrocells cannot be used when operating in mmwave band, also we will be explaining the beamforming and massive MIMO deployment in small cells to get the best performance from small cells. After it, we’ll introduce the basic concepts and scenarios of spectrum sharing explaining the differences between them and the motivation for the mobile operators to co-operate with each other. To finalize the literature review, we explain the different strategies that can be used to compare the economic benefits that operators can use to evaluate the possibility of any cost saving in the chosen strategy.

After, we’ll show the related work to the previously mentioned concepts and how they were applied while showing the results that we will use in our work.

Next we will show our contribution in this thesis and the results that we will try to achieve. Last, we’’ introduce our contribution to the previous work mentioned and the results that we’ll obtain from it.

In chapter 3, we introduce our model that will be used in the analysis explaining its different parameters and formulas. We’ll have a system model where we show the design of the network and evaluate its output performance.

After it, we’ll move to the economic model where we explain how they are both related and that we’ll need the technical model to get a clear result from the economic one.

In chapter 4, we apply the mentioned system model to analyze the

performance by comparing the output of different strategies of downlink data

rate and SINR coverage. Next we’ll use the obtained results and apply them in

the economic model to compare the different strategies and evaluate their

economic feasibility depending on the cost saving achieved.

8

2 Theoretical background & Related Work

2.1 Mmwave Band

Spectrum usage in the 300 MHz-3GHz range is increasing and it’s becoming overly crowded, while the 3 GHz-300GHz is still unutilized in an efficient way [‎13].

This large piece of spectrum is divided into two bands; the 3-30 GHz is referred to as SHF band (Super high frequency), While the 30-300 GHz is referred to as EHF band (Extremely high frequency) which is also referred to as millimeter-wave band. This part of the spectrum can be used to design an MMB system (millimeter-wave mobile broadband) that utilizes this large part of the spectrum with larger available bandwidth and higher data rates.

One disadvantage of operating in mmwave band is the very low penetration rate compared to lower frequencies that can penetrate buildings while mmwaves can’t penetrate walls [‎14]. As a result, indoor users are expected to be served by different base stations other than outdoor users. For instance, indoor mmwave femtocell or Wi-Fi hotspots can be used to serve indoor users separately [‎7].

Figure 4 Millimeter-wave Spectrum

Out of the 3-300 GHz portion, some bands will be excluded due to limitations in their transmission range. For instance, the 57–64 GHz can reach an attenuation level 15 dB/Km as the electromagnetic energy is absorbed by the oxygen molecules. Also in the range of 164-200 GHz, which is called the water vapor absorption band, the attenuation level depends on the amount of water vapor so it’s not reliable to transmit in this band [‎4].

After excluding these 2 bands, the total available bandwidth will equal in 252 GHz to operate in it. An expected share of 40% can be used for MMB systems, which results in about 100 GHz with an expected 500 MHz Bandwidth per operator.

In Sweden, the spectrum auction for 4G license began in 2008 with the participation of 9 operators for a total bandwidth of 190 MHz in the 2.6 GHz band. The auction ended in a total of approximately 2 billion SEK [‎15].

In mmwave band, unlicensed spectrum usage is highly promoted than

licensed usage but in case of the licensed one, sharing the license among

operators will be a good strategy to reduce the cost while trying not to affect

9

the achievable data rate per user with the expected increased interference.

One way to maintain that is by coordination between operators that shares the spectrum to reduce the resulting interference.

2.1.1 Poisson Point Process in Stochastic geometry

As mentioned earlier, the interference is a major limitation and an important parameter to evaluate the network performance, with the expected increase in the user density and base stations which will cause higher interference.

Studying the signal-to-interference-plus-noise ratio (SINR) is inevitable to know the channel capacity as the interference value will affect the ratio heavily.

Stochastic geometry is defined as the study of random spatial patterns and has applications in many areas like Astronomy, Communications, Image analysis, etc...[‎16]

In wireless networks, stochastic geometry has been used to obtain a better understanding of different types of networks and their performance bounds by developing tractable models that provides clear insights into the design of next-generation mobile networks.

A new model was developed for SINR using stochastic geometry that gives expressions for the downlink SINR CCDF [‎17] which is equivalent to the coverage probability and downlink rate coverage. The main idea is by randomizing the base stations position instead of placing them deterministically on a regular grid. In fig.5&6, we see a comparison in cellular networks modeled via grid-based model and PPP model (Poisson distributed base stations), where the cell boundaries are forming a Voronoi tessellation.

Figure 5 Grid-based model

10

Figure 6 Poisson distributed base stations and mobiles, the cell boundaries form a Voronoi tessellation.

Choosing this model is based on the model developed in [‎7] for sharing the spectrum in mmwave cellular systems, it presents effective methods for measuring the performance in mmwave systems.

The modeling for the locations will be a homogeneous Poisson point process (PPP) with a certain density where the base station position will be an independent parameter. PPP is the model most widely used in wireless networks for spatial location of base stations with no dependence between them in a random number and it can be defined on the entire plane. A PPP Φ of density λ has the following properties [‎16]:

 The number of points in a bounded set A ⊂ R

has a Poisson distribution with mean λ|A| , meaning that the points are independently and uniformly distributed in the set A

 P(Φ(A) = n) = exp(−λ|A|) (λ|A|)

/n!

 A ⊂ R

, B ⊂ R

and A ∩ B = Φ

Mobile users can also be placed according to a homogeneous PPP with a different density where each user is associated with the nearest base station with distance r and other base stations will cause interference. Noting that all other base stations should have a radius larger than r and the interference value in SINR expression will be a cumulative value of interference resulting from all other base stations except the one serving the user [‎17].

Due to strong interference generated by close base stations that affects the users SINR, the PPP model is expected to give a pessimistic result compared to grid model interference from other BS is partially neglected due to the optimality of the coverage plan.

In wireless networks, the distance between base stations and users can be set

as Euclidian distance when they are distributed as Poisson Point Process

distribution. Euclidian distance offers new parameters for distance

distribution that can be used to enhance wireless networks planning by the

joint distribution between the center of the typical cell and its neighbors [‎18].

11 2.2 Small cells

The concept of small cells was introduced as a low-power, short range wireless access point. A network composed of only macro cells will not satisfy the increasing data traffic [‎19]. Small cells have advantages like high data rate, reduced network cost and energy saving, they can be deployed in areas with users that require high data rates or needs enhancement in coverage.

Small cells can operate in high frequency bands beyond 10 GHz to gain more bandwidth. As a result, path loss is expected to increase proportionally square with the carrier frequency [‎20], the attenuation level in the signal can reach 20dB [‎19] and the network coverage will be reduced.

The term “small cells” was originally used as an “umbrella term” to refer to any of the cellular base stations that were used in mobile communications with smaller coverage area and low power consumption [‎21].

With the development in the cellular industry, cell sizes were determined in a hierarchy to specify the features of each cell and the possibilities of using them as shown in table 1.

Table 1 Different Cell Size’s properties

Macrocell Microcell Picocell Femtocell Application Outdoor

Rural areas Outdoor

Urban areas Outdoor &

Indoor Urban areas Campus

Indoor Outdoor(in future ultra- dense

deployment) Capacity of

users Up to 1000

users Average 100

users 20-30 users 5-20 users Range Up to 30 Km Up to 1km 100m-1 km Maximum

100m

Power 20-40W 1-2W 0.1-0.5W 20-100mW

In our research, we’ll focus on femtocells as they have a suitable range for the mmwave frequency band with their low penetration rate and high data rates provided.

Femtocells are wireless base stations with limited range (maximum 100 m), high data rates compared to traditional macrocells [‎10]. In general, femtocells are a lower version of macrocells in terms of power consumption and expenses. The technology used is nearly the same which gives the possibility of integrating femtocells into a wider area network of macrocells. While this integration offers complexity in hardware and software components, it also offers better management of the network resources like spectrum, backhaul network and mobility management. Femtocells require lower CAPEX and OPEX to establish and operate compared to macrocells as seen in fig.7. In an LTE network, a macrocell costs 10 times more OPEX and 8 times more CAPEX than femtocells as it consumes more electricity and requires a bigger physical location [‎22]. On the other hand, macrocells will require less number of base stations to cover a certain area compared to femtocells depending on the available bandwidth and the required data rates.

12

Figure 7 Macro & Femtocells cost comparison

In addition, femtocells support a variety of deployment scenarios including different user access methods like open or closed access, and different coverage plan to be indoor or outdoor. The new expected deployment scenario in the next generation cellular network is to deploy femtocells in an outdoor environment for wide-area deployment [‎23]. Indoor users can be served by either indoor femto cells or Wifi APs, note that they will not interfere with outdoor base stations due to low penetration rate of wave at large frequencies.

Femtocells will help in filling the coverage gaps that will be hard to cover by macrocells.

In general, the market drivers for small cells are shown in fig.8.

Figure 8 Market drivers for small cells

Small Cells can also be installed in existing infrastructure as they do not need a dedicated site of their own [‎24] due to their small size. They can be placed in street lamps, bus stops, billboards and kiosks which are called Street Furniture Sites, also underground sites below street level where fiber and power cables already exists. Ericsson in cooperation with Philips developed an advanced LED street lighting system that contains small cells which provided results of 50 to 70 percent in energy savings [‎24].

0 1000 2000 3000 4000 5000 6000 7000 8000

LTE macrocell LTE femtocell

U SD 8yrs OPEX/Mbps

CAPEX/Mbps

• Ultra Dense Deployment

• Outdoor Coverage

• Higher Data Rates

• Better Coverage

• Larger Bandwidth

• Increased number of Cells

• Real-Time Applications

• Location Finder enhancment

Value Added Services

Capacity Enhacment

Deeper Coverage Improved

User

Experience

13 2.3 Beamforming and massive MIMO

Multiple-Input Multiple-Output (MIMO) techniques are generally used by Base Stations (BS) and user equipment (UE) to improve the robustness and data rates in the new generations of mobile networks. BS and UE are equipped with multiple antenna elements to benefit from MIMO capabilities in both uplink and downlink. Base stations are usually equipped with antenna arrays composed of 100 or more small antennas [‎25]. The massive MIMO deployment consists of equipping the base stations with antenna arrays composed of a large number of small antennas plugged together so it will be possible to serve different users during the same time-frequency block [‎26].

Figure 9 MIMO system with M Tx and N Rx antennas

In beamforming technique, the signal strength is focused in a specific direction as it creates narrow radiated beams with small angle of width (between average of 5

and 30

). As a result, this will maximize the signal energy and minimize the interference with the nearest users, which will provide better signal for cell edge users.

Figure 10 Use of User’s RS in Beamforming

14

For the UE to communicate correctly (fig.10), it needs its specific signals to be able to demodulate the PDSCH (Physical downlink shared channel) while other common channels like PDCCH (Physical Downlink Control Channel) can be transmitted without the small beamwidth and reduce complexity at both ends [‎25]. This way guarantees better spectrum usage and minimizing interference with no change in the Backhaul network. On the other hand, it can also be considered as a costly solution due to the higher complexity level at the Base Station side [‎27].

2.4 Spectrum Sharing

In wireless communications, the spectrum is considered the most important and valuable resource [‎28], especially in the next generation mobile network where the number of devices is rapidly increasing as well as the need for higher data rates. An efficient usage of the spectrum will result in a decrease in the CAPEX (Capital Expenditures) of the network while increasing the performance and the quality of service; spectrum sharing is expected to give an average gain in spectrum between 50 and 100 percent depending on the level of sharing [‎28].

Sharing among operators has basically 2 forms:

 Infrastructure sharing: Where the operators share their physical resources. For instance; equipment, sites or cables. Users of other operators can connect to Base stations of other operators and as a result, get better coverage by co-locating sites. In addition, capacity in dense and congested areas are increased also coverage in remote areas will be better as a few number of sites will be enough for the small number of users. Also site sharing will result in a large cost reduction in both Capex and Opex. It can reach 15% in 3G network [‎29].

 Radio Spectrum sharing: Different operators or different services use the same frequency band in either a licensed or unlicensed form. As a result of the increase in spectrum utilization and the need for a cheaper way to gain more spectrum [‎30]. Mobile operators will be able to introduce better services that require high data rates and users will have a better experience in rush hour times in congested areas.

While spectrum sharing provides larger bandwidth for the user, it will experience a lower signal-to-interference-plus-noise ratio as the noise power increases with the bandwidth along with cross-network interference.

One way to combat the increased interference is by coordinating among operators.

Coordination includes sharing relevant information among the operators to control the inter-operator interference [‎32].

The higher the level of coordination is, the higher the accuracy and frequency

of information exchange which results in an enhanced global knowledge of the

entire system [‎31]. On the other hand, Coordination has some disadvantages

that it requires dedicated Backhaul installation and highly centralized

architecture which will be costly and requires high Capex [‎32]. The different

types of coordination are as follows:

15

 Full Coordination: A tight coordination (noted as Full Coordination) will give the highest performance as Inter-Operator Interference is minimized. A disadvantage is that it requires expensive dedicated backhaul installation and has low scalability.

 No Coordination: The performance will be at a lower level and less adaptability to any variations in the user demand and density; on the other hand, it will not need a dedicated backhaul so it will be less expensive and less complicated as it will not require additional configuration. Also as the network will be self-organized without inter- cell signaling, it will have high and simple scalability with independence in BS distribution.

 Moderate Coordination: There’s also an intermediate level of loose coordination or ‘moderate coordination’ which will not require a fully dedicated backhaul and it will be less expensive compared to tight coordination [‎33].

The decision of the operators to coordinate with each other or not will depend on the expected benefits compared to no coordination case if they will be high enough to overcome the difficulties of applying it.

2.5 Breakdown of economic gain

Mobile telecom market has always been very competitive; the competition is between different operators to provide the best service with the lowest price, and the vendors to provide the operators with the best solutions and equipment. As a result, failure of certain technologies is not just due to performance issues, but also due to economic reasons as maximizing revenues is the best motivation for operators to upgrade to a new technology.

Economic gain usually consists of comparing revenue and costs. The revenues section differs from one market to the other due to market dynamics and different business models; the cost section is more useful to study and gives applicable results.

A techno-economic assessment is a good tool to predict the economic viability, total expenditures and revenues of new technologies in comparison with the technical performance. As shown in fig., before deciding the network deployment plan, several parameters have to be taken into consideration including cost analysis [‎34].

Figure 11 Network Planning

The cost analysis must depend on calculating the TCO (total cost of ownership) to get a clear vision on the expenditures of the new technology.

TCO is divided into two main categories, Capital expenditures (CAPEX) and

16

operational expenditures (OPEX). The budget that the operator holds to acquire and deploy new equipment, sites, spectrum license is noted as CAPEX as it’s usually considered as an investment which can be obtained through a loan, while the budget for maintenance and operation activities is noted as OPEX. The equipment part is usually the BS itself including antennas “last mile cost” and the network equipment used to route the traffic to the switching center and the core network “common infrastructure cost” [‎35]

[‎36][‎32]. By estimating the number of base stations needed in the network and the total cost per base station including CAPEX and OPEX, we can reach an approximate value for the total deployment cost.

The cost of a mobile network usually consists of infrastructure and spectrum cost. C

=C

+C

which includes all capital expenditures (Capex) and operational expenditures (Opex) aspects. When applying spectrum sharing, some additional inter-operator costs can be added which is noted as coordination cost [‎32], which depends on the level of coordination among the operators.

C

= C

+C

+C

.

Coordination cost results from the fact that the coordination between different operators requires additional complexity at infrastructure side, which results in additional cost. Also the limited network controllability and management overhead resulting from sharing among the operators will make a difference.

The extra cost depends on the sharing strategy taken by the operators whether it’s loose cooperation or tight cooperation.

2.6 Related Work and Contribution

In this section, we introduce the previously done work related to our research while analyzing the presented results to justify our methodology and the parameters that we’ll consider in our system model.

2.6.1 PPP for Base Stations positioning

The Stochastic geometry was first introduced in [‎17][‎37] as a tool for analyzing the coverage and rate in cellular networks by randomizing the Base stations positions and the cell boundaries will be equivalent to a Voronoi tessellation.

The tractable analysis for a typical user considered that the user is connected to the closest base station while other base stations are considered as interference and measuring both the worst case scenario noted as cell edge user and the average rate noted as median rate. Results showed that the stochastic-based model with PPP distribution is more tractable compared to traditional grid-based models [‎38]. The analysis focused on the coverage and rate, the proposed model offered higher coverage probability and higher data rate.

2.6.2 Beamforming in small cells

Most of the studies on the beamforming focused on a single operator market with studies on the interference between small cells with no clear results regarding interference between several operators.

In [‎39], using beamforming at 15 GHz showed high improvement in cell edge’s

throughput, 5 to 10 times higher compared to LTE without beamforming

which leads to capacity improvements despite the high propagation loss. The

17

study assumed an increasingly traffic demand per area till 4000 Mbps per Km

in a single operator market.

In [‎19], beamforming was introduced as a suitable solution to be used with small cells deployment along with massive MIMO deployment. The direction of arrival (DOA) was introduced as another resource dimension where the transmission beam from the base station can be steered to maximize the transmitted signal. Different beamforming methods were studied in [‎25], the common result that they will require a special antenna array with a distance d≤ λ/2 (where λ represents the signal’s wavelength) which will be more expensive than usual transmission modes.

2.6.3 Massive MIMO

As mobile networks evolve, transmission techniques had to evolve along with it to provide the required data rates. In [‎40], Enhanced MIMO was introduced to be part of 4G evolution so-called LTE-Advanced. There are three operating modes that are essential in deploying LTE to maximize the performance as shown in fig.12.

 Single-user MIMO (SU-MIMO): The classic MIMO deployment where the transmit diversity and spatial multiplexing techniques are used to achieve an increase in the user data rates.

 Multi-user MIMO (MU-MIMO): different number of streams is allowed to reach different users and achieve an increase in the cell average data rate.

 Cooperative MIMO: As similar to Cooperative Multi-point, cell-edge user throughput is increased by using coordination among different base stations in transmission and reception of signals.

Figure 12 Main MIMO modes

2.6.4 Sharing Scenarios

Spectrum sharing when used with beamforming is an important factor that

needs to be studied explicitly. In [‎30], benefits of spectrum sharing technique

were studied with a 2 LTE operators scenario operating at 2.6 GHz carrier

frequency, sharing 2x10 MHz of the spectrum in an orthogonal way (50

resource blocks per operator). The propagation model considers a frequency

selective channel with path loss and fast fading. Results showed that with

spectrum sharing, users were able to gain more data rate from 10 to 100 %. It

18

was noted that even when both operators fully exploit their bands, frequency diversity still enables a better selection of the resource blocks for the users. In [‎28], different sharing options were studied to evaluate the feasibility of sharing the spectrum in mmwave different from the classification mentioned previously where the sharing was either spectrum or infrastructure. The study mentioned another type of sharing which is the access right where users have full access to all the operators. The different scenarios are as shown in fig.13.

 No Sharing: The basic scenario where each operator has its own infrastructure and spectrum and its users don’t have access to other operators. This basic architecture will provide lower data rate and coverage but less interference between base stations of each operator and users.

 Spectrum + Access: Operators have a wider bandwidth as they share their spectrum. They don’t share their infrastructure but users can be associated to any operator. Access sharing requires full coordination among the operators to minimize the resulting interference. As a result of this scenario, operators will have double the bandwidth and double the base stations and users densities.

 Spectrum: users associate to their operator only but the operators share their bandwidth.

 Spectrum + Infrastructure: Almost same as the previous scenario but Base stations are equipped with antennas for each operator. This scenario is characterized by enhancement of the coverage and SINR, as well as reducing cost in terms of CAPEX and OPEX. This requires that operators have equal base stations density before sharing their infrastructure to ensure the equality among them.

Figure 13 Sharing scenarios

Results showed that in the 2nd scenario, with spectrum and access sharing

provides the best SINR coverage rate as a result of the variety of the user

association opportunities. On the other hand, scenario 4 (Spectrum +

Infrastructure) represents a better solution from economic perspective while

the loss in the performance and SINR coverage rate are limited compared to

scenario 2. As a result, operators will be encouraged to follow the

19

infrastructure sharing scenario for its economic benefits as it will compensate for the low performance loss.

2.6.5 Measuring Cost Savings

Economic viability can help the operators deciding which strategy to follow, whether it will cause loss in performance or coverage, the operator can make a compromise to decide which make the network deployment cost considered as a major constraint in this investment to make an accurate business case.

Market studies were made for different techniques like beamforming, spectrum sharing and small cells to study the economic benefits of each solution individually and help in making a cost-performance tradeoff analysis.

In [‎8], the cost for spectrum sharing with different strategies was studied and the benefits behind every degree of cooperation between the operators.

Results showed that making no cooperation has less cost in lower traffic demand (GB/mo/user) but with the increase in the demand, the costs are heavily incrementing after 40 GByte/month so it costs way more than with spectrum sharing. The coordination cost is a parameter that can easily be manipulated between the operators in the total cost formula. The mentioned study included two operators with basic LTE operation without beamforming.

The specific costs for each small cell were mentioned in [‎10] as follows:

 Capital cost of individual small cell

 Licensing cost

 Annual maintenance

 Site acquisition and installation

 Operational costs including electricity and site rental.

In [‎42], total cost calculation was simplified such that OPEX can be calculated as 10% of CAPEX. No price erosion is expected to occur as the technology is expected to enhance and the equipment will be cheaper as price level of electronic equipment is constantly falling [‎44]. One disadvantage of this method that it requires applying specific values in the applied model, which is not the goal in our thesis, which aim to develop a model for estimating the cost savings from the transition of one scenario to the other.

In [‎43], a novel concept of permissible coordination cost was introduced. This concept was a trial to estimate the cost factors associated with the coordination among the operators by solving a dimensioning problem. The dimensioning problem will consist of a linear cost model used for modeling a coordination cost related to a unit deployment cost which will be the number of base stations. A simple linear cost model that was popular for cost analysis in wireless systems [‎44] is used; it consists of cost coefficient that includes CAPEX and OPEX per base station, as number of deployed base stations.

i b basic b i

tot

c N

C  (2.1)

In [‎32], 2 cases were considered in the economic model, one where the coordination cost is dominant compared to base stations cost.

The other is when coordination cost is smaller than base stations cost, this case is applicable when deploying large number of base stations.

Our goal will consist of getting estimation for the cost savings resulting from

adopting different scenarios depending on the number of deployed base

stations.

20

When the operator decides to share its spectrum or perform coordination, the operator can decrease its base station density as the downlink data rate will increase. Added costs can be modeled by a new cost item which can be labeled as ∆c

so the total deployment cost for coordination strategy i will be

i b i b basic b i b i

i

c N c c N

C   (   ) (2.2)

In the Non-coordination dominated case, where is marginal so the relative total cost saving can be expressed as

%

 100

 



b i b basic b

N N

C N (2.3)

In the coordination dominated case, will have a larger value, a new cost item will be defined as permissible coordination cost

which will be defined as

basic b

i b i

b basic i b

perm

c

c N

c  N   

 1 (2.4)

The permissible cost value can be obtained when knowing the number of base stations in the basic case as a reference and in the coordination case which can be obtained from a simulation result that solves a dimensioning problem. The higher permissible cost we have, the cheaper is the new system compared to the basic one.

2.7 Our Contribution

As mentioned in the related work, beamforming was studied with small cells and how it increased their performance. Also studies were made for mmwave band and how the regulators can organize the operation in this band whether it will be licensed or unlicensed. Our model is based on the previous work done in applying stochastic geometry in network planning while we add the beamwidth as an important parameter in measuring the performance.

Our contribution is related to the efficiency of controlling the beam focus and

the best performance we can achieve without sharing the spectrum. We then

add a comparison of the performance when spectrum sharing is added

between the operators while modifying several parameters with different

strategies. The parameters include different degrees of beamwidth, levels of

Inter-operator coordination and base station densities. The comparison

should be useful to the regulatory body to decide the way mmwave band will

be licensed. Next, we extend our analysis to focus on a cost saving comparison

based on the base station density as a reference to reach our KPI which is set

as downlink data rate per user. The result shows the number of base stations

required by each scenario to reach a required data rate.

21

3 System Model

In this section, we explain the model used for evaluating of mmwave network using beamforming and spectrum sharing. We consider a system consisting of 2 different operators that operate in mmwave band (28 GHz) and each operator owns a spectrum license assigned of 250 MHz as Bandwidth with total system bandwidth of 500 MHz assigned equally for mobile operators [‎2].

Mmwaves are highly sensitive to blockages effects which mean that indoor users will not be covered by the outdoor Base Stations. As a result, our analysis will focus on the downlink rate and SINR provided by the outdoor base stations only and experienced by outdoor users using stochastic geometry.

3.1 Methodology

The used system model is related to Andrews model that uses stochastic geometry for Base station distribution and operation in mmwave band [‎7].

First, we’ll consider a system consisting of two mobile operators operating in the same physical area with identical parameters. The analysis will consider a typical user at the origin and measure the downlink rate transmitted from the randomly generated base stations randomly to the typical user with omnidirectional properties. When running different iterations for different user locations, the outputs that we’ll get are:

 Probability that the user is associated with LoS BS

 Probability that the user is associated with NLoS BS with available LoS BS

 Average path loss through base station

 Downlink SNR and SINR CCDF

 Downlink data rate

These outputs give us a primary look at the system performance and how the users can be affected.

Secondly, a density of users is set for each operator (users/km

) to be added to the same scenario while running the simulation with different beamwidths to measure its effect on the system performance and to get a better estimation of the interference between base stations. The base stations density parameter (Base stations per Km

) is an important parameter to measure the economic performance as well.

Third, sharing the spectrum between the two operators is applied with different levels of coordination.

The metrics used to compare the systems are mainly composed of 2 metrics:

 SINR coverage probability: The probability that the SINR at the user from the associated BS is above a certain threshold (CCDF of SINR).

The purpose of using this parameter is that it shows the serving link quality.

 Downlink rate coverage: Probability that the rate of a certain user is above a certain threshold. It represents the data bits received per second per user so it’s a realistic indicator of the system performance.

The parameters that we’ll be useful in optimizing the previous metrics will

consist of transmitting antenna’s beamwidth and base stations density.

22

By analyzing the gathered data, we will be able to determine the suitable density for base stations that leads to the desired data rate with minimum interference.

Next, we start our economic model. The economic gain of mobile operators consists of two parts: revenues and costs.

For the next-generation mobile networks, revenues part is unpredictable and hard to estimate due to market dynamics. As a result, we focus in our analysis on the cost part. Setting specific cost figures for spectrum sharing, coordination and decreasing the beam will be out of our scope as we aim to make a relative cost comparison where the cost is not represented by numerical values but normalized by a reference. That’s when we use the estimated number of base stations as the reference used for cost savings calculation.

In ultra-dense networks, the total deployment cost is approximately liner to the number of deployed base stations. Thus, we formulate a dimensioning problem at a certain scenario and determine the required number of BSs to support our target data rate.

3.2 Operation in 28 GHZ

As mobile operators have a great motivation to work in mmwave band, choosing the specific range of frequencies to operate in can be tricky. 3GPP defined three main levels in mmwave to operate in, 28, 38 and 73 GHz [‎57]. In 2015, Samsung performed channel measurements on the entire mmwave band. Results showed that 28 GHz is the most effective for mobile communication. The measurements included path loss for urban environments with path loss exponent of 3.3 in NLOS links which gives a communication link of up to 200 meters [‎55]. Qualcomm also ran experiments with 128 antennas to show that massive MIMO deployment with directional beamforming can be used in dense urban environments and also in NLOS communications [‎56]. As a result, we’ll operate in the 28 GHz band with 500 MHz allocated for mobile communications. The 500 MHz should be assigned equally to the existing operators.

3.3 Locating Serving Base Station

Base Stations locations are modeled using a homogeneous Poisson Point Process (PPP) Φ with density λ in an area of 4 Km

with specific densities per Km

.

We consider the exponential model for blockage, which has an average LOS

distance of 144m. and will be used to determine whether the link between the

user and BS is LOS or not as a function of the link length, the distance

between them will be set using Euclidian distance based on the location of

each one of them. The probability of having a LOS link inside the LOS distance

will be set to 11% while having it outside the LOS distance will be zero. A

region where no blockage occurs to the user is defined as LOS region having a

stochastic characterization where the cell radius should be scaled with it to

maintain the same coverage probability. The Φ set of base stations are either

LOS or NLOS to a certain user. Let Φ

be the point process of LOS base

stations and Φ

of NLOS base stations = Φ/ Φ

. The probability function will

be noted accordingly as p(R) and will depend on length of the link which is

23

defined as the probability that the link of length R is LOS. NLOS probability function will be noted as 1- p(R).

In other references, p(R) was derived by stochastic blockage models [‎45] such that p(R) was expressed by e

, the blockage parameters were characterized by random distributions and β is set by average size and density of the blockages. As it’s a tractable analysis, LOS probabilities are considered independent between different links so that the correlations of the different links are ignored as SINR evaluation will not be affected heavily [‎46].

LOS association probability is noted as A

which is related to the base stations associated to the user whether it’s in LOS region or outside of it. The decision is based on the distance between the user and the base station and the path loss value. First each user sets a value of the distance to each base station inside the LOS distance. After it a flag will be raised on each BS indicating whether it’s a LOS or NLOS depending on the probability of getting a LOS BS inside the LOS distance and setting the path loss depending on the result. As A

will have an initial value of zero, its value will increase with a value of 1 each time a user is associated with a LOS BS in each iteration and then diving the total value by the number of the total iterations.

AL=0;

if Nflag(AssocBSInd) ==0 AL=AL+1;

end AL = AL/iter

Users select their serving base station based on highest signal provided to them. As we are considering a single-tier network and the transmit power is the same to all the base stations, the distance between a certain user and the nearest interfering base station will sure be greater than the distance between the user and the serving base station.

Figure 14

3.4 Main lobe Beamwidth

Base stations and mobile devices are assumed to deploy antenna arrays that

perform directional beamforming with θ as beamwidth of the main lobe

varying from 5

to 20

. Main lobe directivity gain noted M is assumed to be