Mobility Optimization for Energy-Efficient 5G Networks : Optimering av Mobilitet för Energieffektiva 5G Nätverk

(1)

Linköpings universitet

Linköping University | Department of Computer and Information Science

Master thesis, 30 ECTS | Datateknik

2019 | LIU-IDA/LITH-EX-A--19/085--SE

Mobility Optimization for

Energy-Eﬃcient 5G Networks

Optimering av Mobilitet för Energieﬀektiva 5G Nätverk

Oskar Gustafsson

Supervisor : Klervie Toczé Examiner : Simin Nadjm-Tehrani

(2)

Upphovsrätt

Detta dokument hålls tillgängligt på Internet - eller dess framtida ersättare - under 25 år från publicer-ingsdatum under förutsättning att inga extraordinära omständigheter uppstår.

Tillgång till dokumentet innebär tillstånd för var och en att läsa, ladda ner, skriva ut enstaka kopior för enskilt bruk och att använda det oförändrat för ickekommersiell forskning och för undervisning. Över-föring av upphovsrätten vid en senare tidpunkt kan inte upphäva detta tillstånd. All annan användning av dokumentet kräver upphovsmannens medgivande. För att garantera äktheten, säkerheten och till-gängligheten ﬁnns lösningar av teknisk och administrativ art.

Upphovsmannens ideella rätt innefattar rätt att bli nämnd som upphovsman i den omfattning som god sed kräver vid användning av dokumentet på ovan beskrivna sätt samt skydd mot att dokumentet än-dras eller presenteras i sådan form eller i sådant sammanhang som är kränkande för upphovsmannens litterära eller konstnärliga anseende eller egenart.

För ytterligare information om Linköping University Electronic Press se förlagets hemsida http://www.ep.liu.se/.

Copyright

The publishers will keep this document online on the Internet - or its possible replacement - for a period of 25 years starting from the date of publication barring exceptional circumstances.

The online availability of the document implies permanent permission for anyone to read, to down-load, or to print out single copies for his/hers own use and to use it unchanged for non-commercial research and educational purpose. Subsequent transfers of copyright cannot revoke this permission. All other uses of the document are conditional upon the consent of the copyright owner. The publisher has taken technical and administrative measures to assure authenticity, security and accessibility. According to intellectual property law the author has the right to be mentioned when his/her work is accessed as described above and to be protected against infringement.

For additional information about the Linköping University Electronic Press and its procedures for publication and for assurance of document integrity, please refer to its www home page: http://www.ep.liu.se/.

(3)

Abstract

With the upcoming of the fifth generation of cellular networks there are several perfor-mance requirements that need to be satisfied. This thesis focuses on the topic of mobility which allows users to move through the network using the concept of handovers to switch between base stations. However, the thesis also keeps the energy efficiency in mind and investigates a strategy of reducing the energy consumption. Moving across base stations will inevitably lead to some handover failures, a goal of the system developers is to reduce these, but there exists a tradeoff between too early and too late handover failures. This thesis investigates two approaches of lowering the number of failures by letting the net-work self-optimize parameters in the handover procedure based upon the tradeoff. The first approach includes a parameter adaption to an estimated velocity of the users and the second one making a parameter more granular. Simulating different scenarios in a detailed network simulator shows performance gain by adapting handover parameters to the esti-mated velocity, but gathering more data regarding failures is needed before generalizing the conclusions.

(4)

Acknowledgments

I would like to thank my examiner Simin Nadjm-Tehrani and supervisor Klervie Toczé at IDA. They provided excellent feedback and discussion throughout the work of this master thesis. I would also like to thank Kristina Zetterberg and Pradeepa Ramachandra, my super-visors at Ericsson who have contributed with a lot of knowledge and input on my work. At last I would like to thank my opponent Johan Lindqvist for his input on the report.

(5)

List of Figures

1.1 Overview of the method . . . 4

2.1 The frequency spectrums of LTE and 5G . . . 7

2.2 Illustration of the direction of signals with and without beamforming . . . 8

2.3 Beam setup within a cell . . . 8

2.4 The overall procedure of a handover in 5G . . . 9

2.5 The functionality of the Offset and TTT parameters . . . 10

2.6 The concept of the too late and too early handover illustrated where the dashed lines represent signals not reaching its target . . . 11

3.1 The high level architecture of the simulator . . . 14

3.2 Overlook of the SON functionality in the simulator . . . 16

4.1 The impact of TTT on the different kind of RLFs . . . 19

4.2 The impact of TTT on the number of RLFs and Ping Pong handovers . . . 19

4.3 The impact of offset on the different kind of RLFs . . . 20

4.4 The impact of offset on the number of RLFs and Ping Pong handovers . . . 20

5.1 The initial idea of estimating the UE velocity . . . 24

6.1 The mean estimated velocity and standard deviation for 100 users each with speed constant at 1, 3, 5, 15, 30 (m/s) . . . 31

6.2 The number of RLFs for each of the five iterations without UE speed adaption . . . 32

6.3 The number of RLFs for each of the five iterations with UE speed adaption . . . 32

6.4 The number of RLFs for each of the five iterations changing the offset on a cell-specific level . . . 33

6.5 Number of handovers for each of the five iterations changing the offset on a cell-specific level . . . 33

6.6 The number of RLFs for each of the five iterations changing the Offset on a beam-specific level . . . 34

6.7 Number of handovers for each of the five iterations changing the offset on a cell-specific level . . . 34

6.8 Two networks deployed with different frequency configurations relation between RLF-rate and average resource utilization . . . 35

(8)

List of Tables

2.1 Measurement report triggering events . . . 10 3.1 Common parameters for all scenarios unless it is the investigated parameter . . . . 17

(9)

Acronyms

MRO Mobility Robustness Optimization.

QoS Quality of Service.

RLF Radio Link Failure.

SON Self Organizing Network.

TTT Time To Trigger.

(10)

1 Introduction

With the fifth generation (5G) of networks right around the corner, the capabilities of 5G compared to earlier generations must be extended greatly. The overall purpose of 5G is to make ubiquitous connectivity available for anyone. The internet of things is one main objec-tive where not only traditional devices such as cellphones and computers will be connected, but all things that may be benefiting of a mobile connection [1]. With all focus on 5G, it is also important to note that the introduction of a new generation does not mean that this will be the end of 4G and Long-Term Evolution. 5G will be built on top of the already existing functionality [2].

The capabilities which have to be provided include higher data rates, lower latency, ultra high reliability [3]. The network should also support massive numbers of connected devices and fulfill the requirements from applications such as precise industrial work or critical infrastructure [4][2].

Mobility is an important topic during the development of 5G networks. A handover is the network function that allows for a user to move inside a geographic area and switch the connection from one base station to another. The concept of handover has been around for a long while, but as the networks are evolving and requirements on the network are getting higher, there is still a challenge to reduce the number of failures that exists due to different movement patterns of users and challenging radio environments. The handover procedure includes several different parameters that affect the performance of the network, which may not be optimally configured on all base stations.

Due to the complexity of network architectures, the need for Self Organizing Network (SON) has arisen. The effort of manually installing and optimizing the network setup of base stations in order to deliver the Quality of Service (QoS) promised are increasing. Compared to 4G, 5G will integrate the concept of SON even further as the number of parameters has increased and thus the complexity of configuration as well. With SON self-optimization functionality, the configuration effort as well as the energy footprint is reduced significantly thanks to lower signal overhead. However many parameters included in the handover pro-cedure cause tradeoffs between different kinds of failures. Due to these tradeoffs, algorithm

(11)

1.1. Motivation

development for a SON is a challenging task.

Some problems may not be difficult to solve independently, but seen in a wider context and while trying to minimize the signaling overhead, there may be limitations which make the problems more complex and hard to solve. In the mobility domain, the velocity of a moving User Equipment (UE) is of high importance. However, periodically requesting the UE for the velocity would be too expensive both in terms of signaling as well as putting further requirements on the UE’s capabilities. In this report the problem of acquiring the UEs velocity by estimating a velocity based on the UEs movement pattern will be studied.

1.1 Motivation

Mobility robustness is a term for describing the performance of the network in presence of mobility, with regards to how frequent failures occur in the network. Two metrics for mobility robustness are the handover failure rate as well as the total number of failures. The robustness of a system is of high importance as failures do impact the end user directly. A system having a high robustness means that there are not many failures affecting the per-formance of the system. Even though there exists a lot of research in the subject of Mobility Robustness Optimization (MRO) there still exists unwanted behavior such as Radio Link Failure (RLF) caused by too early or too late handover triggering, or Ping Pong handovers which is unnecessary handovers that puts extra signal costs on the network. Currently, there is no method to completely remove all occurrences of these behaviors. With the new genera-tion of networks, there will be new features introduced such as beamforming signals where signals are transmitted in a different way from the traditional, with purpose of increasing signal conditions. These techniques have not yet been investigated in combination with the mobility robustness research area. Therefore, more work is needed in order to increase the mobility robustness in the new setting.

With 5G networks, there may also be multiple cells transmitting at different frequencies in the same coverage area for a UE to establish a connection with. This provides higher per-formance in the covered area due to load balancing, the use of higher frequencies and less interference. However, this comes at the cost of energy consumption as there may not be a need for multiple frequencies all the time. The load in the cells varies a lot from time to time during a day. One possible way to decrease the energy consumption in the network would be to turn off certain frequencies during times when the load can be handled by a single fre-quency. To ensure the promised QoS these frequencies must stay active until a certain level of mobility robustness is guaranteed for the coverage area.

1.2 Aim

This thesis aims to decrease the RLF rate and the number of unnecessary handovers by in-troducing new variables into the handover decision making. Simulations will be made with a SON that uses an algorithm to update handover procedure parameter values with the goal of increasing the network’s mobility robustness. The thesis will also try to identify a strategy to make the network more energy-efficient by investigating the possibility of turning off base stations operating on different frequencies. A requirement to turn off a frequency would be that another base station covers the area of the turned off cell. A desirable outcome will also be to quantify at what level of resource utilization there need to be several frequencies.

(12)

1.3. Research Questions

1.3 Research Questions

To fulfill the aim of the thesis the following research questions will be answered.

1. With the beamforming technique in 5G, how can the UE velocity be estimated and how accurate is the estimation compared to the actual velocity?

2. With an algorithm changing the cell-specific Time To Trigger (TTT) parameter between each SON iteration, does an adaption of the TTT value to the estimated UE velocity increase performance in terms of number of RLFs and Ping Pong handovers?

3. Using beam specific offset parameters instead of cell-specific, can a SON decrease the number of RLFs and Ping Pong handovers?

4. Given an accepted level of handover failure rate and a strategy of turning off frequen-cies based on resource utilization in order to lower energy consumption, what is the level of resource utilization that a base station can handle before reaching the accepted failure rate?

1.4 Method Overview

An overview of this thesis’s method is shown in Figure 1.1. The workflow starts with a short description of the software implemented in order to simulate a SON. Following is a description of some common parameters that are set in order to create a realistic scenario for the topic of mobility. These common parameters are then used for a prestudy made about the tradeoffs created by changing handover parameters in the simulator, this prestudy is made in order to determine two hypotheses that two SON-algorithms will be based upon. After the prestudy, these algorithms are created alongside the final scenario design. Finally, simulations are made leading to the result of this thesis.

Figure 1.1: Overview of the method

1.5 Delimitations

Although the thesis tries to create an environment as similar to reality as possible the aspects of simulation and handmade scenarios will remain. This means that the conclusions made from these results are specific for this simulator and scenarios created, and can not be fully generalized.

The simulator used in this thesis is not publicly available and therefore details of the simula-tor will not be described. However, some of the results may be affected by how the simulasimula-tor is implemented. In those cases, the issue will be briefly described but not fully motivated with simulator details. The result presented in this thesis is also commercially sensitive data, therefore the presented result will be normalized in order to not show concrete numbers.

1.6 Thesis Outline

Following this introduction, Chapter 2 will cover the background of the thesis work and intro-duce some of the new technology used in 5G together with some theory regarding signaling

(13)

1.6. Thesis Outline

and the handover procedure. Chapter 3 will discuss the simulator extensions made and a common parameter setup made. Chapter 4 will discuss the choice of investigated parame-ters followed by Chapter 5 that will discuss the design of the algorithms and experiments performed in this thesis. Chapter 6 will present the results of the simulations followed by chapter 7 that discusses the result and the work in a wider context. Finally, chapter 8 states the conclusions of this work.

(14)

2 Background and Related Works

This chapter describes the background of 5G as well as the QoS it is supposed to deliver. It introduces some of the new technology that will be used in order to provide the promised QoS and describing the mobility subject for this thesis in detail. Section 2.4 presents the related works on the subject.

2.1 Thesis Environment

The work for this master thesis was conducted at Ericsson AB research department during spring 2019. Ericsson is a world-leading company in telecommunications with about 40% of the world’s cellular traffic1. In Linköping where this thesis was carried out, their focus is on research and development of the upcoming 5G network.

Ericsson has software for simulation of networks. However, it is not fully equipped for sim-ulating all aspects of the SON functionality, therefore a part of this master thesis is to get to know the simulator and extend it with some functionality regarding the topic of the thesis.

2.2 5G Technologies

In order to provide the promised QoS, there are several new technical concepts that are intro-duced in 5G. These concepts include making use of higher frequency bands, device-to-device communication, flexible spectrum usage, multi-antenna transmission [4]. This section will introduce some of these concepts on a high level.

2.2.1 Frequency Bands

The frequencyspectrum used for 5G wireless access will be increased compared to LTE as shown in Figure 2.1. The relevant spectrum ranges from about 1 GHz up to 100 GHz. As higher frequencies carry higher data rates, these will be used to fulfill the data rate require-ments of 5G. However, these frequencies cannot be used on their own as they have a high path loss and air interference. This means that the usage of high frequencies, especially

(15)

2.2. 5G Technologies

above 10GHz, can only act as a complement to lower frequencies to increase system capacity in certain areas [4][5].

Figure 2.1: The frequency spectrums of LTE and 5G

2.2.2 Heterogeneous Network

There are two scenarios of deployment in the current generation of networks. These are often referred to as HetNet (Heterogeneous Networks) and MoNet (Homogeneous Net-works). In a MoNet the deployment typically consist of one type of cells. The borders of the cells are neighboring to the next cell creating a coverage area without much overlap. A HetNet is, as the name suggests, an architecture with different kind of cells. All cell types have different characteristics to suit different environments. In a HetNet there is also most of the times more sharing of the coverage areas. The smaller cells often provide extra throughput for certain areas inside bigger cells bounds and operate on a higher frequency [6]. The cells in a HetNet are usually of four different kinds, macro, micro, pico, and femto cells. The different kinds range from the biggest most power-consuming to smaller ones with smaller coverage area and less energy consumption. Macro cells are typically deployed in rural areas meanwhile micro/pico are deployed at hotspots such as malls or similar. Femto cells are the smallest type which can be used within a personal space and be deployed by an end-user.

2.2.3 Massive Multiple Input Multiple Output

The concept of multi-antenna transmission already exists in current network generation but will be an even more central concept in 5G. Since there are physical limitations such as the non-linear path loss between transmitter and receiver, the base stations have to compensate for the losses. This compensation will be made through beamforming signals directly to-wards certain areas instead of the traditional way of transmitting. Beamforming means that the signals are directed in a specific way as shown in Figure 2.2. There are several benefits of beamforming the signals. It will reduce the interference around the transmitter since the signal only will spread in one narrow direction. Beamforming will also be able to extend coverage and provide higher data rates in areas where the base stations are not as frequently deployed.

Since the base station can only form one beam at a time, this creates a need for some beam management. The general idea of how to handle this is beamsweeping. Beamsweeping forms one beam during a fixed period of time before forming the next one in a different but close direction. By doing this for all the beams the antenna will scan through the whole coverage area in a short period before starting over again. Using this technology also means that the base station needs to be aware of which beam a UE is listening to in order to transmit correctly. For this beam management, there are limitations to how often and how quickly this sweeping can be performed due to signaling overheads and standardization.

(16)

2.2. 5G Technologies

Figure 2.2: Illustration of the direction of signals with and without beamforming

For each cell, there is a set number of beams that together make up the coverage area of the cell, an example is shown in Figure 2.3. As the figure shows, there can be two levels of beams, wide or narrow, represented as the blue and white circles. Wide beams cover a larger area than narrow beams at the cost of less focused signals, meaning that the narrow beams will most of the time have a better signal quality. In this example, the cell has five beams as shown by the figure. The base station is located at the intersection of the arrows, and the five beams projection covers the base stations coverage area. Depending on the height the antenna is deployed at as well as the sector width and height the beams are divided into, the beams may not be as uniformly sized as in the figure. The further away from the base station they are projected, the larger the size of the coverage area on the ground.

Figure 2.3: Beam setup within a cell

2.2.4 Ultra Lean Design

5G intends to have an ultra-lean design, meaning that there should be as little periodic sig-naling as possible. This means that most sigsig-naling should be event-driven and the overhead from periodic signals should be reduced to mainly when a device needs information. This is not an easy task but every step in the right direction means a lower energy consumption, due to less signal overhead as well as increased performance thanks to less interference [7]. By working towards this, there will also be an increase in the complexity of other aspects of the network. At the moment, periodic signals are key concepts in how today’s network act as it is the way different devices can find each other. An example of these periodic signals is the reference signals transmitted from the cells which the UE measures a signal strength from. These signals cannot be removed completely but reduced as much as possible.

(17)

2.3. Mobility in 5G

2.3 Mobility in 5G

Mobility in networking is the capability of a UE to be able to be mobile and still have the same performance as a static UE. One main functionality needed in mobility is the concept of a handover. A handover is the procedure of a UE switching connection from one base station to another.

2.3.1 Handover Procedure

Figure 2.4 shows the main steps of a functional handover procedure made to switch the connected base station. As shown, it all starts with the periodic reference signals transmitted from the base stations from which a UE calculates the signal strength. Upon these data, there is a decision whether or not the UE should send a measurement result report to the base station. This decision is made upon certain triggers shown in Table 2.1. After the report is transmitted it is up to the base station to make the decision whether a handover should be made or not, followed by setup and synchronization between the UE and the new base station [8].

Figure 2.4: The overall procedure of a handover in 5G

There are two central parameters used in the handover procedure that are of importance for this thesis. These are the Offset and TTT parameters. The parameters of interest are used at the "Report trigger check" step. As shown in Figure 2.5 we can see the signal strength of the source cell and a target cell as a UE is moving from the source into the target cell. When the target cell’s signal strength is an offset value better then the source cell, a measurement report triggering event called A3 is triggered. When this happens the UE waits for the TTT value before sending the measurement report. After the report is received at the base station, a decision whether to initiate a handover or not is made. The value of the TTT parameter is not strict, but 3GPP has defined a range of values between 0ms and 5120ms [9].

(18)

2.3. Mobility in 5G

Figure 2.5: The functionality of the Offset and TTT parameters

As described in the 3GPP standard specification [9], there are several report triggering events that all couse the measurement report to be transmitted from the UE. The ones of most interest for this thesis are presented in Table 2.1. The term PCell refers to the primary cell or source cell that the UE is connected to.

Table 2.1: Measurement report triggering events

Event Reason

A3 Neighbour becomes offset better than PCell

A5 PCell becomes worse than threshold1 and neighbour becomes better than threshold2 All events described in the 3GPP standard specification [9] are connected to the concept of mobility. However, A3 and A5 which is shown in table 2.1 are the ones triggering the han-dovers, the rest of the events have different purposes. Even though both A3 and A5 trigger the handovers, event A3 is more adaptive and can be changed while A5 is using global thresholds, named threshold1 and threshold2 in the tables, that should not be changed. Therefore the A3 event is the main focus in this thesis. Equation 2.1 shows the triggering condition for the A3 event. In this equation there are some different variables, Mn and Mp standing for the measured signal strength for the neighbouring and primary cell. All variables starting with the letter O represents different offset parameters. However, for simplicity, they can be seen as one big offset grouped together.

Mn+O f n+Ocn ´ Hysteresis ą Mp+O f p+Ocp+O f f (2.1)

2.3.2 Mobility Difficulties

Even though there are lots of research in the field of MRO there still exists failures and unwanted behaviors due to the difficulties of matching the mobility requirements as well as existing tradeoffs in the network. The RLF is one major issue as it disconnects the user from the network denying its services. One way to define the RLF is when the UE has a signal strength below a certain threshold for a period of time. This time period is defined as a parameter named T310 in the 3GPP standard. In the handover procedure a RLF can occur

(19)

2.4. Related Work

due to three reasons. The following list describes the three different kind of failures and one behavior that is not a RLF occuring in today’s network which are the target of the MRO [10].

• Too early handover - When a handover triggers too early and a RLF occurs between the UE and the target cell after the UE has finished the handover or during the handover procedure. The UE then reestablishes a connection to the source cell. The concept of the too early handover is illustrated in Figure 2.6 where the UE receives the handover command before the signal strength is good enough at the target cell.

• Too late handover - When a handover triggers too late and a RLF occurs between the UE and the source cell. This can happen both during the handover procedure or before a handover was initiated. The UE then reestablishes a connection to the target cell. The concept of the too late handover is also illustrated in Figure 2.6 where the handover command cannot reach the UE as it has moved out of the cells reach.

• Handover to wrong cell - When there is a RLF at the target cell after or during the han-dover procedure but the UE reestablishes a connection to a third cell. This failure can happen due to both reasons illustrated in Figure 2.6. However, instead of reconnecting to the source or target cell, the UE reconnects to a third cell.

• Ping Pong handover - When handovers trigger too easily and the UE switches between cells back and forth. When there are two successful handovers between the two cells within one second, it is classified as a Ping Pong. This is not a RLF but the behavior causes unnecessary signaling.

Figure 2.6: The concept of the too late and too early handover illustrated where the dashed lines represent signals not reaching its target

2.4 Related Work

There are a lot of research with the aim to deliver the QoS promised by 5G. However, there are different approaches such as more high-level designs of the network architecture to low-level algorithm design of handover decisions. There are also different aspects and metrics to con-sider, such as the MRO and the energy efficiency. Some of these works are made simulating 5G and some 4G, but thanks to the different network generations similarities, both are still of interest. This thesis also uses simulation as a tool to gain result as this is the most common and realistic way used in related work.

(20)

2.4. Related Work

2.4.1 Mobility Robustness Optimization

Increasing the mobility robustness is an important aspect in SON, where self-optimization plays a big role in order to satisfy the performance requirements in both 4G and 5G. In mo-bility scenarios, the SON needs to change its parameters over time following some algorithm to optimize the wanted behavior. The following research works are in the topic of MRO and all,similar to this thesis, use the different kinds of RLFs and the RLF-rate as evaluation metric. Mal et al. [11] introduce an algorithm for MRO based on a prediction of RLFs, using positions of UEs to calculate new values for the A3 handover triggering offset. In order to do this, they add a periodically signaling containing the position of the UE. This periodic signal strives against the ultra lean design which this thesis will strive to adopt. They evaluate the per-formance with the different kinds of RLFs, but not considering ping pong handovers. They also have around ten percent RLF-rate as the baseline of their work. They use simulations to evaluate their SON, showing a decrease of RLF-rate with 1.08%.

Nguyen et al. [12] try to decrease the number of RLFs due to its high effect on QoS. The algorithm changes the time-to-trigger (TTT) and Cell Individual Offset (CIO) parameters according to the dominant handover failure reasons. They evaluate the performance by looking at the RLF-rate and mention that there is a tradeoff between the RLFs and ping pong handovers. They simulate several scenarios including small cells and pedestrian UE velocities. The results of their simulations showed a performance gain for each time they updated the parameter value. Their algorithm decreased the parameter values for every iteration which could be due to a reason that their starting values are far away from the best. A difference from this thesis is that in this thesis the simulations intend to have a high-performing baseline.

Klein et al. [13] investigate the self-optimizing and self-healing aspects of SON in LTE. They use the machine learning algorithm Q-Learning as their method. They use their Key Per-formance Indicators as the RLF-rate, ping pong-rate and a rate of the number of dropped connections to the number of total connections. They define their actions to the Q-learning algorithm as the different ways of changing the TTT and hysteresis values. The actions are limited by a few rules, as the states would increase drastically otherwise. They compare their simulated result to benchmarks which shows a performance gain by the algorithm.

Muñoz et al. [14] make a sensitivity analysis of the different parameters included in the handover procedure. They show with simulations that the offset and hysteresis parameters have greater benefits for the network than the TTT parameter. Out of this, they state that a simple but successful SON would be made by only tuning the cell individual offset.

Zetterberg et al. [15] performs a study on the A3 events offset and TTT parameters impact on the mobility robustness in an HetNet. Their study varies the TTT parameter from 40ms to 480ms and the offset from -3 to 3. Their result shows how the different RLF kinds act accordingly to the parameter values between different cell types in a HetNet.

Laakso et al. [16] introduce the framework SONS3 developed with a base in an open source network simulator, adding functionality that enables a realistic scenario for a SON. They also introduce a SON algorithm that can act as a baseline in the new simulation framework. Their algorithm is based upon decreasing/increasing the CIO depending on the dominant failure kind, not taking handover to the wrong cell into consideration.

Weidong et al. [17] introduce an algorithm in the MRO area that is based on the input of the UE velocity and the cell type. The algorithm is simulated in a HetNet scenario, where

(21)

2.4. Related Work

the target cell’s type are of importance. Their result shows a performance gain in terms of RLF-rate.

These research papers all use simulation as a method to evaluate their work. They have created different scenarios matching their interest to see whether their algorithm increases performance. These works are evaluated in terms of the RLF-rate and most of these research works showed an increase in performance. These works show a lot of similarities with this thesis. They all simulate their designed scenarios and evaluate with almost the same metrics as this thesis. The parameters investigated are also often the same, but using them in combi-nation with others or in different algorithms make all these works differ. However, when not taking all parameters affected into consideration and looking at the network on the whole, there may be some tradeoffs not taken into consideration such as energy efficiency.

2.4.2 Energy Efficiency

In comparison to the previous section’s works, which did not evaluate the energy efficiency of the network, this section focuses on works that keep the energy efficiency in mind. Qiu et al. [18] propose a new architecture and handover procedure for 5G networks to lower the energy consumption. Their proposed handover procedure takes advantage of direct com-munication between the source and target base station. This means that the comcom-munication does not go through the core network. They simulate their proposed architecture and shows that this solution only has 65% of the signal overhead compared to LTE.

Dong and Yun [19] introduce a high-level solution separating the data plane and control plane in order to reduce the number of handovers made and the energy used by the base stations. They also present an idea of separating different UEs to handover to small or big cells depending on their velocity and data rate. They evaluate the performance in simulations using the metrics total handovers made and total energy consumption. Their result ends up in different outcomes depending on the data rate and speeds of the UEs.

Boujelben et al. [20] propose an algorithm for an energy-efficient SON, focusing on Self-optimization in the handover process. The algorithm focuses on the parameters of UE velocity and load on a cell. This algorithm tries to put high-velocity UEs into macrocells only to reduce the number of handovers. The algorithm also prioritises to maximize resources on already loaded cells and minimize the usage of low loaded cells. By doing this energy can be saved by turning off the low loaded cells. This approach of turning off low loaded cells is the same strategy as in this thesis. What is different in this thesis, is that this thesis quantifies at what resource utilization the RLF-rate gets too high.

All these works [18][19][20] approach the energy efficiency problem in a similar way to this thesis, proposing changes at an architectural level and not only in the handover procedure. In contrast to this thesis, they make the investigation much more in depth and not only iden-tifying a strategy. One major difference from these works compared to this thesis is also the evaluation metrics, the related works only consider the energy consumption and not the RLF-rate of the handovers performed. Dong and Yun [19] and Boujelben et al. [20] also use the UE velocity as in this thesis, but, there is no description of how this data is gathered.

(22)

3 Simulator and Common

Parameter Setup

In order to perform simulations with SON and to create scenarios to answer the research questions, some development and scenario design had to be made. This chapter describes the functionality with which the simulator was extended along with reasoning for some decisions made in the process. This chapter also provides the method used for setting parameters to create a realistic scenario where the mentioned RLFs occur. This parameter setup is used later in a prestudy as well as a base for the specific scenario design related to each research question.

3.1 Simulator

The simulator provided by Ericsson is a detailed network protocol simulator that follows the 3GPP standard, making the simulations as realistic as possible. It covers all layers of the network stack, which also makes it very complex and computation demanding. However, it is also these aspects that make the simulations realistic and interesting to perform studies with.

Figure 3.1 shows the architecture of the simulator at a high level. The flow of a simulation is that you first need to define the environment of the network by setting parameters such as the number of users, antenna radius, etc. With a complete parameter setup, the simulation can be run and will produce results in self created metrics. In this thesis, the metrics used are connected to each research question and will contain the estimated UE velocity, the amount of RLFs and Ping Pong handovers as well as the load on the base stations.

Figure 3.1: The high level architecture of the simulator

The simulator makes it possible to change a lot of different parameters and algorithms that can differ in networks, enabling the possibility to create scenarios that are designed to ad-dress certain problems. For example, it is possible to change the deployment of the network,

(23)

3.2. Simulator Extensions

varying how many base stations and users there are in a network. The simulator is not only limited to these factors but can change more detailed subjects such as which frequencies and protocols are used in the underlying communication.

The use of seeds is a way to control the random generation of numbers in the simulator. Using the same seed input to the simulator will produce the same random generated num-bers over and over again as long as the seed is the same. This is used in order to be able to reconstruct simulations which will generate the same result as a previous simulation. The amount of load a base station is under is also used in this thesis. The approach for calculating the resource utilization on a cell in the simulator is based on the bandwidth it is transmitting on. The full bandwidth of a cell is divided into 100 subbands which can be either used or not used. Each millisecond, the number of occupied subbands used for downlink transmission is logged. Adding these numbers together and dividing by the time spent in the simulation, we can achieve the average resource utilization of the cell, which is later used for the fourth research question.

3.2 Simulator Extensions

The choice of simulating result instead of only analyzing the impact theoretically or creating the work in a real environment was never really a question. The costs and time consumption of trying out ideas in a real environment are way out of scope for this thesis and only analyz-ing the algorithms theoretically would be too complex a task.

This thesis investigates the area of mobility and takes usage of new functionality that was not yet fully implemented in the simulator, which also created a need for further development. Functionality which the simulator was extended with in order to be able to simulate scenarios for this thesis included:

• An automation of parameter calculation between simulator iterations with the purpose of introducing SON functionality.

• A UE velocity estimation algorithm which uses data provided from the beam manage-ment as described in Section 5.2.1.

• Introduction of the new parameters of UE velocity and beam-specific offset into the measurement report triggering events.

In the creation of the SON functionality in the simulator, there were a few things that had to be decided. The first decision was whether the handover parameters should be updated during simulation runtime or if it should set the values in between every SON-iteration. The decision landed in updating the values between iterations and not during runtime. As the simulator has a complex structure and a long runtime, a server cluster is used where different instances of the simulator can be running separately. The different servers have their own instance of the simulator and therefore can not update the same values. This also implied that an event-driven SON-cycle would be difficult to implement and therefore, the decision was made to update the affected parameters periodically. The structure of the SON functionality implemented is shown in figure 3.2 where one SON-iteration is defined as the period between two "SON Algorithm" in the figure.

A second decision that had to be made was whether the different iterations should be started with the same seeds or not. The decision was made to run the same seeds for every iteration, which comes with both pros and cons in terms of keeping the simulation realistic. Running

(24)

3.3. Common Parameter Setup

Figure 3.2: Overlook of the SON functionality in the simulator

the same seed implies that all randomness in the simulation will be the same for each it-eration, which may seem too strict for a realistic behavior. However, this thesis intends to compare two different algorithms with each other and compare the behavior of an algorithm with different parameter setups. Therefore the need of creating identical scenarios made the decision of using the same seed.

Introducing the new parameters into the handover decision and implementation of the al-gorithms did not include any major decisions that would affect the scenario design. Some algorithm implementation and data processing were also performed outside the simulator. No major design decisions affected this implementation since the functions involved were not complicated.

3.3 Common Parameter Setup

To simulate a SON a common parameter setup was needed. This setup is used for all four research questions investigating different parameters, meaning that it can not be exactly the same, but only differ in the investigated parameters in order to be comparable. There were five different aspects that needed to be carefully considered in the scenario design in order to make the network behave realistically and make the failures we were interested in appear.

• The load on the network • Frequency interference

• Amount of relevant data produced from simulations • Simulation runtime

• Starting values of TTT and Offset

These aspects cannot be studied independently, they are in different ways influenced by each other. During the process of designing the scenarios, there have been a lot of shorter simulations ran in order to see the behavior with different setups.

The first major issue that showed up was regarding the amount of RLFs. Without any load on the base stations or without any interference at the UE there were a lot fewer RLFs than expected. The main parameters that affect these factors are the number of users to-gether with how many beams there were in use as the beams narrow down the area that interference may occur at. Unfortunately, just increasing the number of users increases the

(25)

3.3. Common Parameter Setup

simulation runtime a lot. Instead, changing the amount of data that a UE requests peri-odically was increased which occupies the subbands at the base station and creating load without affecting the runtime or the interference in the same way. As for the interference, we want to have multiple UEs connected to closely located beams in order to create the be-havior. So instead of increasing the number of users, the number of beams used was lowered. As the number of users was a drastic reason for how long the simulation runtime was, shorter simulations were run to see what was the lowest number of UEs that could be used and still see occurrences of RLFs. At a number of 100 UEs, the failures still occurred and this number was therefore chosen as the number of UEs to be used in the simulations.

The next decision was how long the simulation time should be. The simulation time needs to be long enough for all UEs to perform enough handovers in order to produce relevant data, but short enough to make the simulation time realistic in terms of the thesis limitations. A simulation time of 80 seconds was therefore decided. The simulation time may seem short as the lowest velocity used is 1m/s and the SON-cycle should work on a cell or beam-specific level, but it should be enough for all different UE velocities to at least perform some han-dovers.

There were also some more handover specific values that had to be chosen with care. Using absurd values of the parameters of interest would give some initial readings with a trivially low performance and it would be an easy task to increase the performance. Therefore values where the network already has overall good performance were set as the initial performances for the SON experiments. The values that were decided were a combination of related works on the topic and a sample of shorter simulations made to investigate at what values this specific setup had a relatively good performance.

In Table 3.1 some of the common parameters that were used in the scenarios are presented. As mentioned before, there were also several simulation instances running in parallel to decrease the total runtime. As the table shows there are five different UE velocities, these will all be separated into different simulations with 100 UEs of each speed. But in a single simulator instance, there will only be a total of 100 UEs with the same velocity. It is possible to mix the speeds for a single instance of the simulation, but the choice was made not to. Depending on the experiment design there will be a different number of parallel instances. For example, running the five different speeds with two different seeds would imply running 10 different instances simultaneously.

Table 3.1: Common parameters for all scenarios unless it is the investigated parameter

Parameter Value

Simulation time 80 sec

Number of cells 21 Cell radius 166m Number of UEs 100 UE velocity 1, 3, 5, 15, 30 m/s Offset -1.0dB TTT 100ms T310 (RLF timeout) 0.5s Carrier Frequency 2GHz

(26)

4 Choice of Parameters

In order to create a SON that tries to optimize the robustness of the network, the network needs to know which parameters to change and when. The first section describes how this thesis measures robustness. In Section 4.2 a prestudy is made in order to determine which parameter should be used in the SON. Finally, Section 4.3 describes the intended way to lower the energy consumption related to the fourth research question.

4.1 How To Measure Robustness?

As the thesis focuses on evaluating performance in terms of the different RLFs that can occur as well as Ping Pong handovers a definition of how a RLF is determined is needed. There are four different ways a RLF can be determined. Too many attempts on random access or too many Radio Link Control (RLC) retransmission failures are two ways of determining a RLF. However, using the T310 timer is probably the most common RLF source. Whenever a UE has triggered that the signal quality is below the defined threshold, the T310 timer starts, and if the condition is still met when the timer expires, a RLF has occurred.

As described earlier there are four behaviors that are unwanted in a network, three different kinds of RLFs and the Ping Pong handover. These are the too early handover, too late han-dover, handover to the wrong cell and Ping Pong handovers. The amount of these failures compared to the total number of handovers is the commonly used evaluation metric in most research in the area of MRO. In 3GPP, the ways to define these different failures are described by looking at the cell in which the reconnection is established. The number of failures and the failure rate will be used to evaluate robustness in this thesis.

Another aspect that was considered in the early stage of the thesis was whether to consider the Ping Pong handovers or not in the algorithms. The Ping Pong handovers are not consid-ered as a big problem for the network at the time of writing as it does not affect the end users in the same way as a RLF. However, as the number of Ping Pong handovers is significantly higher than the RLFs it would be of interest to try to reduce the number of these as well. After all, they still affect the overall performance even though Ping Pong handovers do not create RLFs directly. In this thesis, the algorithm designed for the third research question will

(27)

4.2. Choice of Investigated Parameters

take the Ping Pong handovers into consideration.

4.2 Choice of Investigated Parameters

The section shows the prestudy made to decide which parameters should be used in the SON. This section will also describe the different granularity levels that can be used for the parameters in the network.

4.2.1 Prestudy on TTT vs Offset

The result of adjusting the TTT parameter is expected to be dominated by the too late han-dovers until a low TTT value [12] is used. Both due to the fact that fast UEs trigger more handovers and therefore more failures with the wrong parameters settings, but also due to the fact that too early handovers are not as common unless there are very bad radio condi-tions. The simulated behavior of changing the TTT is shown in Figure 4.1. This figure shows the result of the shorter simulations performed to understand the dependencies following the common parameter setup described in Section 3.3. All different speeds are used and the only thing different is the length of the simulation. Figure 4.2 shows the same data with an addition of the Ping Pong handovers, but as the number of Ping Pong is much larger than the others they are separated into different figures. The y-axis numbers are left out intentionally since the idea is just to show the overall trends.

Figure 4.1: The impact of TTT on the dif-ferent kind of RLFs

Figure 4.2: The impact of TTT on the num-ber of RLFs and Ping Pong handovers

Following this behavior would mean that there is a point close to the too early and too late curves intersection where the number of RLFs is close to optimum, not taking the Ping Pong or handover to wrong cell into consideration. That intersection point is the intended parameter value the SON algorithm will try to achieve. However, this intersection point is not necessarily the proved optimum in terms of RLFs.

The offset parameter would, in theory, impact the mobility robustness in a similar way as the TTT parameter where there is a tradeoff between the too early and too late handovers. In Figure 4.4 and 4.3 there are results of shorter simulations in order to see how the network behaves when changing the offset parameter. Compared to Figure 4.1 and 4.2 we can tell that the offset has a bigger impact on the number of Ping Pong handovers. The second thing to

(28)

mention is that the too early handovers are more or less the same for all offset values in the range from -3 to 3.

Figure 4.3: The impact of offset on the dif-ferent kind of RLFs

Figure 4.4: The impact of offset on the number of RLFs and Ping Pong han-dovers

The offset parameter was shown by related work to be the one with the most impact on the mobility robustness [14]. However, even though the offset parameter is of high interest for mobility scenarios, I do not think the offset will interact with the choice of offset strongly. The prestudy shows that the TTT parameter is more correlated to the UE velocity when looking at the handover RLFs within different mobility scenarios. In Figure 4.1, the tradeoff between the too early and too late failures is shown. In this figure, the too late failures come from high-speed users with a high TTT value, and the too early mainly comes from low-speed users with a low TTT value. Therefore, the TTT parameter was chosen as the parameter that should be adapted according to the estimated UE velocity.

As the usage of beamforming is introduced with 5G there exist several areas that still have not been investigated. Therefore the question is whether we can use data provided from the beam management in order to further improve the MRO without adding a lot of complexity. The offset value used in event A3 is on a cell-specific level and is investigated a lot to improve the network performance. Therefore the offset parameter by itself is not so interesting to investigate for this thesis. However, making the offset parameter beam-specific introduces a possible improvement by making the parameter of lower granularity, especially in urban environments where there can be huge differences in signal quality between two neighboring beams in a cell. One example is where there may be a tight corner around a building, where the connection may drop really quickly compared to other zones in the sector meaning that one beam has a good signal strength and the next one a really bad value. Adapting the offset values accordingly could improve the performance of the network instead of having the same value for both.

4.2.2 Granularity Level

In today’s networks, many parameters used in the handover procedure are network specific and configured uniformly for a lot of cells. This is a way to reduce configuration costs and still provide an acceptable level of performance. Due to the challenges with variations in environments and UE behaviors that cells are facing, these parameters could probably be

(29)

optimized. There are multiple ways of setting these parameters which can be summarized as follows:

• Network specific - The same value for all cells in the network • cell-specific - Each cell has its own independent value

• Cell relation specific - Each cell has its own individual value relative to each neighboring cell

• beam-specific - Each beam inside the cell has its own specific parameter value.

• Beam relation specific - Each beam has its own individual value relative to each neigh-boring beam.

Theoretically, there would be performance improvements in the network if lower granularity parameters are used, where network specific are of the highest granularity and beam relation specific the lowest. There are really no direct cons of using beam-specific values in terms of the mobility performance which makes the use of this really interesting. The cons of using lower granularity parameters are the amount of data that is needed for the SON-algorithms as well as the configuration costs of installing a base station. Just taking the step from cell-specific to beam-specific would, with Figure 2.3 as an example, need 5 times the data in comparison. With deployment into a real network, a manual setup with beam-specific values would be extremely costly. This motivates a SON that configures the parameters itself. As the beams are introduced with the new technology of 5G, the beam-specific level is of interest to this thesis, but as it requires a lot more data it also requires a lot more time in simulation. However, the choice was made to use beam-specific offset parameters for the third research question.

As the behaviors in each cell may differ, it felt necessary to look at the TTT at a cell-specific level and not only as a network specific value, even though this requires more simulation data. With only network specific values, there might be a tradeoff between cells. One cell might have failures mainly consisting of too late handovers and another mostly due to too early handovers. This means that the two cells would probably want to change the parameter in different ways. By using cell-specific values this can be avoided on a cell level. Therefore the decision was made to use cell-specific TTT parameters for the second research question.

4.2.3 UE Velocity

As the difficulties of mobility prediction arise, due to the different movement patterns of a user, it makes sense to consider the UE velocity in a handover decision. Too early and too late handover issues are strongly connected to the UE velocity and the event parameters used in Equation 2.1. As mentioned above we could be provided with the velocity by a UE if the cost of signaling was ignored and directly used a value periodically calculated and reported by the UE. But due to this signaling and the additional requirement on the UE, estimating the UE velocity at the network side would be a much more efficient and realistic option.

In this context, the actual velocity a UE is moving with is not really of interest. The informa-tion of interest can rather be seen as a relative speed or the probability of the UE to perform a handover soon. With this information, whether it comes in the unit of m/s or a probability be-tween 0 and 1, the UE can be classified into different categories with different characteristics eg. high or low speed. Theoretically, the more accurate the estimation is, the more different classifications can be made and lead to more precise parameter adjustment. However, de-pending on how accurate the estimation of the UE velocity is, it may be better to not use a too high number of classifications.

(30)

4.3. Energy Efficiency and Load

4.3 Energy Efficiency and Load

As the load is directly connected to the usage and need of the base stations there is a clear strategy of trying to adapt a network to the load. With 5G there will be multiple different cell types as presented in Chapter 2. In a HetNet there can be several cells that operate on different frequencies but on the same coverage area. The option of turning off all cells in the area is not a possibility as it would ruin the QoS that has been promised. However, reducing the number of frequencies working in the same area would be a way to save energy as long as we can guarantee a certain QoS.

As the interference of users transmitting and receiving on the same frequency causes more RLFs to occur during handovers we need to know how much load a base station is able to handle by itself before allowing to turn off another cell. Different frequencies have different characteristics such as path loss and setups. These characteristics imply that depending on which frequency is left active, it has to be known how much load that frequency can handle before the others are turned off.

As the load varies a lot from time to time during a day, there could theoretically be big savings from turning off cells during night time until the load starts to increase again. A hypothesis would be that the load is significantly higher between 08:00 to 20:00, meaning that the remaining twelve hours of the day could be a period where some of the active frequencies could be turned off.

As different cell types have different characteristics, it also means that they consume differ-ent amounts of energy. In general, the bigger the cell the higher the consumption, meaning that what is of most interest is turning off macrocell frequencies, even though turning off small cells contribute as well. Therefore this thesis investigates how macrocells operating on different frequencies differ with respect to impact on RLF and also quantifies the load they can handle independently.

Load can be created in two different ways, either there is a lot of users in the coverage area or fewer users who require a lot of throughput. Having many users causes a lot more interfer-ence which affects the signal strength as well as increasing the usage of the cell’s resources. Increasing users’ throughput mainly occupies the resources at the cell without causing so much interference. As the interference affects the signal strength which will cause more RLFs, the way of simulating load for the fourth research question was mainly made by increasing the number of users.

(31)

5 Design of Algorithms and

Experiments

First in Section 5.1 two hypotheses are presented that shows the intended solution of increas-ing the robustness for the network. In Section 5.2 the algorithms that are used for the velocity estimation and the SON are presented. Finally in Section 5.3 the detailed scenario design specific to each research questions are presented.

5.1 Hypotheses

A SON needs to be based upon an algorithm that changes the investigated parameters ac-cordingly. In this thesis, two SON-algorithms will be developed, one changing the TTT and the other changing the offset. These algorithms need to be based upon some hypotheses of how the parameters affect the mobility robustness of the system. Out of the prestudies presented in Section 4.2.1, two working hypotheses have been made.

The first hypothesis is regarding the tradeoff caused by changing the TTT. The prestudy showed there is a tradeoff between the too early and too late handovers when changing the TTT value. A TTT value which is considered good is according to the preliminary result found close to where the number of too late and too early handovers are equal. Therefore the working hypothesis for the first SON-algorithm will be to decrease the TTT value if the too late handover is dominant, and increase the value if the too early is dominant.

The second hypothesis is regarding the tradeoff caused by changing the offset. In this case, the prestudy did not show any tradeoff between the too early and too late handover as was expected. Instead, it showed a bigger impact on the number of Ping Pong handovers than expected. Due to this, not only the too early and too late failures are considered in the work-ing hypothesis. The second workwork-ing hypothesis will therefore also consider the Pwork-ing Pong handovers, decreasing the offset value if the too late handover is dominant, and increase the value if the too early plus one percent of the Ping Pong is dominant.

(32)

5.2. Algorithm Design

5.2 Algorithm Design

This section describes four algorithms: the UE velocity estimation, the calculation of a new TTT and offset value for each cell/beam per SON iteration and the scaling of the TTT value depending on the estimated UE velocity.

5.2.1 UE Velocity Estimation Algorithm

The creation of the velocity estimation algorithm was an iterative process of trying out dif-ferent ideas in order to see their accuracy and determine whether the estimation was good enough or whether it had to be improved.

As an initial thought, the estimation of the UE velocity based upon the beam management seemed like a trivial problem. The idea of estimating the UE velocity is shown in Figure 5.1. Imagine a UE moving in the direction of the arrow passing by the different beams, the UE would probably during that time have updated its best beam to beam number 4, 5, 2, and 3. With this scenario the UE would then need to update its best beam three times. Dividing the number of updates with the time spent moving withing the scenario would give an indicator that is proportionally dependent on the speed. That is, the faster the UE makes these updates, the higher velocity the UE would have.

Figure 5.1: The initial idea of estimating the UE velocity

A linear regression model was chosen as a basis of the estimation problem. The input to train the regression model were data pairs containing the number of beam switches per second related to their actual velocity while running the simulator. During runtime, the model would then only need an input of the UE’s beam switches per second to make an estimation of the velocity. However, this approach makes the model sensitive to the deployment con-figurations of the individual cell. Generating training data with, for example, 9 beams, and using that training data for a cell deployed using 16 beams would give a misfit model. The approach described seemed straightforward, but there are some physical aspects mak-ing the number of beam switches not as obvious as Figure 5.1 shows. There are three major aspects that impact the number of switches.

The first aspect was the beams projection size on the ground. The further away from the base station a beam is projected, the larger its covered area is. This fact means that some beams will cover a wider area than others. Putting this fact into a scenario where one UE moves close to the base station and another one where the UE moves close to the cell border, the user close to the base station will update its beams more often than the user on the cell border even though they move at the same velocity.

(33)

5.2. Algorithm Design

The second aspect was raised by the limitations in how many beams that a base station is able to transmit. There can only be one beam formed at a time, and the beam management needs to sweep through all the beams and transmit their reference signals in order for a UE to measure the signal strength. If there is a high number of beams, the base station could theoretically be occupied by only trying to transmit these signals for all the beams. Therefore, there is a limitation to how many of the beams’ reference signals can be transmitted in a given period. This periodicity is often set to 20 ms. To solve this issue, there exists algorithms that choose in what order the different beams reference signals will be transmitted.

This beam management has an impact on the number of beam updates the UE makes. Imag-ine in Figure 5.1 that the UE is located inside beam number four’s coverage area, but due to the limitations in beam management the UE only receives signals from a total of 4 beams, in this case, beam number 6, 7, 8, and 9. This would make the UE choose beam number 7 as the best available beam, until the beam change period, where beam number 4 will be chosen. The described behavior will end up in more beam switches than needed, meaning that some randomness affects the number of beam switches. In conclusion, due to the beam managements physical limitations and choice of beam selection algorithm, the number of updated beams may not match the initial idea.

The third issue comes from the environment that the UE is located in. This thesis focuses on an urban scenario, which means that there exist obstacles such as buildings that may block or bounce the signals. The effect of this is that the locations of the beams are not as nicely lined up as in Figure 2.3, but rather having gaps between them and overlapping each other. Simulations were made to create data for the regression model. However, due to all the mentioned difficulties, only small differences were shown in the number of beam updates per second for low and high-speed UEs. Therefore a filtering function was created in order to decrease the impact of the physical aspects. The filtering function is shown in Algorithm 1. The first action in this algorithm is on row 4, where a beam switch made just after a previous beam switch is ignored. This reduces the impact on the second aspect described where the UE sometimes makes beam updates that are due to the limitations in the beam management and not the UEs movement. On row 6, we can see another case where we ignore the beam switches. As shown in Figure 2.3, there are both wide and narrow beams, switching between them does not necessarily imply a change in position from the UE. Therefore, switches made from or to a wide beam are ignored by this function. The last if statement reduces the impact of the varying size of the beam projections. Here, a handover, which is happening on the cell’s border, is valued as three beam switches. If it was not a handover but a beam switch in the outer half of the cell’s coverage area, the switch counts as two. Last on row 12 is the standard case of the beam switch, valued as a single beam switch.

Mobility Optimization for Energy-Efficient 5G Networks : Optimering av Mobilitet för Energieffektiva 5G Nätverk

Linköping University | Department of Computer and Information Science

Master thesis, 30 ECTS | Datateknik

2019 | LIU-IDA/LITH-EX-A--19/085--SE

Mobility Optimization for

Energy-Eﬃcient 5G Networks

Optimering av Mobilitet för Energieﬀektiva 5G Nätverk

Oskar Gustafsson

Upphovsrätt

Copyright

Acknowledgments

Contents

List of Figures

List of Tables

Acronyms

1

Introduction

1.1

Motivation

1.2

Aim

1.3

Research Questions

1.4

Method Overview

1.5

Delimitations

1.6

Thesis Outline

2

Background and Related Works

2.1

Thesis Environment

2.2

5G Technologies

2.2.1

Frequency Bands

2.2.2

Heterogeneous Network

2.2.3

Massive Multiple Input Multiple Output

2.2.4

Ultra Lean Design

2.3

Mobility in 5G

2.3.1

Handover Procedure

2.3.2

Mobility Difficulties

2.4

Related Work

2.4.1

Mobility Robustness Optimization

2.4.2

Energy Efficiency

3

Simulator and Common

Parameter Setup

3.1

Simulator

3.2

Simulator Extensions

3.3

Common Parameter Setup

4

Choice of Parameters

4.1

How To Measure Robustness?

4.2

Choice of Investigated Parameters

4.2.1

Prestudy on TTT vs Offset

4.2.2

Granularity Level

4.2.3

UE Velocity

4.3

Energy Efficiency and Load

5

Design of Algorithms and