Energy Eﬃcient Cellular System Design with QoS Assurance

(1)

Energy Efficient Cellular System Design

with QoS Assurance

CHANGYANG ZHANG

Master’s Degree Project

Stockholm, Sweden October 2014

(2)

Abstract

Results from the smart-phones’ ubiquitous Internet access and diverse multi-media applications, there has been an explosion in mobile data communication, this incredible increase will necessitate continual high energy consumption and leads to more CO2 emissions. It is crucial to develop more energy-efficient

systems because of the potential harmful effects to the environment caused by CO2 emissions. It is also significant for the cellular network operators, since

the electricity bills are a considerable portion of their Operational Expenditure (OPEX).

The power consumption at base stations accounts around 60-80% of the total power consumption in a cellular network [1, 2]. Potential energy savings can be expected by implementing BSs sleeping mode, according to the traffic demands and user activity factors.

There are many performance trade-offs in optimizing BSs sleeping modes [3]. One of them is the trade-off between service delay and power consumption of cellular networks. In this thesis, the service delay is considered as a measure of Quality of Service (QoS) user experiences.

Based on the conception of cell wilting and dynamic base stations switching, BS sleep switching algorithms are developed while the QoS of User Equipments (UEs) is guaranteed in the meantime. The switching algorithms are decen-tralized which means no central controller is needed. The BSs can make the switching decisions based on the feedback from UEs and its neighboring BSs. Furthermore, the implementation of the proposed algorithms is also comprehen-sively described at the protocol level.

An urban micro BSs network is built and used as the simulation scenario. Simulations are done with the help of the rudimentary network emulator (RUNE), a network simulator tool developed by Ericsson in MATLAB environ-ment.

Simulation results show that up to 62% of power consumption can be saved by implementing the QoS guaranteed BS switching algorithms. Furthermore, comparisons show that the QoS guaranteed BS switching algorithms have higher performance in terms of power consumption savings and QoS to the transitional BS switching algorithms.

(3)

analysis of the performance is challenging because the required parameters for analysis are dynamically changing during the switching processes. In this the-sis, we develop a rough analythe-sis, which gives an insight into some key factors affecting the performance.

It shows that high system traffic load results in high power consumption and poor system performance. Either a very small or very large cell radius also leads to high power consumption and poor system performance for different reasons. In addition, choosing different values of the switching threshold affects the system performance in different ways.

(4)

Acknowledgments

First and foremost, I have to thank my thesis supervisors, Dr. Guowang Miao. Without his assistance, this thesis would have never been accomplished. I would like to thank you very much for your support.I would also like to show gratitude to Mr Amin Azari, who also gave many valuable advices to my research for this thesis.

I began my master study at KTH Royal Institute of Technology since Aug 2012. I began my new life and made many great friends in Stockholm. I cannot begin to express my gratitude and appreciation for their friendship. Mr. Gustavo Cunha Cintra and Mr.Leo Carlsson have been unwavering in their personal and professional support during the time I spent at KTH. And we had much memorable time together.

Most importantly, none of this could have happened without my family. My parents, Mr. Xilong Zhang and Mrs. Fang Zhao, who supported my study and life abroad for the past two years. My grandmother, who offered her encourage-ment through phone calls every week. This dissertation stands as a testaencourage-ment to your unconditional love and encouragement.

(5)

List of Tables

3.1 List of Notations . . . 14

6.1 General Simulation Parameters . . . 35

7.1 Simulation Parameters Section I . . . 37

7.2 Simulation Parameters Section II . . . 44

(9)

List of Figures

1.1 Effect of smart mobile devices and connection growth on traffic . 2

1.2 Cost makeup of OPEX . . . 3

1.3 Normalized real traffic load during one week[4] . . . 4

1.4 Breakdown of power consumption in a typical base station [5] . . 5

2.1 An example of cellular network with seven hexagon cells . . . 8

2.2 An example of cells wilting [6] . . . 9

2.3 Hand over procedure of cells wilting [7] . . . 10

2.4 Dynamic BS sleep control and it effect [7] . . . 11

3.1 Integration of the base station power model for system level per-formance evaluations[8] . . . 16

3.2 Service Rate of UE m for long time period T . . . 17

4.1 A cellular network unit . . . 19

4.2 An illustrative example of switching-off BS . . . 20

4.3 The switching-off procedure . . . 22

4.4 The different thresholds for switching on/off . . . 24

4.5 The switching-on procedure . . . 25

6.1 The Simulation scenario . . . 32

6.2 The wrap Around Technique . . . 33

6.3 A Simulation Example . . . 36

7.1 Energy Saving of BS Switching Algorithm . . . 38

7.2 Average number of Switching-off BS . . . 39

7.3 Average service delay of UEs . . . 40

7.4 Comparison of Energy Saving . . . 41

(10)

7.6 Required Number of BSs and Traffic Load . . . 44

7.7 Average Service Delay under Different Cell Radius . . . 45

7.8 Total Energy Consumption under Different Cell Radius . . . 46

7.9 Energy Efficiency under Different Cell Radius . . . 47

7.10 Power Consumption and Average Service Delay under Different Switching Threshold . . . 48

(11)

Acronyms

BSs Base Stations. 2

COSOFF Confirmation Of Switching-OFF. 21 CTSOFF Clear To Switch-OFF. 21

EE Energy Efficiency. 4

ICT Information and Communication Technology. 1

OPEX Operational Expenditure. i, 3

PS Processor Sharing. 26

QoS Quality of Service. i

RF Radio Frequency. 15

RTSOFF Request To Switching-OFF. 21 RTSON Request To Switching-ON. 23

SINR Signal to Interference plus Noise Ratio. 28

(12)

Chapter 1

Introduction

Contributed by the significant environmental footprint and eventual exhaustion of traditional energy resources, global energy consumption becomes a major issue. According to current estimates, Information and Communication Tech-nology (ICT) is responsible for a fraction ranging between 2% and 10% [9] of the word energy consumption. And the overall ICT footprint will double between 2007 and 2020, while the footprint of cellular networks is predicted to triple [7]. Wireless access networks, as a branch of the ICT sector, are responsible for 0.5% of the energy consumption. It is crucial to find a way to cut down the energy consumption of the cellular networks. One solution is to use the base station sleep model control.

In this chapter, we present some background knowledge of this field. The current problems motivating this work will also be reviewed prior to introducing the structure and the contents of this thesis.

1.1 General Background

Because of the smart-phones’ ubiquitous Internet access and diverse multimedia applications, there has been an explosion in mobile data recently. One can see from the mobile data traffic forecast shown in Figure 1.1, globally, traffic from smart devices is going to grow from 88 percent of to 96 percent by 2018. This is significantly higher than the ratio of smart devices and connections (54% by 2018), because on average a smart device generates much higher traffic than a non-smart device. This incredible increase will necessitate continual high

(13)

energy consumption and leads to more CO2 emissions. It is crucial to develop

more energy-efficient systems because of the potential harmful effects to the environment caused by CO2 emissions.

Figure 1.1: Effect of smart mobile devices and connection growth on traffic

The problem is also significant for the cellular network operators. Figure 1.2 shows the cost makeup of a mobile operator’s OPEX, which vary with the network architecture and infrastructure. The electricity bills are a considerable portion of their OPEX. The huge savings in capital expenditure and operational expenditure can be implemented through reduced energy needs [10].

The power consumption at base stations accounts around 60-80% of the total energy consumption in a cellular network [1, 2]. Hence, many studies have been done on energy efficiency in wireless network focus on the radio access at base stations. In the litterateurs, several ways to reduce energy consumption in Base Stations (BSs) are proposed. One of them is hardware design. For instance, more energy efficient power amplifiers and natural resource for cooling can be used. The other way is topological management such as the deployment of relays and micro BSs [11, 12, 13].

(14)

Figure 1.2: Cost makeup of OPEX

1.2 Motivation

In this thesis, we introduced the concept of sleep mode in base stations. The main motivation behind base stations sleeping model is the potential energy saving we can get by switching off base stations during low traffic periods.

1.2.1 Traffic Dynamics

Studies have shown that there are high fluctuations in traffic demand over dif-ferent space and time [14]. For instance, there are huge differences between the traffic demands in urban areas and rural areas. Figure 1.3 shows the normalized real traffic load during one week, which are recorded by a cellular network oper-ator. The data captures voice calls information over one week with a resolution of one second in an urban area, and are averaged over 30 minute time-scale [4]. One can see that there is a high traffic load during the evenings on weekdays, while the traffic is low during nights and weekends.

Another fact about the mobile traffic is that data and video will dominate whole networks. On one hand, compared with voice traffic, data and video traffic is typically more dynamic, and therefore consumes more spectrum and energy resources [15]. On the other hand, it can tolerate some delay in general. Moreover, it is common that point-to-multipoint communication will happen since many people may be interested in the same content in a short time period.

(15)

Figure 1.3: Normalized real traffic load during one week[4]

In that case, it will not be energy-efficient to provide mobile data and video services in a real-time point-to-point way as we do for voice traffic [16].

1.2.2 Potential Energy Savings

Figure 1.4 shows a breakdown of power consumption in a typical cellular network at active mode. It can be seen that the power amplifier and the air conditioning consume most part of the energy of a typical base station. For that reason, a cellular base station at zero traffic loads consumes about 70-90% of the energy consumption when it is full loaded [17]. Since the power amplifier and air conditioning a BSs always consume energy. Therefore, potential energy saving can be expected by switching base stations according the traffic demands and user activity factor.

1.2.3 Quality of Service

There are many networks performance trade-offs in optimizing sleeping modes. For instance, Development efficiency-Energy Efficiency (EE) trade-off, Spec-trum efficiency-Energy efficiency trade-off, Bandwidth-Power trade-off and

(16)

Figure 1.4: Breakdown of power consumption in a typical base station [5]

Delay-Power trade-off [3].

The service delay of UEs, also known as service latency, is a measure of QoS and user experience in a cellular network. Furthermore, the service delay is a measure of QoS from UE perspective instead of from a network perspective.

In early mobile communication systems, such as GSM, the service type is very limited and focuses mainly on voice communications. The traffic generated in voice service is characterized by its continuous and constant. In this case, fixed rate coding and modulation schemes are good enough. The service delay between the transmitter and the receiver mainly consists of signal processing time and propagation delay[3]. However, with the development of mobile In-ternet access technology and diverse multimedia applications, future networks must deal with various applications and heterogeneous service delay require-ments. Therefore, in order to build a green cellular network, it is significant to consider the trade-off between service delay and power consumption.

In this thesis, service delay is considered as a measure of QoS and user experience. And the trade-off between service delay and power consumption is discussed in this thesis.

(17)

1.3 Thesis Structure

The problem addressed in this thesis is building and performance evaluation of a BS sleeping algorithm for energy efficient green cellular networks. The objective is to provide insights on the behavior of the algorithms and how it affects network performance. Numerical results for performance evaluating are obtained by simulations.

This thesis is organized into seven chapters. Chapter 2 introduced the related work and the contribution of this thesis. Chapter 3 gives system models of the cellular network scenario. Chapter 4 explains the BS sleep algorithm. Chapter 5 gives some first-order analysis of the algorithms. Chapter 6 describes how the simulations are conducted, including how the system is modelled. Chapter 7 presents the numerical results and analysis. Finally, summary and conclusions are drawn in Chapter 8.

(18)

Chapter 2

Related Works and

Contribution

In this chapter, we will introduce the main related works in the field. The contribution of this thesis is also mentioned.

2.1 Cellular Networks

A cellular network provides wireless connectivity to users in a certain area by di-viding the area into cells. The shape of cells is commonly modelled as hexagons. This geometrical model is chosen because it can cover the service area without overlaps which is shown in Figure 2.1.

Each cell is served by one antenna system at the base station. Each user is associated a certain BS, and radio link is connected between BS and UEs. A radio link corresponding to transmission from the base station to the UE is called down link. In this thesis, only down link communication is considered because of the complexity in uplink communication.

UEs are generally mobile phones or other wireless device. When a UE in some cased cannot be serviced by the current BS anymore, a handover mecha-nism is carried out to assign a different serving base station to the UE.

(19)

Figure 2.1: An example of cellular network with seven hexagon cells

2.2 Base Station Sleep Modes

As mentioned in Chapter 1, the potential energy saving can be expected when the traffic load is low in a cellular network. The unnecessary BSs are one of the main sources of energy waste in current wireless networks because the BSs are designed to continuously listen to the radio environment in order to detect incoming UEs, which consume energy even when no UE is using the BS service. Considering the facts mentioned above, the application of sleep modes of BS decreases the energy consumption. BSs can be switched off when and where they are not necessary. Furthermore, interference between cells is also reduced since the radio emission is interrupted during sleep modes.

The three requirement for BS sleep mode design are [6]: 1. Minimize the consumed energy.

2. Guarantee the availability of wireless access over the service area. 3. Minimize the perceived degradation of user experience.

2.3 Cells Wilting and Blossoming

When a BS is decided to be switched off, its coverage will be covered by its neighboring BSs. All the UEs served by this BS will be handed over to its

(20)

Figure 2.2: An example of cells wilting [6]

neighboring BSs. In this case, neighboring BSs will take responsibility of the traffic of the switched off BS. Similarly, when a sleeping BS is decided to be waked up, UEs are handed over back to the active BS. One example of BS wilting is shown in Figure 2.2 and Figure 2.3.

2.4 Dynamic Base Stations Switching

The traditional network planning and operation is mainly based on the assump-tion that user requests may happen anytime and anyplace. As a result, the existing cellular networks were mostly designed to keep the transmitting power always on in order to guarantee cell coverage as well as provide the appropriate

(21)

Figure 2.3: Hand over procedure of cells wilting [7]

services if requests occurred. It is not energy-efficient since the user requests occur only sometimes and some places in practice. It is reasonable to find a way to keep the cell coverage by a minimum number of BSs on demand based on the service requirements. In other words, the network resources need only be available rather than always on if the coverage can be guaranteed. However, such BS switching algorithms require a high level of BS co-operation.

Cell size in cellular networks is in general fixed based on the estimated traffic load. However, as mentioned in 2.3, the BSs are switched off when the QoS of UEs is satisfied in their coverages. In other words, the cell size can be adaptively adjusted according to traffic conditions as well as the situation of neighboring BSs in a collaborative way. A corresponding cell planning results are shown in Figure 2.4.

Intuitively, the object of BS energy saving is to dynamically minimize the number of active BSs to meet the QoS requirement in the network. This requires the traffic information of the whole network.

There are two ways to implement the BS switching algorithms: centralized switching algorithm and decentralized switching algorithm.

For a centralized switching BSs algorithm, all the channel information, traffic requirement and QoS information are known at the network side. In this case,

(22)

Figure 2.4: Dynamic BS sleep control and it effect [7]

a centralized control system is needed in order to control the status of all the BSs in the system. The optimal set of active BSs can be selected by the central controller in order to save energy. However, more energy consumption is needed for the calculation of the central controller, and it is hard to implement in a real network.

On the other hand, for a decentralized BS switching algorithm, channel information, traffic requirement and QoS information are only shared among a BS and its neighbor BSs. In other words, no central controller is required in this case. BSs can make the switching decision on their own based on the shared information. The optimal set of active BSs can be selected for a specific area.

The author in [18] concluded that the centralized algorithm saves more en-ergy by switching more BSs off, however the QoS of UEs is better when decen-tralized switching algorithm is implemented.

Moreover, based on the conception of the decentralized BS switching al-gorithm, the author in [4] proposed a BSs switching procedure which can be implemented in this thesis.

(23)

2.5 Contribution

In order to save energy consumption in a cellular network, base stations switch-ing algorithms needed to be designed. Meanwhile, the QoS of end users durswitch-ing the switching procedures is also taken into consideration.

When we switch off a base station, the end users being served by this base station will be handed over to its neighbor base stations. Similarly, users will be handed over back when we switch on a sleep base station. The QoS of those users are impacted by the switching actions. In this thesis, the service delay is considered as a measure of Quality of Service (QoS) and user experience. A BS switching algorithm is proposed which can guarantee the QoS of UEs in the system.

In this thesis, a distributed switching-on/off energy saving algorithm is pro-posed without the requirement of a central controller.

Moreover, a hysteresis to mitigate the inefficient repetition of switching off and on due to the high variation of traffic load is considered.

(24)

Chapter 3

System Models

The system models used in this thesis are discussed in this chapter. They are: cellular network model, traffic model, power consumption model, user equip-ments’ association rules and Quality of Service measurement.

3.1 Cellular Network Model

We consider a wireless cellular network in an urban micro-cell system where the set of base stations, denoted by B, lies in a two dimensional area. We only consider down link communication, i.e. from BSs to UEs.

3.2 Traffic Model

Users are assumed to arrive according to a Poisson process with arrival rate λ. Each user requires a random amount of down link service. The file size is assumed to be an exponentially distributed random variable with mean value L.

3.3 Channel Model

We assume that the channel between BSs and UEs only consists of distance-based path loss and shadowing. In other words, the slow-varying channel con-dition is taken into consideration.

The service rate of BSs is equally shared by all users being served, i.e. fair share scheduling is used in the system. If multiple UEs are attached to the same

(25)

BS, round robin time scheduling is implemented.

For convenience, we show the notations to formulate the problems in Table 3.1.

b ∈ B Base station index Bon∈ B The set of active BSs

N Bb The set of neighboring BSs of BS b

Rm(t) The serving rate of UE m got from the network at time t

W Bandwidth of the system λ UEs arrival rate

L The mean value of files size generated by UEs

M The set of UEs

Mb The Set of UEs connecting to BS b

Nb(t) Number of UEs connecting to BS b at time t

Pb Transmission power of BS b

gbm Channel gain between BS b and UE m

N0 Noise level

Im(t) Interference experienced by UE m at time t

η A constant related to bit error rate requirement PT b The total power expenditure of base station b

P0 The static base station power consumption in the active mode

Psleep The static base station power consumption in the sleep mode

∆P The slope of transmission power

Es A fixed switching energy cost for each mode transition

Table 3.1: List of Notations

We assume that there are Nb UEs connecting with BS b. The interference

Im(t) that UE m experiences at time t is calculated as:

Im(t) =

X

b∈N Bb−{b}

Pbgbm ∀b ∈ Bon (3.1)

As mentioned above, the service rate of BSs is equally shared by all users being served, i.e. fair share scheduling is used in the system. When a UE m is selected to be served at time t, the service rate can be computed:

Rm(t) = W log2(1 +

ηPbgbm

(26)

where η is a constant related to the bit error rate requirement when adaptive modulation and coding is used [19].

3.4 User Equipments Association Rules

We assume UEs in the system will select the BS with the strongest signal strength, i.e, the BS b will be selected to be associated which can:

max

b∈BonPbgbm (3.3)

3.5 Power Consumption Model

The base station power model maps the Radio Frequency (RF) output power radiated at the antenna elements,Pout, to the total supply power of a base station

site,Pin. Figure 3.1 illustrates how the base station power model is integrated

into such an existing evaluation framework[8].

The base station power model constitutes the interface between component and system level, which allows quantifying how energy savings on specific com-ponents enhance the energy efficiency at network level. The characteristics of the implemented components largely depend on the base station type, due to constraints in output power, size and cost[8].

We assume ab(t) = {0, 1} is the activity indicator of BS b at a certain time

period, which is determined by the base stations switching strategy, and a is a vector of the activity indicators of all base stations.

The BS has active mode and sleep mode. The total power expenditure of a base station b at time t can be found[14, 8, 20]:

Pin= PT b(ab(t)) =      P0+ ∆PPb ab(t) = 1 Psleep ab(t) = 0 (3.4)

where P0is the static base station power consumption in the active mode, Psleep

is the static base station power consumption in the sleep mode and ∆P is the

slope of the load dependent power consumption.

Note that BSs will only use transmission power Pb when there is at least one

(27)

Figure 3.1: Integration of the base station power model for system level perfor-mance evaluations[8] Pb=      0 Nb = 0 Pb Nb> 0 (3.5)

Furthermore, we also assume that there is a fixed switching energy cost Es for each mode transition[21].

3.6 Quality of Service Measurement

As mentioned above, the service rate of a UE can be computed by 3.2. However, since a fair time-sharing scheduling (e.g. Round robin) is used when multiple UEs are attached to the same BS, for a long time period, say t ∈ (t0, t00), the average service rate of a UE m can be computed:

Rm(t00) =

W

N (t00₎log2(1 +

ηPbgbm

(28)

where Nb(t00) is the number of UEs which are associated with BS b during time

period t ∈ (t0, t00), we assume it is a constant in this period.

Figure 3.2: Service Rate of UE m for long time period T

For a longer period, which contains several time slots as discussed above, which is shown in Figure 3.2, the total bits which can be transmitted in this period can be computed:

Sm(T ) =

Z T

0

Rm(t)tdt ∀m ∈ M (3.7)

The service delay is considered as a measure of QoS and user experience in this thesis. The service delay is defined as the average response time from the user’s service request arriving at the BS until this request is finished. One should notice that, we assume that all the files generated by the UEs can be successfully transmitted through the network as long as the UE is connecting to a base station.

3.7 Energy Efficiency Model

In order to analysis the performance of BS switching algorithm from network perspective, energy efficiency is introduced.

EE = St(T ) Et(T ) = St(T ) RT 0 Pttdt [bits/J oule] (3.8)

where Et(T ) is the energy consumption of the whole network during time period

T .

One can say a network is more energy efficient when EE of the network is higher.

(29)

Chapter 4

Switching Algorithms

In the chapter, we will introduce the background and details of the switching algorithms used in this thesis, which included the BS switching off algorithm and the BS switching on algorithm. Moreover, a piratical switching procedure is introduced. In order to compare the performance of different switching methods, a traditional BS switching algorithm is also mentioned.

The main motivation of BS switching algorithms is the potential energy saving when low traffic BSs are switched off. However, in order to cover the service area, some BS cannot be switched off even there is no traffic currently in the area.

4.1 QoS Guaranteed BS Switching Algorithms

4.1.1 A Notion of Switch-effect

Let us consider a simple case shown in Figure 4.1, where the central base station (BS 1) is turned off. Since the conception of BS wilting is implemented, the UEs serviced by BS 1 will be handover to its neighboring BSs (BS 2-7). Turning off the central base station will increase the traffic load of the other active neigh-boring BSs. However, on the other hand, it will save the energy consumption of the system and may bring positive impact due to reduced inter-cell interference. In Figure 4.2, we provide an example to illustrate how the UEs are trans-ferred to neighboring BSs when the central BS is switched off. Now let us exam-ine the possibility whether a particular BS can be turned off or not. We defexam-ined the set of neighboring BSs of BS b as N Bb, and further denote m ∈ MN Bb as

(30)

Figure 4.1: A cellular network unit

the UEs in the system.

The service rate of UEs in the system can be computed by 3.6. One can see that the service rate is strongly related to the number of UEs being serviced by a BS. The service rate of UEs in the system will be influenced since the UEs will be rearranged to BSs after the hand-over procedure.

During the hand-over procedure, the UEs serviced by BS b will select the BS which can provide the second strongest signal strength to hand-over to.

The BS b will be able to be switched off only if the service rate of all the UEs in the coverage of its neighboring BSs satisfies the following feasibility constraint after the hand-over procedure:

Rm> Rth m ∈ MN Bb (4.1)

To simplify our algorithm, a notion of switch-effect is introduced. Math-ematically, the switch − ef f ect for the decision of the switching-off BS b is defined by:

(31)

Figure 4.2: An illustrative example of switching-off BS

SEb= min m∈Mb

Rm (4.2)

The value of the switch-effect will be used in the switching algorithm which will be mentioned in the later section.

4.1.2 Switching Off Algorithm

The switching-off decision criterion is based on the 4.2. It only depends on information feedback from neighboring BSs and serving UEs. Thus, it is possible that the switching-off algorithm can be localized as a problem at each base station and its neighboring base stations. The system information such as signal strength and number of UEs being serviced by base stations are periodically shared among base stations and UEs. It should be highlighted that the proposed algorithm does not require a centralized controller.

The switching-off algorithm involves three parts[4]:

Pre-processing State

UEs periodically feedback information about the received signal strength from different base stations in typical cellular networks such as IEEE802.16 [22]. That information can be used for resource management. The UEs reports its second

(32)

choice base station ID. If base station b is turned off, the UEs in its coverage will be handed-over to the BS which can provide the second strongest signal strength. The number of UEs being serviced by base stations are shared among neighboring base stations periodically.

Decision State

Each base station first calculates the switch-effect 4.2 based on information received from its neighboring BSs and UEs. It determines whether or not to be switched off after receive confirmation from all its neighboring BSs:

1 Switch Off Algorithm

1: if SEb> Rof f_th && Gb≤ 0 then

2: Send the request to switching-off to neighboring BSs, N Bb

3: else

4: Gb= Tb

5: end if

Note that a protection period Gb is introduced in order to avoid inefficient

repeatedly switching. It is a timer located in each base station which will countdown time continuously. It will be reset every time when the base station is failed to be turned-off.

Since it is a distributed algorithm running without a centralized controller, it might be possible that two or more base stations with overlapping neighbors simultaneously switch off. To avoid such situation, each BS first broadcasts Request To Switching-OFF (RTSOFF) and can only be switched off when it receives Clear To Switch-OFF (CTSOFF) from all its neighbors. Then, it will be turned off and send Confirmation Of Switching-OFF (COSOFF) to all its neighbors again[4].

Note that, before the COSOFF can be sent out, the neighbor BS must make sure that the service area can still be covered when the target BS is switched off.

Post-processing State

The base station b switches off when it received CTSOFF from all its neighbor base stations. UEs served by the switching-off base station are transferred to the neighboring base stations. The procedure requires the technology of group

(33)

hand over since there will be a group of UEs be handed over simultaneously. One of the keys for efficient group hand over is to predict the hand over procedure before it happens[23, 24]. The information about the control signaling related to the switching off decision, such as RTSOFF and CTSOFF can be used in the prediction of group hand over procedure[4].

Figure 4.3: The switching-off procedure

The three parts of the switching-off procedure can be summarized in Figure 4.3.

4.1.3 Switching On Algorithm

The switching on algorithm could be the reverse of the switching off algorithm. The basic concept of the switching on algorithm is that the BS should be switched on when the QoS of UEs in the network unit (including the switched off BS and its neighbors BSs) is less than a threshold. However, the turned-off BS cannot make a switching on decision by itself because it does not have in-formation about the current system inin-formation. For that reason, the switching on process needs to rely on neighboring BSs. Before a base station is turned off, the BS and its neighboring BSs exchange the information. So each BS has the information about how many neighboring base stations are active.

(34)

Similar to the switching off algorithm, the switching on algorithm also in-volves three parts as follows[4]:

Pre-processing State

Assume base station b is turned off. The neighboring base stations of b know that base station b is switched off. And the neighboring base stations, say base station b0 will collect and record the QoS of its own serving UEs including the handed over UEs from BS b.

Decision State

Since the low QoS may result from other reason besides slept BSs like shadow fading. When the QoS of a UE in MN Bb serviced by BS b

0 _{reaches the}

thresh-old, BS b0 _{wakes up base station b for a short period by sending Request To}

Switching-ON (RTSON) to see if the QoS of that UE is improved. The sleeping BS will only be waked up when it receive the switching on conformation. The decision for the switching on is determined as follows:

2 Switch On Algorithm

1: if Rb≤ Ronth then

2: Send Request To Switching On to neighboring BS b which was switched off

3: UEs are associated to the BS which can provide the strongest signal strength

4: if Rb> Ronth then

5: Send Conformation Of Switching ON

6: end if

7: end if

Due to the high variability of the traffic, BSs might repeat switching off and on inefficiently, which is similar to the Ping-Pong effect. To avoid this problem the switch on/off threshold should be set differently.

Figure 4.4 shows an example of setting the different thresholds. We introduce a hysteresis margin ∆hfor piratical implementation. The threshold for switching

(35)

Figure 4.4: The different thresholds for switching on/off     

Rof f_th = Rth+ ∆h/2 for the switching-off

Ron_th = Rth− ∆h/2 for the switching-on

(4.3)

For instance, a BS b will be switched off at time t1, the service rate of a

UE,Rb decrease since it have been handed over to other BS with less signal

strength. At this time, the QoS is under the threshold. If we have the same threshold, BS b will be switched on again immediately, which will lead to a ping-pong effect. In our case shown in Figure 4.4¨ı1₄Œhowever, the BS b will only be switched on again when the service rate is under the switching on threshold at time t2.

Post-processing State

If a turned off base station receives RTSON, it will be woken up. Accordingly, UEs located in its possible serving area will re-select their serving base stations. The three parts of the switching-on procedure can be summarised in Figure 4.5.

4.2 Transitional Switching Algorithm

In order to compare the performance of the switching algorithm mentioned above, the traditional BS switching is introduced here.

(36)

Figure 4.5: The switching-on procedure

1. Providing the service area can be covered by the other active BSs, a BS will be switched off when it is idle, in other words, there is no UEs serviced by this BS.

2. The sleeping BS will be woken up by any its neighboring BS when the traffic of that BS is larger than a certain threshold.

As one can see, the QoS of UEs is not considered directly in the traditional BS switching algorithm. Furthermore, no hand-over procedure is taken in this switching algorithm.

(37)

Chapter 5

Theoretical Background

In this chapter, we will discuss the theoretical background and some mathemat-ical analysis of the switching algorithms.

5.1 Processor-Sharing Queueing System

As mentioned in Chapter 3, the service rate of BSs is equally shared by all users being served, i.e. fairly share scheduling is used in the system. If multiple UEs are attached to the same BS, round robin time scheduling is implemented.

On a single server Processor Sharing (PS) queue model, the capacity C of the server is equally shared between the customers in the system,

1. If there are n customers in system each receives service at the rate C/n. 2. Customers do not need to wait; the service starts immediately upon arrival

The PS queue model is an idealized model, since in general the capacity of the server cannot be divided in continuous (real valued) parts, however, it is a good approximation.

5.2 M/G/1 PS Queueing System

Users are assumed to arrive according to a Poisson process with arrival rate λ. Each user requires a random amount of down link service. The file size is assumed to be an exponentially distributed random variable with mean value L.

(38)

When a UE gets all the capacity of the server, the service rate can be com-puted by 3.2. Since the UEs are uniformly distributed in cells. The mean service rate of a UE is defined by E(Rm). The mean departure rate of UEs is E(µ) =

E(Rm)/L, and the mean service delay of UEs is E(d) = 1/E(µ) = L/E(Rm).

From the Little’s Law, we know that the mean service delay is directly related to the average number of UEs serving by a BS.

E(Nb) = λE(d) (5.1)

According to the property of M/G/1 PS queue, the average service delay of UEs in a cell is

E(d) = L

E(Rm) − λL

(5.2) Because of Little’s Law, the average number of UEs in a cell is

E(Nb) =

λL E(Rm) − λL

(5.3)

5.3 First Order Analysis

The analysis for the amount of energy saving is challenging because the required parameters for analysis are dynamically changing during the switching process. Furthermore, many parameters are related with each other. At the given time instance t, our problem becomes to determine the set of active BSs subject to the QoS guarantee and energy minimization. Note that this problem can be reduced from a vertex cover problem which is NP-complete[25].

In this section, we develop a rough first-order analysis under our BS switch-ing strategy, which gives an insight into key factors affectswitch-ing the energy savswitch-ing.

Let us define the energy saving ratio as:

S = P rof f· |B| (5.4)

where Pof f is the probability of a BS can be switched off and |B| is the total

number of BSs in networks.

According to our switching algorithms mentioned above, consider a BS b in the network. The probability of switching off BS b will be:

(39)

P rb_{of f}= Y

m∈Mb

P r(Rm> Rth) (5.5)

where Rmis the service rate the UEs can get after the hand-over procedure.

In other words, the BS b can be switched off when the switching effect introduced in 4.2 is larger than the threshold.

P rbof f = P r(SEb> Rth) = P r( min m∈Mb

Rm> Rth) (5.6)

One can see from Equation 3.2, the service rate of UEs is strongly related to Nb, and Signal to Interference plus Noise Ratio (SINR). The first one Nb is

the number of UEs serving by a BS, which is directly related to the traffic load of a BS. While the second one is related to the location of the UEs, and it will also be influenced the system traffic load. In order to simplify the analysis, we will discussed the influence of this two parameters separately.

5.3.1 The Influence of Cell Radius

The radius of cell can influence the performance of the network in many ways. Firstly, smaller cell radius results in more number of BSs are required to cover a certain area, which also leads to more energy consumption.

Secondly,since the UEs are assumed to be uniformly distributed in cells, the cell radius influences the SINR experienced by UEs.

The SINR experienced can be computed:

SIN Rb=

ηPbgb0_m W N0+ Im

(5.7) When the cell radius is small, the interference, Im, experienced by UEs is

very large, which will lead to small switching off probability and small energy saving ratio.

On the other hand, when the cell radius is large, the path gain, gbm, will be

also small due to the long distance between BS and UEs. This will also results in small switching probability and energy saving ratio. In summary, a optimal cell radius exist for certain other parameters in a network.

Thirdly, the cell radius influences the UEs arrival rate in each cell. The UEs arrival rate is higher when the cell radius is larger, since more area needs to be cover. The UEs arrival rate is one of the key factors to the system traffic load, more details will be discussed in the next subsection.

(40)

5.3.2 The Influence of Traffic Load

One can see from Equation 5.7 that the switch effect of BS b is strongly related to Nb0 which is the number of UEs connecting to the neighboring BSs of BS b. The number of UEs in a cell depends on many other factors, including UEs arrival rate, package size generated by UEs, service capacity of BSs and etc. It is dynamically changing during the switching process.

Based on the M/G/1 PS queueing model and Little’ law we discussed above, one can easily conclude that the traffic load is directly related to the average number of UEs serving in cells. The higher traffic load leads to a higher number of UEs staying in the system.

When there are more UEs staying in the system, the switching probability will be affected in two ways. Firstly, UEs will experience lower service rate due to the round robin scheduling. It will decrease the probability of switching off BSs according to 5.6. Secondly, since the minimized UE service rate is used in the switching algorithm, the high number of UEs in the system will also lead to a higher probability that a UE is located in a bad location and experiencing bad QoS. It also results in a ower energy saving ratio.

In summary, when the traffic load is higher in cells, the probability of switch-ing off BSs is lower. It results in higher power consumption and UEs in the cells experiences higher service delay.

5.3.3 The Influence of Switching Threshold

When a high switching threshold Rthis chosen, it leads a low traffic load

thresh-old ρth. It results in a low probability of switching BSs and high power

con-sumption. However, when the switching threshold is high, UEs will experience low service delays, since UEs are guaranteed to get a high service rate before BS switching procedures.

On the other hand, when a very low switching threshold Rth is chosen, it

leads a very high traffic load threshold ρth. The probability of switching off the

first several BSs is higher because of the high traffic load threshold. The first several BSs are allowed to be switched off even when there is a high traffic load. More traffic load is distributed to the active BSs, the value of the hand-over effect functions increase significantly, which results in the considerab drop in the probability of switching off more BSs. As a result, only a few BSs can be switched off and the other active BSs keep servicing very high traffic load and

(41)

cannot be switched off. The power consumption of the system is also high. More analysis detail can be found in the appendix.

(42)

Chapter 6

Simulation Environment

Simulation is done with the help of a rudimentary network emulator (RUNE), a network simulator tool developed by Ericsson in MATLAB environment. RUNE is able to create a cellular network with several hexagonal cells and calculate path loss according to the positions of BSs and UEs. Additional MATLAB scripts and functions are added to integrate base stations switching algorithm and cells wilting into the RUNE environment. Furthermore, the energy con-sumption and the QoS measurement model are also added to the RUNE envi-ronment.

6.1 Simulation Scenario

The simulation scenario is considered to be a micro cell cellular network with three layers (37 base stations) in an urban area, which is shown in Figure 6.1. The base stations are set in the middle of each cell. Several UEs are generated in each hexagonal cell with a random size of files needed to be transmitted in the initial network. All the BSs are active and the UEs are associated to BSs according to Equation 3.3.

We assume the cell radius is 100m, the maximum cover radius of a BS is 300m. In other words, each cell can be covered by its neighbor BS. For instance, in Figure 4.1, the area can all be covered by the center BS.

(43)

Figure 6.1: The Simulation scenario

6.2 Wrap Around Technique

In the scenario in Figure 6.1, one can see that the cells at the outskirts of the scenario experience less interference than those in the middle. Since some cells are surrounded by six neighboring cells, while others have fewer neighbors. This simulation artifact is called border effect. However, we cannot afford to study a very big scenario where the problem can be ignored. Instead, we would like to model the system as being homogeneous, both in terms of traffic load and cell sized. Wrap around is a simulation technique used to compensate for the border effects. One can see from Figure 6.2 that each cell is virtually replicated around the simulated system six times to avoid the border effect. After that all the cells in our scenario will experience the same interference since all the cells have six neighboring cells.

6.3 Shadow Fading Model

When a large object, such as a hill, a building or some woods, is located between the sender and the receiver, the waves are more attenuated than in the case that

(44)

Figure 6.2: The wrap Around Technique

two units are in line of sight.

Wireless channel also experiences shadowing effects due to blockages of the paths between the transmitter and receiver by various obstacles. Usually shadow effect is modelled using log normal distribution and this model has been con-firmed empirically to accurately model the variation in path loss or received power in both outdoor and indoor radio propagation environments. The shadow fading gain model as follow is used:

gshadow=√p Gcorrel+

p

1 − p Guncor (6.1)

where the first parameter Gcorrelrepresents the situation when UEs are located

inside buildings, on the tops of buildings or in basements. In this case, the shadow fading is correlated. On the other hand, the second parameter Guncor

represents the situations when UEs are located in the street or at the corner of the buildings. The shadow fading effect is uncorrelated. The p is the parameter to balance those two different shadows fading.

(45)

due to this object is quite similar in a neighborhood. This is modeled a spatial correlation of the shadow fading random variable. If the object is large, for instance a hill, the correlation is strong in the sense that one needs to move a large distance in order to obtain uncorrelated shadow values. If the object is small the correlation distance is smaller.

A convenient way to model the space-correlation and to achieve repeatability in the generation of the shadow fading is to use pre-generated fading maps.

6.4 Path Gain Model

After the UEs are generated according to the arrival rate, the path gain between UEs and BSs are computed [26]:

gbm[dB] = gbmshadow− 32.4418 − 20log10(f ) − 0.0174d − 20log10(max(

0.013d f , 1))

(6.2) where d in meters is the distance from UE m and BS b and f in GHz is the carrier frequency. The channel propagation and BS characteristic we used here are according to the suggestions in the IEEE802.15m evaluation methodology document as urban micro model. In order to simply the problem, only LOS channel propagation is considered.

More details of the parameters used to build the simulation scenario can be found in Table 6.1:

Note that η = 0.283 corresponds to the BER requirement of ε = 10−3[19] and we take the micro BS energy consumption parameters[20].

6.5 Simulation Initiation

As shown in 6.1, a cellular network is built with 37 cells. All the BSs are active at the beginning, and several UEs are randomly generated in each cell. All the UEs generate files with exponentially distributed random variable with mean value L. And all the UEs are associated with the BS which can provide the strongest signal strength.

(46)

Carrier frequency 2.5GHz

System bandwidth 10MHz

Thermal noise per MHz -174dBm

Number of cells 37

Standard deviation (in log-domain) of the shadow fading 5dB

Correlation distance in the shadow fading map 15m

The correlation coefficient of the shadow fading from two cells 0.5

Transmission power of BS 10W

The mean value of files size generated by UEs 50 Mbits Static base station power consumption in the active mode 100W Static base station power consumption in the sleep mode 20W

The slope of transmission power 5

Fixed switching energy cost for each mode transition 25J The constant related to bit error rate requirement 0.283

Simulation duration 60min

Simulation accuracy Every 0.1s

Table 6.1: General Simulation Parameters

6.6 Base Stations Switching

If a BS is switched off, UEs serviced by that BS will be handed over to its neighboring BSs. One example can be seen in Figure 6.3, where four BSs have been switched off. One can see that the UEs in those sleeping cells are handed over to their neighboring cells. Furthermore, since the wrap around technique is used, the locations of the users at the outskirts are ”fold back” to six other locations. In that case, the UEs are handed over to the cells on the other side of the scenario.

(47)

(48)

Chapter 7

Simulation Results and

Performance Analysis

In this chapter, we show the main results from our simulations and give some performance analysis base on the results.

7.1 Energy Saving by the QoS guaranteed BS

Switching Algorithms

In this section, we discuss the energy saving performance by implementing the QoS guaranteed BS switching algorithms, we also compare it with the traditional switching algorithms.

More details of the parameters used in this section can be found in Table 7.1:

Cell Radius 100m

Switching off threshold 20.5 Mbit/s Switching on threshold 19.5 Mbit/s

Table 7.1: Simulation Parameters Section I

Note that a traffic load is assumed to be spatially homogeneous and varies by scaling the traffic arrival rate. With the increasing of traffic arrival rate, if the system load for any BS reach the critical point between stable and unstable status, we treat this point as normalized traf f ic load = 1. When a BS is in

(49)

unstable status, the UEs serviced by the BS will experience unlimited service delays.

7.1.1 Energy Saving Ratio

Figure 7.1: Energy Saving of BS Switching Algorithm

Figure 7.1 shows the amount of energy saving for the QoS guaranteed BS switching algorithms under synthetic traffic profiles, i.e., varying the normalized traffic load from zero to one. We also include the performance of a cellular network without implementing BS switching algorithms as a reference. As one can see, the lower normalized traffic load is, the higher energy savings ratio can be expected. The result can also be expected from theoretical background we discussed above. About 62% energy saving ratio can be expected when the normalized traffic load is low. However, when the normalized traffic load is higher, less BS can be switched off due to the QoS assurance, and less energy saving ratio can be expected.

Furthermore, the energy consumption grows faster when normalized traf-fic load is higher. It results from the influence of the inter-cell interferences. When the normalized traffic load is higher in each cell, more BSs keep using transmission power which leads to worse inter-cell interference to its neighboring BSs.

(50)

7.1.2 Average Switched-off BSs

Figure 7.2: Average number of Switching-off BS

Figure 7.2 shows the average amount of switched-off BSs for the switching algorithms under synthetic traffic profiles. The figure explains the performance of energy saving in the above figure. When the normalized traffic load is lower, more BSs can be switched off by the algorithms, which leads to more energy saving. Up to 30 out of 37 BSs can be switched off by the algorithms while the other seven BSs are active to ensure the coverage of the network. On the other hand, less BSs can be switched off when the normalized traffic load is close to one.

7.1.3 Service Delay

One can see the average service delay of UEs in the network from Figure 7.3. When the normalized traffic load is lower, the USs in the system without imple-menting switching algorithms experience a higher service delay. It contributed by the fact that more BSs can be switched off by the algorithms when the normalized traffic load is lower and the UEs will experience less cell inter-ference. As can be seen from Equation 3.2, UEs will have a higher service rate with less interference.

However, on the other hand, because of the hand-over procedure during the switching algorithms, UEs will be hand-over to the neighboring BSs of its

(51)

Figure 7.3: Average service delay of UEs

originally servicing BS. Because of the high path lose and other factors, UEs will experience a higher service delay. And when the normalized traffic load is close to one, there is no difference between the performance of the two networks since none BS can be switched off by the switching algorithms.

7.1.4 Summary

In conclusion, up to 62% of energy saving can be expected by implementing the BS switching algorithms. And the energy saving ratio is strongly related to the normalized traffic load of the network. A higher energy saving ratio can be found when the normalized traffic load is lower since more BSs can be switched off.

Another advantage of implementing the BS switching algorithms is the de-crease of inter-cell interference when the normalized traffic load is low. As a result, the UEs experience less service delay. On the other hand, higher service delay is expected when the normalized traffic load is high due to the hand-over procedure during the switching procedure.

There is no difference between network with or without switching algorithms both in energy consumption and service delay of UEs when the normalized traffic load is close to one since no BSs can be switched off.

(52)

7.2 Comparison of different BS Switching

Algo-rithms

In order to compare the performance of the QoS guaranteed switching algorithm with other BS switching algorithms, we introduce the traditional BS switching algorithm which mentioned in Chapter 4. The parameters used in the section is the same in Table 7.1.

7.2.1 Comparison of Energy Saving

Figure 7.4: Comparison of Energy Saving

Figure 7.4 shows the amount energy savings for different algorithms under synthetic traffic profiles. One can see both BS switching algorithms can save energy for cellular networks. However, the QoS guaranteed switching algorithms can save more energy when the normalized traffic load is around 0.45 to 0.75. It results from the fact that the QoS guaranteed switching algorithms are more adaptive to the change of traffic load based on the QoS requirement. More BSs can be switched off and more energy savings can be expected since the hand-over procedure is processed. While the BS can only be switched off when it is idle in the transitional switching algorithms. When the normalized traffic load

(53)

is close to one, neither of the algorithms can save energy since no BSs can be switched off.

7.2.2 Comparison of Service Delay

Figure 7.5: Comparison of Average Service Delay

Figure 7.5 shows the comparison of average service delay experienced by UEs under different switching algorithms. It can be seen that the QoS guaranteed switching algorithm and traditional switching algorithm have the same perfor-mance when the normalized traffic load is either close to zero or one. It results from the fact that the same amount of BSs can be switched off under those traffic loads. However, the UEs under QoS guaranteed switching algorithms are expected to experience less service delay when the normalized traffic load is very between 0.3 and 0.8. Because the QoS is considered from the UEs perspec-tive in the QoS guaranteed switching algorithm and BSs can be switched more adaptable to the changes of traffic load. While in the traditional BS switching algorithm, the QoS of UEs is not considered. Sleeping BS may not be woken up in time even if the UEs in those cells are experiencing high service delays. Note that the advantage of reducing the inter-cell interference can also be achieved in the traditional switching algorithm.

(54)

7.2.3 Summary

In conclusion, both the traditional switching algorithm and the QoS guaranteed switching algorithm can save energy for cellular networks. The two algorithms have similar performance when either the normalized traffic load is close to zero or one. More energy savings are expected when the normalized traffic load is lower.

The QoS guaranteed switching algorithm saves more energy when the nor-malized traffic load is between 0.45 and 0.75. It results from that BSs can only be switched off when it is idle in the traditional switching algorithm while more BSs can be switched off under the QoS guaranteed switching algorithm since the hand-over procedure is processed during BS switching.

Furthermore, the UEs under QoS guaranteed switching algorithm are ex-pected to experience a lower service delay when the normalized traffic load is between 0.3 and 0.8. It is contributed by the fact that the QoS of UEs is con-sidered from the UEs perspective in the QoS guaranteed switching algorithm. While under the traditional switching algorithm, sleeping BSs may not be woken up in time even if the UEs are experiencing high service delay in those cells.

7.3 Characteristics of the QoS Guaranteed

Switching Algorithm

As discussed above, there are many factors which affect the performance of the switching algorithms. In this section, we discuss the influence of some key factors which include cell radius, traffic load and switching thresholds.

7.3.1 The Influence of Cell Radius

In this section, we discuss the influence of cell radius to the performance of switching algorithms. As mentioned above, the cell radius parameter influences the performance of switching algorithms in many different ways, we will discuss it in three different perspectives.

We assume there is a certain area needed to be coveedr by a cellular network. The arrival rate of UEs is also assumed to be spatially homogeneous. Some details of the parameters in this section can be found in Table 7.2:

(55)

Area to be covered 0.96 km2

Switching off threshold 20.5 Mbit/s Switching on threshold 19.5 Mbit/s UEs arrival rate 0.185 U Es/min/m2

Table 7.2: Simulation Parameters Section II

Required Number of BSs and UEs Arrival Rate per Cell

Figure 7.6: Required Number of BSs and Traffic Load

Figure 7.6 shows the required number of BSs to cover the area and the UEs arrival rate per cell per min under different cell radius. Smaller cell radius results in high required number of BSs, which also leads to more system energy consumption. However, each cell has less UEs arrival rate.

On the other hand, large cell radius can reduce the required number of BSs but it also results in a high UEs arrival rate for each BSs.

Signal to Interference plus Noise Ratio

Since the UEs are assumed to be uniformly distributed in cells, the cell radius influences the SINR experienced by UEs.

(56)

differ-Figure 7.7: Average Service Delay under Different Cell Radius

ent cell radius. One can see that, When the cell radius is small, the interference, Im, experienced by UEs is very large. UEs experience higher average service

delays. Moreover, there is no difference between the systems with and without implementing sleep mode. When the BS density is very high, the UEs can be handed-over to the BSs which are very close.

On the other hand, when the cell radius is large, the path gain, gbm, will be

also small due to the long distance between BS and UEs. This will also result in high average service delay.

By implementing the BS switching algorithm, less average service delay can be expected when the cell radius is between 40m and 85m since cell inter-ference is reduced by the switching algorithm. However, more average service delay is expected when the cell radius is larger because of the hand-over proce-dure.

Figure 7.8 shows the total energy consumption of the network under different cell radius. The performance of the networks with or without BS switching algorithm are compared.

When the cell radius is low, more energy is needed because of the high required number of BSs to cover the area.

(57)

Figure 7.8: Total Energy Consumption under Different Cell Radius

saved by the BS switching algorithm since UEs experiences very low service rate because of the low path gain, and less BSs can be switched off. Optimal cell radius can be founded when the networks consume the lest energy.

Energy Efficiency

In order to compare the performance of networks from the network perspective, Figure 7.9 shows the energy efficiency of the system under different cell radius. One can see the huge energy efficiency improvement by implementing the BS switching algorithm. The optimal cell radius also exists in terms of energy efficiency.

7.3.2 The Influence of Traffic Load

As shown in above Figures in Section 7.1 and the discussion in Chapter 5. Higher traffic load in cells leads to lower probability of switching off BSs, higher power consumption and higher service delay the UEs experience.

(58)

Figure 7.9: Energy Efficiency under Different Cell Radius

7.3.3 The Influence of Switching Threshold

In this section, we discuss the influence of choosing different switching threshold to the performance of switching algorithm.

Some details of the parameters fused in this section can be found in Table 7.3:

Normalized traffic load 0.66

Cell radius 100 m

Table 7.3: Simulation Parameters Section III

Figure 7.10 shows the power consumption and average service delay of UEs in the network under different switching threshold. The switching-on and switching-off threshold can be computed by Equation 4.3.

As discussed before, the switching threshold is one of the key factors which influence the performance of the algorithms.

As one can see from Figure 7.10,when a very low switching threshold Rth

is chosen, the probability of switching off the first several BSs is very high because of the high traffic load threshold. The first several BSs are allowed to be switched off even when they are serving very high traffic load. More

(59)

Figure 7.10: Power Consumption and Average Service Delay under Different Switching Threshold

traffic load is distributed to the active neighboring BSs, the value of the hand-over effect function increase significantly, which results in the big drop in the probability of switching off more BSs. As a result, only a few BSs can be switched off and the other active BSs keep servicing high very high traffic load and cannot be switched off for a very long time. The UEs serviced by those active BSs will experience very high service delays. The power consumption of the system is high. Another way to understand that is the low switching thresholds decrease the adaptability of the network. The BSs cannot be switched adaptively enough to the traffic load because of the high traffic burden after the first several switches.

On the other hand, when a very high switching threshold Rth is chosen,

it results in a low probability of switching BSs and high power consumption. However, UEs will experience low service delays, since UEs are guaranteed to get a high service rate before BS switching procedures.

(60)

Chapter 8

Conclusion and Future

Work

In this chapter, we drew the conclusion of this thesis project and discuss some future work of the project.

8.1 Conclusion

Contributed by the significant environmental footprint and eventual exhaustion of traditional energy resources, global energy consumption becomes a major issue. Wireless access networks, as a branch of the ICT sector, are responsible for a big part of the global energy consumption. It is crucial to find a way to cut down the energy consumption of the cellular networks.

Results from the smart-phones’ ubiquitous Internet access and diverse mul-timedia applications, there has been an explosion in mobile data recently. This incredible increase will necessitate continually higher energy consumption and leads to more CO2emissions. The problem is also significant for the cellular

net-work operators since the electricity bills are a portion of their OPEX. The huge savings in capital expenditure and operational expenditure can be implemented through reduced energy needs.

Based on the conception of cell wilting,hand-over procedures and dynamic base stations switching, BS sleep switching algorithms are developed while the QoS of UEs is guaranteed at the same time. The switching algorithm is decen-tralized which means no central controller is needed. The BSs can make the

(61)

switching decisions based on the feedback information from UEs and its neigh-boring BSs. Moreover, the implementations of the proposed algorithms are also comprehensively described at the protocol level.

8.1.1 Findings

1. Implementing BS switching algorithms can save energy for the cellular networks. Potential energy savings can be expected because of the fluctu-ations of traffic demands.

2. Switching off BSs with low traffic load can save energy and it can also decrease the inter-cell interference. However, it may decrease the QoS experienced by UEs in terms of service delay.

3. Up to 62% energy savings can be expected by implementing the QoS guaranteed BS switching algorithms.

4. Compared with the transitional BS switching algorithm, the QoS guar-anteed BS switching algorithm has better performance in terms of power saving and average service delay experienced by UEs, since the QoS guar-anteed BS switching algorithm takes the QoS into consideration from the UEs’ perspective.

5. Many factors influence the performance of the BS switching algorithms. The analysis for the performance is challenging because the required pa-rameters for analysis are dynamically changing during the switching pro-cess. The key factors are the traffic load, cell radius and switching thresh-old.

6. Lower traffic load leads to higher probability of switching of BSs. It results in more energy savings and less service delay experienced by UEs. 7. Cell radius parameter influence the performance of the switching algorithm

in many different ways.

8. Lower cell radius results in higher required number of BSs to cover a certain area, while less BSs are needed when the cell radius is higher. 9. The inter-cell interferences experienced by UEs is very high when cell

Energy Eﬃcient Cellular System Design with QoS Assurance

Energy Efficient Cellular System Design

with QoS Assurance

CHANGYANG ZHANG

Master’s Degree Project

Stockholm, Sweden October 2014

Abstract

Acknowledgments

Contents

List of Tables

List of Figures

Acronyms

Chapter 1

Introduction

1.1

General Background

1.2

Motivation

1.2.1

Traffic Dynamics

1.2.2

Potential Energy Savings

1.2.3

Quality of Service

1.3

Thesis Structure

Chapter 2

Related Works and

Contribution

2.1

Cellular Networks

2.2

Base Station Sleep Modes

2.3

Cells Wilting and Blossoming

2.4

Dynamic Base Stations Switching

2.5

Contribution

Chapter 3

System Models

3.1

Cellular Network Model

3.2

Traffic Model

3.3

Channel Model

3.4

User Equipments Association Rules

3.5

Power Consumption Model

3.6

Quality of Service Measurement

3.7

Energy Efficiency Model

Chapter 4

Switching Algorithms

4.1

QoS Guaranteed BS Switching Algorithms

4.1.1

A Notion of Switch-effect

4.1.2

Switching Off Algorithm

4.1.3

Switching On Algorithm

4.2

Transitional Switching Algorithm

Chapter 5

Theoretical Background

5.1

Processor-Sharing Queueing System

5.2

M/G/1 PS Queueing System

5.3

First Order Analysis

5.3.1

The Influence of Cell Radius

5.3.2

The Influence of Traffic Load

5.3.3