Pico Cell Densification Study in LTE Heterogeneous Networks

(1)

Pico Cell Densification Study in LTE Heterogeneous Networks

Guanglei Cong

Supervisors:

Fredric Kronestedt Systems & Technology Development Unit Radio

Ericsson AB

Ming Xiao

Communication Theory Lab School of Electrical Engineering

KTH

(2)

(3)

Abstract

Heterogeneous Network (HetNet) deployment has been considered as the main approach to boost capacity and coverage in Long Term Evolu- tion (LTE) networks in order to fulfill the huge future demand on mo- bile broadband usage. In order to study the improvement on network performance, i.e. capacity, coverage and user throughput, from pico cell densification in LTE HetNets, a network densification algorithm which determines the placement locations of the pico sites based on pathloss has been designed and applied to build several network models with different pico cell densities. The study has been taken based on a real radio network in a limited urban area using an advanced Matlab-based radio network simulator. The simulation results show that the network performance generally is enhanced by introducing more pico cells to the network.

Keywords: HetNet, LTE, pico cell, densification, capacity, coverage,

throughput.

(4)

Acknowledgements

This thesis was performed in System & Technology, DU Radio, Ericsson AB in cooperation with Communication Theory lab, EES, KTH. First of all, I would like to express my appreciation and gratitude to my super- visor from Ericsson AB Fredric Kronestedt for his great guidance and kindness during the last six months. I would also like to thank my supervisor from KTH Prof. Ming Xiao who gave me a great deal of assistance and provided me with support throughout the entire thesis. I must express my thanks to Dirk Gerstenberger and Fredric Kronestedt for giving me this opportunity to spend six months in Ericsson AB to accomplish my master thesis. My sincere gratitude is also directed to my examiner Prof. Mikael Skoglund for his time and for his help when applying for this thesis.

I would also like to thank the following people from Ericsson AB for

their support and kindness during the last six months: Gunther Auer,

Jason Chen, Peter Björkén, Stefan Ström, Thomas Chapman and Tomas

Lundborg.

(5)

List of Figures

Figure 2.1: Heterogeneous Network ... 3

Figure 2.2: Cell selection in 3GPP LTE HetNets ... 6

Figure 2.3: Users with different radio link condition... 8

Figure 2.4: Average bitrate in full buffer model ... 8

Figure 2.5: Average bitrate in equal buffer model ... 9

Figure 3.1: Analyzed area ... 14

Figure 4.1: Work flow ... 20

Figure 4.2: Flow chart of pathloss based pico sites densification algorithm ... 22

Figure 5.1: Densified networks using pathloss based pico sites densification algorithm (chapter 4.2) ... 25

Figure 5.2: Densified networks using randomly distributed pico sites densification algorithm (chapter 4.2) ... 26

Figure 5.3: Pathloss based pico sites densification algorithm ... 27

Figure 5.4: Randomly distributed pico sites densification algorithm ... 27

Figure 5.5: Uplink comparison ... 28

Figure 5.6: Downlink comparison ... 29

Figure 5.7: Uplink comparison ... 30

Figure 5.8: Downlink comparison ... 31

Figure 5.9: Uplink mean user throughput vs uplink subscriber capacity ... 32

Figure 5.10: Uplink cell edge user throughput vs uplink subscriber capacity ... 33

Figure 5.11: Improvement of uplink capacity ... 34

Figure 5.12: Improvement of uplink coverage ... 35

Figure 5.13: Downlink mean user throughput vs downlink subscriber capacity ... 36

Figure 5.14: Downlink cell edge user throughput vs Downlink subscriber capacity ... 37

Figure 5.15: Improvement of downlink capacity ... 37

Figure 5.16: Improvement of downlink coverage ... 38

Figure 5.17: Uplink user throughput maps ... 39

Figure 5.18: Improvement of uplink mean user throughput ... 40

Figure 5.19: Downlink user throughput maps... 41

Figure 5.20: Improvement of downlink mean user throughput ... 42

Figure 5.21: Downlink geometry distribution ... 42

Figure 5.22: Uplink throughput vs capacity ... 43

(8)

Figure 5.23: Downlink throughput vs capacity ... 44

Figure 5.24: Average uplink utilization of pico and macro cells with the

target of 1 Mbps uplink cell edge user throughput ... 45

Figure 5.25: Average downlink utilization of pico and macro cells with

the target of 10 Mbps downlink cell edge user throughput ... 45

Figure 5.26: Pico and Macro cells served area ... 46

Figure 5.27: Comparison of Pico and Macro cells served area ... 47

(9)

List of Tables

Table 2.1: Guidelines for LPNs deployment [3] ... 5

Table 2.2: Comparison of Full and Equal buffer model ... 10

Table 3.1: LPNs deployment... 15

Table 3.2: Simulation parameters... 16

(10)

Terminology

Abbreviations

LTE Long Term Evolution

LPN Low Power Node

HetNet Heterogeneous Network

3GPP 3rd Generation Partnership Project

CSG Closed Subscriber Group

RRU Remote Radio Unit

UE User Equipment

RSRP Reference Signal Received Power RSRQ Reference Signal Received Quality

eICIC Enhanced Inter-cell Interference Coordination

TCP TEMS CellPlanner

GSM Global System for Mobile Communications

CDMA Code Division Multiple Access

WCDMA Wideband Code Division Multiple Access

CPLM Composite Pathloss Matrix

ACP Automatic Cell Planning

SINR Signal to Interference plus Noise Ratio

ABR Achievable Bitrate

MRC Maximal Ratio Combining

IRC Interference Rejection Combining

(11)

ISD Inter Site Distance

(12)

1 Introduction

1.1 Background and problem motivation

Mobile broadband usage has increased dramatically the last couple of years due to new types of terminals such as smart phones and tablet computers. According to [1], in 2010, wireless devices only accounted for 37% of IP traffic; but by 2015, wireless devices is estimated to con- sume 54% of IP traffic while wired devices will only consume 46% of IP traffic, which means that traffic from wireless devices will exceed wired devices.

To support the huge future demands as both the number of users and the user demand will increase, it is essential to enhance the network capacity and coverage. But with the knowledge that the deviation be- tween Long Term Evolution (LTE) link level performance and Shannon capacity is very small which limits the potential to increase spectrum efficiency [2], forcing us to find other means to meet the future demands.

A key method to fulfill the future needs is network densification through adding smaller low power nodes (LPNs) in traditional high power macro nodes, namely Heterogeneous Network (HetNet), which is expected to boost capacity and coverage beyond what is available in current LTE networks [3].

With the knowledge that HetNet deployment has a large potential to improve the network capacity and coverage, the influence of pico cell densification on the network performance is obviously of large interest.

1.2 Overall aim

In order to see how effective the method of HetNet is to solve the prob- lem of the huge future demand on network performance, the project’s overall aim is to investigate how the network performance will be affected by deploying more and more pico cells in the network.

1.3 Scope

The study has its focus on the impact of pico cell densification on the

network performance, the cost of the pico sites deployment is ignored;

(13)

as a result, the number of pico sites in the network models that have been built in this project might be unrealistic.

The study will be taken based on a real radio network in a limited urban area in a dense major European city, the results might vary for different area and cities due to different terrain features and macro sites deploy- ment.

1.4 Problem statement

To achieve the overall aim stated in chapter 1.2, the study has an objec- tive to respond to the following questions:

• Is it true that adding more and more pico cells will result in larger and larger capacity and coverage?

• Is there any upper limit on how many pico cells can be added to a dense urban macro network and still improves the network per- formance?

To achieve this objective, it is also desired to design a strategy of pico sites deployment:

• How should the pico sites be deployed to achieve better network performance from capacity and coverage point of view?

1.5 Outline

Chapter 2 provides the related theory concerning some aspects of HetNet in LTE and some tools that have been used during this thesis work.

In chapter 3, the network model and simulation parameters of this project are presented, in addition with a brief discussion of an indoor propagation model which has been applied in a part of this study.

Chapter 4 describes how the network models with different pico cell densities have been built and simulated. A pico sites densification algorithm to decide where to place the pico sites has been designed and is also discussed in this chapter.

In chapter 5, the simulation result of this work will be presented.

The conclusions and possible future work will be presented in chapter 6.

(14)

2 Theory

2.1 Heterogeneous networks in LTE

Heterogeneous network (HetNet) has been identified as a key method to fulfill the huge future demands on mobile broadband usage as both the number of users and the user demand will increase. In 3GPP LTE HetNets, traditional high power macro nodes are complemented with low power nodes (LPNs) which cover small areas and offer very high capacity and data rates in these areas [3], as illustrated in Figure 2.1.

Figure 2.1: Heterogeneous Network

Besides simply adding LPNs to the existing macro networks, there are some other approaches to expand network capacity and coverage as well, such as improving macro cells by allocating more spectrum and densifying the macro sites. Compared with these two methods, adding LPNs performs the same in the downlink and better in the uplink [3].

These three approaches can of course be combined together to meet higher demand.

2.1.1 Low power nodes deployment

LPNs deployment is a real challenge in HetNets, many aspects need to be considered [3]:

• Demand: traffic volumes, traffic location, target data rates

• Supply: macro cell coverage, site availability, backhaul transmis-

sion , spectrum and integration with the existing macro network

(15)

• Commercial: technology competition, business models In [3], some guidelines for LPNs deployment are provided:

• Open or closed access

Open access means LPNs are available for all subscribers to ac- cess. Open access should be chosen for public systems deployed by operators.

Closed access refers that LPNs belong to a Closed Subscriber Group (CSG), that is to say access is only available for users in CSG. Closed access is always used in user-deployed cases (by in- dividual, enterprises).

• Indoor or outdoor deployment

Deploying indoor LPNs is suitable in cases when traffic is con- centrated to specific indoor locations such as shopping malls.

Outdoor LPNs deployment that also covers indoor areas is pref- erable in cases such as local traffic hotspots cover a wide area in- cluding several buildings or the macro cells in the existing net- works are too sparse to meet indoor service demand.

• Type of LPNs

There are several types of LPNs: Remote Radio Units (RRUs), conventional pico nodes, relay nodes. RRUs are suitable for net- works with low-latency and high-capacity backhaul; otherwise stand-alone pico base stations should be preferable. Deploying relay nodes is a preferred option for networks without wire backhaul.

• Frequency reuse

The HetNets can be seen as composed of two layers: macro cell

layer and pico cell layer. The two layers can use different fre-

quency band or share the same band. Reusing the frequency band

of the macro cell layer for the pico cell layer is of course spec-

trum-efficient. When spectrum is scarce or capacity is the diver,

frequency should be reused. However, due to the inter-layer in-

terference, elaborate cell planning and interference management

technique is needed.

(16)

Table 2.1 summarizes the rules addressed above.

Choices Guidelines

Access Open access Operator-deployed

Closed access User-deployed

Deployment

Indoor Concentrated large indoor hotspot

Outdoor Outdoor hotspot or many smaller indoor hotspot

Type of LPNs

RRU Low-latency and high-capacity backhaul (fiber) Stand-alone pico base

stations High-latency and low-capacity backhaul (copper/microwave) Relay nodes No wire backhaul

Frequency reuse Reuse macro spectrum Spectrum is scarce / Capacity is driver

Separate spectrum CSG

Table 2.1: Guidelines for LPNs deployment [3]

2.1.2 Cell selection

Conventionally, cell selection is based on the downlink received signal strength which means mobile users will connect to the site from which the received power is strongest. For example, in 3GPP LTE, cell selection is performed according to two parameters measured by a User Equip- ment (UE): Reference Signal Received Power (RSRP) and Reference Signal Received Quality (RSRQ) [4].

“Reference signal received power (RSRP), is defined as the linear av- erage over the power contributions (in [W]) of the resource elements that carry cell-specific reference signals within the considered meas- urement frequency bandwidth” [4].

RSRQ is calculated based on RSRP which provides additional infor- mation and ensures a reliable cell selection decision when RSRP is not sufficient.

In homogeneous networks, RSRP based cell selection guarantees good

channel conditions in both downlink and uplink. But in HetNets, since

the transmission power in the downlink is different between the LNPs

and the macro nodes and this transmission power difference doesn’t

(17)

present in the uplink, RSRP based cell selection only guarantees good downlink channel conditions. As illustrated in Figure 2.2: in the grey area, the macro node is selected based on the downlink RSRP, but for the uplink the LPN is better since the transmission power is the same and the pathloss is lower towards the LPN [3]. That is to say, a better cell selection for the uplink is minimum pathloss cell section [5].

Downlink RSRP cell boundary

Uplink minimum pathloss cell boundary Biased RSRP cell boundary (RSRP + offset)

Figure 2.2: Cell selection in 3GPP LTE HetNets

As shown in Figure 2.2, the optimal cell boundary of downlink and uplink is not identical. To solve this problem, the RSRP cell boundary of the LPN should be extended. The most straightforward way is to extend the RSRP cell boundary by increasing the LPN transmission power, but this method reduces site availability since it affects the site size and the cost [3]. Another means without increasing the output power is to add an offset to the RSRP from the LPNs which will affect the cell selection and increase the pico cell range – Biased RSRP cell selection [3][5][6], see Figure 2.2.

Biased RSRP cell selection mechanism could of course improve the

uplink performance. However, it causes higher downlink interference

for users in the extended cell range area. Some interference management

techniques have been developed to solve this problem such as enhanced

Inter-cell Interference Coordination (eICIC) [7]. Without this kind of

interference management, there should be a tradeoff between downlink

and uplink performance. Biased RSRP cell selection with a modest offset

3-4dB performs the best in many cases [6].

(18)

2.2 Two packet traffic models

In the simulation of LTE networks, there are mainly two packet traffic models: full buffer model and equal buffer model. These two models are both implemented in the simulator used in this project.

2.2.1 Full buffer model Definition of full buffer:

• Static traffic: the number of active users in the system is fixed, no arrivals and departures;

• Infinite sessions: each active user’s session lasts forever creating infinite data volume;

• Best effort: each active user fully utilizes the radio link.

In full buffer model, on one hand, the radio links are always utilized since active users are fixed and last forever.

On the other hand, users with different radio conditions spend same time in the system (infinite session), that is to say, users with poor radio conditions generate less data. As shown in Figure 2.3, user A and B are always sending data, as a result A and B contributes the same to the system throughput as illustrated in Figure 2.4.

As a conclusion, full buffer model gives optimistic performance estima-

tion which may deviate from reality.

(19)

User A: 10 Mbps User B: 1 Mbps

Figure 2.3: Users with different radio link condition

t Bitrate

average

Figure 2.4: Average bitrate in full buffer model

2.2.2 Equal buffer model Definition of equal buffer:

• Dynamic traffic: the number of active users in the system is not fixed, new session arrivals and complete session departures;

• Finite sessions: each active user’s session ends when all data have been transmitted, all sessions have the same volume of finite data to send;

• Best effort: available link bitrate is utilized.

In equal buffer model, on one hand, radio links will be idle when all

users are inactive.

(20)

On the other hand, users with different radio conditions have same finite data volume to send (finite session), that is to say, users with poor radio conditions spend more time in the system. As shown in Figure 2.3, user A and B are sending the same volume of data, say 100 Mbits, user A needs 10s and user B needs 100s. As a result, B contributes 10 times more than A to the system throughput as illustrated in Figure 2.5.

As a conclusion, equal buffer model brings down the system capacity but it is more realistic. As a result, equal buffer model is chosen for this study and the simulation results shown in chapter 5 are all from equal buffer model.

t Bitrate

average

Figure 2.5: Average bitrate in equal buffer model

2.2.3 Comparison

Some main differences between full buffer model and equal buffer

model are listed in Table 2.2.

(21)

Full buffer Equal buffer Queue Fixed active users, no

arrivals and departures Random dynamic active users, arrivals and departures Volume per

user Infinite Finite, fixed

Session time Infinite Depend on radio condition of the user

Stability Always utilization < 100%

User arrival No arrivals e.g., Poisson arrival Table 2.2: Comparison of Full and Equal buffer model

2.3 Related Tools

As mentioned before, the study in this project will be taken based a real radio network in a limited urban area. To achieve this, several tools will be used during the whole process of this project.

2.3.1 TEMS CellPlanner

In some studies of radio network, the simulations based on the calculat- ed on-grid site locations and simplified propagation models are not sufficient compared with the realistic site data and propagation predic- tion. TEMS CellPlanner (TCP), a commercial cell planning tool, can be used to export such realistic data [9].

“TEMS CellPlanner is a graphical PC-based application for design- ing, implementing, and optimizing mobile radio networks. It assists you in performing complex tasks, including network dimensioning, traffic planning, site configuration, and frequency planning, and network optimization” [10].

TCP supports multiple mobile technologies such as GSM (Global Sys- tem for Mobile Communications), CDMA (Code Division Multiple Access), WCDMA (Wideband Code Division Multiple Access) and WiMax. The main features of TCP that will be used in this project are pathloss analysis and Composite Pathloss Matrix (CPLM) analysis.

The propagation models TCP used for these analyses are [11]:

• 9999 model

• Urban model

• Okumura-Hata model

(22)

• Walfish-Ikegami model

Pathloss analysis in TCP is performed for each individual sector: a bin matrix is built for each sector with the pathloss from the sector antenna to each bin within the prediction radius set manually. And TCP gives pathloss predictions very close to realistic data based on the inputs of TCP such as [11]:

• Terrain data: elevation, etc.

• Land use data: buildings, trees, open area, etc.

• System data: technology, frequency, etc.

• Site data: site location, etc.

• Antenna data: height, gain, etc.

CPLM analysis in TCP merges the pathloss results from individual sectors into a composite dataset based on the following parameters [11]

[12]:

• Max. pathloss (dB): largest pathloss which is considered to be rel- evant and allowed for the calculation. Any pathloss higher than this value is not included in the resulting CPLM calculation;

• Max delta pathloss (dB): largest difference between lowest and largest pathloss values in the resulting CPLM calculation. The lowest pathloss value is searched. Any value higher than lowest pathloss + maximum delta pathloss is not included in the calcula- tion;

• Max. number of cells: maximum number of cells from which the pathloss values are considered in CPLM calculation at any one bin.

Another useful module of TCP is Automatic Cell Planning (ACP) which

optimizes the network performance in a limited area by modifying the

parameters of the sector antennas. ACP runs under some user-defined

performance targets on coverage, quality and capacity, as well as some

user-defined configuration constraints on the sector antennas, e.g.,

constraints on power, electrical tilt, mechanical tilt and azimuth.

(23)

2.3.2 Elin

Elin is a program package which can be integrated with the TCP instal- lation. Elin is used to export system data, site data and calculated prop- agation data (see chapter 2.3.1) from TCP into some wireless network simulators (e.g., LTE Astrid which will be introduced in the following chapter) [9]. The exported data is stored in a so called proj mat-file (Matlab data file) created by Elin.

The pathloss values are stored in a g-matrix in proj. The pathloss values are sorted to have the best values of each bin in the first column of the g- matrix, second best in the second column and so on. Another matrix, namely cellno-matrix, with the same size of the g-matrix contains the cell numbers corresponding to the pathloss values in the g-matrix [9].

These two matrixes together can be used by the wireless network simu- lators to calculate received signal strength, Signal to Interference plus Noise Ratio (SINR) and so on, and to do cell selection for each bin dur- ing the simulation.

2.3.3 LTE Astrid

LTE Astrid is a static LTE network simulator in Matlab which conducts Monte Carlo method [13] in the simulation. The realistic network data exported from TCP via Elin can be analyzed with respect to capacity and coverage by LTE Astrid.

The entire simulation in LTE Astrid comprises of several Monte Carlo runs. Each run is a network snapshot with randomized distributed users in the area [14]. The number and locations of the users follows the pre- defined parameters, i.e., traffic distributions and utilization.

LTE Astrid simulates the network using RSRP based cell selection. With LPNs deployed, the simulator should be modified to apply Biased RSRP cell selection with an offset added to the RSRP from LPNs as discussed in chapter 2.1.2.

The two packet traffic models, full and equal buffer models, addressed in chapter 2.2 are both implemented in LTE Astrid. Achievable bitrate (ABR) is the main output from LTE Astrid Monte Carlo simulation which is the way to calculate system throughput (STP) for full and equal buffer model:

• Full buffer model estimate: STP

_FB

= mean ( ABR ) × u

(24)

• Equal buffer model estimate: ^STP

EB

= 1 / ^mean ( 1 / ^ABR ) × ^u

• STP

_FB

≥ STP

_EB

Here, u means utilization which defines the probability that a cell has a user to schedule. In LTE Astrid, STP is actually calculated in a similar but more advanced approach.

One problem in wireless network simulation is “border effects”, i.e., the

number of cells in an area might be very large which makes the simula-

tion of all the cells infeasible, but simulating a part of them would

underestimate interference for the users in the border cells. One method

to avoid this problem is called “wrap-around” which models the system

as homogeneous and connects the edges of the simulation area in a

torus fashion. But this method is not realistic in HetNets for analyzing

real networks. Another solution is to analyze a small area while simulate

a relatively larger area. The smaller area is denoted active area while the

larger one is denoted supporting area. The cells in the active area are

active cells while surrounding cells are supporting cells [14]. These two

areas should be defined in TCP by polygons in advance and network

data in the supporting area should be exported via Elin.

(25)

3 Models

3.1 Studied Network

As mentioned before, the study will be taken based on a real radio network in a limited urban area in a dense European city as shown in Figure 3.1.

[m]

5.28 5.285 5.29 5.295 5.3 5.305

x 10⁵ 1.802

1.804 1.806 1.808 1.81 1.812 1.814 1.816 1.818 1.82 1.822

x 10⁵

Figure 3.1: Analyzed area

The studied network models are built based on existing WCDMA 2.1 GHz site grid:

• Macro cells: 228 cells in total (83 sites)

• Macro inter site distance(ISD): 250-350 m

• 60% of study area is indoor area

• Macro sites average antenna height ~33 m

(26)

• Macro sites average antenna electrical tilt: ~5.15 deg

• Macro sites average antenna mechanical tilt: ~1.46 deg Analyzed area:

• Area: ~1.4 km

²

• Macro cells: 24 cells (9 sites) High resolution maps are used:

• 3D building data bases

• 5x5m bin resolution 3.2 Simulation parameters

According to chapter 2.1.1, the choices of LPNs deployment in this project are listed in Table 3.1. Lamp post deployment means the LPNs are deployed several meters away from the buildings. And in the project, they are deployed 3 meters away.

Access Open access

Deployment Outdoor (lamp post deployment, ~3 m)

Type of LPNs RRU

Frequency reuse Reuse macro spectrum

Table 3.1: LPNs deployment

All the results demonstrated in this report, unless otherwise indicated,

are obtained based on simulation parameters in Table 3.2.

(27)

Parameter Value

Network

Carrier frequency 2.6 GHz

Bandwidth 20 MHz

Cell selection Biased RSRP cell selection with 4dB offset (Chapter 2.1.2)

Traffic distribution

80% of the traffic is generated from indoor area (all buildings in the area), the rest 20% is distributed

outdoor

Utilization 1, 5, 10, 20, 30, 50, 70 , and 95 % Packet traffic model Equal buffer model

Macro Sites

Output power 60 W

Average antenna height ~33 m Average antenna gain ~16 dBi

Average antenna tilt ~5.15

^o

(Electrical tilt)

~1.46

^o

(Mechanical tilt)

Tx/Rx 2 Tx/2 Rx

Diversity combining Maximal Ratio Combining (MRC) in uplink

Pico Sites

Output power 5 W

Antenna height 5 m

Antenna gain 12 dBi

Antenna half power beam width (HPBW)

63

^o

(Horizontal) 28

^o

(Vertical)

Tx/Rx 2 Tx/2 Rx directional antenna Diversity combining MRC in uplink

UE

Max output power 21 dBm

Min output power -40 dBm

Antenna height 1.5 m

Antenna gain -1 dBi

Tx/Rx 1 Tx/2 Rx omni antenna

Body loss 3 dB

Diversity combining Interference Rejection Combining

(IRC) in downlink

Table 3.2: Simulation parameters

(28)

3.3 Propagation models 3.3.1 Urban model

As addressed in Chapter 2.3.1, there are several propagation models which are implemented in TCP. Since this study will be taken in an urban environment, the urban model is the one which will be mainly used for pathloss analysis.

In an urban environment, radio wave propagation has two dominant paths which are over the rooftops and along the street. The first path dominants when the UE is far from the site, while the second path dominants when the UE is near to the site [15].

The urban model is valid under the following conditions [10][15]:

• Frequency from 450 MHz up to 2200 MHz;

• Receiving antenna at distance to the base station antenna from 0 m up to (at least) 50 km;

• Base station antenna heights between 5 m and 60 m and antennas placed below as well as above rooftops;

• Large receiving antenna height from 1.5 m up to 5 m.

The urban model consists of three wave propagation algorithms [15]:

• Half-screen model: calculates propagation above the rooftop and generates pathloss L

^above

;

• Recursive micro cell model: calculates propagation between buildings, i.e. along the street, and generates pathloss L

^below

;

• Building penetration model: calculates propagation from an out- door base station antenna to an indoor UE and generates pathloss L

^inside

;

The urban model pathloss is expressed as:

) , min(

_above _below

urban

L L

L =

(29)

As mentioned before, the radio waves propagation has two dominant paths and the received signal strength is the sum of both. In most situa- tions, one of the paths will dominate, so L

^urban

which takes the minimum is justified [15].

The building penetration model pathloss is determined by [15]:

s

α

w outside

inside

L L d

L = + +

in which

L

^outside

is the pathloss from the base station antenna to a point just outside the external wall;

L

^w

is the penetration loss through the external wall;

d

^s

is the distance inside building [m];

α is the building penetration slope [dB/m].

In this project, L

^w

=12 dB and α = 0.8 dB/m.

3.3.2 ITU indoor propagation model

ITU (International Telecommunication Union) indoor propagation model estimates the pathloss of radio propagation in the indoor envi- ronments. This model is applicable to frequency from 0.9 up to 5.2 GHz and to buildings with 1 to 3 floors [8].

According to ITU indoor propagation model, the indoor propagation pathloss is [8]:

28 ) ( log

log

20 + + −

= f N d P n

L

_indoor _f

,

in which,

f is the transmission frequency [MHz];

d is the transmission distance [m];

N is the distance power loss coefficient;

P

^f

(n) is the floor loss penetration factor;

n is the number of floors via transmission.

(30)

In this project, ITU indoor propagation model is applied in pico sites placement algorithm which will be addressed in chapter 4.2. It is just used for rough estimation of the pathloss from the newly added pico site. After pico cell placement is done, the pathloss will be calculated in TCP.

Since only rough estimation is needed, the floor loss penetration factor P

^f

(n) is ignored here. And the distance power loss coefficient N is chosen to be 28.

Taking into account of the exterior wall loss L

^wall

=12 since the pico sites will be placed outdoor, the total pathloss estimation can be expressed as:

16 log 28 log

20 + −

= +

= L L f d

L

_indoor _wall

(31)

4 Implementation

4.1 Work flow

TCP Project (WCDMA)

LTE Astrid (Matlab)

Performance Evaluation

Elin Project

Simulation Results Elin

Project TCP

ACP

New Cells

Pico Sites densification

(Matlab)

4 pico cells/macro cell 8 pico cells/macro cell 12 pico cells/macro cell

. . . Desity

Figure 4.1: Work flow

The whole project process is organized as Figure 4.1:

1. Export Elin project of the original network with macro cells only from TCP;

2. Based on the Elin project, densify the network with new pico sites in Matlab using the pico sites densification algorithm which will be discussed in chapter 4.2. Network densification should be done with several pico cell densities;

3. Import the new pico sites to the original TCP project, run ACP to optimize the network and export Elin projects of the networks af- ter densification;

4. Run LTE Astrid for the networks with different pico cell densities;

5. Evaluate simulation results.

(32)

4.2 Pico sites densification algorithm

As mentioned in chapter 1.4, how should the pico sites be deployed to achieve better network performance is of the interest of this project as well, a pico sites densification algorithm based on pathloss is designed.

Briefly speaking, the core of the algorithm is as following:

1. Find the indoor bins with worst pathloss;

2. Find the closest outside-wall bin;

3. Find an outdoor bin 3 meters away from this outside-wall bin (lamp post deployment);

4. Place the pico site in this outdoor bin and point the sector anten- na to the worst pathloss indoor bin.

The details about how the algorithm works are illustrated in Figure 4.2:

1. Find the worst pathloss indoor bin based on the first column of the g-matrix from the Elin project file exported from TCP;

2. Find the closest outside-wall bin to this indoor bin, then find an outdoor bin 3 meters away from this outside-wall bin;

3. Check if this outdoor bin fulfills the distance constrain predefined, minpico2pico: if not, redo the previous steps for the next worst pathloss indoor bin; if yes, continue;

4. Place a pico site in this outdoor bin and point the antenna to the worst pathloss indoor bin currently considered;

5. Calculate pathloss vector for the indoor bins within 100 meters to the new pico site;

6. Compare the new pathloss vector with the first column of previ- ous g-matrix and modify the pathloss values of the indoor bins covered by the new pico site;

7. Check if the picodense target is fulfilled: if not, redo the previous

steps to find the next pico site position; if yes, stop searching.

(33)

Start

Find the worst pathloss indoor bin

Find the closest outside-wall bin

Find the next worst pathloss

indoor bin

pico2pico

>minpico2pico

₁

no Place a pico in this

outdoor bin

yes Point the sector antenna to the worst

pathloss indoor bin

Calculate pathloss for the indoor bins within 100 meters to the new

pico site

₂

Modify pathloss values of the indoor

bins covered by the pico site

End

1. minpico2pico: minimum distance between pico sites, a constraint to reduce interference between pico sites.

2. Pathloss calculated based on ITU indoor propagation model.

3. picodense: pico cell density target, e.g., 4 pico cells/macro cell.

picodense

₃

fulfilled?

no

yes Find an outdoor

bins 3 meters away from this outside-wall bin

Figure 4.2: Flow chart of pathloss based pico sites densification algorithm

In this project, the parameter minpico2pico has been chosen to be 30 meters. This might not be a perfect choice. And the parameter picodense has been chosen to be 4, 8, 12, 16 and 20 in order to build network mod- els with different pico cell densities; an extreme case with picodense being 40 has also been studied.

In the following chapter, the network performance will be compared

between the network where pico sites are placed using this pathloss

(34)

based algorithm and the network where pico sites are placed randomly.

In the randomly distributed algorithm, the pico sites are also deployed

outdoor (lamp post deployment), but the placement positions are ran-

domly selected without the constraint of the parameter minpico2pico.

(35)

5 Results

In this chapter, the results of pico sites densification will be shown followed by the simulation results for the network models with different pico cell densities.

5.1 Network densification

Figure 5.1 and Figure 5.2 illustrates the densified networks: the results of the pathloss based pico sites densification algorithm are shown in Figure 5.1 while the results of randomly distributed algorithm are presented in Figure 5.2. Compare these two figures, it can be seen that some of the pico sites in Figure 5.2 are very close to each other since there is no constraint on the minimum distance between pico sites (minpico2pico) in the randomly distributed algorithm.

In Figure 5.1 and Figure 5.2, only network models with pico cell density

up to 20 picos/macro are shown since the sites in the 40 picos/macro case

are too dense and won’t be very clear to be shown on the map here.

(36)

Macro only 4 picos/macro cell

8 picos/macro cell 12 picos/macro cell

16 picos/macro cell 20 picos/macro cell

Figure 5.1: Densified networks using pathloss based pico sites densification

algorithm (chapter 4.2)

(37)

Macro only 4 picos/macro cell

8 picos/macro cell 12 picos/macro cell

16 picos/macro cell 20 picos/macro cell

Figure 5.2: Densified networks using randomly distributed pico sites densification

algorithm (chapter 4.2)

(38)

Figure 5.3 shows the average cell density (per km

²

) and average ISD of the networks where pico sites are placed using the pathloss based algo- rithm, while Figure 5.4 shows the average cell density and ISD of the networks where pico sites are placed randomly. Here one extreme case with pico density to be 40 picos/macro is also illustrated. As shown in these two figures, the average cell density presents no large difference;

however the ISD of the pathloss based algorithm is longer since there is a constraint on the minimum distance between pico sites (minpico2pico) in this algorithm.

0 4 8 12 16 20 40

0 100 200 300 400 500 600 700

18 74

129 191

251 303

579 Average cell number per km²

Pico cell density (picos/macro) Average cell number [/km2]

0 4 8 12 16 20 40

0 50 100 150 200 250 300 350

305

110 86

70 60 55

38 Inter site distance [m]

Pico cell density (picos/macro)

ISD [m]

Figure 5.3: Pathloss based pico sites densification algorithm

0 4 8 12 16 20 40

0 100 200 300 400 500 600 700

18 70

126 177

258 299

616 Average cell number per km²

Pico cell density (picos/macro) Average cell number [/km2]

0 4 8 12 16 20 40

0 50 100 150 200 250 300 350

305

93

65 53 48 40

27 Inter site distance [m]

ISD [m]

Figure 5.4: Randomly distributed pico sites densification algorithm

(39)

5.2 Comparison of Pico sites densification algorithms

In order to choose a better pico sites densification algorithm for the main parts of this project, in this chapter the network performance is com- pared between the pathloss based algorithm and the random one.

Figure 5.5 and Figure 5.6 show the user performance (y-axis) versus capacity (x-axis) for both uplink and downlink. Two types of user per- formance measures are used: mean and 10% worst user throughput. In this project, the cell edge user throughput is defined as the 10th percen- tile of the user throughput in the analyzed area.

0 50 100 150 200 250

0 5 10 15 20 25 30

Uplink indoor

Subscriber capacity [GB/month/subscriber]

Mean user throughput [Mbps]

4 picos/macro(Pathloss based) 4 picos/macro(Random) 12 picos/macro(Pathloss based) 12 picos/macro(Random) 20 picos/macro(Pathloss based) 20 picos/macro(Random)

0 50 100 150 200 250

0 5 10 15 20 25 30 35 40 45 50

Uplink outdoor

0 50 100 150 200 250

0 0.5 1 1.5 2 2.5 3 3.5

Uplink indoor

Throughput 10% worst uers [Mbps]

0 50 100 150 200 250

0 5 10 15 20 25 30 35 40

Uplink outdoor

Figure 5.5: Uplink comparison

As illustrated in Figure 5.5, the two pico sites densification algorithms

result in almost the same uplink performance for lower pico cell densi-

ties, e.g., 4 picos/macro. That is to say, for low pico cell densities the

uplink performance is mainly related to the number of cells in the net-

work rather than the position of cells. However for higher pico cell

densities, e.g. 20 picos/macro, the pathloss based algorithm performs

better than the random one according to Figure 5.5. As a conclusion, for

the uplink, the advantage of the pathloss based algorithm compared

(40)

with the random one becomes clearer for higher pico cell densities. The difference between the algorithms increases with pico cell density.

0 50 100 150 200 250 300

0 10 20 30 40 50 60 70 80

Downlink indoor

0 50 100 150 200 250 300

0 20 40 60 80 100 120

Downlink outdoor

0 50 100 150 200 250 300

0 5 10 15 20 25 30 35

Downlink indoor

0 50 100 150 200 250 300

0 10 20 30 40 50 60 70 80 90 100

Downlink outdoor

Figure 5.6: Downlink comparison

Figure 5.6 illustrates the downlink performance comparison. It is obvi-

ous that the pathloss based algorithm results in better performance than

the random one on the downlink.

(41)

5.3 Comparison of RSRP and Biased RSRP

According to the results in chapter 5.2, the pathloss based pico sites densification algorithm will be chosen for the following simulation and discussion.

As shown in Figure 5.7 and Figure 5.8, biased RSRP cell selection with 4dB offset presents a better improvement than RSRP based cell selection on both uplink and downlink performance. But according to some previous studies, biased RSRP cell selection with higher offset will degrade the downlink performance.

0 50 100 150 200 250

0 0.5 1 1.5 2 2.5 3 3.5

Uplink indoor

4 picos/macro(RSRP+4dB) 4 picos/macro(RSRP) 12 picos/macro(RSRP+4dB) 12 picos/macro(RSRP) 20 picos/macro(RSRP+4dB) 20 picos/macro(RSRP)

0 50 100 150 200 250

0 5 10 15 20 25 30 35 40

Uplink outdoor

0 50 100 150 200 250

0 5 10 15 20 25 30

Uplink indoor

0 50 100 150 200 250

0 5 10 15 20 25 30 35 40 45 50

Uplink outdoor

Figure 5.7: Uplink comparison

(42)

0 50 100 150 200 250 300 0

5 10 15 20 25 30 35

Downlink indoor

0 50 100 150 200 250 300

0 10 20 30 40 50 60 70 80

Downlink indoor

0 50 100 150 200 250 300

0 20 40 60 80 100 120

Downlink outdoor

0 50 100 150 200 250 300

0 10 20 30 40 50 60 70 80 90 100

Downlink outdoor

Figure 5.8: Downlink comparison

5.4 Network Performance

Based on the comparisons in chapter 5.2 and 5.3, the following simula- tion results are obtained from the network models where pico sites are placed using the pathloss based algorithm, and biased RSRP cell selec- tion with 4dB offset has been applied during the simulation.

In this chapter, the uplink and downlink performance for the indoor and

outdoor users will be analyzed separately.

(43)

5.4.1 Capacity and Coverage

As illustrated in Figure 5.9 and Figure 5.10, for the uplink, with the same target of mean user throughput or the same target of cell edge user throughput, the subscriber capacity keeps increasing with pico cell densification.

0 50 100 150 200 250

0 5 10 15 20 25 30

Uplink indoor (RSRP+4dB offset)

Macro only 4 picos/macro 8 picos/macro 12 picos/macro 16 picos/macro 20 picos/macro

0 50 100 150 200 250

0 5 10 15 20 25 30 35 40 45 50

Uplink outdoor (RSRP+4dB offset)

0 50 100 150 200 250

0 5 10 15 20 25 30 35

Uplink all area (RSRP+4dB offset)

Figure 5.9: Uplink mean user throughput vs uplink subscriber capacity

(44)

0 50 100 150 200 250 0

0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

Uplink all area (RSRP+4dB offset)

0 50 100 150 200 250

0 5 10 15 20 25 30 35 40

Uplink outdoor (RSRP+4dB offset)

0 50 100 150 200 250

0 0.5 1 1.5 2 2.5 3 3.5

Uplink indoor (RSRP+4dB offset)

Figure 5.10: Uplink cell edge user throughput vs uplink subscriber capacity

Figure 5.11 illustrates one example of the improvement of uplink capac-

ity: with the same target of 1 Mbps uplink cell edge user throughput,

uplink capacity improves with pico cell densification. For the network

with pico cell density less than 4 picos/macro, the cell edge throughput

can never reach 1 Mbps, thus the capacity is shown to be 0.

(45)

0 4 8 12 16 20 0

5 10 15 20 25 30 35 40

Uplink cell edge throughput=1Mbps, RSRP+4dB offset

Uplink subscriber capacity [GB/month/subscriber]

Pico Cell Densification Study in LTE Heterogeneous Networks