Juni 2018
Propagation Modeling and LTE
Network Performance in Real City
Scenarios
Teknisk- naturvetenskaplig fakultet UTH-enheten Besöksadress: Ångströmlaboratoriet Lägerhyddsvägen 1 Hus 4, Plan 0 Postadress: Box 536 751 21 Uppsala Telefon: 018 – 471 30 03 Telefax: 018 – 471 30 00 Hemsida: http://www.teknat.uu.se/student
Abstract
Propagation Modeling and LTE Network Performance
in Real City Scenarios
Sebastian Vestberg
Maps of chosen areas in Chicago, San José, London and Shibuya, are imported from Open Street Map into matlab in order to run LTE network simulations for various scenarios. Firstly, two path loss models are compared, the empirically based WINNER model and a set of site-specific model. Secondly, low load network simulations are run separately at two different carrier frequencies, 700MHz and 2GHz, for city specific base station deployments. Simulation results show that user performance is quite unique for each city and that deployment strategies and city environments are strongly influencing path gain, SINR and throughput. In general, user performance in UL is
significantly worse at 2GHz than at 700MHz, whereas DL performance is not as affected by the change in carrier frequency.
Populärvetenskaplig sammanfattning
I detta projekt har kartdata från Open Street Map använts i en matlab-baserad nätverkssimulator med syfte att analysera och jämföra effektförluster och täckn-ing både inomhus och utomhus mellan städer från olika kontinenter och med olika
stadsstruktur. Följande stadsdelar importerades och simulerades; 1000x1000m2 av
Chicago, 4000x4000m2av San José, 600x600m2av London samt 400x400m2av Shibuya. Eftersom effektförluster för radiovågor är större för högre frekvenser så under-söktes hur två vanliga LTE-frekvenser 700MHZ och 2GHz skiljer sig vid samma basstationsplacering i de utvalda städerna. Dessutom undersöktes hur två olika propageringsmodeller, en statistisk och en deterministisk, skiljer sig vad gäller beräk-nandet av effektförluster.
Den deterministiska modellen som är utvecklad hos Ericsson kräver mycket ge-ometrisk information och de importerade byggnaderna, som representeras av poly-goner, måste följa simulatorns riktlinjer till punkt och pricka. Denna modell är valid-erad mot mätdata och ger väldigt bra indikation på hur stor signaldämpningen är mellan basstation och användare i verkligheten. Den statistiska modellen är snab-bare men ger inte ett lika pålitligt resultat.
ii
Acknowledgements
Firstly I would like to send a million thanks to Gunther Auer for being the best supervisor a student could ever wish for. Your knowledge, patience and positive attitude have been extremely valuable for me throughout this thesis.
I would also like to express my sincere gratitude to Dirk Gerstenberger. Your kindness and humbleness are striking and I am very proud to have been a part of your brilliant team at Ericsson.
Contents
Acknowledgements ii
1 Introduction 1
1.1 Historical Background of Mobile Communications . . . 1
1.2 Objective . . . 2
2 Theory 3 2.1 Propagation Models . . . 3
2.1.1 Links and Paths . . . 3
2.1.2 Path Loss and Path Gain . . . 3
2.1.3 Free-space Path Loss . . . 4
2.1.4 Deterministic Models . . . 5
Diffraction . . . 5
2.1.5 Statistical Models . . . 5
2.1.6 WINNER Path Gain Models. . . 5
2.1.7 A Set of Site-Specific Path Gain Models . . . 7
Model for propagation above terrain and buildings . . . 7
Model for propagation around buildings . . . 8
Foliage model . . . 8
Outdoor-to-indoor model . . . 9
Additional stochastic model. . . 9
2.2 SNR, SINR and Channel Capacity . . . 9
2.2.1 SNR . . . 9
2.2.2 SINR . . . 9
2.2.3 Channel Capacity . . . 9
2.3 Long Term Evolution (LTE) . . . 10
List of Figures
2.1 Illustration of Half Screens . . . 8
2.2 OFDM . . . 10
2.3 TDD and FDD . . . 11
3.1 Deployment in Chicago . . . 15
3.2 Deployment in San José . . . 16
3.3 Deployment in London. . . 16
3.4 Deployment in Shibuya . . . 17
4.1 3D view of PG in Chicago . . . 20
4.2 2D view of PG in Chicago . . . 21
4.3 CDF of PG in Chicago . . . 21
4.4 2D view of PG in San José . . . 23
4.5 CDF of PG in San José . . . 23 4.6 3D view of PG in London . . . 25 4.7 2D view of PG in London . . . 25 4.8 CDF of PG in London. . . 26 4.9 3D view of PG in Shibuya . . . 27 4.10 2D view of PG in Shibuya . . . 28 4.11 CDF of PG in Shibuya . . . 28
4.12 UL and DL performance for Chicago in 3D . . . 30
4.13 UL and DL performance for Chicago in 2D . . . 31
4.14 CDF of SINR and Throughput in Chicago . . . 31
4.15 UL and DL performance for San José in 2D . . . 32
4.16 CDF of SINR and Throughput in San José . . . 33
4.17 UL and DL performance for London in 3D . . . 34
4.18 UL and DL performance for London in 2D . . . 35
4.19 CDF of SINR and Throughput in London . . . 35
4.20 UL and DL performance for Shibuya in 3D . . . 37
4.21 UL and DL performance for Shibuya in 2D . . . 37
vi
List of Tables
2.1 Path loss models for C2- and D1 WINNER propagation scenarios . . . 6
List of Abbreviations
AGL above ground level
bps bits per second
BS base station
BW bandwidth
CDF cumulative distribution function
DL downlink
FDD frequency division duplex
FFT fast fourier transform
FSPL free space path loss
GP guard period
IFFT inverse fast fourier transform
ISD intersite distance
ISI intersymbol interference
ITU international telecommunications union
LOS line of sight
LTE long term evolution
NLOS non line of sight
OFDM orthogonal frequency division multiplexing
OFDMA orthogonal frequency division multiple access
OSM open street map
PG path gain
PL path loss
SINR signal to interference and noise ratio
SNR signal to noise ratio
TDD time division duplex
UE user equipment
1
Chapter 1
Introduction
Today billions of smartphones are connected through our complex mobile communi-cation networks across the globe. For the new generation of users it’s almost difficult to imagine a life without these devices providing us with fast internet access. Many could surely think of situations and locations when mobile broadband coverage is extra poor. Rural areas with base stations deployed far from each other sometimes provide low data rates, but also dense venues such as crowded sport arenas cause large problems in our existing networks. By combining available systems, such as 2G, 3G and LTE-advanced, and optimally deploy broadband antennas for places with certain local traffic demands could help overcome these problems. Currently the world is preparing for the upcoming 5G release including many new features likely to enhance user experience greatly, but since older releases will remain in our networks for a long time it’s also important to continue analyzing possible improve-ments for these systems. This chapter will give the reader a brief background of mobile communications followed by a description of the thesis objectives.
1.1
Historical Background of Mobile Communications
Ever since the 1G cellular telephones were standardized in the 1980s, new techniques and features have constantly been added to earlier releases to improve and reshape wireless networks to what we currently have. Earlier mobile communication stan-dards rarely occupy frequency bands above 1GHz and when the first European GSM 2G networks were released in the early 1990s, peak data rates hardly reached over a couple of hundreds of kbps. Enhancements of 2G, such as GPRS and EDGE, in-creased data rates up to almost 400 kbps by improving network architecture as well as modulation and coding techniques.
In the early 2000s the international telecommunications union (ITU) released their standard, IMT-2000, including technical specifications for the 3G standards. 3G systems have peak data rates of a few Mbps, and mainly two incompatible standards are dominating the market, cdma2000 and W-CDMA, both using CDMA-techniques.
3GPP is an organization working towards backward and forward compatibility between different standard releases worldwide. With the 3GPP long term evolution (LTE) release 8 in December 2008, many of the 4G requirements were fulfilled. LTE-Advanced, however, is the system closest to the "true 4G". For the LTE system the 3GPP requires, for instance, peak data rates of 100Mbps, increased spectral efficiency
and OFDM modulation. LTE is described in more detail in section 2.3, and is the
1.2
Objective
The main objective of this thesis work is to investigate LTE network performance in real cities with realistic building data imported from Open Street Map (OSM). Two different path gain (PG) models, the statistical WINNER model and a set of site-specific models, are estimating signal strengths in four city areas from various continents; Chicago, San José, London and Shibuya. The, by the site-specific model, calculated PG is further used as input in the throughput simulations. Two common carrier frequencies for LTE systems, 700MHz and 2GHz, are simulated separately for each city to capture understanding in how local city environments and base station (BS) deployments are affecting network performance differently.
The data from OSM is copyrighted c OpenStreetMap contributors and is
avail-able under the Open Database License. For more information, see https://www.
3
Chapter 2
Theory
In order to prepare the reader for the network analysis part a few fundamental con-cepts will be introduced in this chapter. The choice of propagation model is one of the core aspects in system simulations so theory concerning channel prediction and path loss modeling will be covered in the first section. Also, this section includes a detailed description of the two propagation models used in this thesis work. The second section in this chapter is aiming at the throughput analysis of the simulation scenarios, and therefore includes important concepts in wireless communication the-ory along with an overview of the LTE systems.
2.1
Propagation Models
Maxwell formulated his famous electromagnetic equations, in detail describing ra-dio wave propagation, already in the late 19th century. Solving these equations ana-lytically is often not possible, and approximating propagation models must in such cases be implemented instead. Propagation models mainly estimate signal strength at the receiver antennas in environments that can be represented either stochastically or deterministically. Many models are also combining stochastic models with deter-ministic models in order to increase accuracy, such as the set of site-specific models used in this thesis (2.1.7).
2.1.1 Links and Paths
In wireless communications, a link is the channel connecting two nodes within a network. A link is most often between a user equipment (UE), such as a mobile phone or tablet, and its serving base station (BS), but it could also include a relay that simply acts as a mid-link between UE and BS that only receives and retransmits the radio signal towards the desired destination. A path between a UE and BS could therefore consist of multiple links depending on the number of relays. However, relays are not included in this work.
A signal sent from a UE to a BS has an uplink (UL) direction and a signal trans-mitted from a BS to a UE has a downlink (DL) direction.
2.1.2 Path Loss and Path Gain
Assume that the transmitted signal s(t)and the received signal r(t)are modeled as
s(t) = <{u(t)ej2π fct} (2.1)
where u(t)is a baseband signal known as the complex envelope of s(t), fcis the carrier
frequency and n(t)is the noise. The real part of u(t)is called in-phase component, and the imaginary part is called quadrature component. In the received signal, v(t) is channel dependent, and in the most simplistic channel models only a complex rescaling of u(t).
If s(t)has power Ptand r(t)has power Pr, then path loss (PL) and path gain (PG)
in dB are defined as
PL=10 log10 Pt Pr
(2.3)
PG = −PL. (2.4)
Thus, received signal power Prcan be calculated according to the following formula
when Ptand PL are known:
Pr = Pt
10PL/10. (2.5)
Signal power variation on the receiving side is in this work calculated in PG, although channel model descriptions perhaps more often refer to PL. Note also that the noise term is not included in PL.
If no obstacles are nearby a signal that is propagating through free space, then the signal is attenuated according to the free-space propagation law. In our real-world networks, however, all types of objects (houses, trees, hills etc.) attenuate the transmitted signal on its way towards the receiver. Diffractions, reflections, wall penetrations, shadowing effects etc. must therefore be taken into account in the PL models.
The simplest of all propagation models is the Free-Space Path Loss (FSPL), ex-plained in next subsection , which is assuming a line-of-sight (LOS) channel between the transmitter and receiver antennas.
2.1.3 Free-space Path Loss
If the signal model in previous subsection is used, the following formula is describ-ing v(t)when the transmitted signal is propagating through free space over a dis-tance d without any obstructions:
v(t) = λ √
Gle−j2πd/λ
4πd u(t), (2.6)
where λ=c/ fcis the wavelength and Gl is an antenna gain factor depending on
the radiation patterns. This channel model introduces only a phase- or amplitude shift on the modulated baseband signal u(t)depending on λ and d and the resulting power ratio between Ptand Prbecomes:
Pr Pt = λ √ Gl 4πd 2 . (2.7)
Chapter 2. Theory 5 FSPL=20 log 4πd λ √ Gl (2.8) where Gl =1 for isotropic antennas.
2.1.4 Deterministic Models
Deterministic propagation models are site-specific and therefore require geometrical information of, for instance, buildings, terrain and foliage. Ray-tracing methods are essentially used for urban areas where site-specific propagation modeling is needed for performance evaluation of local areas with high densification [1]. One of the most fundamental parts of site-specific modeling is knife-edge diffraction, which is approximating the loss in signal strength due to diffraction around buildings.
Diffraction
Diffraction occurs when a signal propagates over an edge, which is causing PL that is increasing with the angle of the bent signal. This is often approximated in determin-istic propagation models with the Fresnel knife-edge diffraction model. If the diffracted signal travels an additional distance∆d relative the LOS-path component, then the Fresnel-Kirchhoff diffraction parameter v is described as
v=2
q
(∆d/λ). (2.9)
The PL, in this case called Fresnel-loss, can be calculated as a function of v [4].
2.1.5 Statistical Models
Although deterministic site-specific models are more accurate, they are often more time consuming and computationally heavy compared to stochastic propagation models. The statistical models are based on empirical measurements to fit param-eters for certain scenarios and are often limited to specific environments, path dis-tances, frequencies etc. For instance, shadowing effects are modeled statistically using log-normal distributions in both the empirically based WINNER model and the set of site-specific models used in this thesis.
2.1.6 WINNER Path Gain Models
The WINNER, in this case WINNER II, channel model is a geometry based stochastic propagation model that can enable both link and system level simulations [2]. All channel parameters, such as delay spread, angle-of-departure, angle-of-arrival and shadow fading (all large scale parameters), are stochastic and based on parameter distributions. Channel models are defined for several different propagation sce-narios, each containing specific parameter distributions derived from real channel measurements.
TABLE 2.1: Path loss models for C2- and D1 WINNER propagation scenarios
Path loss models for WINNER C2 and D1 scenarios
Scenario d LOS
or NLOS
Path Loss
C2 d> dBPC2 LOS 40 log d + 13.47 − 14 log hBS − 14 log hUE +
6 log fc
5
C2 d< dBPC2 LOS 26 log d+39+20 log f5c
C2 all d NLOS (44.9 − 6.55 log hBS)log d + 34.46 +
5.83 log hBS+23 log f5c
D1 d> dBPD1 LOS 40 log d + 10.5− 18.5 log hBS − 18.5 log hUE +
1.5 log fc
5
D1 d< dBPD1 LOS 21.5 log d+44.2+20 log f5c
D1 all d NLOS 25.1 log d + 55.4 − 0.13(hBS − 25)log100d −
0.9(hUE−1.5) +21.3 log f5c
stations are assumed to be placed just above average building height although high-rise buildings in the area can exceed this height. For the urban macro-cell scenario (C2) the base stations are assumed to be deployed above the roof tops in a city area with homogeneous building heights.
In this work, the propagation scenario used for the cities Chicago, London and Shibuya is the urban macro (C2) scenario for WINNER PL calculations. For PL cal-culations in the chosen residential part of San José, rural macro (D1) scenario is used. Although each scenario has its unique formula for PL calculation, all models are of the following form:
PL= A log d+B+C log fc
5 +X, (2.10)
where d is the distance in meter between the transmitter and receiver, and fc is
the carrier frequency in GHz.
The variables A, B, C and X are scenario specific. For instance, FSPL is modeled with A=20, B=46.4, C=20 and X=0. The term X is in many cases set to zero. Table2.1 summarizes the PL models for the two scenarios, C2- and D1, used in this thesis. For each scenario, LOS and NLOS links are modeled differently. Additionaly, a scenario specific breakpoint distance dBP is determining the choice of PL model for a certain
link. Clearly, base station antenna height hBSand user equipment height hUEare two
main variables in the PL models.
The two breakpoint distances dBPC2and dBPD1are calculated from:
dBPC2=4(hBS−1)(hUE−1)fc/c (2.11)
dBPD1 =4hBShUEfc/c, (2.12)
where c=3×108is the free-space propagation velocity in [m/s].
Chapter 2. Theory 7 in Table2.1 as well as shadowing losses generated from log-normal distributions. These losses are together with the antenna gain determining the total link PG that is further used in the network simulations.
2.1.7 A Set of Site-Specific Path Gain Models
A set of site-specific path gain models that are much more computationally expen-sive than WINNER is also used in this work [1]. The total PG for each link is a sum of all components from the following models:
• Model for propagation above terrain and buildings • Model for propagation around buildings
• Foliage model
• Outdoor-to-indoor model • Additional stochastic model.
Model for propagation above terrain and buildings
In this set of site-specific models, the general PL model for propagation above terrain and buildings is:
PL[dB] =20 log(4πd/λ) +D++D−, (2.13)
where d is the distance between transmitter and receiver and both D+ and D− are knife-edge diffraction losses based on half-screen representations in the verti-cal plane of the building and terrain data. The D+ term is a sum of Fresnel-losses F+(v+i )for all half screens between the transmitting and receiving antennas. For the ith half screen, the Fresnel-loss Fi+is calculated using the Fresnel-Kirchhoff diffrac-tion parameter:
vi+=2 q
(s+i /λ), (2.14)
where s+i is the string distance that is defined as the difference between the sum of left and right convex hulls and the convex hull between transmitter and receiver.
The D−term is calculated as:
D− =
3
∑
i=1
F−(v−i ), (2.15)
where each loss term Fi−(v−i )comes from the dominating screen of all screens that creates NLOS between transmitter and receiver. The dominating screen has the largest difference between the geometric distance and the distance transmitter-screen-receiver. The diffraction parameter v−i is defined as:
vi−=2 q
FIGURE 2.1: Illustration of the half screen model including three screens for plus diffraction and the convex hull between transmitter
(left) and receiver (right).
where s−1 is the difference between the geometric distance and the transmitter-screen-receiver distance, s−2 is the difference between geometric distance and the transmitter-screen distance, and s−3 is the difference between geometric distance and screen-receiver distance. All distances in the D−term are calculated as if the dom-inating screen is isolated. Fig.2.1illustrates a path that includes three half screens for plus diffraction and the convex hull between transmitter and receiver.
For a link, various types of paths are calculated; direct paths, backscatter paths and specular reflection paths are all calculated separately using the half screen ap-proach.
Adding antenna gain from both transmitting and receiving antennas with respect to path angles gives the total "above" PG.
Maximum PG is selected for all calculated paths for a certain link.
Model for propagation around buildings
A 3D model that is combining the half-screen model with a recursive 2D model is used to trace paths and calculate PG around buildings. For cities with high-rise buildings that reach above antenna heights, diffraction around corners into street canyons is a very common phenomenon.
Foliage model
Chapter 2. Theory 9 represented in the OSM maps used in this thesis. Additionally, neither oxygen loss nor rain loss is modeled.
Outdoor-to-indoor model
For indoor users, outer wall losses together with a loss per meter factor are added to the outdoor losses.
Additional stochastic model
Stochastic shadowing and multipath components are also included to generate a more realistic propagation behavior.
2.2
SNR, SINR and Channel Capacity
Three important concepts regarding user throughput are introduced in this section. The signal to noise ratio (SNR) and signal to interference and noise ratio (SINR) are both closely related to the maximum bitrate that the user actually can experience. This upper boundary bitrate (often measured in bits per second (bps)) is referred to as channel capacity.
2.2.1 SNR
Although noise is not included in the PG calculations, it plays a major part in the channel capacity evaluation. If the received signal power (calculated by PG) for a certain link is Prand the noise Power is Pn, then SNR is defined as
SNR= Pr
Pn
. (2.17)
2.2.2 SINR
Since a mobile network often contains many active links, various types of signal interference could decrease, along with noise, the channel capacity. Therefore, SINR is a good indicator of user throughput and is defined as
SI NR= Pr
Pi+Pn
, (2.18)
where Piis the power of the total interference.
2.2.3 Channel Capacity
The channel capacity, is described as the maximum data rates that can be transmit-ted over a channel when the error probability approaches zero. The capacity limits are upper bounds and can therefore not be reached. The evolution of turbo-codes, however, has made it possible for real systems to achieve data rates very close to capacity. In particular, a single-user system with single antennas on both transmitter and receiver sides has a capacity limit that is described by the well known Shannon Capacity:
where B is the bandwidth for an AWGN-channel. Capacity, however, is in most cases not explicitly defined.
2.3
Long Term Evolution (LTE)
This section covers fundamental principles of the system that is simulated in this thesis, Long Term Evolution (LTE). The LTE-framework is described in order to un-derstand the setup outlined in the next chapter, and what capacity limits the LTE-systems can have.
2.3.1 LTE-framework
The core of the LTE DL transmission scheme is the Orthogonal Frequency Division Multiplexing (OFDM), which is a multicarrier modulation technique based on di-viding the total system bandwidth into many narrow-band subchannels that are or-thogonal to each other. The data rates on each subchannel are much lower than the total data rate and by choosing the subchannel bandwidth to be smaller than the coherence bandwidth, each subchannel becomes (somewhat) flat fading which thus mitigates the effect of intersymbol interference (ISI). A drawback, however, with OFDM is that is leads to high peak-to-average power ratio (PAPR) and therefore a single carrier OFDM technique (SC-OFDM) is used in UL.
The multiuser scheduling principle is called orthogonal frequency division
mul-tiple access (OFDMA), since it’s based on OFDM. As shown in Fig. 2.2
OFDM-subcarriers are divided into blocks, known as resource blocks, each of bandwidths 180kHz. Each resource block is scheduled for a specific user within one subframe duration of 1ms. The spacing between the orthogonal subcarriers is 15kHz, resulting in 12 subcarriers per resource block. LTE supports 6 different system bandwidths; 1,4MHz, 3MHz, 5MHz, 10MHz, 15MHz and 20MHz, leading to a maximum num-ber of 100 user resource blocks for a bandwidth of 20MHz in each subframe. Within a subframe, either 12 or 14 OFDM-symbols is transmitted depending on the length of the cyclic prefix (described in2.3.1), which is inserted into every OFDM symbol to eliminate ISI.
Chapter 2. Theory 11
OFDM
OFDM is based on Inverse Fast Fourier Transform (IFFT) on the transmitter side and Fast Fourier Transform (FFT) on the receiver side. The input bit-sequence is modu-lated into a stream of parallel complex symbols of some length N. IFFT is converting each symbol X[n]in frequency domain into a discrete sample x[n] in time domain by x[n] =√1 N N−1
∑
i=0 X[i]ej2πni/N. (2.20)The full sequence x[0], ..., x[N−1]is known as the OFDM symbol. After adding the cyclic prefix to this OFDM symbol, the discrete time sequence is sent through a parallel to serial converter, D/A converter and then finally upconverted to carrier frequncy f0. This signal is then transmitted into the channel. On the receiver’s side,
FFT is used in order to recover the original data stream.
Cyclic Prefix
The cyclic prefix in each OFDM-symbol is a known sequence that is appended to the beginning of the OFDM symbol in order to mitigate the ISI. If the length of the cyclic prefix is µ, then this sequence often consists of the last µ numbers in the orig-inal input sequence. The simulator used in this thesis assumes that the cyclic prefix completely eliminates the ISI.
TDD and FDD
All terminals in LTE are supporting two duplex schemes for UL and DL
transmis-sions; frequency division duplex (FDD) and time division duplex (TDD). Fig. 2.3
shows the structure in one radio frame (10 subframes) for FDD and TDD respec-tively. In FDD, UL and DL data is transmitted simultaneously on two different car-rier frequencies. In TDD, the UL and DL transmission is on the same carcar-rier fre-quency and must therefore be separated in time. LTE supports 7 different TDD con-figurations. For TDD, guard periods (GP) are needed to ensure a non-overlapping switch between UL and DL.
Chapter 3
Method
In this chapter the steps towards the simulations will be described. Firstly, import and treatment of OSM-data is presented along with an overview of the geometrical objects that represent buildings in the simulator, polygons. Thereafter the simulation scenarios and various necessary network setup, including system parameters, are described.
3.1
Map Generation
3.1.1 OSM import
Open Street Map (OSM) is an open source database founded in England that is sup-porting free editing of the world map. One of the main goals of this thesis is not only to successfully import maps from OSM into matlab, but also to treat and prepare the data for error free simulations.
There are two Overpass API query languages, Overpass XML and Overpass QL, used for reading OSM data from a specific region of interest and subsequently export this data into a specific file format. A helpful tool for quick export and display of the returned OSM data is Overpass turbo. In Overpass turbo there is a wizard that helps the user building Overpass queries by simply converting a single search term, such as "building", into a script. More information and language guides for the two query languages can be found in [9].
The OSM data consists of three element types; nodes, ways and relations. • A node element is a point representation with a latitude and a longitude
coor-dinate.
• A way element is a list of two or more nodes that represents linear features or areas such as building polygons or roads.
• A relation element is a list of nodes, ways or other relations, that is defining a relationship between its including elements.
Chapter 3. Method 13 The following Overpass XML script is used in Overpass turbo for all map im-ports in this thesis:
1 < !−−
This has been g ene rat ed by t h e overpass−turbo wizard .
3 Gets back b u i l d i n g s
−−>
5 <osm−s c r i p t output=" j s o n " timeout=" 25 ">
< !−− g a t h e r r e s u l t s −−>
7 <union>
<query type="way">
9 <has−kv k=" b u i l d i n g "/>
<bbox−query { { bbox } } / >
11 </query>
<query type="way">
13 <has−kv k=" b u i l d i n g : p a r t "/>
<bbox−query { { bbox } } / >
15 </query>
<query type=" r e l a t i o n ">
17 <has−kv k=" b u i l d i n g "/>
<bbox−query { { bbox } } / >
19 </query> </union> 21 < !−− p r i n t r e s u l t s −−> < p r i n t mode=" body "/> 23 < r e c u r s e type="down"/> < p r i n t mode=" s k e l e t o n " order=" q u a d t i l e "/> 25 </osm−s c r i p t >
This script is querying for ways including either tag "building" or "building:part" and relations with tag "building" within the bounding box bbox that is manually selected with the web-gui. After running the query the returned data can be down-loaded into a kml-file that is further imported into matlab for map generation.
In matlab, the kml-file is opened with fopen which generates an integer, FID, that is used as input to fread according to the following lines:
1 [ FID msg ] = fopen( kmlFile ,’ r t ’) ;
t x t = f r e a d( FID , ’ u i n t 8 =>char ’) ’ ;
The return variable txt is a string that includes all information from the kml-file. In order to extract and sort useful data stored in txt, the matlab function regexp can return a substring that matches the input requirement. For instance, finding a height value in a string variable objectString that includes all information about a specific building object can be done by using the following line since the height value (if it exists) is somewhere between the strings ’height"’ and ’</value>’:
h e i g h t = regexp ( o b j e c t S t r i n g ,’ h e i g h t " >.+? </ value > ’,’ match ’) ;
3.1.2 Polygons
The imported and filtered building data is stored as vectors in a matlab struct that must be further corrected before the simulator can use this data as geometrical infor-mation in the PG calculations. The buildings will be defined as polygons that must follow the same set of rules as outlined in [8].
A polygon will be represented as a number of rings that each has an orientation defining if the surface is on the outer or inner part of the ring. A ring is a sequence of at least 4 vertices in which the first vertex equals the last and with a direction spec-ified as the order the vertices from the first to the last vertex. A clockwise direction sets the interior of the polygon inside of the ring and vice versa for a counter clock-wise orientation. Hence, the outer ring of a polygon must have a clockclock-wise order of vertices and that a ring inside another ring (defining, for instance, building court yards) must have counter clockwise orientation to not overlap the interiors.
3.2
Network Setup
This section covers the network setup, including BS and user deployments for each city scenario and the system specific parameters. Four cities are simulated, Chicago, San José, London and Shibuya. The deployments are scenario specific and two LTE-systems, that only differ in carrier frequency, are simulated for each city.
3.2.1 Deployment
The simulation scenarios include two layers in the network, one UE layer and one macro BS layer. In order to implement the set of site-specific propagation models outlined in2.1.7, both UE and BS positions must be represented by xyz-coordinates within the boundaries of the generated map. Additionally the deployment coordi-nates must be a subset of the mapbins, defined by the binsize. The binsize is the distance in meters between the mapbins. For each of the following four maps, UE deployment will be uniform within a chosen user polygon. The result area is defined as the same as the user polygon.
As urban areas could look very different around the world, it is of great interest to analyze network behavior in cities from various continents. In the analysis part, Chicago, London and Shibuya have been chosen to represent typical urban environ-ments in North America, Europe and Asia respectively. Also, deployment strategies are different in these three continents which is accounted for here.
Since residential areas with buildings that rarely reach above 2 floors mostly look similar all over the planet, only one city, San José, is representing this type of city area.
Note that all deployments in this thesis are fictional, so the base station locations are not derived from real deployments.
Chicago
The deployment of macro BS and users in Chicago is shown in Fig.3.1. The macros
Chapter 3. Method 15 The user area is 500×500m2 and users will be deployed on every xy-bin within
this area, but only on every 4th floor in the z-direction. Hence, users will be sampled for heights 1.5m, 17.5m, 33.5m, etc. inside buildings.
FIGURE3.1: The Chicago map and its deployment. Base station loca-tions are shown in red. Horizontal direcloca-tions of the sector antennas are also visualized. User positions are included in the right figure and
represented as blue points. The binsize in Chicago is 5m.
San José
In the residential area of San José, almost all buildings are of height 5m (partly due to the default building height in the map generation). The macro sites are therefore
placed on poles 20m above ground level as seen in Fig.3.2. In American
residen-tial areas, macros are mostly deployed with 2-3km ISD. However, ISD is chosen to 1900m to fit the map of size 4×4km2. In order to deploy all sites on poles with height 20m, two site positions are adjusted.
The users will be deployed on every xy-bin of the map and every 4th floor (users will therefore only be located on the 1st floor) within the 4×4km2large user polygon.
London
In London, and Europe in general, it is most common to place macro BS on roof tops in urban areas. ISD is chosen to be 300m and the hexagonal deployment resulted in that most BS are placed on roof tops (2m above building height) and the rest on
poles 20m AGL. Fig.3.3 shows both site deployment and user deployment (result
FIGURE3.2: Base station and user deployment in San José.
Chapter 3. Method 17
Shibuya
Shibuya is a well known part of Tokyo, and is here chosen to represent a dense urban area in Asia. In this region of the world, macro BS are deployed with a shorter ISD than in both North America and Europe. Therefore, ISD is set as 200m as shown in Fig.3.4. Since no manual adjustments are made, two of the sites are placed on roof tops; one at height 41m and one at height 22m AGL. The other five macros are placed on poles 20m AGL. Users are located within a 200×200m2area (users are marked by blue dots in Fig.3.4(right)) and sampled for every xy bin, and for every 4th floor in z-direction just as in the other scenarios.
TABLE3.1: Parameters for two LTE systems with different carrier fre-quencies and bandwidth
System and Propagation Parameters
Bandwidth for fc= 700MHz 20MHz
Bandwidth for fc= 2GHz 20MHz
Common Parameters for Both Systems Macro BS Layer
Duplex Scheme FDD
Highest Modulation 64QAM
DL Maxpower 40W
UL Noise Figure 2.3
No. DL TX Antennas 2
No. UL RX Antennas 2
Electrical Down Tilt 0◦
Mechanical Down Tilt 9◦
UE Layer
Duplex Scheme FDD
Highest Modulation 64QAM
UL Maxpower 0.2W
No. DL RX Antennas 2
No. UL TX Antennas 1
DL Noise Figure 9
Noise Floor -204dB
Antenna Model isotropic
3.2.2 System Parameters
Along with the BS and UE deployment setup, general system parameters must be defined before simulations can run. In particular, each layer has its own param-eters for each system to be simulated. Since only two LTE systems with different carrier frequencies are simulated, each layer will define parameters for these
sys-tems separately. However, most parameters are common for both syssys-tems. Table3.1
19
Chapter 4
Results
4.1
Path Gain
This section is presenting results of PG for all links in the systems specified in chap-ter3. A chosen WINNER scenario (see 2.1.6) and the set of site-specific propaga-tion models (see2.1.7) are applied to each system separately. The set of site-specific models will in this chapter be referred to as "the site-specific model" to simplify the reading.
The realistic scenarios for the city areas in Chicago, San José, London and Shibuya showcase on differences between the models for various scenarios and environ-ments. PG results for each city are shown in 3D views, 2D views and cdfs that are based on the PG distribution. The 3D views include all indoor users whereas the 2D views only include the 1st floor users.
4.1.1 Chicago
The PG results for Chicago are presented in Fig.4.1, Fig.4.2and Fig.4.3.
Indoor
WINNER PG at 700MHz is likely between -110dB and -130dB for the lower floor users and for users deployed higher up in sky scrapers PG is likely as low as -140dB. The site-specific model is more dependent on the local environment and although PG is also likely between -110dB and -130dB at 700MHz, this model generates a larger spread in PG and is overall more optimistic.
Link PG is worsened at 2GHz, and the site-specific model is also at this carrier frequency slightly more positive than WINNER, although both models predict that indoor PG is likely between -120dB and -140dB.
Outdoor
Outdoor PG is, obviously, overall higher than indoor user PG. The site-specific model is significantly more optimistic than WINNER for outdoor users. WINNER PG at 700MHz is likely between -60dB and -80dB only for users with LOS links near the BS antennas. The site-specific model, however, is predicting that PG at 700MHz is likely between -60dB and -80dB for users that not necessarily are close to a BS but are located in street canyons. For the most faded links with lowest PG at 700MHz, the site-specific model in fact predicts PG to be as low as -150dB, whereas no link PG is lower than just below -120dB with WINNER.
model very likely between -70dB and -90dB, whereas the WINNER model predicts PG for these users to be more likely between -80dB and -100dB. Also at this fre-quency the site-specific model is more pessimistic for the most attenuated links, and some links can according to the site-specific model have PG as low as -160dB. Lowest outdoor WINNER PG is about -130dB.
Chapter 4. Results 21
FIGURE4.2: 2D view of PG in Chicago.
4.1.2 San José
PG results for San José are presented in Fig.4.4 and Fig.4.5. Only 1st floor indoor users and outdoor users are included in the calculations, since the residential area of San José is homogeneous in low height (about 5m) and doesn’t have any high rise
buildings. Rural macro scenario D2 (see Table2.1) is here used in the WINNER PG
calculations.
Indoor
Due to the large result area and the small building buildings, indoor user results are only understood from the cdfs in Fig.4.5. The site-specific model is clearly more op-timistic and at 700MHz it has more than 10dB higher mean and smaller spread than PG from the WINNER model. Indoor PG from the site-specific model at 700MHz is most likely between 90dB and 110dB, whereas WINNER PG is likely between -90dB and -130dB. At 2GHz the site-specific PG is mostly between --90dB and -120dB, whereas WINNER PG is more spread out and predicts that indoor PG is likely be-tween -90dB and -140dB.
Outdoor
Also for the outdoor user links the site-specific model is more optimistic than the WINNER model. PG at 700MHz is according to the site-specific model most likely between -70dB and -100dB. Again, the WINNER model predicts PG to be more spread out than the site-specific and is likely between -70dB and -110dB. Note that the WINNER model is more pessimistic for LOS links and PG falls off faster with distance than the site-specific PG. Also NLOS links are modeled more pessimistic with WINNER.
Chapter 4. Results 23
FIGURE4.4: 2D view of PG in San José.
4.1.3 London
Fig.4.6, Fig. 4.7 and Fig.4.8 show the PG results for London. Since the building height in this area of central London is homogeneous, apart from a few taller build-ings, PG for mostly 1st floor and 4th floor indoor users are calculated and presented along with outdoor users.
Indoor
The WINNER model and the set of site-specific model are generating more similar indoor results for London than San José. The 2D views in Fig.4.7show that PG at 700MHz for users located on the 1st floor is very likely between -90dB and -110dB
for both models. Fig.4.8 shows, however, that the site-specific PG at 700MHz is
having about 5dB higher mean and generates somewhat more optimistic results. The indoor links with the highest PG can be found for the 4th floor users, and these can experience PG up to -70dB.
At 2GHz, WINNER and the site-specific model have even more similar PG re-sults. 1st floor PG are likely between -100dB and -140dB for both models and 4th floor users are likely having PG between -80dB and -100dB.
Outdoor
The site-specific model is generating more positive outdoor user PG results than WINNER at 700MHz in London also. Just as in Chicago, the site-specific model is predicting significantly higher PG for users located in street-canyons. If BS is near a street canyon, then PG is very likely between -60dB and -80dB in a large part of that street according to the site-specific model. PG generated with the WINNER model fall off more rapidly with distance from BS, and only LOS links for users less than about 100m from BS are likely to have PG between -60dB and -80dB. Note that although the site-specific model is overall more optimistic than WINNER for outdoor users, it is also more pessimistic for the most attenuated links.
Chapter 4. Results 25
FIGURE4.6: 3D view of PG in London.
Chapter 4. Results 27
4.1.4 Shibuya
The PG results for the 400x400m area in Shibuya are shown in Fig.4.9, Fig.4.10and Fig.4.11.
Indoor
The site-specific model is generating overall higher indoor user PG than the C2 WIN-NER scenario. The cdfs show that WINWIN-NER PG has mean of about -100dB and site-specific PG has mean of about -90dB for 700MHz. For 2GHz, WINNER PG has mean just over -110dB and site-specific PG has decreased its mean to about -100dB. The WINNER model also generates overall larger spread for both frequencies than the site-specific models.
Outdoor
Due to the short ISD, outdoor users in Shibuya are very likely to have high PG. WINNER is again more pessimistic than the site-specific model at both frequencies. Whereas WINNER has a mean of about -80dB at 700MHz, the mean of the site-specific PG is near -70dB. Users located in the open areas experience highest PG, and the cdfs show that PG at 700MHz can be over -60dB for the site-specific models. At 2GHz, LOS links in the open areas are more likely to have PG between -70dB and -80dB for the site-specific models, and between -70dB and -90dB for WINNER. The means have now dropped to about -90dB for WINNER and just over -80dB for the site-specific model.
FIGURE4.10: 2D view of PG in Shibuya.
Chapter 4. Results 29
4.2
Network Performance
The simulator is using all predetermined system parameters, together with PG, in order to generate bitrates, SINR and interference for each link in both UL and DL direction. It is important to stress that the traffic load in these simulations is near zero. Hence, each user is allocated the full available spectrum (BW).
Since the network consists of two layers, performance is only evaluated for links between the user layer and macro layer. Since BW is 20MHz for every scenario, 100 RB is allocated to the users. In DL direction, the total BS power of 40W is divided equally for each RB. Thus, power per resource block is 0.4W in DL.
The DL and UL performance evaluations follow similar procedure, and thus al-most the same algorithm is used. PG results from the site-specific model serves as input to the simulator. The path with the strongest PG is determining the serving BS-UE link, but other paths with lower PG are also used in interference calculations in DL. The received signal strength is together with all modeled interference and noise determining the link SINR. Since the BS transmit power is much larger than the UE transmit power, the received signal power will be much larger in DL than in UL.
Besides carrier frequency, the choice of highest modulation is an important vari-able in how many bits per RB that can be received for each direction. As described
in Table 3.1, both DL and UL are using FDD and 64QAM as highest modulation,
but the DL transmission has a higher average number of bits per RB than the UL direction since MIMO is used in DL.
The final bitrate for each link is the number of assigned resource blocks (100 for DL and 96 for UL) multiplied with the bitrate per RB.
4.2.1 Chicago
Throughput and SINR results for Chicago are presented in Fig.4.12, Fig. 4.13and Fig.4.14. Both UL and DL results are presented at fc=700MHz and fc=2GHz
respec-tively.
Indoor
The 3D views shown in Fig.4.12illustrate the indoor user bitrate results from the two simulations. Fig.4.13shows throughput for 1st floor users. As illustrated in the figures, throughput can vary between everything from below minimum bitrates up to near peak rates inside of a single building.
In DL at 700MHz, 1st floor users located near the outer wall closest to the serving sector antenna are likely experiencing bitrates over 50mbps. On this floor, through-put rarely drops below 10mbps. For users located on top floors in the high rise buildings, bitrates are likely between 10mbps and 50mbps. In UL, 1st floor users located near outer walls can experience rates between 40mbps and 50mbps. How-ever, bitrates are also likely to drop below 10mbps for user positions on the side of the building furthest away from nearest BS. On top floors in high rise buildings, UL throughput is almost always below 10mbps.
throughput at 2GHz compared to 700MHz is perhaps even more prominent. Most 1st floor users have UL bitrates under 10mbps and higher up in sky scrapers, users are likely out of coverage.
The cdfs of SINR in UL are less smooth than those of DL. This is due to UL power control for a minimum SINR of -3dB and target SINR of 10dB. Hence, it’s much more likely for indoor links at 2GHz to have SINR below -3dB. Note that this is reflected in much worse indoor throughput at 2GHz.
Outdoor
It it obvious that SINR and throughput in UL is much more affected by the change in carrier frequency than DL. In fact, DL performances are nearly identical for both car-rier frequencies. Keep in mind, though, that PG results are not identical for 700MHz and 2GHz.
In UL, LOS links in open areas and street canyons are likely experiencing rates over 50mbps for both frequencies. For more attenuated links, however, rates are much lower at 2GHz. Rates can likely be below 30mbps at 2GHz, but not at 700MHz. Note that this is a consequence of that SINR falls below target SINR.
Chapter 4. Results 31
FIGURE4.13: UL and DL performance for Chicago in 2D.
4.2.2 San José
Fig.4.15and Fig.4.16show the SINR and throughput in UL and DL for San José.
Indoor
Indoor user results are presented in cdfs. DL SINR and throughput results are al-most identical for both carrier frequencies. In UL, however, SINR and subsequently throughput results are more pessimistic at 2GHz than 700MHz. The mean of UL SINR at 700MHz is about 13 and almost all links have SINR over target SINR. Since mean of SINR at 2GHz is just above 10, many links have SINR under target SINR. As a consequence, throughput at 2GHz becomes much worse for those links.
Outdoor
Also for outdoor users DL results are close to identical for both frequencies. Out-door user SINR and throughput in UL direction, though, is clearly worse at 2GHz. Although SINR is overall lower at 2GHz, almost all links have SINR over 10. UL throughput is therefore not as prominently decreased at 2GHz as in, for instance, Chicago. Spread is very low in SINR at both frequencies, and the throughput cdfs are therefore also having a low variances.
Chapter 4. Results 33
4.2.3 London
Indoor and outdoor user performance results in London are presented in Fig.4.17, Fig.4.18and Fig.4.19.
Indoor
Best indoor DL throughput rates are provided to 4th floor users, and Fig.4.17shows that bitrate results on this floor are similar for 700MHz and 2GHz. For 1st floor users,
2GHz generates lower DL bitrates than 700MHz, as shown in Fig.4.18. Although
spread in throughput is large at both frequencies, the cdfs tell that bitrates at 2GHz is spread out between 0mbps and 100mbps whereas DL throughput at 700MHz is mostly between 20mbps and 100mbps.
In UL, both 4th and 1st floor users are experiencing worse performance at 2GHz. The cdfs show that SINR at 2GHz is more likely to be below target SINR, which again affects throughput negatively. Throughput decrease is therefore most prominent for indoor user UL transmissions.
Outdoor
DL throughput for outdoor users are almost identical for both frequencies. In fact, some parts of London provide higher bitrates at 2GHz than at 700MHz. In UL, however, throughput is also in London worse at 2GHz. The cdfs show that the spread of SINR at 700MHz is between 10 and 20 for almost all users. 2GHz has lower SINR mean, resulting in that SINR are more likely to fall below target SINR and UL throughput is subsequently decreased significantly for the worst percentile users.
Chapter 4. Results 35
FIGURE4.18: UL and DL performance for London in 2D.
4.2.4 Shibuya
The throughput rates for DL and UL transmissions in Shibuya are displayed in Fig.4.20and Fig.4.21. The cdfs of SINR and throughput are presented in Fig.4.22.
Indoor
DL indoor user performance is nearly identical for 700MHz and 2GHz. Bitrates in the DL direction are mostly between 20mbps and 100mbps as a result of that SINR is mostly between 5 and 25.
In UL, however, the change in carrier frequency affects SINR and subsequently also throughput. Mean of SINR is decreasing from about 14 at 700MHz to about 12 at 2GHz, and the number of users with SINR below target SINR is therefore increasing. Fig.4.20and Fig.4.21show that indoor links that experience rates between 30mbps and 40mbps at 700MHz can have rates as low as 5mbps at 2GHz.
Outdoor
DL bitrates are also for outdoor users not much affected by the change in carrier frequency. Some parts of the more open areas, in fact, have slightly higher rates at 2GHz than at 700MHz. Note that the outdoor DL performance is much worse in Shibuya than in any of the other cities, although PG is high. This is because CRS interference is increased, since the ISD is shorter than in the other cities.
Chapter 4. Results 37
FIGURE4.20: UL and DL performance for Shibuya in 3D.
39
Chapter 5
Conclusions
In this thesis it has been shown that OSM maps can successfully be imported and used in the matlab based simulator. Buildings are often not as detailed as in pro-fessional maps, but the data import can be further developed in order to use more available information and by that make the scenarios more realistic.
The separate PG analyses of the four chosen cities are giving us guidelines about when the WINNER model and the site-specific model can provide quite different results. A general observation for both indoor and outdoor users is that the site-specific model is more optimistic for the top percentile users. Street canyons are common outdoor user positions in urban cities, and the site-specific model is much more optimistic than WINNER for such areas. Also, LOS link PG is likely to be higher for the site-specific model.
For low percentile users, the site-specific model can even be more pessimistic than WINNER. For a very nonuniform dense urban city like Chicago, the spread of PG is larger for the site-specific model. On the other hand, the spread of PG is smaller for the site-specific model in both rural cities with very uniform building heights and in urban city areas with large open areas such as this part of Shibuya.
The capacity analyses tell us that DL and UL transmissions are responding dif-ferently on the change in carrier frequency. UL bitrates are in general much more negatively affected by a change from 700MHz to 2GHz than DL, since UL transmis-sions are more affected by the larger pathlosses. Largest differences in SINR and capacity in UL between 2GHz and 700MHz are for indoor users in Chicago, since a majority of the users have SINR around or below minimum SINR at 2GHz. For the other cities UL SINR for indoor users is in general higher, but since SINR is more likely to be below target SINR at 2GHz than 700MHz, also the indoor UL throughput is becoming much worse at 2GHz for all four cities. Although UL SINR for outdoor users is worse at 2GHz than 700MHz, the number of users that fall below target SINR is low, even for Chicago and London, and therefore the throughput becomes only slightly worse.
DL throughput is in general much less affected by the change in carrier frequency than UL throughput. The most significant difference in DL bitrates between 700MHz and 2GHz is for indoor users in Chicago. A lower mean and larger spread of SINR at 2GHz leads to much worse throughput at this frequency. DL bitrates are also slightly worse at 2GHz than at 700MHz for indoor users in London. However, indoor users in San José and Shibuya are experiencing almost identical bitrates at both frequen-cies. Outdoor users in all four cities, however, are not experiencing any difference in neither SINR nor throughput when carrier frequency is changed from 700MHz to 2GHz.
41
Bibliography
[1] H. Asplund, M. Johansson, M. Lundevall, N. Jaldén, ”A set of propagation mod-els for site-specific predictions”, accepted for publication in 12th European Con-ference on Antennas and Propagation (EuCAP 2018), London, UK, April 2018. [2] Pekka Kysti and et al. IST-4-027756 Winner II. In WINNER II Channel Models,
2007.
[3] A. Goldsmith, Wireless Communications. Cambridge University Press, 2005. [4] Recommendation ITU-R P.526-13 Propagation by diffraction (11/2013).
[5] D. Astély, E. Dahlman, A. Furuskär, Y. Jading, M. Lindström, and S. Parkvall, Ericsson Research, “LTE: The Evolution of Mobile Broadband”, IEEE Communi-cations Magazine, April 2009.
[6] URL:http://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=7152831
[7] URL:https://ieeexplore.ieee.org/document/6782076/authors
[8] URL:https://www.esri.com/library/whitepapers/pdfs/shapefile.pdf
