UPTEC F13 036
Examensarbete 30 hp Oktober 2013
Deployment of Indoor Small-Cells for 4G Mobile Broadband
Patrik Ek
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
Deployment of Indoor Small-Cells for 4G Mobile Broadband
Patrik Ek
This thesis presents an investigation of the impact of indoor small-cells. It is expected that small-cells will be able to increase the throughput and capacity for the existing networks. A deployment algorithm is presented with focus on offloading traffic from the macro layer. The performance of the deployments created with the proposed algorithm, is compared with a reference deployment. The different deployments are then simulated in a real network simulator, which performs static simulations in 3 dimension using the theory of multiple knife-edge diffraction. The small-cells increased the throughput and capacity remarkably and additional gains were obtained with the proposed algorithm. The thesis also includes strategies for small-cell deployment.
ISSN: 1401-5757, UPTEC F13 036
Examinator: Tomas Nyberg
Ämnesgranskare: Mikael Sternad
Handledare: Gunther Auer
Sammanfattning
Anv¨ andandet av mobilt bredband har aldrig varit s˚ a h¨ ogt som idag. Dessutom verkar det som om anv¨ andandet kommer att ¨ oka drastiskt de kommande ˚ aren. Vissa f¨ orutsp˚ ar en ¨ okning p˚ a s˚ a mycket som 1000 procent. Det h¨ ar s¨ atter f¨ orst˚ as h¨ oga krav p˚ a utformningen av det mobila n¨ atverket. Samtidigt som anv¨ andningen har ¨ okat har det s˚ a kallade ”makro n¨ atet”
blivit t¨ atare. Det ¨ ar det n¨ at som uteslutande har anv¨ ants tidigare och de basstationer som placeras ut s¨ atts oftast p˚ a hustak eller p˚ a radiomaster. Dessutom har utvecklingen g˚ att fram s˚ a mycket att det nu g˚ ar att s¨ anda med samma frekvens fr˚ an alla basstationer. Mobila s¨ andningar ¨ ar popul¨ ara och det g¨ or att radiokanaler ¨ ar begr¨ ansade resurser. Olika n¨ atverk f˚ ar inte s¨ anda p˚ a samma frekvens och de lediga frekvenser som finns tillg¨ angliga ¨ ar f˚ a och dyra.
Genom att flera s¨ andare i ett n¨ atverk kan anv¨ anda samma frekvens, kan bandbredden ¨ oka f¨ or varje station. Det finns ett linj¨ art samband mellan bandbredd och den m¨ angd data som kan
¨
overf¨ oras och det g¨ or att ˚ ateranv¨ andningen av samma frekvens kan ¨ oka kapaciteten f¨ or varje basstation. Trots alla de ˚ atg¨ arder som hittills har genomf¨ orts f¨ or att f¨ orb¨ attra kapaciteten f¨ or det mobila n¨ atverket finns det fortfarande risk f¨ or att n¨ atet kommer att bli ¨ overbelastat i framtiden. D¨ arf¨ or kommer det att beh¨ ovas nya l¨ osningar. Makro n¨ atet kan byggas ut till en viss del, men tillslut kommer man att komma till en punkt d¨ ar st¨ orningarna blir f¨ or stora f¨ or att kunna n¨ atet ska kunna f¨ orb¨ attras mer. Det har lett fram till forskning p˚ a s˚ a kallade sm˚ a-celler. Dessa celler har f˚ att sina namn fr˚ an storleken p˚ a de ytor som de t¨ acker. De s¨ andare som servar dessa celler har betydligt l¨ agre effekt och det g¨ or att de t¨ acker en mindre yta. Dessa s¨ andare kan s¨ attas in p˚ a de ytor d¨ ar internet anv¨ andandet ¨ ar h¨ ogt och p˚ a grund av den svaga s¨ andeffekten kommer dessa att st¨ ora omgivande anv¨ andare mindre. Det ¨ ar ¨ aven ett k¨ ant problem att inomhus anv¨ andarna ¨ ar de som skapar mest datatrafik och samtidigt
¨
ar det d¨ ar som det ¨ ar sv˚ arast att f˚ a t¨ ackning. Om man skulle s¨ atta in s¨ andare i de st¨ orsta
husen skulle man mer effektivt kunna ge dessa anv¨ andare bra t¨ ackning. Dessutom skulle
de resurser, som dessa anv¨ andare beh¨ ovde, kunna ¨ overl˚ atas ˚ at andra anv¨ andare. Eftersom
kommer det alltid att finns en v¨ agg mellan inomhus s¨ andarna och utomhus s¨ andarna. Det g¨ or
att signalen som kommer utifr˚ an d¨ ampas n¨ ar den kommer in i byggnaden och signalen inifr˚ an
kommer att d¨ ampas f¨ or alla anv¨ andare som befinner sig utanf¨ or byggnaden. Allts˚ a kan dessa
s¨ andare s¨ anda samtidigt, utan att st¨ ora varandra alltf¨ or mycket. F¨ or nuvarande finns det
inte n˚ agra bra metoder f¨ or hur dessa s¨ andare ska placeras. Det finns inte heller n˚ agon bra
kunskap om hur m˚ anga s¨ andare man beh¨ over f¨ or att f˚ a t¨ ackning i en hel byggnad. I det
h¨ ar examensarbetet kommer jag att beskriva en algoritm, som har m¨ ojlighet att minimera
resursf¨ orbrukningen f¨ or makro n¨ atverket. Det vill s¨ aga, maximera de tillg¨ angliga resurserna
f¨ or makro n¨ atverket och p˚ a s˚ a s¨ att ¨ oka den maximala kapaciteten f¨ or dessa s¨ andare.
Contents
Contents 1
List of Figures 3
List of Tables 5
List of Abbreviations 6
1 Introduction 7
1.1 Background . . . . 7
1.2 Simulations in 3 dimensions . . . . 7
1.3 Scenario . . . . 8
1.4 Objective . . . . 8
1.5 Summary of results . . . . 8
2 Propagation models 9 2.1 Channel model . . . 10
2.1.1 Free Space Path Loss . . . 10
2.1.2 Antenna gain . . . 11
2.1.3 Multi-path fading . . . 11
2.2 Ray tracing models . . . 11
2.2.1 Diffraction . . . 11
2.2.2 Propagation paths . . . 13
2.2.3 BEZT model . . . 13
2.2.4 Extensions to BEZT . . . 14
2.3 Statistical propagation models . . . 14
2.3.1 Shadow Fading . . . 15
2.3.2 Statistical models with map information . . . 15
2.4 Indoor propagation models . . . 15
2.4.1 Indoor models to consider . . . 16
2.4.2 Model selection . . . 18
3 Basics In Digital Communication and LTE 19 3.1 Terminology . . . 19
3.2 Orthogonal Frequency-Division Multiplexing . . . 21
3.3 LTE Basics . . . 21
3.4 Heterogeneous Networks . . . 21
4 Small-cell deployment 23
4.1 Deployment Algorithm . . . 23
5 Experimental setup 26
5.1 Building Information . . . 26
5.2 System Parameters And Network Parameters . . . 30
5.3 Experiments . . . 31
6 Visualization 32 7 Results 34 7.1 Analysis of the algorithm . . . 34
7.1.1 Building 1 . . . 34
7.1.2 Building 2 . . . 37
7.1.3 Building 3 . . . 40
7.2 Analysis of the performance for different number of cells . . . 44
7.2.1 Results Building 1 . . . 44
7.2.2 Results Building 2 . . . 45
7.3 Coverage for the Small-Cells . . . 45
7.3.1 Coverage Per Floor . . . 46
7.3.2 Coverage for Selected Deployment Configurations . . . 46
7.3.3 General Discussion and Deployment Strategies . . . 46
8 Conclusions 52 8.1 Open Issues . . . 53
A Bin Plots for Throughput and Bitrate 54 A.1 Building 1 . . . 54
A.2 Building 2 . . . 55
A.3 Building 3 . . . 56
Bibliography 58
List of Figures
2.1 A multi-user mobile communication network. . . . 9
2.2 The wave front when it diffracts on a knife edge . . . 12
2.3 Geometry for the fresnel zone and for the knife edge diffraction. . . 12
2.4 The geometry of WINNER case B1. . . 17
3.1 This is an illustration of a heterogeneous network . . . 22
5.1 Rsrp interference to noise ratio and reference deployment for building 1 . . . . 27
5.2 Rsrp interference to noise ratio and reference deployment for building 2 . . . . 28
5.3 Rsrp interference to noise ratio and reference deployment for building 3 . . . . 30
6.1 User throughput building 1 with a low load, from Google Earth. . . 33
6.2 User throughput for building 1 with a low load, from MATLAB. . . 33
7.1 Throughput vs served traffic inside building 1, for a heterogeneous network with 56 indoor nodes. . . 35
7.2 Throughput and coverage for building 1 for the 56 small-cells reference deploy- ment. . . 36
7.3 Throughput vs served traffic for the macro layer outdoor surrounding building 1 with, for a heterogeneous network containing 56 indoor nodes. . . 36
7.4 Throughput vs served traffic inside building 1, for a heterogeneous network with 28 indoor nodes. . . 37
7.5 Throughput vs served traffic for the macro layer outdoor surrounding building 1, for a heterogeneous network containing 28 indoor nodes. . . 37
7.6 Throughput vs served traffic inside building 2, for a heterogeneous network with 56 indoor nodes. . . 38
7.7 Throughput for the macro network in building 2 for a low load. . . 39
7.8 A comparison between small-cell coverage and the bitrate for the macro layer in building 2. . . 39
7.9 Throughput vs served traffic for the macro layer outdoor surrounding building 2, for a heterogeneous network containing 56 indoor nodes. . . 40
7.10 Throughput vs served traffic inside building 2, for a heterogeneous network with 28 indoor nodes. . . 40
7.11 Throughput vs served traffic for the macro layer outdoor surrounding building 2, for a heterogeneous network containing 28 indoor nodes. . . 41
7.12 Comparison in small-cell coverage between the reference deployment and the deployment made with the deployment algorithm. . . 42
7.13 Throughput vs served traffic inside building 3, for a heterogeneous network with 30 indoor nodes. . . 42
7.14 Throughput vs served traffic inside building 3, for a heterogeneous network
with 60 indoor nodes. . . 43
7.15 Throughput vs served traffic for the macro layer, outdoor surrounding building
3, for a heterogeneous network containing 60 indoor nodes. . . 43
7.16 Throughput vs served traffic outside building 1 with different node number and weights w 1 = 0.99 and w 2 = 0.01. . . 45
7.17 Throughput vs served traffic outside building 2 with different node number and weights w 1 = 0.99 and w 2 = 0.01. . . 45
7.18 The figure shows coverage per floor and number of nodes per floor. . . . 47
7.19 Coverage for building 1 with with 56 pico cells for different configurations. . . 48
7.20 Coverage for building 2 with with 56 pico cells for different configurations. . . 49
7.21 User throughput for building 1 for the macro network with a high load. . . 50
7.22 User throughput for a high load with a heterogeneous network with 56 pico cells in building 1 . . . 50
7.23 Comparison between pico node positions and macro layer received signal power. 51 A.1 Macro network low load. . . 54
A.2 Macro network high load. . . 55
A.3 Macro reference network for a low load. . . 55
A.4 Macro reference network for a high load. . . 56
A.5 Macro reference network for a low load. . . 56
A.6 Macro reference network for a high load. . . 57
List of Tables
5.1 Table of system parameters and network parameters . . . 31
5.2 The different weights for the deployment algorithm that are investigated . . . 31
7.1 Different number of indoor nodes for building 2 . . . 44
List of Abbreviations
BS base station
DFT discrete Fourier transform DL downlink
FSPL free space path loss LTE long term evolution
MIMO multiple input multiple output OFDM orthogonal frequency multiplexing PG path gain
PL path loss
UL uplink
Chapter 1 Introduction
1.1 Background
The demands for mobile broadband services over wireless networks are expected to increase in the future as the use of terminals such as smartphones and tablets will continue to grow.
Services like streaming of music and movies over wireless networks will increase and will need to be executed in a reasonable time. The services that today are done mostly by other means, for example payment, are expected be done by smartphones to a larger extent. This means that in the near future we will require higher capacity and better coverage than we have today. To meet the future requirements new solutions needs to be developed. It is known that most of the traffic is generated indoors and also this is also where the coverage is poor.
One way to increase both the capacity and the coverage at the same time is to deploy indoor small-cells. The advantage with this solution is that it makes it possible to deploy cells locally near the places where they are needed. These cells will also have a lower transmit power and will so have a limited impact of the existing macro network. Another advantage is that they are easy to deploy since the supporting structures are easier to get. Site acquisition is easier as well.
1.2 Simulations in 3 dimensions
Most radio propagation models that exist now are constructed for 2 dimensions. Some models have a height correction that can account for height differences to a limited extent. However, when the networks becomes more dense a new problem has arisen. The path gain may change quite drastically with the height of a building. This has led to the development of a new simulator, to model the radio propagation in 3 dimensions. This also leads to a need to visualize the result in 3 dimensions. One part of this thesis will then be about visualizing the result in a good way. It is already possible to visualize the results in Google Earth in in 2 dimensions. The job that needs to be done is then to add support for 3D visualization.
Also some plots are made in the MATLAB. Here it will also be needed to create support for 3D visualization.
Chapter 2 will give an overview of the radio propagation models. This explains the theory behind the 3D model and contains a discussion about which indoor model have been selected.
Chapter 3 will give the the basic theory and terminology in digital communication and LTE.
Then follows a description of a deployment algorithm for maximum offload of macro cells in
Chapter 4. In Chapter 5 the experiments and investigations that will be done in this thesis
will be explained. The experiments will first deal with finding the most effective way to
deploy a fixed number of indoor small-cells and then to find out how the performance varies
with a different number of small-cells. Chapter 6 explains a little about how the visualization is done and will also describe some features.
1.3 Scenario
The simulations are performed on a major American city. This means that the centre mostly consists of high rise buildings where the path gain varies with height. Earlier studies is mostly done on ground level and some uses a height correction factor to account for changes in height. However, the fact is that we do not know much about how the path gain varies with height. This is what we expect to find out with this simulator. What we know is though that the signal is stronger a couple of floors up in the building and we expect this depends on shadowing from adjacent buildings. We also know that the top floor in high rise buildings have problems with low path gain and expect that the problem is due to the the antenna directivity. The antennas is in general directed downward and this is a problem for the tallest buildings. In the suburban areas the buildings are instead low and spread out over larger areas. This give a problem with interference instead. The base stations (BS) need to be deployed sparsely too avoid interfering with each other. These suburban BSs do also provide the most of the coverage in high rise buildings. The problem is that these buildings have line of sight to all the sites and this gives a problem with interference.
1.4 Objective
The main objective is to investigate how the deployment of indoor small-cells will impact on a wireless network in terms of capacity and coverage. Of particular interest is how the performance of the wireless network varies as a function of the building height. It is suggested by previous studies that the lower floors will be shadowed by the surrounding building and will thus gain more from the deployment of indoor small-cells. On the upper floors on the other hand the signal from the small cells will interfere with the macro network and compromise the benefit from the small cells.
To remove as much traffic as possible from the macro network by deploying indoor small- cell, an algorithm for maximal offload is implemented. The algorithm focuses on minimizing the resource consumption for the macro network. The first issue is to find an implementation of the algorithm, which minimizes the macro layers resource consumption for a fixed number of small-cells. The second issue is to find out what happens when different numbers of small-cells are deployed.
1.5 Summary of results
The first part of the results compare the deployments created by the deployment algorithm
for maximal offload with a reference deployment. The second part of the results deals with
deployments different numbers of small-cells. After that follows a discussion about coverage
and a deployment strategies. The results show that comparing with a reference deployment,
the algorithm for maximal offload could improve the capacity of the network. It was also
shown that when applying the algorithm on different numbers of small-cells in a building the
capacity for the network did not change much until there were few enough small-cells. The
results regarding coverage and node placement showed that the upper floors required more
nodes and had in general a lower coverage than the bottom floors. It also showed that a
denser deployment was needed in areas with a strong macro signal.
Chapter 2
Propagation models
A mobile communication network is displayed in Figure 2.1. The connection between one device and another is called a link. This link will be affected by the geometry of the sur- rounding area. For networks with multiple base stations (BSs) and multiple users, the links will interfere with each other. This arises from a phenomena called multi-point to point links.
On the downlink one user will receive a signal from one BS, but because of its location it may receive an unwanted signal from other BSs as well. An example is user A and B in Figure 2.1. The users are served by different BSs, but they will also receive a strong signal from each others servers. On the uplink the BS will be interfered by the users that are served by the surrounding cells and is scheduled on the same time-frequency block. This is illustrated by user B and C in Figure 2.1. Since the network uses frequency reuse 1, there will be strong interference from the neighbouring cells. More information about the LTE network can be found in Section 3.3.
Interference Link
A
B C
Interference
Figure 2.1: A multi-user mobile communication network.
2.1 Channel model
Let s ij (t) be the transmit signal from BS i to user j. The signal is assumed to have a transmit power P tx in Watts. The received signal can then be written as
r ij (t) = h ij (t)s ij (t) + n(t), (2.1) where h ij (t) is called the channel coefficient. The variable n(t) is the sum of the noise that arises from the receiver and thermal noise. When the path gain is calculated the noise term is normally disregarded and this gives
r ′ ij (t) = h ij (t)s ij (t). (2.2) Furthermore, P tx is known and constant. Assuming isotropic antennas for the transmitter and the receiver, the path gain(PG) of the channel is defined as
PG = P rx
P tx
(2.3) where P rx is the square of the expectation value of r ij ′ (t). When P rx is assumed constant this would yield
PG = E 2 [h ij (t)]. (2.4)
For directive antennas the same approach is used and the antenna gain, which is explained in Section 2.1.2, is then multiplied with the path gain for the isotropic antenna. A more thorough explanation of the channel model can be obtained in [1].
The PG can be modelled in different ways. The simulator used in this thesis uses ray- tracing to model the outdoor and outdoor to indoor propagation from the macro-cells. For the indoor propagation and indoor to outdoor propagation for the small-cells a statistical model is used. These two approaches are explained in Sections 2.2 and 2.3.
Some simulators explicitly model multi-path fading to include effects that arises from local scattering of the signal. Multi-path fading is discussed in Section 2.1.3.
2.1.1 Free Space Path Loss
The PG is a term that explains how the large scale propagation behaves. More common is however to use the term path loss (PL). The PL is simply 1/PG for a linear scale. The free space path loss FSPL is a function of the distance between the transmitter and the receiver d, the frequency f and the speed of light c. For isotropic antennas the FSPL is calculated as [1]
FSPL = 4πdf c
2
, (2.5)
The FSPL can also be written in dB and is then calculated as FSPL = 20log 10 (d) + 20log 10
4π c
+ 20log 10 (f ). (2.6)
The FSPL is usually seen as a lower bound of the PL and the real PL is in most simulators set as
PL ≥ FSPL. (2.7)
2.1.2 Antenna gain
An antenna can be directed to transmit with higher power in any direction or capture more of the transmitted signal on the receivers side. The gain that arises from antenna directivity is called antenna gain. The antenna gain from an antenna that transmits with the same power in all directions is called an isotropic antenna and the antenna gain for it is defined as one. If an antenna is directed to have an increased gain in any direction, then the gain will also decrease in another direction. An example is a parabolic antenna that receives the TV signal from satellites. It is really effective in the direction that the signal comes from, but does not catch much signal power from any other directions.
2.1.3 Multi-path fading
In a real network the signal will scatter on different objects. This leads to signal delay, which will create impulse responses of different length. This part of the signal is frequency selective and in general impossible to predict. Multi-path fading is normally simulated from a statistical distribution. In the used simulator, the multi-path fading is not simulated, but accounted for in the bitrate calculations.
2.2 Ray tracing models
All electromagnetic phenomena can be described using Maxwell’s equations together with the correct boundary conditions. However this would be too complicated which means that some simplifications need to be done. Assuming that the wave length is small compared to the surrounding objects makes it possible to look at micro waves in similar manner as light rays. This allows us to use ray theory to make accurate path loss predictions, taking into account diffraction, reflection, scattering, absorption and other ray phenomena. The simulator used for this thesis uses a model called BEZT to model the PL and is named after its creators. The abbreviation stands for ”BErg Zofia Thiessen”. This model is based knife edge diffraction and is also extended to be able to calculate reflections, but it neglects absorption, and scattering.
2.2.1 Diffraction
Diffraction occurs when a wave meets an obstacle or aperture. If the obstacle is much larger than the wavelength the rays will diffract around the edge of the object as in Figure 2.2 according to Huygens-Fresnel principle [2]. The principle states that any disturbance of an electromagnetic wave will become a point source of a spherical electromagnetic wave. This will reduce the energy density of the wave since it spreads out.
Knife edge diffraction
Diffraction can be computationally heavy to model so the calculation efficiency is an impor- tant issue here. A comparably easy method to implement is called knife edge diffraction.
The geometry of the problem is described in Figure 2.3a. The knife edge is assumed to be
infinitely thin which is normally not the case. The method is based on the grade of obstruc-
tion in the first Fresnel zone. The first Fresnel zone is a cross section of the antenna radiation
pattern at the knife edge, in which the phase difference of the transmitted signal is smaller
than half a wavelength. The radius of the Fresnel zone is elliptic in the vertical plane and
Figure 2.2: The wave front when it diffracts on a knife edge
d
1d 2
h
(a) The geometry of the knife edge diffrac- tion.
(b) The geometry of the fresnel zone.
Figure 2.3: Geometry for the fresnel zone and for the knife edge diffraction.
is illustrated in Figure 2.3b. The electric field strength of the diffracted electric field E is calculated with an integral called the Fresnel integral and is written as [3]
E = E 0
1 + i 2
Z ν
−∞
exp
−i π 2 t 2
dt, (2.8)
where E 0 is the incident electrical field strength and ν is the Fresnel-Kirchhoff diffraction parameter. The parameter ν can be calculated as
ν ≈ h
s 2
(d 1 + d 2 ) λd 1 d 2 ≈ 2 r ∆d
λ , (2.9)
where d 1 , d 2 and h is defined as in Figure 2.3a. The variable ∆d is the difference in distance that the wave has to travel to reach the receiver relative to the geometric distance between the nodes. Here h and thus ν is defined as negative if the obstacle extends above the knife edge. Equation (2.8) makes it possible to define a diffraction loss ℓ as the ratio of the initial electric field and the diffracted field,
ℓ = |E|
E 0 . (2.10)
The Fresnel integral is a complicated function to work with and is therefore normally ap-
proximated with simpler functions. When there is line of sight conditions, diffraction losses
will go to 1, which means that there is only FSPL. The knife edge diffraction model has
the drawback that it does not take into account parameters as reflections, surface roughness,
polarization and conductivity of the diffractor [1]. Knife edge diffraction can be expanded to
include multiple knife edges. This is often referred to as multiple knife-edge diffraction.
2.2.2 Propagation paths
For a ray tracing method to work well, the most likely propagation paths needs to be found.
In general it is said that distant users get their strongest signal from over roof top propagation.
This is because the signal that propagates around the corners will lose too much signal power due to reflections on the walls. Users close to the macro station will still most likely get the strongest signal from over roof top propagation if the station rises above the closest buildings. However, for a Manhattan-like grid describing centres of larger cities, the macro stations would in most cases be placed below the roof tops and for these sites the strongest signal will probably come from around corner propagation for the closest users. Micro stations are almost always placed below the roof tops and in many cases these will even be placed just a few meters above ground. Also small-cells have a much shorter range than the macro cells. This means that for small-cells the dominant propagation path is in general around corners.
2.2.3 BEZT model
The BEZT model covers the case over roof top propagation. The model itself does not consider reflections, so that other solutions need to be applied. BEZT is based on multiple knife-edge diffraction and the model is executed in two steps. For the reader to fully under- stand why this has to be the done it is important to note that there are actually two cases where knife edge diffraction applies. One where the signal propagation path is shorter than the path the signal would need to travel to connect with any of the knife edges. In other words the user has line of sight to the BS. This case is referred to as plus diffraction. Also there is a second case where the diffracted path is equal to the propagated path which in turn is longer that the geometric distance between the transmitter and the receiver. This corresponds to the non-line of sight case and is referred to as minus diffraction. For multiple knife-edge diffraction some knife edges have plus diffraction between each other and some have minus diffraction, except for the line of sight case where only plus diffraction exists.
This makes it important to model both cases.
Step 1: Plus Diffraction Loss
Deploy screens where there is need to insert knife edges and calculate the plus diffraction loss for every screen according to (2.10). The distance difference ∆d would in this case be the shortest distance the electromagnetic wave needs to travel to reach the receiver without penetrating any screens (d min ) minus the distance it needs to travel to connect with screen i (d i ),
∆d = d i − d min . (2.11)
The screens closest to the transmitter or the receiver will be the most likely screens to be hit first by the signal ray and thus have the greatest impact on the plus diffraction. Because of that an update algorithm will be needed not to overestimate the diffraction loss of the screens in the middle of the transmitter and the receiver. The plus diffraction loss ℓ +,tot is then the product of all the plus diffraction losses,
ℓ +,tot = Y
i
ℓ + (i), (2.12)
where ℓ + (i) is the diffraction loss from screen i. The path loss of the plus diffraction is then calculated as,
PL = 1
ℓ 2 +,tot FSPL. (2.13)
For line of sight conditions this is the only path loss that you need to calculate.
Step 2: Minus Diffraction Loss
For non-line of sight conditions another step is required to get the correct path loss. For this step ∆d is defined as in Section 2.2.1. Calculate the diffraction losses as in Section 2.2.1 for every screen separately and then select the dominating screen. The diffraction loss for this screen is then called ℓ − ,max . Then calculate the diffraction losses for the left ℓ − ,left and the right side ℓ −,right of the dominating screen, where the dominating screen is the right boundary for ℓ −,left and the left boundary for ℓ −,right . The total minus diffraction loss ℓ −,tot is then the product of all the minus diffraction losses
ℓ −,tot = ℓ −,max ℓ −,left ℓ −,right . (2.14)
The total path loss for the non-line of sight case is then
PL = 1
ℓ 2 −,tot ℓ 2 +,tot FSPL. (2.15)
To get the total gain, you also need to add the antenna gain.
2.2.4 Extensions to BEZT
The BEZT model can be used for outdoor macro cells and it supports outdoor propagation only. An extension to calculate reflections has been implemented. Then there are limitations regarding tall buildings, street-crossings and some other cases where the signal in reality will propagate around corners instead of above roof tops even for the macro BSs. To solve the around corner problem, a penalty algorithm is implemented. To make it possible to also simulate indoor users, the model calculates indoor loss in a dB per meter fashion.
2.3 Statistical propagation models
When there is no ray tracing method developed or when ray tracing would be too time consuming a statistical method can be easily implemented. A statistical method is simple and in general less accurate, but if it is carefully elaborated it can still provide good results.
For propagation inside buildings, ray tracing can be hard to implement and even if it is implemented it may need a lot of tuning to work properly. Also a floor plan will be needed, which in most cases it is not. For indoor cells the transmit power will also be too weak to be able to transmit any data after a fairly short distance which means that the cells coverage area will not range much further than just outside the walls of the building it is deployed in.
In this case a statistical model can be of good use.
The general form of a statistical path loss prediction can be written as [4]
PL = A log 10 (d) + B + C log 10 f c
f 0
+ X, [dB] (2.16)
where f c is the carrier frequency and f 0 is a reference frequency. The parameters A, B and
C are constants that are adjusted to a set of empirical measurement data. Excess losses, for
example indoor losses, can be added to the parameter X. As it can be seen from Equation
(2.16) A considers the distance dependence and C considers the frequency dependence. B is
a constant that determine the minimum path loss in the sense that even if d is small there
will still be an appropriate loss. If the distance would be infinitely small the path gain would be infinite according to the equation. This is however not the case and this gives a limitation to how close the equation is valid. Depending of how far from the transmitter measurements could be done and how accurately the model takes long distance effects account, the model has also an upper limit for how far it can be used. Thus the model is only defined for a finite range of distances d.
2.3.1 Shadow Fading
A statistical path loss model of the form of Equation (2.16) does mainly depend on distance and does not model the random variations that may arise when users are shadowed by a building or another large obstacle. To also be able to take this into account when doing a statistic simulation a random element needs to be added. It has been empirically determined that these variations are log-normally distributed for a constant distance from the transmit- ter. This means that for a circle with constant radius around the transmitter the random variations are normally distributed on a logarithmic scale. The variance of the distribution is environment dependent. The shadow fading is linear for the logarithmic scale which gives us the total path loss PL tot as
PL tot = PL + Shadow Fading. (2.17)
for this thesis the shadow fading has not been modelled.
2.3.2 Statistical models with map information
While the BEZT model is using map information to calculate the gain from the macro network, the implementation of a statistical indoor model has to be thought through carefully.
Also I need to consider the environment. The same model may not be valid in all the different areas. Since I will consider two scenarios, indoor propagation and indoor to outdoor propagation, I will most likely need two different models for the different scenarios, one that applies for indoor only and one that consider indoor to outdoor propagation.
2.4 Indoor propagation models
Indoor models may in some cases use the additional parameter X in (2.16) to model the indoor loss. The parameter will then look like
X = L indoor , [dB] (2.18)
where L indoor is the indoor loss in addition to the PL. The indoor loss may include both the floor loss and the wall loss depending on how the coefficients A, B and C is generated.
Normally the losses that comes from the characteristics of indoor propagation is included in the wall loss. If the signal passes an outdoor wall, the wall loss for this wall should be modelled by an additional loss factor. The outdoor walls has in general a higher loss than the indoor walls and needs to be treated separately. Also if the signal does not come from the same building it is most likely that it comes in through the outdoor wall on the same floor.
For some links where the height difference is large the strongest signal could pass through
one or two floors, but this signal will be weak and there is really not way to tell when this
will happen without thorough measurements.
2.4.1 Indoor models to consider
In this thesis a number of different models have been considered,
• WINNER case A1, B1 [4]
• 3GPP case 1 [5]
• COST 231 indoor models [6]
WINNER case A1
WINNER case A1 refers to the indoor office case. Here the signal is assumed to propagate in a typical office environment. There are two implementations for the non-line of sight propagation for this case. The first implementation gives empirical values for the constants A, B and C in Equation (2.16) and the value for X from Equation (2.18) contains only the floor loss. This is the actual scenario that was measured and assumes a corridor to room propagation. An alternative method is to identify the coefficients A, B and C from the FSPL and then use X to model the indoor loss. The PL is then calculated as
PL = FSPL + n int L int + FL, [dB] (2.19) where n int is the number of internal walls between the transmitter and the receiver and L int
is the loss per internal wall. Here the indoor losses that arises from the indoor environment is included in the wall loss. The parameter FL is the floor loss. The FL is modelled as
FL = 17 + 4(n fl − 1), [dB] (2.20)
where n fl is the number of floors between the transmitter and the receiver. This model assumes that the signal will go through the floor for the first floor and if it is more than one floor between the transmitter and the receiver the strongest signal will instead be reflected on a nearby building. If it is possible the model from Equation (2.19), which explicitly models the floor loss and wall loss, should be preferred. This is partly because many measurements that have been done actually confirm that the indoor loss is more or less linear in dB, but also because the optional method gives an increased freedom. For example the case corridor- room-room would not be the same as corridor-room even if the distance would be the same.
WINNER Case B1
This case refers to outdoor micro, but it can still be useful since most of the propagation from the indoor nodes will be outdoor propagation. The model assumes around corner propagation and assumes a geometry as in Figure 2.4. This is however a model that is complicated to implement because a multitude of different paths needs to be investigated. The model also assumes a square Manhattan like grid, which is not the case in all cities.
3GPP Case 1
This is a model that has worked well in the past and is definitely worth considering. The model to consider here would be the urban micro cellular model. This would be considered as the dominant propagation path for users that are not in the same building as the indoor node. The model has an equation
PL = 36.7 log 10 (d) + 22.7 + 26 log 10 (f c ), [dB], (2.21)
Figure 2.4: The geometry of WINNER case B1.
where f c is the carrier frequency in GHz. Since the frequency correction factor used in this thesis is 23, Equation (2.21) is modified. The equation is then normalized to 2 GHz. The equation is then instead written as
PL = 36.7 log 10 (d) + 30.6 + 23 log 10 f c 2
. [dB] (2.22)
COST 231 Indoor Models
In COST 231 three different indoor models are considered. The models are listed below.
One Slope Model This is a simple model that that uses Equation (2.16) with X set to 0.
The floor loss problem is solved by selecting different values for the constants A, B and C. The drawback for this model is that the frequency parameter C and the minimum PL parameter B are not separated. The measurements are done for 900 MHz and 1800 MHz and are only valid for those frequencies. The frequency correction factor for different environments is not considered here either, though it is not unnoticed that the difference between the frequency bands is slightly higher than for free space propagation. This could be adjusted with data from other investigations, but it is still quite tedious since the environment there may be different.
Multi-wall Model The model suggests a problem similar to WINNER case A1 and indoor loss is calculated as
X = L indoor = n int L int + n
nfl+2 nfl+1
−b
fl L fl , [dB] (2.23)
where the constant b is a tuning constant. This model uses the same assumption as WINNER case A1 for the PL and it is calculated as
PL = FSPL + X. [dB] (2.24)
The assumed floor loss is slightly higher for the COST multi-wall model than for WINNER
case A1.
Linear Attenuation Model This model is a simplification of the multi-wall model. The model has a dB per meter approach which means that the loss per meter in dB is constant.
The model can be written as
PL = FSPL + αd, [dB] (2.25)
where α is the loss per meter and d is the distance between the transmitter and the receiver.
The constant α has different values depending on what frequency is used and how many floors separating the transmitter from the receiver.
2.4.2 Model selection
The model selection requires some discussion. All models should be assumed to be equally correct for their respective site. However some of the models are more preferred than others because of the environment where the measurements was performed and also because some methods have more freedom in the way that they can be used. Some methods can be discarded directly: The method with explicit walls and also WINNER Case B1. The multi-wall model (or corresponding WINNER model) is not is not consistent with the already implemented outdoor to indoor propagation model from BEZT which uses the linear attenuation model.
Also there are no good ways to represent walls in the simulator and just randomly deploy walls would not give BEZT the same conditions as the indoor model. However, the linear attenuation model is quite similar to the multi-wall model except that it gives the PL as a mean per meter instead of adding additional loss after each wall. If α is given for the transmitter and the receiver on the same floor then the linear attenuation model is compatible with any floor loss model. The conclusion is then that the linear attenuation model from COST 231, with α defined as the indoor loss per meter for transmitter and receiver on the same floor, should be used. To this a floor loss must be added. Since the WINNER floor loss is lower than FSPL for large buildings this model will be discarded and the floor loss model to be used in this thesis will be the floor loss from COST 231 multi-wall model. The indoor model will then be
PL = FSPL + αd + n
nfl+2 nfl+1
−b
fl L fl . [dB] (2.26)
For the indoor to outdoor model the 3GPP case 1 micro cellular model will be used. The
reason for this is that WINNER B1 is too restrictive and the COST 231 micro cellular model
works bad for users far below roof tops. The second statement is also confirmed in [6].
Chapter 3
Basics In Digital Communication and LTE
This chapter deals with explaining some basic principles about digital communication and LTE.
3.1 Terminology
Here follows some important concepts for digital communication.
Signal to Noise Ratio
Let the received signal be defined as in Equation (2.1). Define P rx as the received power of the first term in Equation (2.1) and the noise power P n as the received power of second term in Equation (2.1). The signal to noise ratio (SNR) can then be calculated as
SNR = P rx
P n
. (3.1)
Signal to Interference Ratio and Geometry
Define the downlink interference for user i as the received signals from all downlink transmis- sions received by user i that is not sent to this user. The signal to interference ratio (SIR) is the ratio between the received signal power and the total interference power P I calculated as
SIR = P rx
P I . (3.2)
Uplink interference for BS j can be defined as the received signals from all uplink transmissions on the same time-frequency block received by BS j . This interference arises from the out of cell users that are scheduled on the same time-frequency block as an in cell user. Due to the orthogonality in LTE, this is at most one user per cell. More information about LTE can be found in Section 3.3. The ”geometry” is defined as the SIR when all the transmitters transmits at the same time and can be seen as the worst case of the SIR.
A term closely related to how much data that can be received is the Signal to Interference and Noise ratio γ, which is calculated as
γ = P rx
P i + P n
. (3.3)
Channel Capacity and Bitrate
The channel capacity is measured by bit per second and is the upper bound for how much data that can be transmitted without detection errors. In year 1949 Claude Shannon [7]
proved that the channel capacity C can be calculated as
C = B log 2 (1 + γ), (3.4)
where B is the bandwidth and C has the unit bit/second. Expanding this to also be valid for MIMO systems, Equation (3.4) can be written as
C ≤ min(M, N )B log 2 (1 + γ), (3.5)
where M and N is the number of transmit and receive antennas and B is the channel bandwidth. The maximum achievable bitrate (b) requires rigorous calculations and because of that, they are in most cases taken from lookup-tables, where the upper bound is the channel capacity,
b ≤ C. (3.6)
The maximum achievable bitrate is the bitrate for a particular user, assuming the user gets all the resources.
With this theory it is then possible to define the spectral efficiency for the channel capacity and for the bitrate. The spectral efficiency is defined as
R C = C
B , (3.7)
and have the unit bit/second/Hz. The spectral efficiency is the best way to measure how effective a channel is. The channel capacity is relative, since a channel with low spectral efficiency still can have a high capacity, given much bandwidth. Given Equation (3.7) it is possible to define an upper bound of the spectral efficiency for the bitrate
R b = b B ≤ C
B , (3.8)
Coverage
When the bitrate is below a specified limit, then no data will be transmitted to that user.
Then you say that this user is not covered. The coverage area is defined as an area around a BS, where all the users are served by this particular BS. Some ways increase the coverage are to deploy more BSs, increase the transmit power or add receiver antennas.
Throughput
The throughput is what the user will experience. If the number of active users is low, then
the throughput will be close to the maximum achievable bitrate. If there are many active
user in the system, the user will be placed in a queue and this will significantly lower the
throughput. Depending on what queueing model is used the throughput will be different. In
the simulations done in this thesis, all the data packets that are transmitted are assumed to
be received by the user. This is not the case in a real network since some packets will need to
be retransmitted. Also mobility is not considered. These assumptions will result in a slight
overestimation of the throughput.
3.2 Orthogonal Frequency-Division Multiplexing
A Orthogonal Frequency-Division Multiplexing (OFDM) system, divides the total bandwidth into number of narrow-banded sub-carriers. The sub-carriers are orthogonal toward each other if the maximimum excess delay of the channel response is shorter than the cyclic prefix of the signal.
3.3 LTE Basics
The LTE system uses OFDM on the downlink and DFT-spread OFDM on the uplink. LTE downlink transmissions are ordered in 10 ms radio frames which consists of 10 subframes of 1 ms. In LTE a resource block consists of 12 sub-carriers, of 15 kHz each, in the frequency domain. The number of resource blocks in the frequency domain in an LTE system depends on the system’s bandwidth.
By scheduling the transmissions it is possible to reduce the uplink interference since the users only transmits in their scheduled time frequency-block. This means that the only users that can create interference on the uplink is the users that are scheduled on the same time- frequency block in another cell. The interference is also reduced compared to if the BS should transmit with full power since the power is split up on all the sub-carriers.
The downlink transmissions are dynamically scheduled and to inform the user of which resource block it gets, the first 1-3 symbols in each subframe are control symbols. Also there are common reference symbols spread out in the resource block, which are known to the user and helps the user to estimate the channel.
For downlink transmission the BS always transmits with maximum power. For the uplink however a power control is applied on the user side. This is to limit the exposure of radiation towards the head and to increase the battery time of the user.
3.4 Heterogeneous Networks
A heterogeneous network is a network that contains more than one kind of cells. These can be for example macro cells (BSs) and micro cells or macro cells and pico cells. The difference is the mainly the output power, which affects the cell size. The macro cells have a larger coverage area than the micro cells and the micro cells have a larger coverage area than the pico cells. However since the coverage area is much smaller for the micro cells than for the macro cells, they serve a more local area which means that they in most cases are placed below roof top to give a better coverage. This means that the micro cells also need another channel model. There are not much difference between micro cells and pico cells, why it is usual to refer to them as small-cells. 3.1 shows a picture of a heterogeneous network.
This thesis will deal with co-channel deployed pico cells, which means that they share spectrum with the macro cells. This means that we will create additional interference, macro BS to pico user, pico BS to pico user and pico BS to macro user. One way to reduce the effects of this interference is to deploy the pico cells indoors. Then the signals from the interfering layers would have to pass an outer wall before they start to interfere with each other. Pico nodes have a low transmit power and after passing a wall the most of the signal is lost. This will however also mean that they will have a limited coverage area, which gives them problem to capture a significant amount of traffic. Also since the traffic will increase in the future there are of great importance to offload the macro networks as much as possible.
However, the more traffic that is moved from macro-cells to pico-cells, the higher utilization
Interference
Interference
Interference
Figure 3.1: This is an illustration of a heterogeneous network
of the pico-cells. In chapter 4 an algorithm is described, which provides a method to give
maximum offload for the macro network.
Chapter 4
Small-cell deployment
There are no exact methods for pico cell deployment. There are certainly a multitude of reasons for that. The different indoor propagation scenarios and the fact that indoor pico cell deployment is a relatively new strategy are two reasons. However, increased computation power and more accurate simulation models makes it possible to do some estimations of how deploy the pico cells. Some guidelines to consider when deploying the pico cell are:
• Pico cells should be placed where the macro cell has the highest resource consumption, which means that the small cells should be placed where the macro cells has a low bitrate or where the traffic is high.
• Small cells should be placed to create dominance. In other words the pico cells should be placed where the macro-cell have a high path gain to remove as much traffic from the macro network as much as possible. This scenario is mainly applicable for indoor deployment and only when the first scenario already applies. Since the traffic is expected to escalate under the closest years, as much of the traffic as possible need to be removed from the macro cells, which means that the small-cells should be deployed to have a large serving area.
• Pico cells should not be placed in the same room as a window. This is because the windows may have a low path loss. The signal is then reflected on the wall and out through the window. This leads to that the most of the signal will end up outdoor creating interference for the macro cells instead of serving the indoor users which they are intended to. This criterion will not be tested since the simulator disregards windows and indoor walls.
Also there are already some a priori information about how the small cells should be deployed and this information can always be used as a base for the optimization.
4.1 Deployment Algorithm
The resource consumption RC explains in some sense how much time it takes to transfer one bit. A BS is able to transmit a limited amount of data for a given time and to maximize the capacity of the BS, the transmission time per bit needs to be as short as possible. The resource consumption can be calculated as
rc = 1 R b
, (4.1)
where R b is the spectral efficiency of the the channel. However if all the users are given the same bandwidth it is then possible to use Equation (3.8) define the resource consumption as
rc = B
b , (4.2)
where B is a constant and takes the same value for all the users. Since the only bitrate considered in this thesis is the maximum achievable bitrate, the bandwidth is the same for all the users and will not have any effect on the deployment. This makes it possible to define RC as
RC = B × rc. (4.3)
The rest of this report RC will, a bit sloppy, be called the resource consumption as well.
To remove as much traffic as possible from the macro cell, the small-cell have to be placed on a location where it can serve an as large area as possible, which also have a high resource consumption. Removing traffic from a cell can also be called offload and will be called this for the rest of the report. To find this location, begin by defining an area which is then split up into a grid of equally sized bins. For an indoor situation, which we have in this thesis, all the floors need to be split up in bins as well. It is important that the bitrate is known for each of the bins. Then deploy a small-cell on location ~x and define its coverage area as A(~x).
The resource consumption for the area A(~x) can then be the calculated as RC(A(~x)) = X
i∈A(~ x)
1 b i
, (4.4)
where b i is the bitrate for bin i and i refers to a bin served by a macro cell. The small-cell is then to be placed at ~x opt , where ~x opt fulfils the criterion
RC(A(~x opt )) = sup RC(A(~x)). (4.5)
By deploying small-cells according to this criterion, each small-cell that is deployed will give a maximum offload to the macro cells.
However, a problem with this approach is that only the macro cells are considered. Equa- tion (4.4) does only take the macro performance into account and this will in some cases affect the performance of the newly deployed small-cells. A sparse deployment of small-cells in areas where the received macro signal is already weak may not improve the bitrate as much as expected. Since the received macro signal is weak, the small-cell signal can be weak as well and still overpower the macro signal. Some users may then be served by a small-cell and still have a weak received signal. This problem can however be solved by adding another term to Equation (4.4). Let j denote the bins that is served by other small-cells and redefine Equation (4.4) as
RC(A(~x)) = X
i∈A(~ x)
w 1
b i
+ X
j∈A(~ x)
w 2
b j
, (4.6)
where w 1 and w 2 are weighting constants that can be adjusted and w 1 + w 2 ≡ 1. The second term in Equation (4.6) will then serve to offload the small-cells. We know however, that the small-cells in general will have a low utilization due to their small serving area and thus they will not get overloaded so easily. Then it is clear that the second term in Equation (4.6) has a different purpose. The only purpose of that term is to improve the bitrate for the small-cells.
Further it is a fact that the area closest to a small-cell will have a high bitrate so these bins should be excluded from the second term of Equation (4.6). Hence the equation could be rewritten as
RC(A(~x)) = X
i∈A(~ x)
w 1
b i
+ X
j∈A(~ x)
w 2
b 10% j , (4.7)
where b 10% j is the worst 10 th percentile of the small-cell users. It is not necessary to have a limit at the 10 th percentile, but a too high limit will cause the small-cells too stack up in some locations.
For best performance an update algorithm may be needed to recalculate the bitrate after each deployment. This can be done by first creating an initial deployment. The deployment can be done either by hand according to some given guidelines, or by running the algorithm described above with an expected serving area per small-cell and with w 1 set to 1 and w 2
set to 0. Then remove each of the small-cell nodes one at the time and replace it according to Equation (4.5). When the deployment converge, which means that the small-cell node is replaced on the same location from where it was removed, the algorithm has reached an optimum. It may not be the global optimum, but it will at least be a local optimum. This is however to be expected. In a large building the furthest small-cell nodes will not correlate with the closest since the path loss will be too high. However the small-cells are not bound to one area. Moving one cell will then not affect the whole building. This means that the global optimum will not be reached until all the local optimums are reached. To achieve this an initial deployment is required, which makes it possible to reach all these local optimums.
Problem with a limited impact on the result is that it would require infinite resolution maps to find the optimal positions. This makes it hard to reach complete convergence.
However it is not necessary, since small movements of the small-cell nodes will have a minor
effect on the global performance.
Chapter 5
Experimental setup
To investigate the deployment algorithm in Chapter 4, it have implemented it for three different buildings with different properties. The algorithm is the used to find out what happens when different numbers of nodes are deployed. The loads have been simulated for 8 different average loads for the complete map. The lowest load is 1 Mbps/km 2 and then the load increases in steps of 6 Mbps/km 2 until it reaches the maximum load on 43 Mbps/km 2 .
5.1 Building Information
The section below gives a short description of each building and its properties. The reference deployments are listed as well.
Building 1
The building is 65 m tall (14 floors of 4.5 m each) and has a footprint of 120 m×120 m. This can make it hard for the macro network to reach the users in the middle of the building. Here one pico cell will have a too small coverage area to be able to cover the a whole floor which means that multiple pico cells needs to be deployed on every floor. This building has a weak received signal strength on the bottom floors which can be seen in Figure 5.1c. On the upper floor the received signal is quite strong from all directions. Also the interference is higher which can be seen in Figure 5.1d. This building is highly loaded which means that the user throughput can be very low. User throughput and maximum achievable bitrate for all three buildings can be seen in Appendix A. For this building two different reference deployments have be used. Deployment number 1 is dense and will be referred to as the dense reference deployment. Deployment number 2 is sparse and will be referred to as the sparse reference deployment.
1. Four pico cells per floor and 56 in total. This deployment is dense and has the ability to cover most of the building if the pico cells are placed correctly. Here it is important that the pico cells have a good indoor performance. There can be no other motivation to why so many pico cells are to be deployed in this building. The nodes are placed in each quarter of the building and on the same place on every floor. This reference deployment can be seen in Figure 5.1a.
2. Two pico cells per floor and 28 in total. This is a good estimation to how many cells
that are needed to get enough indoor coverage. Here offload is more important than
indoor performance. The cells are deployed to complement each other and users on the
adjacent floors are in some places getting the strongest signal through the floor. This
deployment can be seen in Figure 5.1b.
−480
−430
−380 550
600 650
10 20 30 40 50
m
node positions
m m
(a) The dense pico cell reference deployment
−480
−430
−380 550
600 650
10 20 30 40 50
m
node positions
m m
(b) The sparse pico cell reference deploy- ment.
−480
−430
−380 550
600 650 10 20 30 40 50 60
m
Strongest Macro Received Power
m m
−120 dBm
−110 dBm
−100 dBm
−90 dBm
−80 dBm
−70 dBm
−60 dBm
(c) The rsrp for the macro network.
−480
−430
−380 550
600 650 10 20 30 40 50
m
IoT [dB] DL, network load 19 Mbps/km2
m m
0 dB 10 dB 20 dB 30 dB 40 dB
(d) The interference to noise ratio for the macro network for a high load.
Figure 5.1: Rsrp interference to noise ratio and reference deployment for building 1
Building 2
This building has the footprint in the same size as building 1, 120 m×120 m. However the building is not a square block as the top floors have a different area than the rest of the floors.
The lowest height of the building is 60 m while it is 70 m on the highest point. Also the
building has a strong server on one side of the building. This gives that side of the building a
higher user throughput and bitrate than the other side. Appendix A shows user throughput
and bitrate for all three buildings. The received signal power of the building is in general
higher than for building 1 and is displayed in Figure 5.2c. The interference is also higher on
that side but in general the interference is low as seen in Figure 5.2d. This building has two
different reference deployments which can be seen in Figure 5.2a-5.2b. Also here deployment 1 is dense and deployment 2 is sparse. These deployments will be referred to as the dense reference deployment and the sparse reference deployment as well.
1. Four pico cells per floor and 56 in total 2. Two pico cells per floor and 28 in total.
−430
−380 400
450 500
10 20 30 40 50 60
m
node positions
m m
(a) The dense pico cell reference deployment
−430
−380 400
450 500
10 20 30 40 50 60
m
node positions
m m
(b) The sparse pico cell reference deploy- ment.
−460
−410
−360 400
450 500 10 20 30 40 50 60
m
Strongest Macro Received Power
m m
−120 dBm
−110 dBm
−100 dBm
−90 dBm
−80 dBm
−70 dBm
−60 dBm
(c) The rsrp for the macro network.
−460
−410
−360 400
450 500 10 20 30 40 50 60
m
IoT [dB] DL, network load 19 Mbps/km2
m m
0 dB 10 dB 20 dB 30 dB 40 dB
(d) The interference to noise ratio for the macro network for a high load.
Figure 5.2: Rsrp interference to noise ratio and reference deployment for building 2
Building 3
The building has a footprint of around 1500 m 2 . The building is more than 250 m tall (64 floors of 4.5 m each). The upper floors have a higher receiver signal power since the bottom floors are shadowed by other buildings which can be seen in Figure 5.3c. The received signal power in this building is not that low, but instead this building is covered by many macro servers. Because of its height this building can be served by distant BSs which have a high antenna gain and line of sight to the upper floors. This means the building experience a high interference which will lead to low bitrates for highly loaded networks. The Figures 5.1d, 5.2d and 5.3d shows the interference to noise ratio and there it can be seen that while the other buildings are noise limited this building is rather interference limited. The small footprint of the building suggests that more than one pico cell per floor will be too many.
For this building two different reference deployments are used, where deployment 1 will be called the dense reference deployment and deployment 2 the sparse reference deployment.
1. One cell per floor which means 60 pico cells. The cell is deployed in the middle of each floor and above each other. The deployment can be seen in Figure 5.3a
2. One cell per every second floor which means 30 pico cells. The cells are placed in the
middle of the building here too as in Figure 5.3b.
720 −120 740 10 40 70 100 130 160 190 220 250
m
node positions
m m
(a) The dense pico cell reference deployment
720 −120 740 10 40 70 100 130 160 190 220 250
m
node positions
m m
