Performance Evaluation of Self-Backhaul for Small-Cell 5G ...

UPTEC E 18018

Examensarbete 30 hpJuni 2018

Performance Evaluation of Self-Backhaul for Small-Cell 5G Solutions

Martin Hellkvist

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

Performance Evaluation of Self-Backhaul forSmall-Cell 5G Solutions

Martin Hellkvist

This thesis evaluates the possibility of using millimeter waves of frequency 28GHz for the use of wireless backhaul in small cell solutions in the coming fifth generation mobile networks. This frequency band has not been used in preceding mobile networks but is undergoing a lot of research. In this thesis simulations are performed to evaluate how the high frequency waves behave inside a three dimensional grid of buildings. The simulations use highly directive antenna arrays with antenna gains of 26dBi.A main results of the investigation was that a high bandwidth of 800MHz was not enough to provide 12Gbps in non line-of-sight propagation within the simulations. Furthermore, without interference limiting techniques, the interference is probable to dominate the noise, even though the high diffraction losses of millimeter waves propose that interference should be very limited in urban areas.

Tryckt av: UppsalaISSN: 1654-7616, UPTEC E 18018Examinator: Tomas NybergÄmnesgranskare: Mikael SternadHandledare: Erik Larsson

Sammanfattning

Framtidens mobila natverk kommer erbjuda uppkopplingshastigheter upp till 100 gangerde hastigheter som vi anvander idag. En central del i utvecklingen av de snabba natverkenar anvandandet av sandningsfrekvenser som 28 GHz. Dessa frekvenser ar mer an 10ganger hogre an de som anvands idag.

Det som ligger i grunden for forvantningarna pa sa hoga hastigheter ar den vidabandbredden i det nya frekvensbandet. Bandbredder forvantas vara upp till 800 MHz ellermer. En annan forvantan pa de nya natverken ar att natverksmaster kommer installerasmycket tatare an vad de gjorts tidigare, och avancerade antenn- och sandtekniker kommeranvandas for att isolera dem fran varandra. For att halla nere kostnaderna for de nyatatt placerade natverken forvantas tradlosa lankar installeras mellan masterna, for attbegransa installationskostnaderna for fiberoptiska kopplingar mellan dem.

Denna exjobbsrapport behandlar just den tradlosa kopplingen mellan de nya hogpresterandenatverksmasterna. Rapporten beskriver simuleringar som har gjorts av exjobbaren foratt utvardera prestanda och analysera vilka utmaningar som finns i detta hogfrekventaband.

Simuleringar har gjorts i en virtuell tredimensionell miljo i en MATLAB-baserad sim-ulator som utvecklats av Ericsson AB i Kista, Stockholm.

En egenskap hos de hoga frekvenserna ar att radiovagorna inte tar sig igenom ellerforbi byggnader och andra fasta material lika latt som lagre frekvenser. Detta foreslar attvi borde kunna installera master i stadsmiljoer sa att de isoleras naturligt av byggnaderna,och pa sa satt ger lag storning sinsemellan. Simuleringarna har visat att detta intenodvandigtvis ar enkelt att forverkliga, utan hoga storningar kan komma att existera pagrund av korta avstand och valdigt fokuserade antenner.

Generellt, sa visar simuleringarna att hoga hastigheter av flera Gigabits per sekundkan uppnas i stadsmiljoer, aven dar natverksmaster inte har klar sikt sinsemellan. Dettahar observerats i simuleringarna i avstand upp till 250 meter i stadsmiljon.

3

Contents

1 Introduction 11.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.2 Evaluation by Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . 21.3 Contribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21.4 Summary of Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

2 Theoretical Background 42.1 Path Loss and Propagation . . . . . . . . . . . . . . . . . . . . . . . . . 4

2.1.1 Path Loss Models . . . . . . . . . . . . . . . . . . . . . . . . . . . 52.1.2 Ray Tracing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52.1.3 Diffraction and Scattering . . . . . . . . . . . . . . . . . . . . . . 6

2.2 Noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62.3 SNR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72.4 Antenna Gain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72.5 Capacity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

2.5.1 SNR as Function of Capacity . . . . . . . . . . . . . . . . . . . . 102.6 Link Margin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102.7 Cellular Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

2.7.1 Backhaul . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

3 Millimeter Wave Networks 133.1 Challenges of Millimeter Wave Transmission . . . . . . . . . . . . . . . . 13

3.1.1 Smaller Antennas Increase Path Loss . . . . . . . . . . . . . . . . 133.1.2 Phase Noise in High Frequency Oscillators . . . . . . . . . . . . . 14

3.2 Simulation Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143.3 Standardized Path Loss Model . . . . . . . . . . . . . . . . . . . . . . . . 143.4 5G System Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

4 Simple Scenarios, Empirical Models 164.1 Single-hop Scenario . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164.2 Shared Access and Backhaul Scenario . . . . . . . . . . . . . . . . . . . . 19

5 Evaluation By Simulations 255.1 Deployment Performance . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

5.1.1 Link Budget . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 255.2 Simulation Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

5.2.1 Channel Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 265.2.2 General Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . 26

5.3 Evaluations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

5.3.1 One Base Station . . . . . . . . . . . . . . . . . . . . . . . . . . . 295.3.2 Comparison to 3GPP’s Empiric Models . . . . . . . . . . . . . . . 325.3.3 Deployment Proposals . . . . . . . . . . . . . . . . . . . . . . . . 35

6 Discussion And Conclusions 406.1 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40

6.1.1 Range And Backhaul Bandwidth Allocation . . . . . . . . . . . . 406.1.2 Interference Between Base Stations . . . . . . . . . . . . . . . . . 416.1.3 Interference Between Base Stations and Users . . . . . . . . . . . 426.1.4 Performance In Deep NLOS . . . . . . . . . . . . . . . . . . . . . 426.1.5 Deployment Proposals . . . . . . . . . . . . . . . . . . . . . . . . 436.1.6 Comparison to Empiric Models . . . . . . . . . . . . . . . . . . . 436.1.7 Base Stations Mounted on Lamp Posts . . . . . . . . . . . . . . . 436.1.8 Highly Directive Antennas . . . . . . . . . . . . . . . . . . . . . . 44

6.2 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 446.2.1 Open Issues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

Abbreviations

3D three dimensional

5G fifth generation wireless systems

CoMP coordinated multipoint

EIRP effective isotropic radiated power

EMR electromagnetic radiation

FBR front-to-back ratio

FDM frequency division multiplexing

FTTC fibre to the curb

HPBW half-power beam width

IAB integrated access and backhaul

ISD intersite distance

LOS line-of-sight

mmW millimeter waves

MBB mobile broadband

NLOS non-line-of-sight

SNR signal-to-noise ratio

SINR signal-to-interference-plus-noise ratio

TR technical report

UE user equipment

UMi urban micro

Chapter 1

Introduction

To respond to the as ever increasing demand of data rates and broadband availability, thefifth generation wireless systems (5G) are under way. Examples of use cases for 5G includesmart vehicles, media streaming everywhere, critical process control and interaction-human Internet of Things [9, 8].

A central part of 5G is operational frequencies in the millimeter waves (mmW) bands,e.g. 28-39 GHz. The benefit of these frequencies is the large available bandwidths, inthe magnitude of 1 GHz, facilitating very high data rates. Lower frequency bands offermuch lower bandwidths, limiting the achievable data rates. However, the weak propaga-tion properties of mmW limits the operational ranges, especially in urban environments,where there are many blocking objects hindering the path of mmW. This would make itproblematic to deploy large cells in such environment, due to the dependency of line-of-sight (LOS) propagation. An interesting opportunity is to employ self-backhaul in smallcells and perhaps to share the bandwidth between backhaul and access.

This thesis presents a high-level feasibility analysis of the concept self-backhaul insmall cell deployments with respect to coverage and capacity. The main part includesmodeling and evaluations of advanced simulations in a 3D Manhattan grid.

1.1 Background

Backhaul defines the network link that connects a pair of base stations (BS-BS), orconnects a base station to the core network (BS-NW). Self-backhaul in particular is whenthe backhaul and access (BS-User) share channel resources such as time, frequency andspace. In this report, backhaul refers to the wireless link between base stations. Figure1.1 shows a schematic overview of a few base stations with wireless backhaul betweenthem, one of which is connected to the core network with optic fiber. The advantages ofwireless backhaul are mainly two-fold. Firstly, it would serve as a cost effective alternativeto backhaul over optic fiber. A lower need for fiber installation means less cable to digdown in the streets, which would perhaps make the project of deploying a network easier.Secondly, it would be an overall simplification using the same technology for access andbackhaul.

As mentioned, mmW have very weak propagation properties and is also prone toblockage leading to outage. This naturally limits the viable cell sizes for deployments.Small cells means that LOS channels are more likely to occur, or at least non-line-of-sight (NLOS) with strong first component, above street level where the backhaul linkswill be. Assuming high performance wireless backhaul links are available, dense small

1

Figure 1.1: Schematic overview of a network using wireless backhaul between base stations, while theyserve access to users. Only one base station is directly connected to the core network.

cell deployments could be a good solution for increasing data rates without increasingcosts of fiber installations, which can be a large portion of the installation cost.

To compensate for the large path losses, highly directional antenna arrays can beemployed. Due to the small wavelengths of mmW , making antenna elements small, alarge amount of small antenna elements may be used to cover a small area, increasingthe gain. With a high number of antennas comes high directivity (section 2.4), whichcan decrease interference. The fact that mmW penetrates solid materials badly impliesthat the interference will be limited between cells separated by buildings. However, theresults will show reflections to be rather substantial in NLOS .

If the wireless backhaul link can exceed the needed backhaul data rate per site, itwould be viable to construct so-called multi-hop networks. That means base stations orsites are connected with two or more wireless links between them and the core network,further decreasing the need for fiber installations.

1.2 Evaluation by Simulation

The analysis presented uses simplified virtual 3D models of a Manhattan grid cityscape.Base stations and user terminals are placed in the three dimensional space and the pathgains can be computed between these nodes. Path gains are computed by taking intoaccount reflections, diffractions and scatterings (section 2.1.3) on the modeled buildings.

In the presented analysis, the performance of backhaul links are evaluated both bytheir respective capacity and robustness. The robustness is here evaluated in consider-ation to how the capacity or data rate changes if there would be any angular or spatialinaccuracies in the base station placements, as well as if sudden blockage would occur.

1.3 Contribution

The purpose of this report is to assess the feasibility of deploying outdoor (and indoor)mmW systems with integrated access and backhaul (IAB) for mobile broadband (MBB).The evaluations could be used as guidance for development of 5G networks, where mmWis expected to play a key role [2]. The assessment is done through link budget analysisand evaluations of simulations performed in Ericsson’s internal simulation environment.

2

More specifically, the thesis presents an analysis of the characteristics found in simu-lations of the Manhattan grid. The found characteristics can pose as guidance for eval-uating simulations in more advanced environments. Furthermore, the simulation resultsare compared to empirical path loss models presented by 3GPP. The evaluated scenariosare solely outdoor and on street level in between high buildings, and distances are limitedto up to 400 meters.

1.4 Summary of Results

The results presented in this thesis mainly handle multiple simulations of the Manhat-tan grid. The first part evaluates the overall mmW propagation from one specific basestation to others evenly distributed over the streets up to 240 meters away. The secondpart describes comparisons of a few simulation scenarios against the empirical models de-scribed by 3GPP. The last part evaluates four specific deployment geometries of two-hopbackhaul chains.

3

Chapter 2

Theoretical Background

The following chapter explains concepts vital for understanding the simulations and inter-preting the results. Mainly, the report will evaluate backhaul link performances affectedby the following aspects:

� Signal attenuation, captured by modeling path loss effects as diffraction, reflectionand scattering in combination with ray tracing.

� Transmitter attributes, such as antenna array structure, antenna gain, directivityand transmit power.

� Link sensitivity in relation to SNR and capacity requirements.

� Interference between different backhaul links and from users to backhaul links. Thebandwidth allocation will be taken into account here.

2.1 Path Loss and Propagation

As opposed to wired connections, the use of a wireless radio channel for communicationcomes with more susceptibility to noise, interference and blockages. This section aimsto identify and describe how the quality of a wireless channel changes due to path loss,shadow fading, and small-scale fading.

Path loss is a measure of the attenuation occurring when a signal propagates througha medium. If there are no blocking objects, the only path loss is the transmitted signalspreading out in space more the further it propagates. If a directional antenna is used,an antenna gain is used in combination with the path loss model, decreasing the pathloss in the antenna’s direction. A path loss model is most often deterministic, taking nostochastic shadowing effects into consideration.

Shadow fading is caused by electromagnetic radiation (EMR) being absorbed, re-flected, scattered and diffracted due to interaction with materials in its path. Whenmodeled for non-specific environments, the shadow fading is often described by a nor-mally distributed random variable with a certain variance.

Small-scale fading variates over small distances, often in the order of the signal wave-length. This is due to the transmitted signal taking multiple paths to reach the receiver.The different paths are of different lengths, causing constructive and destructive interfer-ence to occur at the receiver. Small-scale fading is often called multi-path fading [7].

4

2.1.1 Path Loss Models

A transmitted signal s(t) has the power Pt. Transmitted through a given channel, thereceived signal r(t) will have the power Pr. The path loss is then defined as

PL =PtPr

. (2.1)

Expressed in decibels, the path loss is the difference in dB between the transmitted andreceived power:

PLdB = 10 log10

PtPr

dB = 10(

log10 Pt − log10 Pr

)dB. (2.2)

A wireless channel cannot amplify the transmitted signal, so the path loss is non-negative.Therefore, the dB path gain is defined as the negative of the path loss [7]:

PGdB = −PLdB = 10 log10

PrPt

dB. (2.3)

In general, it is the direct path, or LOS path, and a few strong reflections of it, thatprovide the receiver with sufficient signal power. If there is no LOS component, andreflections are weak, the received signal power can become very low and the signal hardto detect.

Empirical Path Loss Models

The real propagation environment is often very complex to model accurately for mo-bile communication systems. Path loss models have been developed to describe typicalpropagation environments. These models mostly differ over different cell sizes, and if thedescribed environment is urban, suburban or indoor, and are based on empirical measure-ments. Models developed in a certain environment can not be assumed to be accurate ifapplied for another environment [7].

Site Specific Path Loss Models

When a path loss model is created for a specific 3D propagation environment, it is calledsite specific. However, the path loss model is still an approximation since we can onlymodel a finite set of propagation paths. Furthermore, the 3D environment can at its bestalso only be an approximation of reality.

2.1.2 Ray Tracing

To obtain site specific path loss models, a very common technique is to use ray tracingtechniques, tracing the transmitted radio signals wavefronts as simple particles, reflecting,diffracting and scattering them on the well defined objects in the propagation environ-ment. This allows us to see many different copies of the transmitted signal, which we callmultipath components. The different components can have individual path loss, shadow-ing, time delay, phase shift and frequency shift. By representing the signals as particles,simple geometric models can be used to model the reflections, diffractions and scatterings.As described by Goldsmith in [7], ray tracing techniques have shown to be accurate inrelation to actual measurements in terms of received power.

If the surfaces of the defined reflectors are not smooth, it is important to use statisticalapproximations to get a realistic rendering of the received signal.

5

2.1.3 Diffraction and Scattering

Diffraction is the phenomenon of electromagnetic waves “bending” around edges or cor-ners in their paths. In excess of blocking the LOS path of the channel, a corner causingdiffraction further increases the pathloss by several decades of decibels [7]. Dependingon the angle of incidence, more or less power remains around the corner. A more narrowangle of incidence reduces the power more than a wider angle.

Scattering occurs when a signal ray hits an object. It is basically reflection on a nonsmooth surface. Reflections and scatterings can create§ many rays, causing the path lossof a scattered ray being proportional to the squared length of its path, as opposed to onlythe direct length. Also a ray is not singular, but consists of a small cluster of rays whichall will be scattered of the surface.

A signal ray can be diffracted several times. This is in addition to being scatteredand reflected. Signal components that experience multiple diffractions or reflections willquickly lose power, so they are mostly negligible in relation to noise.

2.2 Noise

Any given receiving antenna will be affected by noise. The noise disturbs the process ofidentifying the received message. A fundamental contribution to noise in received radiosignals is thermal noise, which is generated by random thermal movement of electronsin the receiver circuits. Its power, PN , is a function of the antenna temperature T andchannel bandwidth B:

PN = kB T B, (2.4)

where kB = 1.38 · 10−23 J/K is Boltzmann’s constant. The noise is often expressedin decibels-milliwatts (dBm). Figure 2.1 shows the noise in dBm as function of thebandwidth in MHz when the temperature T = 290 K.

Furthermore, the noise is amplified by the receiver circuits noise figure. The noisefigure is the ratio between the power noise at the antenna, and the power of the noise afterit has been amplified by the receiver circuit. The magnitude of the noise figure is a resultof building practice of the devices, and a better noise figure implies a more expensiveproduct. In this report, the noise figure is a system assumption and a mathematicalderivation is excluded.

0 100 200 300 400 500 600 700 800

Bandwidth [MHz]

-110

-100

-90

-80

Noi

se p

ower

[dB

m]

Figure 2.1: Received noise power in dBm as function of the channel bandwidth in MHz. The noise powervaries from -94 dBm to -84 dBm over the range of 100 MHz to 800 MHz.

6

2.3 SNR

What determines if a received signal can be identified or interpreted without error, isthe relation between the intended signal and the amount of noise or disturbances on thechannel being recorded by the receiver. This relation is called signal-to-noise ratio (SNR)and is defined as the ratio between the beneficial power Pr and the power of the noisePN , both in watts:

SNR =PrPN

. (2.5)

And if the powers are expressed in dBm the SNR is expressed in dB:

SNRdB = Pr [dB]− PN [dB]. (2.6)

SINR

Often noise only includes background noise or thermal noise from components in thereceiver electronics. To include disturbances from other transmissions or communicationlines, the SNR can be extended to signal-to-interference-plus-noise ratio (SINR) whichrelates the received beneficial power to the received power I of non-beneficial messages:

SINR =Pr

PN + PI. (2.7)

2.4 Antenna Gain

Isotropic Antenna

C. A. Balanis states in [5] that an isotropic antenna is defined as “a hypothetical losslessantenna having equal radiation in all directions”. He continues with explaining that whilethis is not realizable, it is a tool for relating directivity of real antennas.

Directivity

The performance of an antenna can be measured by its antenna gain G [5]. It is oftenexpressed in relation to the performance of an isotropic radiator, denoted as dBi, ratherthan dB, for emphasization. The gain G is a function of spatial direction, hence the termdirectivity. The maximum gain is found in the antennas ‘direction’ and can in dBi beexpressed as a function of the antenna’s efficiency E and directivity D:

Gmax,dBi = 10 log10(ED). (2.8)

Thus, the performance can increase if the antenna focuses its output power in the de-sired direction, increasing D, and if the antenna is more efficient, increasing E. Due toreciprocity, an antenna has the same maximum antenna gain if it receiving as if it wastransmitting.

When multiple antenna elements are used in an antenna array, we assume that themaximum gain grows in proportion to the number of elements N :

GAS = NG ,

GAS,dBi = G+ 10 log10(N).(2.9)

7

The upper equation is in the linear domain, while the lower is the same expression indBi.

The maximum gain per element can according to [11] be approximated as function ofthe element spacing d and radiation wavelength λ:

G0,element =4πd2

λ2(2.10)

Radiation pattern

An antenna radiation pattern is a representation of how the antenna’s gain depends onthe outgoing or ingoing direction of the transmitted or received wave. Parts of the patterncan be divided into lobes which are separated by relatively weak radiation intensity. Asfigure 2.2 depicts, the lobe with highest intensity is called the main lobe while lobes withlower intensity are called side lobes. At an horizontal angle of 180° from the main lobe isthe back lobe.

The angular width of the points of the main lobe where its power is half its maximum,or 3 dB lesser, is called half-power beam width (HPBW). This is a measure of how wideor narrow the beam is. While a narrower main lobe makes it stronger, it will also result instronger side lobes. It is possible to have several side lobes. Side lobes are often radiationin undesired directions.

The front-to-back ratio (FBR) describes the ratio in intensity between the main andback lobe. It is often expressed in dB. [5]

- 0Horizontal angle from front

G-3 dB

G

Rad

iatio

n in

ensi

ty

Back lobeSide lobe

Main lobeHPBW

Figure 2.2: Radiation pattern with denotations of different lobes.

When a large amount of antenna elements are arranged in an array pattern, theradiation pattern becomes very complex. [11] presents rules of thumb for calculating theelement HPBW ΘH in radians as function of the element gain G expressed on the linearscale:

Θ2H,element =

32400

G. (2.11)

For an array, the HPBW is assumed to be proportional to the number of elements ineach direction (Nvert , Nhoriz):

ΘH, array horiz =ΘH,element,horiz

Nhoriz

,

ΘH, array vert =ΘH,element,vert

Nvert

.

(2.12)

8

Simplified Radiation Pattern

To model antennas, or antenna patterns for simulation can be somewhat complex. 3GPPdescribes in [2] a simplified radiation pattern, only defined by the horizontal and verticalHPBW, and the FBR of the radiation pattern. The gain in decibels as function ofhorizontal θ or vertical angle φ, is described by a parabola with global maximum inθ = 90°, φ = 0°:

AdB(θ ∈ [0°, 180°], φ = 0°) = −min

{12( θ − 90°

ΘH,array horiz/2

)2, Amin

},

AdB(θ = 90°, φ ∈ [−180°, 180°]) = −min

{12( φ

ΘH,array vert/2

)2, Amin

},

AdB(θ, φ) = −min{−(AdB(θ,φ = 0°) + AdB(θ = 90°, φ)

), Amin

},

(2.13)

where Amin is the FBR.A visualization of such a pattern will look as a parabola with a floor around it, given

that the HPBW is small enough. An example, with FBR = 30 dB and HPBW = 30° isshown in figure 2.3.

-180 -150 -120 -90 -60 -30 0 30 60 90 120 150 180

Angle [degrees]

-30

-20

-10

0

Nor

mal

ized

ant

enna

ga

in [d

B]

AdB

( =90, )

AdB

( , = 0)

Figure 2.3: The horizontal (red) and vertical (blue) cuts of the antenna radiation model from [2].

2.5 Capacity

Capacity is a measure of how many information bits per time unit can be transferredwithout error over a given channel.

The theoretical maximum capacity a given wireless link can achieve is described bythe Shannon capacity C [7]. It is defined as function of the channel bandwidth B in Hzand received SINR γ :

C = B log2(1 + γ) . (2.14)

Here the SINR is not expressed in dB but on the linear scale.When dimensioning the simulations later in the report, the actual interference power

will be ignored. Instead the SNR is used at first, but adding a so called interferencemargin. This term describes how much the experienced noise increases due to interference.Expressing all quantities in decibels, the expression for SINR with the use of interferencemargin is

γdB = Pr −N − Im dB . (2.15)

9

2.5.1 SNR as Function of Capacity

Given a certain capacity C and bandwidth B, the required SNR γ can be extracted from(2.14):

γ = 2C/B − 1. (2.16)

Which for C >> B (C in bps and B in Hz) can be approximated as

γ ≈ 2C/B , (2.17)

γdB = 10 log10(2)C

BdB , (2.18)

which is a linear relation for a fixed bandwidth B. Figure 2.4 depicts the relationship in(2.16).

2 4 6 8 10 12 14 16 18

Capacity [Gbps]

0

50

100

150

SN

R [d

B]

800 MHz

100 MHz

Figure 2.4: The required SNR for a given capacity at different bandwidths. The relationship is ratherlinear for each bandwidth. The curves represent bandwidths 100, 200, ..., 800 MHz, with 100 MHz beingthe highest curve, and each curve being lower per step in bandwidth.

2.6 Link Margin

Under a constraint on bit rate or capacity C of a given interference free channel, the linkmargin is a measure of how many decibels the SINR can decrease before the capacitybecomes smaller than C. A 20 dB link margin means that the channel can only suffer20 dB more attenuation or interference before its capacity drops below the required value.A negative link margin means the channel is insufficient for its requirement. It is in asense the ‘distance’ to the required SINR.

2.7 Cellular Networks

This thesis discusses cellular networks which are infrastructure-based wireless networks,as opposed to ad-hoc networks, where mobile units set up networks without any underly-ing infrastructure. The following section is a summary of the basics of wireless networksas described by Goldsmith in [7]. I recommend a reader fresh to the subject to readthrough it to get a basic understanding of how a wireless network functions.

Traditionally, an infrastructure-based wireless network employs a wired network ofbase stations to provide mobile terminals, i.e. user equipments, with network or Internet

10

access. It is the base stations which control the data flows between units or betweenunits and servers. Proximate base stations often work in coordination with each other tocontrol mechanisms like scheduling, frequency allocations and user handover, e.g. whena user comes closer to the another base station than the currently transmitting.

Cellular networks are systems where base stations are spatially separated, or geo-graphically separated, in order to reuse frequency bands efficiently. This is possible dueto very fact that the signal power decreases over distance. A cellular network divides aspecific area into non-overlapping sub-areas, often visualized as hexagons, or cells. Eachcell is allocated a certain set of channels that are either separated by frequency (FDMA)or time (TDMA). A specific set of channels is not solely used by one specific cell, but isreused in cells sufficiently far away. This limits the interference that can occur betweencells, the intercell interference.

Goldsmith further mentions that a good cellular system is interference limited, definingit as the interference power being much larger than the noise power. If there was roomfor more interference, more users could be added. In an interference limited system, anyinterference reducing technique would increase the capacity.

2.7.1 Backhaul

The term backhaul in this thesis report refers to the data flow between base stations,i.e. backhaul data links. In addition, the aforementioned coordination between cells insection 2.7 also take place over backhaul control channels. Wireless backhaul replacestraditional wired backhaul links with wireless channels using 5G technologies.

Assuming the total capacity need of users exceeds what the backhaul link can provide,the backhaul link becomes a bottleneck of the system. If the users and the backhaulshare resources, it becomes important to allocate them in a balanced manner, in orderto maximize performance.

Multi-hop Backhaul

In a network of base stations where most base stations are not connected to fiber, therewill exist wireless multi-hop backhaul links. In this section some identified key mechan-ics of multi-hop systems are highlighted in order to provide intuition for the evaluateddeployments later in the report. Assuming that all backhaul links can use the same fre-quency spectrum, and that all access cells connected to the sites have identical trafficdemands, the following points are raised:

� Backhaul links more hops away from fiber connection have lower capacity require-ment. This is because the traffic is evenly distributed over the small sites. Thelower capacity requirement allows a weaker backhaul link, allowing perhaps NLOSchannels relying on the diffraction and reflections creating low enough path loss.

� The first backhaul link will always limit the achievable throughput to all sites furtherdown the multi-hop chain. It therefore becomes very important to have a strongfirst backhaul link.

� Links further down the multi-hop chain can be of decreasing channel strength. Thisis due to the capacity demand decreasing further down the chain.

11

� Users on ground level should cause very little interference to the backhaul linksif the backhaul uses highly directional antenna systems in both horizontal andvertical direction. Problems can occur if frequencies are shared with users on similarelevation levels as the base stations.

12

Chapter 3

Millimeter Wave Networks

In response to the increasing demand of mobile network data rates, higher frequencybands are expected to be allocated for the implementation of coming 5G systems. Inaddition to the sub 6 GHz bands used in 4G LTE and 3G mobile networks, 5G frequencybands will make use of frequency bands of 28 GHz and above [8]. Waves of such frequen-cies (up to 300 GHz) are often called mmW, since their wavelengths are the size of onemillimeter to one centimeter. A key benefit of using mmW is the vast amount of availablebandwidth. The Third Generation Partnership Program (3GPP) Radio Access Nework(RAN) 1 specifies 50 MHz to 400 MHz bandwidth for the 28 GHz band in the first 5GNew Radio (NR) standard.

3.1 Challenges of Millimeter Wave Transmission

It has been shown that the use of mmW can provide high throughput. One exampleis the IEEE 802.11ad standard [1], supporting data transfer rates up to 7 Gbit/s in the60 GHz band for short range WLAN systems. On the contrary of the cases studied in thisthesis, systems like that are often under the assumption of indoor LOS channels. Thisthesis presents evaluation of solely outdoor systems, where LOS channels are possible,but not necessarily deployed.

The high frequencies of mmW suffer more path loss and are more attenuated bymaterials than low frequencies. Furthermore, mmW results in higher Doppler spreads.Consequently, there has been a fear that this could mean that mmW are less favorablewhere long range and high reliability are required [13].

However, these specific characteristics could be of advantage of small cell deployments,where wireless backhaul interference is limited due to the increased shadowing of mmW .Also, since mmW are less diffracted than lower frequency signals, this further limits theinter-cell interference during wireless backhaul. We here identify some key advantages ofself-backhaul for small cell deployments of 5G NR systems.

3.1.1 Smaller Antennas Increase Path Loss

As Goldsmith describes in [7], the received power in free space decreases with the signalwavelength λ:

PrPt

=

[√Glλ

4πd

]2, (3.1)

13

where d is the distance in meters, and√Gl is the the product of the transmit and receive

antenna radiation pattern in the LOS direction.This dependency is not derived from higher frequency signals having weaker propaga-

tion properties, but due to the antenna element size. Antenna elements are dimensionedto be in the same size as the wavelength of the transmitted signals. Therefore antennasfor mmW transmission is in the size of a few millimeters. However, this dependency canbe counteracted by constructing arrays of these small antenna elements. According toequation (2.10), the element gain can become π if the spacing is λ/2.

3.1.2 Phase Noise in High Frequency Oscillators

In modern radio systems, oscillators are used to generate signals at carrier frequencies inthe transmitters. Oscillators are not ideal, and introduce something called phase noise.This is a measure of how much a generated signal will deviate from the desired carrierfrequency. The value of phase noise is given in dBc/Hz (carrier decibels per Hertz) anddescribes how likely the generated signal will deviate by a given offset frequency ∆f fromthe desired carrier frequency fc.

The phase noise increases with the carrier frequency, and an increment from 3 GHzto 30 GHz makes it approximately 20 dB worse for any given offset frequency. The phasenoise will increase the error rate on a given radio channel and therefore limit the actuallyachievable data rate from the Shannon capacity.

The paper [10] released by Ericsson describes more in detail how phase noise can limita given radio system.

3.2 Simulation Method

To evaluate the specific system deployment geometries, an Ericsson internal Matlab-based simulation tool was used. In addition to simpler, empirical path loss models, itoffers site-specific propagation models including ray tracing.

As described in 2.1.2, ray tracing is the technique of following an electromagnetic wavefrom a transmitter to a receiver in a given propagation environment. In the simulations,ray tracing in combination with a half screen diffraction model finds the five strongestpaths between a given pair of a transmitter and receiver. The half screen diffractionmodel is described in [4], which furthermore discusses different concepts of the simulationenvironment used in this thesis work.

From the ray tracing a site specific path loss model is generated. The path loss willdepend on both the effective length of the trace, along with losses due to reflections,diffractions and scatterings.

3.3 Standardized Path Loss Model

Presented here are two empiric mmW path loss models, specified by the 3GPP technicalreport (TR) 38.901 [2]. Models are specified for different scenarios and for frequenciesfrom 0.5 GHz to 100 GHz. One listed scenario is the urban micro (UMi) street canyonscenario, representing a street with relatively high buildings along its sides. This is aninteresting scenario for this project. The model is developed for base station to userequipment (UE) communication. The height of the UE is limited from 1.5 m to 22.5 m

14

from the ground. With this, we can assume the model to be applicable for base stationto base station communication as well. The LOS and NLOS models are defined as

PLUMi-LOS = 32.4 + 21.0 log10(d) + 20.0 log10(fc) + ξ , ξ ∼ N (0, 42) , (3.2)

PLUMi-NLOS = 22.4 + 35.3 log10(d) + 21.3 log10(fc) + ξ , ξ ∼ N (0, 7.822) , (3.3)

where the carrier frequency fc is in GHz and the intersite distance (ISD) d in m. The equa-tions clearly depict that high frequencies are more attenuated than low frequencies. From(3.2) we see that, a carrier frequency of 28 GHz results in 35 dB (20 log10(28.0/0.5) ≈35 dB) more attenuation than for 0.5 GHz.

3.4 5G System Model

The network systems evaluated in this report represent a sub-part of an entire 5G systemas specified by 3GPP in [3]. The subject of evaluation is solely the wireless backhaul links,with reference to available access resources. The thesis does not discuss any underlyingarchitecture, but focuses only on the outermost physical layer of the channels betweendeployed transmitters and receivers, their respective Shannon capacities, and link marginsto given constraints upon the capacity.

The evaluations consider network deployment geometries with the use of highly direc-tional antenna systems. The interest lies in identifying challenging deployment scenariosand discuss these. The performance will be measured by the path loss of the backhaullinks in a given deployment scenario, and also associate this to Shannon capacity undersome assumptions on transmission power, bandwidth and antenna gain. The interfer-ence will not be taken into account here, but separate discussion will arise about howinterference can occur between separate backhaul links.

Antenna Types

Antennas of different complexities differ much in price and efficiency [6]. Thus, a trade-off arises between the two. A more efficient antenna has potential to be capable forlarger ISDs, leading to fewer sites and base stations, which could decrease the overalldeployment cost. On the other hand, it could be cost-efficient to deploy a more densenetwork with smaller and cheaper antenna systems.

This is motivation for the thesis to investigate how different choices of antenna systemscan affect the overall performance of a mmW network. For the simple path loss analysis,two extremes were analyzed to capture the overall characteristics; for a simple singleantenna system, the gain is assumed to be 0 dBi, while for a complex system the gainwas chosen close to 24 dBi.

After the path loss analysis, in the simulations of chapter 5, a more precise antennamodel is derived from the theory in section 2.4. This will show to have a maximum gainof 26 dBi and very narrow HPBWs.

15

Chapter 4

Simple Scenarios, Empirical Models

The 3GPP LOS and NLOS UMi street canyon scenario models were analyzed to get aninitial sense of the scenario and the characteristics of mmWs. Specifically, how the ISD,access and backhaul bandwidth split affect the overall performance of mmW networks.

The analysis was constructed using only the empirical path loss models suggestedby 3GPP in [2]. Two types of setups were analyzed. The first one describes a simpleconnection between two base stations. The other type is a model of multiple base stationsin a multihop system. The central site, i.e. the one connected to fibre to the curb (FTTC),is assumed to have infinite backhaul capacity to the core network, while the others havetheir backhaul capacity limited by the propagation model and antenna systems.

In practice, time-, frequency-, and spatial resources must be distributed betweenaccess and backhaul in order to optimize the particular deployments. The results andevaluations presented in this chapter considers performance of individual time staticbackhaul links. Therefore no assumption are made regarding up- or downlink duplexing,which in practice would divide the here discussed Shannon capacity depending on theratio between them. The following results regarding Shannon capacity only takes SNRinto consideration, leaving possible interference related problems for discussion in section6.1. If one was to assume a factor 0.5 for up- and downlink scheduling, all computedcapacities must be multiplied by 0.5 in order to give the capacity for up- and downlinkrespectively.

4.1 Single-hop Scenario

This scenario was designed to visualize the empiric path loss models in a rather simpleenvironment. This section will provide intuition in how the capacity of a backhaul linkchanges over distance and bandwidth, under different assumptions on environment andantennas.

This model assumes full antenna gain in the link direction. Assumptions of antennasand environment are kept fixed, while the only variating parameters are ISD and accessversus backhaul bandwidth split. The following system assumptions were made:

With the concepts described in chapter 2 together with the path loss models 3.2 and(3.3), the Shannon capacity can be expressed as a function of distance and bandwidth:by including the antenna gain in equation (2.1) we get

Pr(d) = Pt − PL(d) +Gt +Gr ,

16

Parameter Description Value

fc Central carrier frequency 28 GHzPt Transmission antenna output power 45.4 dBmGTx Advanced 64 element transmission antenna gain 24 dBiGRx Advanced 64 element receiver antenna gain 24 dBiT Temperature 290 KNF Noise figure 9 dB

Table 4.1: System assumptions for link budget analysis.

where Gt and Gr are antenna gains of transmitter and receiver, respectively. Exchangingthis for Pr in the SNR expression in equation (2.5) gives

γdB(B, d) = Pt − PL(d) +Gt +Gr − PN(B) ,

where B is the channel bandwidth and PN(B) is

PN(B) = PN0(B) + PNf ,

where PN0(B) is described by equation (2.4) and PNf is the noise figure of the receiver.The formula for Shannon capacity in equation (2.14) then becomes

C(B, d) = B log2(1 + γ(B, d)) = B log2(1 + 10γdB(B,d)/10) . (4.1)

The two empiric path loss models were used in equation (4.1) with the assumptionsin table 5.2 to generate the following results, summarized by figures 4.1 through 4.4.

0 100 200 300 400

ISD [m]

0

5

10

15

20

25

30

C [G

bps]

100 MHz400 MHz800 MHz

Bandwidth

(a) A backhaul link’s Shannon capacity asfunction of ISD in the LOS case. Dashedlines show the capacity with interferencemargin Im = ±3 dB intervals.

200 400 600 800

BW [MHz]

0

5

10

15

20

25

C [G

bps]

10 m50 m100 m

ISD

(b) A backhaul link’s Shannon capacity asfunction of bandwidth.

Figure 4.1: Plots of backhaul LOS link capacity, with use of high gain antenna systems on both ends.

For LOS and high gain antennas (24 dB maximum antenna gains at both transmitterand receiver), see figure 4.1, it was found that channels could achieve capacities above10 Gbps over distances up to 400 meters, given a high enough bandwidth. At given dis-tances (10 m, 50 m, 100 m), the capacity was rather linearly dependent of the bandwidth

17

0 100 200 300 400

ISD [m]

0

5

10

15

20

25

30

C [G

bps]

100 MHz400 MHz800 MHz

Bandwidth

(a) A backhaul link’s Shannon capacity asfunction of ISD in the NLOS case. Dashedlines show the capacity when the Im =±3 dB intervals.

200 400 600 800

BW [MHz]

0

5

10

15

20

25

C [G

bps]

10 m50 m100 m

ISD

(b) Backhaul link Shannon capacity asfunction of bandwidth in the NLOS case.

Figure 4.2: Plots of backhaul NLOS link capacity, with use of high gain antenna systems on both ends.

0 100 200 300 400

ISD [m]

0

5

10

15

20

C [G

bps]

100 MHz400 MHz800 MHz

Bandwidth

(a) A backhaul link’s Shannon capacity asfunction of ISD in the LOS case. Dashedlines show the capacity with interferencemargin Im = ±3 dB intervals.

200 400 600 800

BW [MHz]

0

2

4

6

8

10

12C

[Gbp

s]

10 m50 m100 m

ISD

(b) A backhaul link’s Shannon capacity asfunction of bandwidth.

Figure 4.3: Plots of backhaul LOS link capacity, with use of isotropic antenna systems on both ends.

in the range of 100 MHz to 800 MHz. Here we see the massive benefit of the large band-widths, as was expected from the theory chapter. For short distances this relation holdsfor isotropic antennas as well, which is depicted by figure 4.3b. As the ISD increases, thecapacity decreases less. Also when the bandwidth is low, the capacity is less changed bythe ISD. Between 100 and 800 MHz the relation between capacity and bandwidth wasrather linear in the LOS case.

In NLOS (see figure 4.2) for the high gain antennas, we see a significant decrease incapacity compared to the LOS case. The bandwidth had to be 800 MHz to obtain 10 Gbpsat 200 m distance. Above 100 m, the capacities are around half the values shown in theLOS case. Furthermore, we see a weaker relationship between bandwidth and capacityin figure 4.2b; especially the 50 m and 100 m plots decrease earlier than in figure 4.1b.

Results for the zero gain isotropic antennas are showed in figure 4.3 and figure 4.4.It can be seen how the performance now is much worse than for the high antenna gain

18

0 100 200 300 400

ISD [m]

0

5

10

15

20

C [G

bps]

100 MHz400 MHz800 MHz

Bandwidth

(a) A backhaul link’s Shannon capacity asfunction of ISD in the NLOS case. Dashedlines show the capacity when the Im =±3 dB intervals.

200 400 600 800

BW [MHz]

0

2

4

6

8

10

C [G

bps]

10 m50 m100 m

ISD

(b) Backhaul link Shannon capacity asfunction of bandwidth in the NLOS case.Dotted yellow line represents capacitywith high interference margin Im = 3 dB.

Figure 4.4: Plots of backhaul NLOS link capacity, with use of isotropic antenna systems on both ends.

systems. In LOS , a backhaul link with isotropic antennas should be able to obtain above2 Gbps with 400 GHz up to several hundred meters. Within 100 m, above 3 Gbps shouldbe realistic. In NLOS however, the performance diminishes very quickly with the distancefor all three bandwidths. Furthermore, deploying isotropic antennas withing 50 m fromeach other to obtain several Gbps would probably increase the interference by much morethan 3 dB, and in that case further decrease the capacity of the backhaul links.

Although the system assumptions are quite generous, it seems unlikely to reach20 Gbps Shannon capacity in the backhaul link. However, 10 Gbps seems much morelikely, even in NLOS, given a high enough bandwidth.

To summarize the results:

� Zero gain isotropic antennas seem to need a very high bandwidth to achieve gigabit-speed channels. Assuming a high bandwidth increases interference between back-haul links, the isotropic antennas is a bad choice for wireless backhaul in LOS. InNLOS, the high path loss in combination with low antenna gain will always createtoo weak channels for wireless backhaul, even when disregarding interference andphase noise.

� High gain directional antennas are a good choice for both LOS and NLOS scenarioswhere many Gbps are required. Here we can further assume the capacity to belinearly dependent of the bandwidth.

4.2 Shared Access and Backhaul Scenario

This section presents an analysis of a multi-hop backhaul scenario with respect to accesscapacity. The purpose is to get a sense of the fundamental dynamics of backhaul inrelation to the access, mainly focusing on the split of bandwidth between them.

It is not trivial to set up a model for such an analysis. The practice of networkdimensioning is complex and one has to assume many things and simplify in order tocalculate capacities and data rates by hand. The system analyzed here is a model of

19

ISD

C1

FTTC

C2

ISD

A1 A2 A3 A4

Figure 4.5: Visual overview for simple analysis of access and in-band backhaul. A1 through A4 denotethe access data rate in each cell, and C1 and C2 denote the backhaul data rate of each backhaul link.

communication system in a simple form. Again, no regard is taken to the resourcedistribution between up- and downlink transmission. The computed capacities are thetotal Shannon capacities for the given physical channels.

The analysis uses the results from the previous section 4.1 regarding approximativelinearity between bandwidth and capacity.

Each base station is assumed to handle both one backhaul link to one other basestations, as well as access capacity to a cell with 5G users on the 28 GHz band. Thismeans that the access and backhaul are sharing the available 800 MHz of bandwidth onthis band. There is no overlap in frequency between them, they simply use differentportions of the band at all times.

The achievable access capacity in a cell is in reality not only bounded by the backhaulcapacity, but also by the users’ channel strengths. In the following analysis, it is assumedthat the users have strong enough channels to always make use of the whole backhaulcapacity.

The Model

The system model analyzed here consists of three unique MBB sites in a multi-hopbackhaul chain. The first site, S1, has one base station. This site has connection to aninfinite capacity FTTC link, and is the central and first node of the communication chain.The second site, S2, has two base stations, where both base stations have correspondingaccess cells. The third and final site, S3, consists of one base station that representsanother cell. Thus, there are four access cells in the system, and two backhaul links. Avisual overview of the model is provided in figure 4.5. One important assumption is thatbase stations linked on the same backhaul link must have the same bandwidth allocation.It is assumed that there is no interference between access traffic and the backhaul, orbetween the different backhaul links, which leads to the achievable Shannon capacities ofthe two backhaul links are only dependent on their respective bandwidth.

Algorithm

The method for the analysis is formed as an algorithm, initialized by setting a constrainton the capacity of A1, which limits the amount of bandwidth available for C1. Since thebase station on S2 that handles the backhaul data in C1 will have the same bandwidthallocation, the capacity of A2 must be upper bounded by A1.

20

The algorithm is designed with the goal of giving equal access data rates over all cells,i.e. A1=A2=A3=A4.

If A2 is not limited by C1, it is limited by A1 and therefore equal to it. Then C2+A3= C1–A2. To have an even distribution of access data rate, the aim is to get A3=A4,thus choosing them as A3=A4=(C1–A2)/2. A4 should not become higher than what C2can achieve, since A4 is smaller than C1/2, and C2 should be as potent as C1. Also, A4is limited by A3, due to having the same amount of access bandwidth.

Assuming the cell peak data rate is linearly dependent of the access bandwidth, thebackhaul bandwidth needed to supply S3 can be calculated easily. The last step of theanalysis should be checking the maximum capacity of C2, given the access bandwidth incell 4, to see if it is sufficient.

Both the access CA and backhaul CB capacity is assumed to be linearly dependenton the access bandwidth W . We saw in section 4.1 that for the directive LOS scenario,2.3 Gbps Shannon capacity was achieved at 100 MHz bandwidth and 50 m ISD. Also, itwas seen that the maximum capacity with 800 MHz bandwidth was around 20 Gbps. Thefollowing expression describes an approximation the relation between access capacity CA,backhaul capacity CB and the bandwidth W allocated for access, for a given base station:

CA(W ) =2.3 · 109

100 · 106W = 23W ,

CB(W ) = 20 · 109 − CA(W ) .(4.2)

The following is one example as calculated by hand. The method was implementedas a MATLAB script, and the results are presented after the example.

Example of Shared Backhaul and Access

The initial conditions are

CA1 = 2.3 Gbps, WA1 = 100 MHz ,

which leaves 700 MHz to the backhaul link between S1 and S2 and from (4.2) we get itsbackhaul capacity CC1:

CC1 = CB(100 MHz) = 17.7 Gbps .

And under the assumption that the capacity of cell A2 is equal to that of A1, we getthat:

CA2 = CA1 = 2.3 Gbps .

What now is left for cells A3 and A4 to share is the remainder of CC1 − CA2:

CA3 + CA4 = CC1 − CA2 = 17.7− 2.3 Gbps = 15.4 Gbps .

Assuming cells A3 and A4 have the same capacity:

CA3 = CA4 = 15.4/2Gbps = 7.7Gbps .

From (4.2) we can extract the needed access bandwidth of A3 and A4:

WA3 = WA4 = CA3/23 = 335 MHz ,

21

which allows us to compute the Shannon capacity of the backhaul link C2:

CC2 = CB(335 MHz) = 20 · 109 − 7.7 · 109 = 13.3 Gbps .

We see now that the system dimensioning is valid but unbalanced, giving much moreaccess capacity in cells A3 and A4 than in A1 and A2. The total system capacity was2 · 2.3 + 2 · 7.7 = 20 Gbps.

The next step toward finding a balanced system could be to increase capacity in A1and A2, thus decreasing the bandwidth for C1 and eventually capacities in A3 and A4.

Evaluation using MATLAB

0 100 200 300 400 500 600 700 800BW C1 [MHz]

0

200

400

600

800

BW

C2

[MH

z]

18

19

20

21

22

23

Tot

al c

ap [G

bps]

560 MHz

560 MHz

21.9 Gbps

BW C2Total system capacity

(a) The bandwidth of C2 (solid blue) as function of the bandwidth of C1. The total systemcapacity (solid red) as function of the bandwidth of C1 as well. The dotted red and blue linesmark the bandwidth and capacity at 561 MHz on C1.

0 100 200 300 400 500 600 700 800

BW C1 [MHz]

0

5

10

15

20

Cel

l cap

acity

[Gbp

s]

561 MHz

5.5 Gbps

A1A2A3A4

(b) The individual cell capacities as function of the bandwidth of C1. Between 400 and700 MHz all cells have substantial capacities.

Figure 4.6: The simple backhaul and access scenario system capacity as function of the bandwidth splitof the first backhaul link.

The method as explained and shown in the example above was implemented in MAT-LAB to evaluate different conditions on A1 and how it affects the system in total. Wehere analyze results from setting initial conditions on A1 from 0 to 18.4 Gbps. The resultsare visualized in figure 4.6.

Physical assumptions made for this investigation are:

� Antenna gain of 24 dB both at transmitter and receiver.

22

� Total bandwidth is 800 MHz.

� Inter-site distance is 50 m.

� Carrier frequency: 28 GHz.

� Constant noise power over the whole bandwidth.

� The computed link capacities are the Shannon capacity under full resource usagefor the respective links.

The plots generated by the algorithm show that bandwidths close to 756 MHz for thefirst backhaul link resulted in the maximum total capacity for the system. However thischoice of bandwidth gave very unfair or uneven resource distributions over the differentcells since the first cell A1 has very low access bandwidth available here.

The most fair distribution was found for bandwidth close to 560 MHz on both backhaullinks. Here the total was 22 Gbps, giving equal capacity of 5.5 Gbps in all cells.

It should be noted that there is not much variation in the total system capacity, thelowest being 18.4 Gbps and the highest 22.8 Gbps. These results are depicted in figures4.6a and 4.6b.§

When an additional site was appended on the chain after A4, introducing two addi-tional cells with capacities A5 and A6, it is expected that for an even system, the cellpeak data rate would decrease, since all cells except cell 1 is sharing C1. An evaluationwas done by the with an extended script.

The results of this evaluation shown in figure 4.7 that most fair system capacitywas 22.2 Gbps. Thus, marginally larger than the two-hop system. This gave all cells anavailable capacity of 3.7 Gbps, two thirds of the fair cell capacity in the evaluated two-hopsystem.

Note that the backhaul bandwidths are actually functions of the access data in therespective cells. That is why the bandwidths for links 2 and 3 are equal until A4, A5 andA6 decreases, close to 800 MHz in C1.

Evaluations were also done for a total bandwidth of 400 MHz, where the results weresimilar to the evaluations with the double bandwidth, but the values are halved for allmeasures. This is probably due to the linearity of the models.

One important take-away is that the best bandwidth split is probably above halfthe available bandwidth. This makes it impossible to fully orthogonalize the frequencyresource usage between different backhaul links. Imagine the split being perfectly atat half, then the every second link could use the lower half of the spectrum, and therest could use the upper, this would minimize interference from nearby backhaul links.Furthermore, more hops increases the needed backhaul bandwidth early in the chain.

Conclusions

� A two-hop network with 4 cells can demand higher rate per cell than a three-hopnetwork with 6 cells, if the cell sizes are the same in both systems. This is becausethe first backhaul link limits all cells on the lower side of it.

� The first backhaul link limits the whole system on its low side. If the remainingbackhaul links have alike properties and spatial distance, they will not limit thesystem. An opportunity here is to increase the ISD for these backhaul links, which

23

0 100 200 300 400 500 600 700 800

BW C1 [MHz]

0

200

400

600

800

BW

C2

[MH

z]

18

19

20

21

22

23

Tot

al c

ap [G

bps]

BW C2BW C3Tot sys capacity

22.3 Gbps

638 MHz

(a) The bandwidth of C2 (dashed blue) and C3 (dotted red) as function of the bandwidth of C1. Thetotal system capacity (solid red) as function of the bandwidth of C1 as well. The bandwidths of C2 andC3 are equal for all values of bandwidth of C1.

0 100 200 300 400 500 600 700 800BW C1 [MHz]

0

5

10

15

20

Cel

l cap

acity

[Gbp

s]

639 MHz

3.7 Gbps

A1A2A3A4A5A6

(b) The individual cell capacities as function of the bandwidth of C1. Between 400 and 700 MHz all cellshave substantial capacities.

Figure 4.7: The simple backhaul and access scenario system capacity as function of the bandwidth splitof the first backhaul link.

could increase the coverage area per base station. However, an increased coveragearea would in turn increase the number of users and cell peak rate.

24

Chapter 5

Evaluation By Simulations

This chapter describes the task and results of evaluating wireless backhaul using 28 GHzwaves. Simulations were performed in an Ericsson internal Matlab-based simulator envi-ronment. The simulations concern multiple deployments of base stations in a well definedManhattan grid of large buildings. The chapter starts with defining the relevant criteriafor determining the performance of the evaluated backhaul deployments, and continueswith describing the used simulation setup, and lastly presents results from different sim-ulations.

5.1 Deployment Performance

The end users’ experiences determine the performance of the network they are connectedto. Mainly, the experienced performance is measured by the achievable data rate oftheir channel, latency, error rate and overall connection stability. On a higher level, anindividual user’s capacity is also limited by the amount of users simultaneously sharingthe available resources of the base station.

The following analyses consider cellular systems under the assumption that the totalcell throughputs are only limited by the respective backhaul link. As a result, there is noneed to make assumptions about the users’ specific traffic types or individual capacitiesalthough it was shown in chapter 4 that the channel strength of the users can limit thetotal data rate in a cell.

The performance of the evaluated backhaul links will be measured in path loss andthe associated Shannon capacity of the interference free channel, together with the cor-responding link margin to three stages of target capacities:

� Low: 2.0 Gbps

� Medium: 6.0 Gbps

� High: 12.0 Gbps

The link margins will be used in a later discussion where factors like interference, backhaullink scheduling and physical robustness will be considered.

5.1.1 Link Budget

The link budget is used to supply approximate allowed minimum values of the path gainfor the three levels of target capacities. It can be used as a framework for designing

25

proper backhaul scenarios. A set of link budgets for three bandwidth cases is presentedin table 5.1.

Target capacity Low: Medium: High:

2.0 Gbps 6.0 Gbps 12.0 Gbps

Transmit power 40 dBm

Transmitter gain 26 dBi

Receiver gain 26 dBi

Interference margin 0 dB

Noise figure 6 dB

Bandwidth 100 400 800 100 400 800 100 400 800 MHz

Noise -94.0 -88.0 -84.9 -94.0 -88.0 -84.9 -94.0 -88.0 -84.9 dBm

Min required SNR 60.3 15.0 6.75 180.7 45.2 22.6 301.1 75.3 37.6 dB

Min required PG -67.7 -107.0 -112.2 52.7 -76.8 -96.3 173.1 -46.7 -81.3 dB

Table 5.1: Link budget examples for target capacities. The noise and noise figure was computed asdefined by sections 2.2. The minimum required SNR was computed by solving for γ in equation (2.14).The minimum required path gain PG was then computed by replacing Pr with Pt +PG in equation (2.6)and solving for PG.

The link budget table shows SNR values above 40 dB which would in reality would beimpossible to achieve due to limitations from interference, phase noise and other impair-ments introduced by the radio equipment [10].

5.2 Simulation Methodology

5.2.1 Channel Model

The modeling of the channels is done through ray tracing, described in general in theorysection 2.1.2. A well defined geometrical three dimensional (3D) city model is used aspropagation environment. The model is simply characterized by boundaries of the build-ings of the city. The model represents a large city with wide and long blocks of buildingsseparated by generous streets and sidewalks. The model’s dimensions are inspired by theUpper East Side of Manhattan, New York City. The model approximates each block ofbuildings as only one large smooth building. In reality, the blocks are more complex withvarying building heights with windows, ledges and different facade materials. To limitthe complexity, both in modeling and simulation, the model is a very simplified versionof reality. The large city model together with a map over a part of Upper East side isincluded in figure 5.1.

The Manhattan grid was quite large in relation to evaluated distances. Although noedge effects were ever observed, it seemed like good practice to over dimension it with12x12 building blocks.

5.2.2 General Procedure

This section describes the general procedure of finding the path loss from a given trans-mitter Tx to a given receiver Rx in the simulation environment.

The procedure can be step-wise described as:

(i) Creation of the Manhattan grid environment.

26

(a)

1000

500

y[m]

0

-500400

200

x [m]

0-1000 -200-400

(b)

Figure 5.1: (a) Map over a few blocks of Manhattan, NY. (b) Image over created Manhattan grid forpropagation environment.

(ii) Definition of positions and antenna properties of the isotropic antennas Tx and Rx.

(iii) Identification of the five strongest isotropic propagation path from Tx to Rx.

(iv) Redefining Tx and Rx as directional and aligning their main lobes along the strongestpath.

(v) Applying the respective antenna radiation pattern to the propagation paths andreturn the total path loss of the five strongest paths.

The site-specific propagation modeling used for the simulations is described as a wholein [4], but summarized by the following sections.

Finding the Strongest Paths

Ray tracing is used to find the possible paths between Tx and Rx. Diffractions, scatteringsand reflections are allowed along each ray. Each coordinate where that happens, is definedas an imaginary node, as opposed to the real nodes Tx and Rx. A ray can also be LOS.To limit the number of rays, only two imaginary nodes are allowed along any given ray,and the ray between each imaginary node must be LOS. Eventual diffraction, scatteringand reflection losses are applied and combined with the path losses between the imaginarynodes.

Diffraction

The computation of diffraction losses is performed with a reciprocal half-screen model asdescribed in [4].

27

-4 -3 -2 -1 0 1 2 3 4

Horizontal/vertical angle [radians]

-10

0

10

20

30

Ant

enna

gai

n [d

Bi] Horizontal

Vertical

Figure 5.2: The vertical and horizontal cuts of the antenna radiation pattern used in the simulations.

Scattering from Walls

The building facades are modeled as a mesh of diffuse scatter points. The rays fromdifferent scattering points are modeled individually and are very weak in comparison toLOS or reflection components, but can be combined to create a stronger sum component.In LOS-scenario where a wall is closely behind a receiver, the combination of scatteringscan become stronger than the LOS-component from the transmitter, which is not realistic.Therefore the scatterings are not combined while simulating LOS-scenarios. While inNLOS, not combining the scatterings would result in underestimation of the total pathgain since they are closer in magnitude to the strongest path, so in NLOS the scatteringsare combined.

In some simulations, both LOS and NLOS backhaul links are evaluated simultane-ously, the scatterings are combined for simplification of the simulation.

Antenna Models

The antenna model used for simulation approximates a single polarized planar arrayantenna with 8x16 (vertical x horizontal) elements. The side lobe levels are assumed tobe constant with same level as the backward lobe, so there is only the main lobe withspecified width and the remaining part of the pattern is constantly the minimum gain, asdescribed in the simplified antenna radiation pattern in section 2.4. The antenna modelis summarized in table 5.2, and visualized in figure 5.2, being the result of equation(2.13) which is used in the simulator. The number of elements in the array was chosenarbitrarily to get a narrow beam, while the element separation λ/2 is a common choicein antenna array design [5, 12].

The maximum element gain was computed with the rule of thumb in equation (2.10):

G =4π(0.5λ)2

λ2= π ≈ 4.98 dBi . (5.1)

The maximum array gain was computed as

Garray = 8 · 16 · π ≈ 402.1 ≈ 26 dBi . (5.2)

The array FBR is assumed to be 30 dB, making the minimum array gain

26.1− 30 = −3.9 dBi . (5.3)

28

# Elements 8x16

Element separation 0.5 λ

Max element gain 4.98 dBi

Max array gain 26 dBi

Min array gain -4 dBi

Element HPBW 101.6 Degrees

Array horizontal HPBW 6.3 Degrees

Array vertical HPBW 12.7 Degrees

Table 5.2: Summary of the used highly directive antenna model.

The element HPBW was computed according to the corresponding rule of thumb inequation (2.11):

ΘH =

√32400

π≈ 101.6° , (5.4)

and with equation (2.12), the array HPBWs became

ΘH, array horiz =ΘH

16≈ 6.3° , ΘH, array vertical =

ΘH

8≈ 12.7° . (5.5)

Lamp Posts

As mentioned, the base stations are thought of to be mounted on lamp posts in the model.However, the lamp posts are not modeled as any solid object in the city or propagationmodel. They are only used to reference the mounting height of the base stations. A lamppost is commonly four to eight meters high. In the model we assume the mounting heightof the base stations to be 6 meters. The propagation model used for evaluation does nottake ground reflections into account so the height of the base stations is not expected tohave a huge impact if far enough from the rooftops, since the propagations include aboverooftop diffractions.

5.3 Evaluations

The evaluations performed in this section are in reference to the target capacities Low(2.0 Gbps), Medium (6.0 Gbps) and High (12.0 Gbps), defined in section 5.1.1 and thecorresponding link budget in table 5.1.

The objective of the following evaluations is to see how well different deployments, i.e.base station placements, can perform. While the analysis is on a high level, the evaluationsconsider specific deployments in the Manhattan grid, with objective to capture significantcharacteristics.

5.3.1 One Base Station

Link budgets were created for the city model in order to relate the target capacities totarget path gains. To get an idea or inspiration for good site placements in order to reach

29

these targets, an initial look over the propagation environment was taken, only evaluatingthe propagation from one base station. This section describes the process and findingsfrom that initial look.

The simulations described here use combination of wall scattered ray traces for bothLOS- and NLOS-scenarios. Since these simulations are on a very high level, and theimpact of combination is only visible over very short distances, the effect of combiningscatterings or not should be low.

Sensitivity Of Placement

(a) Base station on coordinate (10,15,6). (b) Base station on coordinate (8,13,6).

Figure 5.3: The 3D-Manhattan grid as seen from above. The colored markers describes the respectivepath gains to the markers’ coordinates in the two mentioned placement-scenarios: when the base station(C) was placed (a) exactly on the building corner and (b) when it was 2 meters out in both x- andy-direction.

Initially, the base station was placed on xyz-coordinate (10,15,6) which is exactly ona building corner. Such a placement is obviously not realistic, since in reality two objectswould not be on the same exact point in space. However, this scenario is interesting fromthe sensitivities point of view.

Figure 5.3 presents the path loss evaluations for the two mentioned scenarios. Thecolored patches represent the path gains (reciprocal of the path loss) in decibels at thecorresponding coordinate. The path loss value for each receiver, i.e. marked coordinate,was computed after the base station (C) was directed along the strongest path to thatreceiver. The strongest paths were found as described in section 5.2.2.

This placement resulted in very high path losses, only sufficient to achieve High back-haul capacity within LOS and with 800 MHz bandwidth. Although, the link margins werevery large, around 25 dB at 50 m, and close to 10 dB at 200 m in the negative y-direction.In the opposite direction and along the x-axis, the path gain was lower; 9 dB at 50 malong the positive y-axis, and 15 dB at 50 m along the x-axis. In NLOS, i.e. around acorner, the path loss was so high only the Low capacity target was reachable with highbandwidth usage.

Low capacity could be achieved almost as far as two buildings away, almost 280 maway. Only very close to the first corners, Medium could be achieved in NLOS, where

30

Figure 5.4: The 3D-Manhattan grid from above, with colored markers representing computed path gainvalues from base station (C) in coordinate (40,13,6) to respective coordinate. In relation to figure 5.3b,this placement gives longer range around the closest corner and across the closest buildings. For eachreceiver, the base station is directed along the strongest found path.

the diffraction angles are still wide.The range could be greatly increased by moving the base station 2 meters out from

the building in both x- and y-direction, placing it in (8,13,6). This placement is moresimilar to as mounting it on a lamp post. The High capacity target was then reachablewith 400 MHz bandwidth, with 4 dB link margin in LOS at 50 meters and 0 dB linkmargin at 70 m. This placement had a more symmetric result around the base station,giving similar path gains along both axes. With 800 MHz bandwidth allocation, Highcapacity could be achieved several hundreds of meters away in LOS. The path loss wasalso good enough to reach High capacity in NLOS; a link margin of 0 dB with 800 MHzwas found 65 m from the first corner in NLOS. Across the buildings (along y = −220 m)in NLOS, Low capacity could be achieved with 400 MHz; 12 dB link margin across onebuilding (x = −80 m), 5 dB link margin across two buildings (x = −160 m), but −1.5 dBlink margin across three buildings (x = −240 m).

With 400 MHz, Medium capacity was achievable in NLOS, but not as far as acrossthe buildings.

That the second placement yields much stronger channels, displays the simulationenvironment’s sensitivity to how the transceivers are placed in relation to building walls.Also from these evaluations, we get an intuition in how far around corners good backhaullinks can be created. The reason for this effect is unclear, but the case of placing a basestation on the same coordinate as a wall is probably undefined, yielding strange results.

Under the assumption that there in a real deployment would be access devices alsousing mmW, we could perhaps predict that they as well would not reach around cornersvery good since they would have lower antenna gain than the base stations. We identifyan opportunity to install cells that are rather isolated, in terms of intracell interferencejust by the shadowing nature of the city environment.

Placement Along Short Wall

For the sake of comparison, the transmitting base station was also placed along one ofthe short walls of the grid, now at position (40,13,6). It is of interest to see how much

31

further the signals reach around the closest corners. The corners are closer here than inthe previous simulations, but that is what makes the angles of incidence wider, whichcreates less diffraction losses.

The LOS performance was comparable to that of placement (8,13,6), achieving Highcapacity with 400 MHz in rather close range, but very far with 800 MHz.

In NLOS this placement performed better than the previous. Around the closestcorner, there is a clear effect of a specular reflection, increasing the path gain by at least15 dB in relation to nearby coordinates. Within this reflection, the link margin to Highcapacity with 400 MHz was −7 dB, while the next sample 10 meters away had −23 dB.With 800 MHz, High capacity could be achieved up to 160 m past the corner, in thenegative y-direction, while in the opposite, it reached 0 dB link margin 85 m past thecorner. Along streets along the negative y-axis in x = −80 and x = 160, High capacitywas attainable up to 40 m past the corner. Medium capacity was here achievable with800 MHz almost all the way to (-80,-220) and (160,-220).

Low capacity was reachable with 400 MHz bandwidth over almost the whole mapshowed in figure 5.4. Some ranges along y = ± 220, where path gains dropped below−112 dB, was not reachable even with 800 MHz.

5.3.2 Comparison to 3GPP’s Empiric Models

For any given base station placement, we can compare the path loss to receivers atdifferent distances to see how similar the results are to the heuristic models in equation(3.2) and (3.3), given by the 3GPP standard 38.901 in [2], as was analyzed initially inchapter 4.

A comparison was made for both the corner and wall placements p1 = (8,13,6) andp2 = (40,13,6). Path gains were sampled both in LOS- and NLOS-scenarios, in a total of

Figure 5.5: Visualization of the three simulation scenarios to compare with the empiric models. Points(C) p1 and p2 are the placements (8,13,6) and (40,13,6), respectively.

32

0 50 100 150 200 250 300

Distance [m]

-150

-140

-130

-120

-110

-100

-90P

ath

gain

[dB

] S1

S2

3GPP Umi NLOS

From base station p2LOS

Figure 5.6: Comparison of the empiric 3GPP model, with two simulation-scenarios S1 and S2 in NLOSfrom base station in (40,13,6). The path gain values for S1 and S2 are the isotropic path gains, i.e. noantenna gain is included. The distance is the direct 3D-distance between transmitter and receivers. Thesteepest part of S1 and S2 are values from receives with LOS channels, which were not quite around thecorner. The first samples in both scenarios are in LOS, which is displayed as a short plateau, which isthen followed by a slope which is the receivers sweeping through the diffraction pattern. Furthermoreanother plateau is visible, which is the effect of the first specular reflection which increases the path gain.The path gain is isotropic, i.e. no antenna gains included.

three scenarios S1 , S2 , S3:

S1 = {(x, y, z) : x = 8m, z = 6m, y ∈ [−10.00,−10.01, . . . ,−250.00]m},S2 = {(x, y, z) : x = −72m, z = 6m, y ∈ [−10.00,−10.01, . . . ,−250.00]m},S3 = {(x, y, z) : z = 6m, y = −13m, x ∈ [0.00,−0.01, . . . ,−250.00]m},

which are present graphically in the Manhattan grid in figure 5.5.

NLOS Scenarios

For the placement p2, along the short building wall, only scenario S1 was close to theempirical model. For both NLOS-scenarios, S1 and S2, it was shown that the buildingcorners along the direct path between the base station and receivers had great impacton the path loss values. We know from [4] that the NLOS propagation model in simula-tion has less direct distance dependency than the 3GPP model, but that the half-screenmodeling is to compensate for further losses. Figure 5.6 presents the comparison betweenthe simulated NLOS propagation and the empiric UMi-NLOS model from 3GPP in [2].Since the empiric model does not include any antenna gains, they are excluded from thesimulation results. Scenario S1 is closer to p2, and has a wider angle of incidence for thediffraction, than in S2. The plots show that while the receivers in S1 have path gainsclose to the empirical 3GPP model, the results for S2 diverge a lot in comparison to S1.The values in S2 were equal to the 3GPP model at d = 132m, and 100 m further thediscrepancy was −7.5 dB, which is less than the 7.82 dB standard deviation of the shadowfading component of the 3GPP model. In NLOS and outside the first specular reflection,the discrepancy never exceeded 7.7dB.

One takeaway from this evaluation is that the 3GPP model is rather pessimistic in thequite interesting case where the receiver is closely placed around a corner, where there

33

0 50 100 150 200 250 300

Distance [m]

-150

-140

-130

-120

-110

-100

-90P

ath

gain

[dB

] S2

3GPP Umi NLOS

From base station p1

Figure 5.7: Comparison of results from simulation of scenario S2 with the base station in (8,13,6) withthe empirical 3GPP model. The distance is the direct 3D-distance between transmitter and receivers.As in figure 5.6, no antenna gain was included here in order to compare it with the empirical model.Similarities to the previous simulations show the first few samples in LOS, diffraction and the plateau ofthe first specular reflection. The path gain is isotropic, i.e. no antenna gains included.

can be a large spread of rays from the diffraction and a strong specular reflection. On thecontrary, the empiric model becomes optimistic in deeper NLOS with narrow diffractionangles.

From placement p1, only S2 represents a NLOS-scenario. Between 150 m and 280 mthe simulated values were within 2 dB of the empirical model. This is similar discrepancyto that of S1 from p2. This scenario closes in on the empirical model sooner than S1 fromp2. To come within 2 dB it took 38 m from the first sample in S2 from p1, while it took42.5 m to S1 from p2. This evaluation is presented in figure 5.7.

0 50 100 150 200 250 300

Distance [m]

-110

-100

-90

Pat

h ga

in [d

B]

S3, p1

S3, p2

3GPP Umi LOS

Figure 5.8: Simulated LOS-scenarios from both base stations show small discrepancy against the empiric3GPP model. The distance is the direct 3D-distance between transmitter and receivers. Apart from thelarge two peaks from the first placement, there is visible ripple along most part of the range. The pathgain is isotropic, i.e. no antenna gains included.

34

LOS Scenarios

The simulated LOS-scenario S3 were very close to the empiric 3GPP LOS UMi modelfor both base station placements. The simulated values were within 2 dB of the empiricvalues over the whole range of S3, except for very close to placement p1. Two localpeaks are seen here, as an effect of the receivers being close to the opposing corner fromthe transmitter. These effects are canceled via the antenna gain, and the sum becomessmooth. The same goes for ripple below half of a decibel, that occurs further away fromthe base stations along the range of S3. This ripple comes from moving along the discretescatter points on the closest building wall. Figure 5.8 presents the simulation result fromthese two scenarios, and comparison to the empiric model.

5.3.3 Deployment Proposals

This section presents multiple suggestions for how a deployment could be executed withone-, two- or three-hop backhaul links. The suggestions are based on the observations inthe previous link budget analysis.

The design of the deployment geometries was based upon each base station servingas backhaul infrastructure for a site it is connected to. Each site in the deployment wasassumed to have similar traffic need, so that a hop further down the chain would needless capacity than one early in it.

The performance of each deployment proposal was measured through the path gainand Shannon capacity of each backhaul link, and also the path gain to nearby basestations not part of a backhaul link. The latter path gain works as a pointer to theamount of interference there could be on that link, if nearby links would reuse time andfrequency resources.

Figure 5.9: Overview of deployment geometry proposal 1. The first link, between stations (C) 1 and 2is LOS, and thus very strong in terms of path gain and Shannon capacity. The second link, betweenstations 3 and 4, only needs to support data for one access point at the site of base station 4. This makesthe link need less capacity than the first, and therefore allows a weaker channel. The red arrows showthe main lobes’ directions.

Proposal 1

This proposed geometry, as visualized in figure 5.9, includes one two-hop chain of threesites with a total of four identical base stations with directional antennas. Two sites

35

are placed on the opposite side along the same vertical street, while the last site withthe fourth base station is placed around a corner from them. This limits the interferencebetween the first and fourth base station. The deployment should allow for good coverageon all four streets (x = 0m, x = −80m, x = −160m, y = 0m), as figure 5.3b suggests.

This geometry allows for a really strong first link between the stations 1 and 2, sincethe channel has a LOS-component. Under the assumptions that the capacity demands aresimilar for all three sites, we can allow a weaker backhaul channel between base station3 and 4, of at the most half the capacity of the first link. This specific geometry hasplaced stations 3 and 4 without a LOS component. The results presented in figure 5.7shows that the path gain between base station 1 and 4 should be very low, limiting theinterference between them.

The following matrix G1 (equation (5.6)) presents the simulated path gain valuesbetween all the base stations, where each row and column represents base station 1through 4, e.g. element (1,2) represents path gain between station 1 and 2, in decibels.The diagonal elements are crossed out, they are of no significant value.

G1 =

− −47.79 −77.81 −104.98

−47.79 − −55.32 −99.93

−77.81 −55.32 − −69.89

−104.98 −99.93 −69.89 −

. (5.6)

Again referring to the link budget example table 5.1, the simulation results show that thechannel between base station 1 and 2 has a path gain of −47.8 dB, giving it a −1.1 dB linkmargin to the High capacity target with only 400 MHz. The second backhaul link, i.e.between stations 3 and 4, achieves −69.9 dB path gain, leaving link margin of 6.9 dB toMedium capacity at 400 MHz, and 11 dB link margin to High with 800 MHz bandwidth.In relation to the Low capacity target, 9.2 dB link margin was achieved at only 100 MHz.From the gain matrix G1 we notice how the path gain from station 1 and 2 to 4 is around30 dB lower than the path gain between 3 and 4.

Base stations 2 and 3 are placed with 2 dm distance from each other, resulting inhigh path gain between them. That path gain is rather large and comparable to the linkbetween 1 and 2, and even larger than the path gain from 3 to 4. The path gain between1 and 3 is also rather substantial, being only 8 dB smaller than the link between 3 and4. Without employing any interference limiting techniques between these two links, thecapacity could be very limited by the interference.

Proposal 2

This geometry is similar to proposal 1, but with base station 4 and its corresponding sitemoved further down the street to y = −80 m. The first link remained unchanged. Inrelation to proposal 1, it is interesting to see how the interference between the backhaullinks change. The deployment is presented in figure 5.10.

While our intuition says that the path gain between base station 1 and 4 shoulddecrease due to the increased distance, compared to proposal 1, simulations actuallyshowed larger path gain. In comparison to the first proposal, the path gain increased by

36

Figure 5.10: Overview of deployment proposal 2. Here the distance has increased to base station 4,in relation to proposal 1. The aim here is to decrease the interference between stations 1 and 4, whilekeeping high path gain between 3 and 4, in relation to proposal 1.

(a) (b)

Figure 5.11: Ray traces from base station 1 to 4 in proposal geometry 1 (a), and in geometry 2 (b).Although the traces are longer in (b), they arrive more central in the main lobe of base station 4’santenna array, than they do in (a). Since the main lobe is only 6.3° wide, this has a lot of impact on theresulting path gain, rendering higher gain in (b) than in (a).

7 dB. Meanwhile, path gains from 2 and 3 to 4 decreased by approximately 12 dB:

G2 =

− −47.81 −77.79 −97.79

−47.81 − −48.37 −112.06

−77.79 −48.37 − −82.09

−97.79 −112.06 −82.09 −

. (5.7)

The changes resulted in that the difference in path gain between station 1 to 4 and from3 to 4 has decreased and is now around 15 dB. To summarize, the change of deploymentincreased the interference by 15 dB, and decreased the path gain by 12 dB, meaning thatthere now is a negative link margin to Medium capacity at 400 MHz and to High at800 MHz. The link margin is positive at 14.2 dB to Medium at 800 MHz.

The reason for the contra intuitive change in path gain, between station 1 and 4,seems to lie in the sensitivity of using highly directive antennas. A closer look at theindividual paths in both cases, show that 4 of the strongest 5 paths experience scattering

37

from close to the building corner in (−150, 15). In the second proposal, those scatteringsarrive more centrally in the main lobe of base station 4, while in the first proposal theyarrive further out in that main lobe. The paths for both geometries are showed in figure5.11.

Proposal 3

Figure 5.12: Overview of deployment geometry 3. As proposal 2 did not decrease the interference betweenstations 1 and 4, here base station 1 is moved further away instead of base station 4.

Another attempt of decreasing the path gain between station 1 and two was the thirdgeometry, presented in figure 5.12. Here base station 1 was moved further from the others,to x = 88 m, and base station 4 moved back to its initial position in y = −40 m. Thisincreased the distance between station 1 and 4 by almost 80 m. The direction of basestation 1 becomes more parallel to the x-axis, causing more gain in the direction of thediffraction point between base station 1 and 4, as well as to the scatter points close tocorner (−150, 15). This leads to more antenna gain at base station 1 instead of basestation 4. Overall, the change increased the path gain even more in relation to proposal1, than proposal 2 did. The gain matrix G3 shows that the path gain is now −91.29 dB:

G3 =

− −53.51 −83.52 −91.29

−53.51 − −55.32 −99.93

−83.52 −55.32 − −69.89

−91.29 −99.93 −69.89 −

. (5.8)

In addition to the interference increasing between base stations 1 and 2, there was adecrease in the path gain of the first link due to the increased distance, stepping evenfurther away from High capacity at 400 MHz.

Proposal 4

To decrease the interference between the links, the fourth geometry proposal places thefirst base station on the next street parallel to the x-axis. The resulting gain matrixG4 showed that the interference between base stations 1 and 4 decreased in relation to

38

Figure 5.13: Overview of deployment geometry 4.

previous proposals, but to the cost of the link margin of the first link decreasing by a lotas well:

G4 =

− −57.08 −86.20 −137.47

−57.08 − −55.32 −109.20

−86.20 −55.32 − −79.28

−137.47 −109.20 −79.28 −

. (5.9)

The link margin to High capacity with 400 MHz on the first link has now decreasedfrom −1.1 dB in proposal 1 to −10.4 dB. Meanwhile the link margin is high at 20 dB toMedium capacity with 800 MHz. This means the first link should end up between 6 and12 Gbps, while the second link would be between 2 and 6 Gbps, rendering a good balancebetween the links.

This deployment offers good coverage on only two y-parallel streets, but two x-parallelstreets, as opposed to the previous deployments, covering three y-parallel, but only onex-parallel.

39

Chapter 6

Discussion And Conclusions

Chapter 4 and 5 presented the outline and results of evaluating 28 GHz wireless back-haul for future 5G systems. Firstly, a simple analysis of standardized empirical modelswas conducted, followed by a rudimentary analysis of the concept of shared backhauland access. Lastly, multiple simulations of deployments in a Manhattan grid were per-formed using one of Ericsson’s internal simulation environments. Large scale propagationoverview, comparisons to the mentioned empiric models and also some specific deploy-ments of multi-hop backhaul chains were simulated and evaluated.

This chapter opens with some discussions about the results overall, and then tries tosummarize that discussion in section 6.2. In the last section of the chapter, Open Issues,problems, possible improvements and investigations are summarized.

6.1 Discussion

6.1.1 Range And Backhaul Bandwidth Allocation

The simulations presented in section 5.3.1 gave a clear overview of what capacity levelscould be achieved at different positions in the Manhattan grid. In relation to the linkbudget examples in table 5.1, it became clear that achieving High (12 Gbps) Shannoncapacity is a very hard task using 400 MHz bandwidth. In the corner placement out inthe street, as well as the placement along the short wall, only a small link margin of 4 dBwas found to High capacity at 50 meters away. However with a much higher bandwidthallocation of 800 MHz the range was increased by several hundred meters. This massiveincrease is due to the required path gain decreasing by almost 35 dB.

The long range with High capacity at 800 MHz is in line with the results in section4.1. It was there shown that 800 MHz bandwidth would give almost 15 Gbps Shannoncapacity in LOS with highly directive antennas. The results also agree that High Shannoncapacity could be achieved in NLOS under the same conditions.

The later simulations of deployment geometries also showed no tendency of achievingHigh Shannon capacity with 400 MHz bandwidth. The path gains on the first backhaullink, between base station 1 and 2 were between -47 dB and -57 dB. Using equation(2.14) for Shannon capacity, with the conditions from the link budget on noise figure andtransmit power, we get that -62 dB pathgain corresponds to 8 Gbps Shannon capacity at400 MHz bandwidth. Using this result together with the fair bandwidth split 560 MHzfor a two-hop system, that was found in section 4.2, and assuming linearity between

40

bandwidth and capacity, we get the Shannon capacity at 560 MHz and -62 dB path gain:

C ≈ 560

4008 Gbps = 11.2 Gbps ,

which is almost the High capacity target. This discussion tells us that the first link in thedeployment geometry proposals should with 560 MHz bandwidth allocation for backhaulbe able to have a few decibels link margin to the High capacity target.

Although the fair bandwidth split stated in section 4.2 showed to be 560 MHz, latterresults in the report have been focused on 400 MHz. The importance of a 400 MHzallocation would allow for the two backhaul links in the system to use the upper andlower half, respectively, of a 800 MHz wide spectrum. This would make the two backhaullinks completely frequency separated, minimizing interference between them. In such acase, the link budget’s assumption on zero interference margin could be close to valid, ifwe can neglect interference from eventual users.

To further discuss the issue of interference between the backhaul links, we can employintuition gained from the Shared backhaul and access scenario in section 4.2. It wasthere stated that the second link would always need lesser capacity demand than thefirst link. And further assuming equal capacity in all four cells of a two-hop system,the link 2 would actually only need one third of the capacity of link 1. The link budgetexample showed how -77 dB is needed to get 6 Gbps at 400 MHz. A link would with thesame linearity assumptions as earlier only need 266 MHz bandwidth to get 4 Gbps at thesame path gain. Geometry proposals 1 and 3 showed path gains of the second link to bearound -70 dB. If the 800 MHz spectrum would be allocated as 560 MHz on the first linkand 266 MHz on the second, only 26 MHz would overlap.

6.1.2 Interference Between Base Stations

The link budget example in table 5.1 is based on zero interference margin. The values onrequired PG and SNR to achieve a certain Shannon capacity could certainly be appliedon a given link, if the noise would dominate the interference. If we assume low to zerofrequency overlap between the backhaul links, this would be the case. However if thesystem was dimensioned with a large or full overlap, as when both link use 800 MHz,the interference could easily become higher than the noise. For example, lets look atgeometry proposal 1, assuming high frequency overlap between the links.

Example

If the received power at a given receiver is Pt−PL, adding the transmit power of 40 dBmwith the path gains PL from G1 in section 5.3.3 gives that the received power P4 at basestation 4 is approximately the path gain from 3 to 4 plus 40 dBm:

P4 = −70 dBm + 40 dBm = −30 dBm .

The interference I4 station 4 experiences from station 1, has a path gain of -105 dB, andwith the transmit power of 40 dBm the received interference power becomes

I4 = −105 dBm + 40 dBm = −65 dBm .

From the link budget table we know that the noise at 800 MHz bandwidth is N = −85 dBm.Hence we get SNR and SIR as

SNR = P4−N = −30 dBm+85 dBm = 55 dB , SIR = P4−I4 = −30 dBm+65 dBm = 35 dB ,

41

which means that the interference is dominating the noise.

In the example we saw that the SNR was 20 dB lower than the SIR, which means theSINR will be approximately the SIR. To relate to the link budget table, we can comparethe SIR to the required SNR, and the difference is the link margin. An SNR of 35 dBis still very high, and we see with the link budget table that it should give a Shannoncapacity between 2 and 6 Gbps.

This thesis has not touched the subject of scheduling in the backhaul links. Nonethe-less, it could be of interest in deployments with much frequency overlap. Taking up-and downlink transmission in consideration has been outside the scope of this thesis, aswell as traffic scheduling. However, one can imagine how the up- and downlink trafficcould perhaps be scheduled so that when base station 1 send downlink, base station 4could send uplink and vice versa. This would allow for some frequency overlap withoutcompromising the capacity due to increased interference.

Antenna models used for simulations have neglected nulls in the radiation pattern.Potentially there is some interest in modeling nulls in order to investigate the possibilityof placing base stations in those nulls to limit received interference.

The gain matrices for the proposed deployment geometries show large interfering gainsbetween base station 1 and 3. The difference against the path gain between 1 and 2 isclose to 30 dB, which was the FBR for the antennas. This aligns with intuition, due tobase station 1 being directed toward station 2 and 3, while station 3 is directed in thealmost opposite direction from 1. These path gains between 1 and 3 are high in relationto the path gain between 3 and 4, giving SIRs of 5 dB to 15 dB. Such SIRs are very low,keeping in mind that the required SNR for Low capacity is 15 dB with 400 MHz.

6.1.3 Interference Between Base Stations and Users

This thesis does not present any analysis of eventual interference occurring between thebackhaul links and the users’ channels. Assuming an access cell in close proximity to abackhaul link would use its spectral complement, interference would occur from the accesscell to another nearby backhaul link. It is then probable that the largest interferencewould occur at base stations somewhat directed toward the user.

The antenna array’s vertical HPBW was assumed to be almost 13°. Assuming suchan antenna is directed parallel to the ground and located 6 m above ground level, basictrigonometry tell us that a user on the ground must be at least 53 m from the base stationto be inside the HPBW:

6m

sin(6.5°)≈ 53m. (6.1)

6.1.4 Performance In Deep NLOS

The ’heatmaps’ of path gain in figures 5.3b and 5.4 showed similar performance in LOS,but quite different in NLOS. The latter figure, with the base station placed along theshort wall, showed many dark spots in the ’heatmap’, while the former figure showed nosuch dark spots but a more evenly decrement of path gain as the distance increased.

The dark spots across the buildings, in 5.4 are generated from the increased diffractionloss. While the corner placed base station have at least one path with only one diffraction

42

point, the other placement must have paths that pass through at least two diffractionpoints.

This extra diffraction loss could be an important aspect when designing a geometryfor a physical deployment.

The same simulations showed how the Low (2 Gbps) capacity target could be achievedalmost the whole map in the figures. Even though it is the lowest of the three target, itis still a substantial amount of capacity that can be achieved deep in NLOS.

6.1.5 Deployment Proposals

Section 5.3.3 showed that with the simulation tools used, it can be rewarding to simulatedifferent deployment geometries. The simulation results were sometimes counterintuitive,or unexpected, which can be helpful to raise important and interesting questions aboutdesigning actual base station geometries.

6.1.6 Comparison to Empiric Models

Section 5.3.2 showed how the simulations in quite deep NLOS could generate path gainvalues quite far off from the empiric NLOS model. As discussed in the result section, thevalues were within the 7.82 dB shadow fading standard deviation of the empiric model.However, we must remember that there is no stochastic shadow fading component in thesimulations, and that it is very deterministic. If a shadow fading component was to beadded, simulated values could quickly be far off from the empiric model.

From this discussion follows that while the empiric models could give a somewhataccurate estimation of what path gain to expect at a certain distance, the site-specificsimulator must be used to get a good intuition in what happens in different scenarios ofthe environment.

Furthermore, the use of highly directive antennas with a high number of antennaelements would in practice come with the possibility of enabling Multiple-Input-Multiple-Output (MIMO) transmission [7]. MIMO transmission is helpful over NLOS channelswhich has a wide range of different physical paths, allowing spatially separated datastreams. This type of MIMO transmission could greatly increase performances overNLOS channels.

6.1.7 Base Stations Mounted on Lamp Posts

In the simulations of chapter 5 all base stations (except the fixed base station mountedon precisely a building corner) are mounted on 6 m high lamp posts. As the simulationsshowed, the performance of a base station in NLOS around a corner can vary very muchover a couple of meters. Specifically, the NLOS simulations in figure 5.6 showed theimpact of a specular reflection. While the receiving base station was placed inside thefirst reflection from the opposite building, the path gain was close to 10 dB higher thanit was outside the reflection.

This is an important highlight because a physical deployment geometry in some envi-ronment similar to the Manhattan grid might not have lamp post available at the mostoptimal coordinates. Other problems could be blockages at certain lamp posts. Theessence of this discussion should be that even though deployment geometries can be high

43

performing in the simulations, it can be far from realizable in a real physical environmentdue to the actual streetscape.

The number of allowed imaginary nodes per ray has been set to 2. In NLOS, the firstreflection will probably be of an already diffracted ray. This means only one reflectionwill be allowed in NLOS. Perhaps the NLOS performance could be increased by allowingadditional imaginary nodes. On the downside, this could drastically increase the simu-lation complexity due to the many more possible signal paths. For systems with a lownumber of base stations it could be feasible.

There are of course other factors of simplification that makes the simulations differfrom reality. Creating a realistic scenario was not the purpose of this thesis, but it isworth discussing the fact that the created scenario is not realistic.

6.1.8 Highly Directive Antennas

The thesis has focused the deployment of highly directive antennas with beamwidth6°and 12° in horizontal and vertical direction. If the models used would redesigned toinclude nulls in the radiation pattern, the overall results in terms of range would notchange by any drastic means. However, a more advanced antenna model could openup for designing deployment geometries which could try to aim the nulls in directionsdecreasing the interference between backhaul links.

6.2 Conclusions

The analyses and simulations in the thesis has shown optimism toward the concept ofmulti-hop backhaul with the use of 28 GHz mmW. The results demonstrates how the useof highly directive antenna arrays can provide high enough path gains to achieve severalGbps Shannon capacity with bandwidth allocations of 100 MHz to 800 MHz. However, theSNRs needed with the use of 100 MHz seem unrealistic, especially due to the assumptionsof low to no interference, and disregard of phase noise.

With a 0 dB interference margin, a very high bandwidth between 400 MHz and 800 MHzis needed to secure a large link margin for a backhaul link with 12 Gbps Shannon capac-ity in LOS up to 200 meters distance. In NLOS, the simulations have shown small linkmargins to 12 Gbps with 800 MHz just around the closest corners from the base stations.Further away in NLOS, around one or two corners, small link margins to 2 Gbps linkswith 400 MHz were found up to 240 meters away.

The theory suggests that mmW should render low interference between backhaullinks not in LOS from each other, due to the weak NLOS propagation. However, thesimulations have shown that interferences in NLOS can easily become large.

The Manhattan grid simulations can be a good complement to empiric models inorder to provide intuition in how mmW behave in similar environments.

With a low number of nodes allowed per ray, one should take caution while simulatingNLOS scenarios where a reflection is one of the strongest components. It is probable thatthe low ray tracing complexity can render pessimistic results in NLOS.

Without any interference limiting techniques, the noise is likely be dominated bythe interference from nearby backhaul links. Further investigation is needed to assessthe interference from nearby users within the same frequency band. It is importantto take interference into account for dimensioning deployment geometries. There are

44

many techniques to limit the interference between backhaul links; scheduling, frequencyallocations and spatial separation.

In a multi-hop chain of backhaul links, the limiting nature of the first link allows forlesser capacity further down the chain. This is under the assumption of the individualaccess networks demanding comparable data rates over all sites in the multi-hop chain.

The employment of highly directive antennas in the simulations have shown counterintuitive examples that increasing the distance of a NLOS backhaul link can also in-crease its path gain. Therefore, dimensioning deployment geometries with high spatialseparation can be tricky.

Comparisons to the empirical models provided by 3GPP suggests that the simulationscan be pessimistic in deep NLOS, while the opposite in short range NLOS. One potentialcause of this discrepancy is the modeling of reflections being too strong in the shortNLOS, while further away they are excluded by the low number of nodes allowed. Moredata could be collected using the same map used in section 5.3.2.

6.2.1 Open Issues

The results presented in the thesis can be further developed by more advanced assump-tions on the interference and eventual interference decreasing techniques. Additionaldevelopments to the Manhattan grid could be to change the dimensions of the streetsand buildings to see how it affects the observed characteristics on reflections and diffrac-tions.

The reasoning behind the Manhattan grid’s simplicity was to keep simulations andanalyses simple and intuitive.

6.2.2 Further Work

A suggestion for further development of this thesis, or another thesis, would be to expandthe deployment geometry analysis by designing more diverse scenarios and classifyingthem by their performance.

45

Bibliography

[1] 802.11ad: WLAN at 60 GHz. White paper, IEEE, Munich, Germany, November2017.

[2] 3GPP. Study on channel model for frequencies from 0.5 to 100 GHz. TechnicalReport (TR) 38.901, 3rd Generation Partnership Project (3GPP), December 2017.Version 14.3.0.

[3] 3GPP. System Architecture for the 5G system. Technical Specification (TS) 23.501,3rd Generation Partnership Project (3GPP), March 2018. Version 15.1.0.

[4] Henrik Asplund, Martin Johansson, and Niklas Jalden. A set of propagation modelsfor site-specific predictions. accepted for publication in 12th European Conferenceon Antennas and Propagation (EuCAP 2018), London, UK, April 2018.

[5] Constantine A. Balanis. Antenna Theory, Analysis and Design. John Wiley & Sons,Inc., 2005.

[6] FCC. Catalog of potential expenses and estimated costs. https://apps.fcc.gov/

edocs_public/attachmatch/DA-17-154A2.pdf, 2017.

[7] Andrea Goldsmith. Wireless Communications. Cambridge University Press, 2009edition, 2005.

[8] Niklas Heuveldop. Ericsson Mobility Report. Ericsson AB, November 2017.

[9] Stefan Parkvall, Erik Dahlman, Anders Furuskar, and Mattias Frenne. NR: The New5G Radio Access Technology. IEEE Communications Standards Magazine, December2017.

[10] R4-168226. On mm-wave technologies for nr. RAN4#79, Ericsson.

[11] R4-168305. On micro-wave and millimetre-wave base station antenna technology.RAN4#80-BIS, Ericsson.

[12] Viktor Rabinovich and Nikolai Alexandrov. Antenna Arrays and Automotive Appli-cations. Springer-Verlag New York Inc., 2012.

[13] Sundeep Rangan, Theodore S. Rappaport, and Elza Erkip. Millimeter Wave CellularWireless Networks: Potentials and Challenges. Proceedings of the IEEE, 102(3):366–385, January 2014.

46