Outdoor to Indoor Coverage in 5G Networks

(1)

UPTEC F 16034

Examensarbete 30 hp Juni 2016

Outdoor to Indoor Coverage in 5G Networks

Vilhelm Rydén

(2)

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

Outdoor to Indoor Coverage in 5G Networks

Vilhelm Rydén

Outdoor to indoor mobile coverage is evaluated for different frequencies in two scenarios, a single building scenario and a city environment. A new model for outdoor to indoor propagation is suggested, connecting existing, highly detailed indoor and outdoor ray-tracing propagation models. The model is compared to previous, site specific as well as statistical, propagation models. Results conclude that the new model gives higher path gain for edge users in the single building scenario, whereas results from the city scenario are inconclusive. Furthermore, results from the single building scenario suggest that indoor coverage is possible at 5 GHz and below for most buildings, whereas for the city scenario indoor coverage at 5 GHz will be possible only for buildings without metally coated windows. Achieving indoor coverage at 30 GHz is highly problematic for all cases, and it is concluded that indoor base stations are necessary if frequencies of 10 GHz and above are to be used in future mobile networks.

In addition, an indoor analysis is made to verify existing loss per meter indoor models. It is concluded that such models are often optimistic, although their assumption of log-normal shadowing remain valid, at least for closed offices. Furthermore, the assumption of loss as a linear function of distance might be unfeasible for higher frequencies, where a breakpoint in the linear model was observed at a distance of roughly 10 meters.

ISSN: 1401-5757, UPTEC F 16034 Examinator: Tomas Nyberg Ämnesgranskare: Mikael Sternad Handledare: Gunther Auer

(3)

Sammanfattning

Det här arbetet undersöker hur inomhustäcknining kan uppn˚as i framtida generationens mobilnät. I synnerhet undersöks utbredningen fr˚an basstationer placerade utomhus till användare som befinner sig i en byggnad. Detta görs för tv˚a fall: en ensam byggnad med en basstation pekad mot sig och ett stadscenario med flera basstationer och intilligande byggnader.

För att uppn˚a högre överföringhastigheter finns en stor enighet om att nya frekvensband kommer att behöva användas, eftersom det är ont om bandbredd p˚a nuvarande band. De nya band man tittar p˚a ligger p˚a högre frekvenser, fr˚an ungefär 5 Ghz upp till kanske 30 GHz. Radiov˚agor p˚a högre frekvenser dämpas till större del av hinder s˚a som väggar och träd, varp˚a det kan bli sv˚arare att n˚a mobiltäckning p˚a m˚anga platser. Därför krävs det ocks˚a nya utbredningsmodeller och simuleringar för att först˚a hur v˚agutbredningen fungerar p˚a dessa frekvenser.

I detta arbete används Ericssons befintliga inom- och utomhusmodeller, som b˚ada baseras p˚a str˚alföljning. De använder modeller av städer för utomhussimuleringar, och modellen tar hänsyn till reflexioner och diffraktion runt byggnader. Inomhusmodellen fungerar p˚a ett liknande sätt, och använder planritningar fr˚an riktiga byggnader i sina beräkningar. I detta arbete har dessa tv˚a modeller kopplats samman, för att ge en utbred- ningsmodell som fungerar hela vägen fr˚an en utomhusbasstation till en inomhusanvändare i en byggnad.

Arbetet visar att när en ensam byggnad simuleras, g˚ar det i de flesta fall att n˚a in- omhustäckning inomhus fr˚an en utomhusstation för frekvenser upp till 5 GHz. Undantaget

¨

ar byggnader med moderna fönster med ett metalliskt skikt p˚a och en planritning med m˚anga väggar inomhus. I vissa falla är det även möjligt att uppn˚a inomhustäckning vid 10 GHz. När en byggnad placerad i en stad simuleras, kan inomhustäckning för 5 GHz uppn˚as i hälften av fallen, men aldrig för 10 GHz. Detta tyder p˚a att placering av separata inomhusbasstationer kan bli en viktig del i att skapa inomhustäckning för höga frekvenser i framtida mobilnätverk.

(4)

Glossary

BS base station.

cdf cumulative distribution function.

EIRP equivalent isotropically radiated power.

FSPL free space path loss.

KM the Keenan-Motley model.

LOS line-of-sight.

PG path gain.

PL path loss.

SNR signal-to-noise ratio.

(7)

Chapter 1 Introduction

There is a general consensus that there will be an enormous growth in the amount of data generated in mobile networks in the near future. New requirements for 5G networks include 1000 times higher data volumes, an order of magnitude lower latencies and a large number of connected devices [1–6].

In order to realize higher data rates, the amount of available bandwidth for each user must increase. As available spectrum in the current license bands are becoming scarce, moving to higher frequencies has gone from a challenging opportunity to being an absolute necessity [7]. The higher attenuation at these bands can possibly be compensated by the use of large antenna arrays [8], a solution which becomes more feasible as each antenna is growing smaller when wavelengths are in the mm range.

Another solution is to use indoor base station to offload outdoor cells and to give coverage to users not reachable from outside base stations. In order to understand how and where to place such indoor cells, good knowledge of the influence of outdoor macro cells is required. Thus, the need to properly modeling outdoor to indoor propagation remains an important challenge when it comes to achieving indoor coverage.

In general, mobile networks are assumed to be interference limited rather than noise limited, meaning that interference from other users rather than propagation characteristics limits the network performance. However, due to the higher attenuation of high frequency signals combined with smaller cells, this is likely to change in future generations [2]. As a consequence, having detailed propagation models becomes increasingly important to accurately simulate and evaluate future mobile communication networks at higher frequencies.

1.1 Goals

This thesis aims to investigate how indoor coverage can be achieved in high frequency ranges. In particular, outdoor to indoor coverage from outdoor base stations will be evaluated using a new propagation model, which connects existing, highly detailed outdoor and indoor models. In addition, existing indoor models are compared to each other, using real building floor plans for different office types.

(8)

1.2 Thesis Outline

Chapter 2 begins by introducing some basic terminology followed by a brief introduction to wireless propagation. In chapter 3, the simulator is briefly described, follow by an overview of the parameters used in the simulations. Then, different propagation models are introduced, including the new outdoor to indoor model developed in this thesis. In order to fully understand the differences between these propagation models, chapter 4 contains a single floor analysis, analyzing the excess indoor loss obtained with the different models. In chapter 5, outdoor to indoor propagation is simulated in a single building scenario, where path gains are analyzed for different building types, propagation models and frequencies. Finally, chapter 6 shows similar results in a more realistic city scenario, before the work is concluded in chapter 7.

(9)

Chapter 2 Technical Background

This chapter intends to provide the reader with the basics of mobile communication and wireless propagation models, including basic terminology and useful formulas.

2.1 Signal-to-noise Ratio

Noise exists in all communication systems, and can be a combination of thermal, cosmic or man-made noise that cannot be controlled. The signal-to-noise ratio (SNR) is a measure of how much stronger a desired signal is compared to the noise, such that

SNR = P_r

P_N. (2.1)

Here, P_r denotes the received signal power and P_N the noise power at the receiver.

2.2 Channel Capacity

Channel capacity is an upper bound of the rate in bits per seconds at which information can be reliably communicated over a wireless channel, closely related to the SNR described in the section above. A theoretical upper bound is given by the Shannon-Hartley theorem, stating that

C = B log₂(1 + SNR), (2.2)

where B is the available bandwidth in Hz. Note the close relationship between channel capacity and bandwidth; for example, inserting an SNR of 1 gives an upper bound for channel capacity of 1 bit/Hz.

Furthermore, note that the capacity increases linearly with available bandwidth, but logarithmically with the SNR. Thus, in the high SNR regime, increasing bandwidth rather than transmit power would give the largest increase in capacity. In the low SNR regime, however, equation 2.2 can be approximated by

C ≈ kB P_r

P_N = kP_r

N₀, (2.3)

(10)

where k is a constant and N₀ represents the spectral density of the noise, such that P_N = BN₀. Since the total noise level P_N is proportional to the bandwidth B, the channel capacity is independent of B in this regime.

2.3 Path Loss and Path Gain

The attenuation of a wireless signal propagating from sender to receiver is known as path loss (PL) and defined as the power received divided by power transmitted. That is,

PL = P_r

P_t. (2.4)

Its reciprocal is known as path gain (PG), which gives the simple relationship

PL_dB = −PG_dB (2.5)

in logarithmic scale.

2.3.1 Free Space Path Loss

If there are no obstacles between sender and receiver, a wireless wave propagates according to the free space path loss (FSPL), given by

FSPL = 4πf d c

2

, (2.6)

where f is the carrier wavelength in Hz, d is the distance in m and c is the speed of light.

In most situations, the FSPL provides a poor estimate of the actual PL. For instance, the line-of-sight (LOS) path might be blocked by obstacles such as buildings and trees such that a reflected or diffracted path provides a better path. The additional loss apart from the FSPL is called excess loss (X), such that

P L_total,dB= FSPL + X_dB. (2.7)

2.3.2 Antenna Gain and EIRP

Antenna gain is a measure of how much the characteristic of an antenna may increase the power transmitted in a certain direction, for instance by having the shape of a horn or by using a parabola. The sum of P_t,dBmand G_t,dBis known as equivalent isotropically radiated power (EIRP), measured in dBi, and can be understood as the transmit power that would have to be used for an antenna with unit antenna gain to match that of a directed antenna in a certain direction. Regulations for EIRP exist, and need to be taken into account when planning cell sites or simulating cellular systems.

(11)

Beamforming Gain

By using an array of antennas, it is possible to steer the beam by adjusting the phases of individual antenna elements. By combining several antennas, a beam forming gain can be achieved, similar to the antenna gain described above. The use of large antenna arrays to achieve a high beam forming gain is considered to be central in future 5G networks [2].

2.3.3 Friis Transmission Equation

Using the definitions above, Friis transmission equation gives an expression for the PG as PG_dB = G_t,dB+ G_r,dB− FSPL_dB. (2.8) When results are presented, both the terms gain and PG in this thesis will refer to the equation above, but with FSPL replaced with the path loss obtained from the propagation model that was used.

2.4 Propagation Models

In most situations, the FSPL provides a poor estimate of the actual PL. For instance, the LOS path might well be blocked by obstacles such as buildings and trees, so that a reflected or diffracted path may provide a better path. For this reason, several statistical as well as site specific models exist which replace the FSPL of (2.8) with a better estimate of the propagation path loss.

A site specif ic model uses a 3D environment with outdoor buildings (or indoor walls) plans along with some ray-tracing based propagation model, and is accurate for the modeled situation. Such models will be described further in chapter 3.

A statistical model relies on measurements from similar environments, and is usually based on the distance between sender and receiver. In contrary to site specific models, it fails to take the actual environment into account.

In addition to the PL obtained from the propagation model used, shadow f ading may be added to the calculated PL. It accounts for random losses by obstacles in the propagation path, and is usually modeled as a log-normally distributed random variable.

2.5 Reflection, Diffraction and Scattering

The main propagation mechanisms for large-scale characteristics are reflection, diffraction and scattering. Together they the determine the received power and thus the PG from one node to another. A brief introduction is given in this section, a more detailed description can be found in textbooks such as [9].

Ref lection occurs when an electromagnetic wave hits a surface which is large in relation to its wavelength. The dielectric properties of the surface along with the frequency of the

(12)

wave determine the amount that is reflected. Reflections may occur from the ground or buildings in a city environment, or from a wall, floor or ceiling inside a building.

Dif f raction occurs when a wave hits a sharp surface, such as a building corner or roof edge. It allows the wave to propagate around corners, and reach a receiver that is not in LOS with the sender. Diffraction is based on Huygen’s principle, which states that every point of a wave front acts as a source for secondary waves. The diffracted wave is then the integral of all secondary waves adding up to the received signal in a shadowed location.

Similarly to reflection, the amount of diffraction depends on the exact geometry of the edge as well as the frequency of the wave.

Finally, scattering occurs when a surface is hit which is small compared to the wavelength of the electromagnetic wave. This could be a lamp post, a tree or a rough surface, causing a diffuse reflection from the surface of the object.

2.6 Required Path Gain

By solving equation 2.2 for SNR and given a certain bandwidth, the minimum required PG for a communication link may be calculated. First, let the noise figure NF be a measure of how much noise is added in the receiver circuits. Then, the minimum PG is given by

PG_min,dB= SNR_dB− P_t,dBm+ kTB_dBm+ NF_dB, (2.9) where P_t is the transmit power and kTB the thermal noise, a product of the Boltzmann constant k, temperature T and bandwidth B. Thus, (2.9) together with (2.2) give a relation between PG and achievable data rate for a user in a noise limited mobile network.

2.6.1 Cell Edge Path Gain

Cell edge PG is defined as the PG achieved by 95 percent of the users, and will be used as the measure of coverage in this work.

(13)

Chapter 3 Methodology

This chapter is organized as follows: section 3.1 briefly describes the Ericsson simulator and the simulation parameters used. Section 3.2 further describes the propagation models used by the simulator, before the new outdoor to indoor model is presented in section 3.3.

3.1 The Simulator

The simulator used in this thesis is an Ericsson internal, time static LTE system level network simulator written in Matlab. It offers support for various propagation models, ranging from fast, statistical models to more computationally demanding, ray-tracing based models, as will be further described in the following sections. The fact that it is time static is not a limiting factor, as the goal of this thesis is to determine indoor coverage for throughout a building floor plan, where users can assumed to be stationary.

3.1.1 Ray Tracing

Ray tracing is a technique to accurately model propagation of high frequency electromagnetic waves, by tracing the wave propagation path from a base station (BS) to a user. In particular, ray tracing may be used to model the reflections with building walls in a city, or indoor walls in an indoor environment [10]. Since this technique approximates the wave front with particle-like ray, wave-like phenomena such as diffraction needs to be modeled separately. This can for instance be done by replacing the terrain profile with absorbing half-screens [11], and using a recursive model [12].

3.1.2 Simulation Parameters

Since the parameters used in future 5G networks are yet to be determined, parameter assumptions have to be made. The most significant parameters when studying proagation is the carrier frequency, for which a few candidates exist. For instance, 28 GHz is used in [13]. In this work, a range of frequencies was selected in order to give an idea of how

(14)

propagation characteristics change for various frequencies. The frequencies chosen in this work are 2, 5, 10 and 30 GHz.

The EIRP for a 5G system is assumed to be 65 dBi, mostly by the use of large antenna arrays to achieve a high beam forming gain. In the simulations, an antenna with 65 degrees horizontal and vertical half power beam width is used to simulate an antenna with many beam forming elements. As only PG is studied, and not interference between users, this corresponds to being able to steer the beam individually to each user with a high gain.

To reach the chosen EIRP, a transmit power of 40 W and antenna gain of 19 dBi was assumed. The actual values for a 5G systems will most likely differ from these, and may also be frequency dependent. However, the EIRP remains the most important measure, and it is assumed to be around 65 dBi for 5G systems. To reach an SNR of 0 dB, (2.9) gives a minimum PG of -131 dB, which will be defined as the threshold for whether a user is in coverage or not. For this reason, PG will be the measure of performance in this work.

The simulation parameters are summarized in table 3.1.

Table 3.1: System parameters.

Frequency [GHz] 2, 5, 10 and 30

Transmit power, BS [dBm] 46

Antenna gain (BS) [dBi] 19

Noise figure, DL [dB] 9

EIRP (BS) [dBi] 65

Horizontal beam width, BS [deg] 65 Vertical beam width, BS [deg] 65

3.2 Propagation Models

This section presents available models for outdoor and indoor propagation losses, including wall penetration losses for different kind of walls. This work is largely based on the values used in [14], which in turn are based on measurements such as [15–17].

3.2.1 Outdoor Propagation Model

The Ericsson outdoor propagation model uses a ray tracing approach to follow the rays from an outdoor BS to a user. During the calculation, beams may be diffracted by roof edges as well as building corners, and reflected on other buildings. The model is site-specific, meaning it uses the environment of a real or fictional city to calculate the propagation, rather than relying on statistical models.

3.2.2 Wall Penetration Losses

When a wireless signal propagates through a wall, the signal will be attenuated by a certain amount, depending on parameters such as wall material and carrier frequency. In [14], the

(15)

building propagation loss in dB scale for a single glass window can be modeled as

L_glass = 0.1f_GHz+ 1 (3.1)

where f_GHz is the carrier wavelength in GHz. Similarly, an inner wall is modeled as

Linner = 0.2fGHz+ 1.7 (3.2)

and a concrete wall as

L_concrete= 4f_GHz + 5. (3.3)

Furthermore, a distinction is made between two building types, ”old” and ”new”. The building propagation loss for the old building type is modeled as

Louter,old = 2aoldLglass+ (1 − aold)Lconcrete, (3.4) where a_old = 0.2 represents the fraction of the wall covered by two-layer windows. The new building type is modeled as

L_outer,new = 3a_newL_glass+ 20 + (1 − a_new)L_concrete, (3.5) where a_new = 0.9 represents the fraction covered by three layer windows, where an additional 20 dB loss has been introduced due to an infrared reflective metal coating commonly used on modern building windows.

The wall penetration losses for a few frequencies are summarized in table 3.2.

Table 3.2: Building losses for various wall types.

Frequency [GHz] 2 5 10 30

Lglass [dB] 1.2 1.5 2 4

L_concrete [dB] 13 25 45 145

L_inner [dB] 2.1 2.7 3.7 7.7

L^old_outer [dB] 8.1 9.9 11 15

L^new_outer [dB] 20.5 24.6 26.5 32.5

In addition to the losses mentioned above, an angular dependence is modeled as

L_ang= 20(1 − cosθ)², (3.6)

motivated in [18]. Here, θ is the deviation of the incidence angle from the normal vector of the wall. For non-LOS paths, an angle of π/3 is used.

(16)

3.2.3 Statistical Indoor Model

When signals propagates through an indoor environment, the excess loss i heavily dependent on floor plan properties such as wall placement and wall materials. If no explicit floor plan is used, implicit models with a frequency dependent loss per meter can be used. They are based on the assumption that walls are uniformly spread across a floor plan, with a certain average wall distance between them.

In [seeman], two indoor models are proposed: one uses the single glass window of (3.1) and the other the inner wall of (3.2). In this work an average of these two is used, such that

Lindoor,dB = L_glass+ L_inner

2d , (3.7)

where d = 4 is the average wall distance in meters. The losses are summarized for a few frequencies in table 3.3.

Table 3.3: Excess indoor loss per meter.

Loss [dB/m] 0.415 0.525 0.713 1.463

3.2.4 Partitioned Indoor Loss Model (Keenan-Motley)

If floor plans are available, a site specific propagation loss may be obtained by the use of a partitioned loss model. Whenever the signal propagates through a wall of a given material, an additional loss is introduced as described in [19]. By summing up the losses of all walls, the excess indoor loss can be obtained such that

L_indoor=

n

X

i=1

L_i, (3.8)

where L_i denotes the loss from the i:th wall crossed when drawing a straight line from BS to user. Thus, no angular dependence is used in the partitioned model, sometimes called the Keenan-Motley model (KM).

3.2.5 Detailed Indoor Model (Spoke)

In addition to the indoor propagation models described in the previous sections, Ericsson has develop a highly accurate, ray-tracing based indoor loss model called Spoke. Apart from evaluating the loss from KM above, it also takes reflections on walls as well as diffraction around wall corners into account. Both specular and diffuse reflections are modeled.

In addition, Spoke adds an angular penalty when the incidence angle of a wall crossing is not perpendicular. Finally, the model chooses the path that gives the lowest PL, which could be a reflected or diffracted path as well as the direct path used in KM.

(17)

3.2.6 Outdoor to Indoor Propagation Model

One model for outdoor to indoor propagation was suggested in [14] and consists of evalua- tion four candidate paths, as shown in figure 3.1. The outdoor part of the path is evaluated using the outdoor model described in section 3.2.1, and the indoor paths may be calculated by either the statistical indoor model or by KM described in sections 3.2.3 and 3.2.4 respectively. Finally, the outer wall loss is calculated using (3.4) or (3.5) in combination with (3.6). However, the Ericsson simulation lacks support for using the outdoor to indoor model together with Spoke, desccribed in section 3.2.5. For this reason, a new outdoor to indoor model was developed during this work, which will be described in the next section.

Figure 3.1: Propagation from an outdoor node to an indoor user [14]. The four paths are chosen such that the indoor paths are always perpendicular to the outer walls.

3.3 New Outdoor to Indoor Propagation Model

The outdoor to indoor model described in the previous section is simple and works well for low carrier frequencies, as the direct path is then the dominant propagation path. However, for higher frequencies with an available floor plan, reflected and diffracted paths fails to be taken into account. Furthermore, the four candidate paths evaluated might not give enough richness in the amount of possible propagation paths, as will be shown in chapter 4.

In order to address these issues, a new outdoor to indoor propagation model was developed in this work, and will be described shortly in this section. The new model should be seen as a way of connecting the existing outdoor model described in section 3.2.1 with the detailed indoor model Spoke, which was not previously possible in the Ericsson simulator.

(18)

3.3.1 Description of the New Model

The new model works by evaluating a number of entry points through the outer wall for each indoor user. Rather than basing the entry points on the position of the indoor user, it uses the same fixed set of entry points for all users inside the building. By doing so, the indoor simulation becomes independent of the outdoor simulation, whereby these can be run simultaneously in order to save computation time. The distance between the entry points can be set independently of other parameters, and a value of 1.5 meters will be user throughout this work. A lower value would likely give more accurate results, at the cost of increased computational time.

Another difference concerns the wall loss which is added when penetrating the outer wall. While the old model uses the angular wall loss dependence in (3.6), the new model traces each ray through the wall, and any change of direction adds an additional loss according to the recursive model in [12]. This is meant to give a more accurate modeling of propagation through the outer walls, such as the ability to capture shadowing by concrete pillar in the outer wall. But most notably, by combining the use of multiple entry points and the detailed indoor model described in section 3.2.5, multiple reflected paths are evaluated for each indoor user. This is assumed to be especially important at higher frequencies, as reflections may play a larger role due to the higher attenuation of propagation through walls.

In this report, the new outdoor to indoor model will be referred to as Spoke, as it allows for the indoor model Spoke to be used together with the existing outdoor model.

(19)

Chapter 4 Single Floor Analysis

This chapter describes the simulations done on a single floor for different office types. The goal is to analyze differences between the different indoor propagation models, as described in sections 3.2.3 - 3.2.5.

4.1 Setup

In this section, floor plans for an open and closed office respectively will be used, shown in figure 4.1. Both floor plans have a footprint of 50 × 40 meters. Here, a dry wall is modeled with twice the loss of (3.2) and a concrete wall according to (3.3). An indoor BS is placed in the upper right corner, and the excess PL is calculated to each user, placed in every 1.5m × 1.5m bin, resulting in a total of 832 users in each floor plan.

(a) (b)

Figure 4.1: (a) Closed and (b) open office floor plan where outer walls are red, inner dry walls blue and concrete walls purple. The indoor BS is marked with a green circle.

(20)

4.2 Results

Figure 4.2 and 4.3 show excess loss per meter as a function of distance between BS and user for the closed office and two different propagation models. The frequencies 2 and 30 GHz where chosen in order to clearly show any frequency dependent characteristics. Note that the two models give similar results for the lower frequency, whereas for the higher frequency Spoke suggests a significantly lower loss per meter. Furthermore, at the higher frequency a breakpoint at a distance of roughly 10m can be seen, after which the inclination of the data points seem to be lower.

0 10 20 30 40 50 60

Distance [m]

0 20 40 60 80 100 120

Excess loss [dB]

Data y = 1.64x

(a)

0 10 20 30 40 50 60

Distance [m]

0 20 40 60 80 100 120

Data y = 1.53x

(b)

Figure 4.2: Closed office: Excess indoor loss for (a) KM and (b) Spoke at 2 GHz, and fitted curve (solid line) plus minus one standard deviation (dotted lines).

0 10 20 30 40 50 60

Distance [m]

0 100 200 300 400 500 600

Data y = 7.72x

(a)

0 10 20 30 40 50 60

Distance [m]

0 50 100 150 200 250 300

Data y = 4x

(b)

Figure 4.3: Closed office: Excess indoor loss for (a) KM and (b) Spoke at 30 GHz, and fitted curve (solid line) plus minus one standard deviation (dotted lines). Note the different scales of (a) and (b).

(21)

The cumulative distribution function (cdf) for the deviation between the data and the fitted curve is shown in figure 4.4, where a normal distribution with the same mean and variance as the deviation is shown as well. The data seem to match well, indicating a log-normal shadowing in linear domain.

-30 -20 -10 0 10 20 30

Deviation [dB]

0 0.2 0.4 0.6 0.8 1

CDF

(a)

-150 -100 -50 0 50 100

Deviation [dB]

0 0.2 0.4 0.6 0.8 1

CDF

(b)

Figure 4.4: Closed office: Histogram of excess loss distribution (boxes) and fitted normal cdf (red) using Spoke, at (a) 2 GHz and (b) 30 GHz.

Results for the open office are presented in figure 4.5 and 4.6. Here, the difference between the two propagation methods is even more pronounced. Compared to the closed floor plan, the acquired loss per meter is generally lower for the open office floor plan. The users with an excess loss of 0 dB are in LOS with the BS. While the standard deviation shown in table 4.2 is similar for the open and closed floor plan, the log-normal shadowing assumption seem less accurate for the open office case, as can be seen in figure 4.7.

0 10 20 30 40 50 60

Distance [m]

0 20 40 60 80 100

Data y = 1.08x

(a)

0 10 20 30 40 50 60

Distance [m]

0 10 20 30 40 50 60

Data y = 0.544x

(b)

Figure 4.5: Open office: Excess indoor loss for (a) KM and (b) Spoke at 2 GHz, and fitted curve (solid line) plus minus one standard deviation (dotted lines).

(22)

0 10 20 30 40 50 60 Distance [m]

0 100 200 300 400 500 600

Data y = 6.22x

(a)

0 10 20 30 40 50 60

Distance [m]

0 50 100 150 200

Data y = 0.98x

(b)

Figure 4.6: Open office: Excess indoor loss for (a) KM and (b) Spoke at 30 GHz, and fitted curve (solid line) plus minus one standard deviation (dotted lines).

-30 -20 -10 0 10 20 30

Deviation [dB]

0 0.2 0.4 0.6 0.8 1

CDF

(a)

-150 -100 -50 0 50 100 150

Deviation [dB]

0 0.2 0.4 0.6 0.8 1

CDF

(b)

Figure 4.7: Open office: Histogram of excess loss distribution (boxes) and fitted normal cdf (red) using Spoke, at (a) 2 GHz and (b) 30 GHz.

The loss per meter values and standard deviations are summarized for all frequencies in tables 4.1 and 4.2, as well as in figure 4.8. The standard deviation for the fitted curves in 4.8(a) were 0.027 dB and 0.059 dB for the open and closed offices respectively.

Table 4.1: Indoor excess loss [dB/m] using Spoke.

Open office 0.54 0.63 0.72 0.98 Closed office 1.53 1.88 2.36 4.00

(23)

Table 4.2: Standard deviation [dB] for fitted loss per meter using Spoke.

Open office 10.1 12.7 16.4 30.6 Closed office 9.53 12.8 16.8 29.7

As seen in figure 4.8, there is some resemblance between the open office Spoke simulations and the statistical loss per meter values, even though the former suggests a weaker frequency dependence. For 10 GHz, the same value is obtained for both methods. The closed office values are seen to be significantly higher than those of the statistical model for all frequencies.

0 5 10 15 20 25 30

Frequency [GHz]

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5

Excess indoor loss [dB/m]

Data, open office Data, closed office y = 0.015x + 0.54 y = 0.087x + 1.43 Statistical loss/m

(a)

0 5 10 15 20 25 30

Frequency [GHz]

5 10 15 20 25 30 35

Standard deviation [dB]

Data, open office Data, closed office y = 0,71x + 8.96

(b)

Figure 4.8: (a) Loss per meter and (b) standard deviation for all frequencies and both floor plans using Spoke. Both (a) and (b) show data points as well as a linear fit.

4.3 Discussion

The fact that KM and Spoke give similar results for the low frequency, whereas a large difference for the higher frequency, is most likely a consequence of the reflected paths of the detailed indoor model, which seem to be increasingly important as the frequency increases.

The horizontal lines which can be seen in the KM plots correspond to the number of crossed walls, which due to the lack of angular dependence give discrete loss levels. For Spoke, the results are more smooth because of the angular dependence of wall penetration as well as the reflected and diffracted paths.

Furthermore, a breakpoint can be noted at a distance of roughly 10 meters in figure 4.3(b). After this breakpoint, the excess loss increases more slowly as a function of distance.

This is likely because at longer distances, the signal tends to find alternate, reflected paths rather than the direct path from sender to receiver.

(24)

For the open office, a comparison with table 3.3 shows that Spoke suggest a slightly higher PL for lower frequencies, but a lower loss for higher frequencies. This can also be seen from figure 4.8(a). However, for the closed office Spoke gives a significantly higher PL for all frequencies. This might partly be explained by the higher loss for indoor walls used by Spoke, but possibly also a shorter average wall distance than the 4 meters used in the statistical model. Furthermore, the large difference between obtained loss per meter for the open and closed offices further motivates the need for site-specific, detailed propagation models.

The standard deviations listed in table 4.2 suggest little or no difference for the two floor plans. This is somewhat surprising, as the cdf curves for the open office (figure 4.7) seem to suggest a poorer match than those of the closed office in figure 4.4. This fact suggests that the same log-normal shadowing could be applied to a loss per meter model, even though the loss per meter itself should be varied to match that of different floor plans.

As shown in figure 4.8, assuming a linear relationship between loss per meter and frequency seems reasonable, at least for this limited set of frequencies. Possibly, a non- linear relationship could be obtained by running simulations on more frequencies.

(25)

Chapter 5 Single Building Scenario

In this chapter, the performance of users in a single building with a single outdoor BS will be analyzed. Results are presented in the form of gain maps as well as gain cdfs.

5.1 Setup

For these simulations, a macro antenna is placed a 100 meters away from a 15 floor single building, as shown in figure 5.1. On each floor, simulations are run with either a closed floor plan, an open floor plan or no floor plan. When a floor plan is available, simulations are run using both KM and Spoke. When no floor plan is available, the loss per meter model from section 3.2.3 is used.

140 120 0 100

25 80

60 30

0 40 60

20 -25 0

Figure 5.1: Single building setup where the BS i marked with a green star and a user with a blue dot.

For the gain maps shown in figure 5.2 - 5.6, a user was placed in every 1.5m × 1.5m bin, resulting in 832 users per floor. For the other simulations, a 5m × 5m sampling was

(26)

used, resulting in 80 users per floor. The full set of building parameters are summarized in table 5.1

Table 5.1: Building parameters.

Building footprint 40m × 50m

Building height 60m

Floor height 4m

Users per floor 832 (gain maps), 80 (cdf:s)

5.2 Results

The gain for users on the 8th floor are depicted in figure 5.2 to 5.6. The 8th floor was chosen since its users are at a height of 30m, which is the same as the macro antenna.

The results for the open office floor plan at 5 GHz are shown in figure 5.2, and are similar for both propagation models. However, Spoke suggests higher gain for the weaker users in the center of the building, whereas a blue, horizontal stripe in figure 5.2(b) indicates that users are shadowed by the concrete walls shown in purple in figure 4.1(b).

(a) (b)

Figure 5.2: Open office: Gain map of 8th floor for (a) Spoke and (b) the KM indoor model with new building type and 5 GHz carrier frequency.

A similar gain map for 30 GHz can be seen in figure 5.3. Coverage is at -140 dB or below for the whole center area of the office, including the shadowed region at the center right. The two propagation methods suggest similar coverage, even though a fewer number of users are shadowed by the small rooms in the upper and lower part of the office when using Spoke.

(27)

(a) (b)

Figure 5.3: Open office: Gain map of 8th floor for (a) Spoke and (b) the KM indoor model with new building type and 30 GHz carrier frequency.

Looking at the closed office in figure 5.4, Spoke once again gives a slightly higher gain for the worst case users (dark blue) in the center right part of the office, while it gives somewhat lower gain for the typical user. The detailed statistics will be shown more clearly in the cdf plots in figure 5.7 to 5.10 below.

(a) (b)

Figure 5.4: Closed office: Gain map of 8th floor for (a) Spoke and (b) the KM indoor model with new building type and 5 GHz carrier frequency.

(28)

For 30 GHz (figure 5.5), indoor coverage is extremely poor for both models. Here, some users in the horizontal corridor above the center using the KM model have a slightly higher PG than the surrounding users, something which cannot be seen for Spoke.

(a) (b)

Figure 5.5: Closed office: Gain map of 8th floor for (a) Spoke and (b) the KM indoor model with new building type and 30 GHz carrier frequency.

Finally, figure 5.6 shows the results of the loss per meter model for reference. Since no floor plan is used, the building specific propagation properties are completely lost. Also for the loss per meter model, indoor coverage at 30 GHz is poor as seen in figure 5.6(b).

Results for other frequencies follow the same pattern, as can be seen from the cdfs described in the next paragraph.

(29)

(a) (b)

Figure 5.6: Gain map of 8th floor using the loss per meter model, with new building type and a carrier frequency of (a) 5 GHz and (b) 30 GHz.

The cdfs from a simulation at 5 GHz are shown in figure 5.7. As expected, the closed floor plan provides a lower gain in general. The curve for the statistical model somewhat resembles that of an open floor plan, whereas it is highly optimistic compared to the closed floor plan, especially when it comes to edge users. Spoke does in turn suggest a higher gain for edge users than KM, even though gain in general is slightly lower, as previously seen in the heat maps.

-180 -160 -140 -120 -100 -80

Path gain [dB]

0 20 40 60 80 100

CDF [%]

Closed office, Spoke Closed office, KM Open office, Spoke Open office, KM Loss/meter

(a)

-180 -160 -140 -120 -100 -80

Path gain [dB]

0 20 40 60 80 100

CDF [%]

(b)

Figure 5.7: PG at 5 GHz for (a) old and (b) new building type for the different floor plans and propagation models.

(30)

A similar curve is shown for 10 GHz in figure 5.8, with similar trends observable for the different propagation methods.

-180 -160 -140 -120 -100 -80

Path gain [dB]

0 20 40 60 80 100

CDF [%]

(a)

-180 -160 -140 -120 -100 -80

Path gain [dB]

0 20 40 60 80 100

CDF [%]

(b)

Results are summarized for all frequencies in figure 5.9, showing that an increase in frequency mainly corresponds to a translation of the cdf curve to the left. Another observation is that for 30 GHz the gain for users below the median value suffer more from the increase in frequency.

-180 -160 -140 -120 -100 -80

Path gain [dB]

0 20 40 60 80 100

CDF [%]

2 GHz 5 GHz 10 GHz 30 GHz

(a)

-180 -160 -140 -120 -100 -80

Path gain [dB]

0 20 40 60 80 100

CDF [%]

2 GHz 5 GHz 10 GHz 30 GHz

(b)

Figure 5.9: PG for new building type, using Spoke for (a) closed and (b) open office.

(31)

Finally, cell edge rates are summarized in figure 5.10(a). For none of the cases, achieving a 0 dB SNR is possible at 30 GHz, and it is only possible for the open office with old building types at 10 GHz. For 5 GHz, however, all cases but the closed office with a new building type achieves this target. As a comparison, figure 5.10(b) shows the results of an indoor deployment with 5 indoor base stations per floor. With a transmit power of 33 dBm for the indoor base stations, (2.9) gives a minimum PG of -118 dB in order to reach 0 dB SNR. As seen in the figure, this target is reached for frequencies of 10 GHz and below for the indoor deployment.

2 5 10 30

Frequency [GHz]

-200 -180 -160 -140 -120 -100 -80

5th percentile [dB]

Open office, old BT Closed office, old BT Open office, new BT Closed office, new BT -131 dB

(a)

2 5 10 15 20 25 30

Frequency [GHz]

-140 -130 -120 -110 -100 -90 -80

Open office, indoor deployment Closed office, indoor deployment -118 dB

(b)

Figure 5.10: PG, 5th percentile, using Spoke. The green, horizontal line shown the minimum required PG in order to reach a SNR of 0 dB. (a) Outdoor base station and (b) indoor deployment with 5 base stations per floor for comparison.

5.3 Discussion

As can be seen from figure 5.2, and to some extent 5.4, using Spoke clearly suggests a higher PG for edge users. The square, rectangular ”boxes” of low gain users (dark blue) are less pronounced, likely due to the additional reflected and diffracted paths. Furthermore, the statistical model in figure 5.6 fails to capture anything but the general trend of users closer to the macro have a stronger PG.

The bright stripes seen in the left part of figure 5.3(a) is likely an effect of the limited number of entry points used by the new propagation model, and can likely be reduced by using a larger number of entry points, though this would also increase computation times.

The difference between the two models in figure 5.5 is possibly due to the same reason, ie.

the limited number of entry points.

Looking at figures 5.7 and 5.8, the detailed model seem to suggest higher PG for cell edge users than what KM suggests for both floor plans, which could also be seen in the heat maps in figure 5.2 and 5.4. However, PG for users in general seem lower with Spoke

(32)

than KM, and the statistical model suggests higher rates for almost all users, except for the best users with KM and the open floor plan.

Note that these results are valid in the case of a LOS macro BS, capable of beam steering to provide a high antenna gain to each user, independent of height in the building.

A more realistic scenario, where several macros exist but none has LOS, is simulated in the city scenario in the next chapter.

(33)

Chapter 6 City Scenario

In this scenario, a building with the same parameters as in chapter 5 is placed in a fictional city environment with a large number of outdoor BSs. Results are presented in the form of gain cdfs, as in the previous chapter.

6.1 Setup

The city center, where the building is placed, is shown in figure 6.1. Users are placed in this building in the same manner as in the single building scenario. A number of macro antennas are predefined in the city, and the PG presented in the cdfs will be to the strongest BS for each user.

(a) (b)

Figure 6.1: (a) The city centre with (b) a zoomed in top view around the actual building (shown in orange), including nearby macro antennas (green stars).

6.2 Results

The results from a simulation at 5 GHz are shown in figure 6.2, and a similar curve for 10 GHz in figure 6.3. Results are summarized for all frequencies in figure 6.4. Once again,

(34)

the closed floor plan provides a lower PG in general. The curve for the statistical model somewhat resembles that of an open floor plan, whereas it is highly optimistic for the closed floor plan, especially when it comes to edge users. In the city scenario, it is no longer clear that Spoke gives a better PG for edge users. For the open office, Spoke and KM perform similarly for edge users, whereas for the closed office, Spoke rather seems to suggest a lower PG for almost all users, including edge users.

-180 -160 -140 -120 -100 -80

Path gain [dB]

0 20 40 60 80 100

CDF [%]

(a)

-180 -160 -140 -120 -100 -80

Path gain [dB]

0 20 40 60 80 100

CDF [%]

(b)

-180 -160 -140 -120 -100 -80

Path gain [dB]

0 20 40 60 80 100

CDF [%]

(a)

-180 -160 -140 -120 -100 -80

Path gain [dB]

0 20 40 60 80 100

CDF [%]

(b)

(35)

-180 -160 -140 -120 -100 -80 Path gain [dB]

0 20 40 60 80 100

CDF [%]

2 GHz 5 GHz 10 GHz 30 GHz

(a)

-180 -160 -140 -120 -100 -80

Path gain [dB]

0 20 40 60 80 100

CDF [%]

2 GHz 5 GHz 10 GHz 30 GHz

(b)

Figure 6.4: PG for new building type, using Spoke for (a) closed and (b) open office.

Finally, cell edge rates are summarized in figure 6.5(a), which can be seen to be more pessimistic than for the single building scenario. Here, only the old building types manage to get a PG above -131 dB for its edge users, at a frequency of 5 GHz. For the higher frequencies of 10 and 30 GHz, none of the parameter sets manage to achieve the minimum PG. As in the single building scenario, figure 6.5(b) shows the results of an indoor deployment for comparison.

2 5 10 15 20 25 30

Frequency [GHz]

-200 -180 -160 -140 -120 -100 -80

Open office, old BT Closed office, old BT Open office, new BT Closed office, new BT -131 dB

(a)

2 5 10 15 20 25 30

Frequency [GHz]

-140 -130 -120 -110 -100 -90 -80

Open office, indoor deployment Closed office, indoor deployment -118 dB

(b)

Figure 6.5: PG, 5th percentile, using Spoke. The green, horizontal line shown the minimum required PG in order to reach a SNR of 0 dB. (a) Outdoor base stations and (b) indoor deployment with 5 base stations per floor for comparison.

(36)

6.3 Discussion

In general, the gain in the city scenario is lower for all propagation methods. This could be due to the lack of LOS path, something which has previously been noted in measurements [20]. Furthermore, the difference between propagation models seems smaller in the city scenario, with the exception of the closed office simulations with Spoke for which gain is considerably lower than for all other methods and parameter combinations. Possibly, this suggests that the new model is pessimistic in non-LOS scenarios, even though further verification and calibration is needed to evaluate the new propagation method.

(37)

Chapter 7 Conclusions

In this work, simulations were run in different scenarios for carrier frequencies of 2, 5, 10 and 30 GHz. The general conclusion is that outdoor to indoor coverage might be possible for frequencies of 5, or possibly 10 GHz, whereas extremely challenging for 30 GHz, even with high beam forming gain. For this reason, indoor base station are likely necessary if frequencies of 10 GHz and above are used for indoor coverage in future networks.

The indoor analysis done in chapter 4 suggests that having a detailed indoor model becomes increasingly important for higher frequencies, likely because reflections and diffrac- tions play a larger role as the wall propagation losses become larger. Furthermore, the indoor analysis suggests that existing indoor loss per meter models might be too optimistic, at least for anything but open office floor plans. Another observation is that, for high frequencies, assuming a linear loss per meter relationship might not be correct. After roughly 10 meters, a breakpoint can be noted after which the loss increases more slowly as a function of distance. Once again, this is likely to due the role of the diffracted and reflected paths at high frequencies. If case linear relationship between excess loss and distance is assumed, the loss as well as its standard deviation both seem to be linear functions of carrier frequency.

For the single building scenario of chapter 5, there is a clear difference between the different propagation models and floor plans. As a general trend, Spoke suggests higher gain for edge users, but slightly lower gain in general. Achieving a SNR of 0 dB is possible in general for frequencies of up to 5 GHz, even though new outer walls with infrared coating as well as a closed office floor plan make coverage harder. Furthermore, the type of office play a large role in achieved gain, and for this reason statistical models should adapt a loss per meter model corresponding to the simulated floor plan.

The conclusions from the single building case do not directly translate to the city scenario in chapter 6. Here, achieving a SNR of 0 dB is challenging also at 5 GHz. Fur- thermore, Spoke does not anymore suggest higher gain for edge users, and does in general give a lower gain. One possibility is that the lack of LOS paths account for these differences, although further verification and calibration of the new model is most likely needed.

Similar to the single building though, the floor plan and building type play an important role in determining whether users are in coverage or not.

(38)

Bibliography

[1] A. Osseiran et al. “Scenarios for the 5G Mobile and Wireless Communications: The Vision of the METIS Project”. In: IEEE Commun. Mag. (May 2014).

[2] J. G. Andrews et al. “What will 5G be?” In: IEEE J. Sel. Areas Commun. 32.6 (June 2014), pp. 1065–1082.

[3] 5g: What is it? Ericsson, Stockholm, Sweden, White paper, Oct. 2014. url: http:

//www.ericsson.com/res/docs/2014/5g-what-is-it.pdf.

[4] 5G Radio Access: Requirements, Concept and Technologies. NTT DOCOMO, Tokyo, Japan, White paper, July 2014. url: https://www.nttdocomo.%20co.jp/english/

binary / pdf / corporate / technology / whitepaper _ 5g / DOCOMO _ %205G _ White _ Paper.pdf.

[5] 5g: A Technology Vision. Huawei Technologies, Shenzhen, China, 2013. url: http:

//www.huawei.com/5gwhitepaper/.

[6] Looking ahead to 5G: Building a virtual zero latency gigabit experience. Nokia Solu- tions and Networks, Espoo, Finland, White paper, 2014. url: http://networks.

nokia.com/sites/default/files/document/5g_white_%20paper_0.pdf.

[7] Estimated spectrum bandwidth requirements for the future development of IMT-2000 and IMT-Advanced. Geneva, Switzerland, Rep. M.2078, 2006.

[8] Z. Pi and F. Khan. “An introduction to millimeter-wave mobile broadband systems”.

In: IEEE Commun. Mag. 49.6 (June 2011), pp. 101–107.

[9] T.S Rappaport. Wireless Communications: Principles and Practice. Prentice Hall, 2002.

[10] M.F Iskander and Zhengqing Yun. “Propagation prediction models for wireless communication systems”. In: IEEE Transactions on Microwave Theory and Techniques 50.3 (2002), pp. 662–673. doi: 10.1109/22.989951.

[11] H. Holmquist J. -E. Berg. “An FFT multiple half-screen diffraction model”. In: Ve- hicular Technology Conference, 1994 IEEE 44th 1 (June 1995), pp. 195–199. doi:

10.1109/VETEC.1994.345139.

(39)

[12] J. -E. Berg. “A recursive method for street microcell path loss calculations”. In:

PIMRC’95. Wireless: Merging onto the Information Superhighway., Sixth IEEE In- ternational Symposium on 1 (Sept. 1995), pp. 140–143. doi: 10.1109/PIMRC.1995.

476420.

[13] C. Larsson et. al. “An Outdoor-to-Indoor Propagation Scenario at 28 GHz”. In:

Antennas and Propagation (EuCAP), Proc. of the 6th European Conference (Apr.

2014).

[14] E. Semaan et. al. “Outdoor-to-Indoor Coverage in High Frequency Bands”. In: Globe- com Workshops (GC Wkshps) (Dec. 2014), pp. 393–398. doi: 10.1109/GLOCOMW.

2014.7063463.

[15] H. Zhao et. al. “28 GHz Millimeter Wave Cellular Communication Measurements for Reflection and Penetration Loss in and around Buildings in New York City”. In:

IEEE International Conference on Communications (ICC) (2013), pp. 5163–5167.

issn: 1550-3607. doi: 10.1109/ICC.2013.6655403.

[16] H. Xu et. al. “Spatial and temporal characteristics of 60-GHz indoor channels”. In:

IEEE Journal on Selected Areas in Communications 20.3 (2002), pp. 620–630. doi:

10.1109/49.995521.

[17] J. Medbo et. al. “Multi-frequency path loss in an outdoor to indoor macrocellular scenario”. In: 3rd European Conference on Antennas and Propagation (2009), pp. 3601–

3605.

[18] J-E. Berg. “Building Penetration Loss Along Urban Street Microcells”. In: Proc.

PIMRC’96 (1996), pp. 795–797.

[19] A.J. Motley J.M Keenan. “Radio coverage in buildings”. In: British Telecom Tech- nology Journal 8.1 (Jan. 1990), pp. 19–24.

[20] T. S. Rappaport et. al. “Broadband Millimeter-Wave Propagation Measurements and Models Using Adaptive-Beam Antennas for Outdoor Urban Cellular Communi- cations”. In: IEEE Transactions on Antennas and Propagation 61.4 (2013), pp. 1850–

1859. doi: 10.1109/TAP.2012.2235056.

Outdoor to Indoor Coverage in 5G Networks

Examensarbete 30 hp Juni 2016

Outdoor to Indoor Coverage in 5G Networks

Vilhelm Rydén

Abstract

Outdoor to Indoor Coverage in 5G Networks

Sammanfattning

Contents

Glossary

Chapter 1 Introduction

1.1 Goals

1.2 Thesis Outline

Chapter 2

Technical Background

2.1 Signal-to-noise Ratio

2.2 Channel Capacity

2.3 Path Loss and Path Gain

2.3.1 Free Space Path Loss

2.3.2 Antenna Gain and EIRP

2.3.3 Friis Transmission Equation

2.4 Propagation Models

2.5 Reflection, Diffraction and Scattering

2.6 Required Path Gain

2.6.1 Cell Edge Path Gain

Chapter 3

Methodology

3.1 The Simulator

3.1.1 Ray Tracing

3.1.2 Simulation Parameters

3.2 Propagation Models

3.2.1 Outdoor Propagation Model

3.2.2 Wall Penetration Losses

3.2.3 Statistical Indoor Model

3.2.4 Partitioned Indoor Loss Model (Keenan-Motley)

3.2.5 Detailed Indoor Model (Spoke)

3.2.6 Outdoor to Indoor Propagation Model

3.3 New Outdoor to Indoor Propagation Model

3.3.1 Description of the New Model

Chapter 4

Single Floor Analysis

4.1 Setup

4.2 Results

4.3 Discussion

Chapter 5

Single Building Scenario

5.1 Setup

5.2 Results

5.3 Discussion

Chapter 6

City Scenario

6.1 Setup

6.2 Results

6.3 Discussion

Chapter 7 Conclusions

Bibliography