Performance of In-Band Full-Duplex for 5G Wireless Networks

(1)

IN

DEGREE PROJECT

ELECTRICAL ENGINEERING,

SECOND CYCLE, 30 CREDITS

,

STOCKHOLM SWEDEN 2016

Performance of In-Band

Full-Duplex for 5G Wireless

Networks

OSAMA AL-SAADEH

KTH ROYAL INSTITUTE OF TECHNOLOGY

(2)

(3)

Performance of In-Band Full-Duplex for

5G Wireless Networks

OSAMA AL-SAADEH

Master of Science Thesis performed at

the Communication Systems Department, KTH.

June 2016

(4)

(5)

Abstract

In-band full duplex is a new duplexing scheme that allows radio nodes to trans-mit and receive, utilizing the same frequency and time resources. The imple-mentation of in-band full duplex was not feasible in practice, due to the effect of self-interference. But then, advances in signal processing made it possible to reduce this effect. However, the system level performance of in-band full duplex has not been investigated thoroughly.

Through computer simulations, we investigate the performance of in-band full duplex, for indoor 5G small cell wireless networks. We examine the perfor-mance of in-band full duplex in comparison to dynamic and static time division duplexing. Additionally, we analyze the performance of the duplexing schemes with two interference mitigation techniques, namely beamforming and interfer-ence cancellation.

Our results indicate that for highly utilized wireless networks, in-band full duplex should be combined with beamforming and interference cancellation, in order to achieve a performance gain over traditional duplexing schemes. Only then, in-band full duplex is considered advantageous, at any network utilization, and any downlink to uplink traffic demand proportion. Our results also suggest that in order to achieve a performance gain with in-band full duplex in both links, the transmit power of the access points should be comparable to the transmit power of the mobile stations.

Keywords: Wireless networks, In-band full duplex, Static-time division du-plexing, Dynamic-time division dudu-plexing, Interference mitigation techniques, small cell, 5G, mmWave bands, Beamforming, Interference cancellation.

(6)

(7)

Sammanfattning

Inomband hel duplex är en ny typ av duplexmetod som till˚ater radionoder att sända och ta emot i samma frekvens- och tidsresurs. Att implementera inom-band hel duplex har fram tills nu inte ansetts praktiskt genomförbart till följd av självstörningar. Framsteg inom signalbehandling har dock gjort det möjligt att begränsa denna självstörningseffekt. Emellertid har systemprestanda av inomband hel duplex inte undersökts tillräckligt noga i tidigare verk.

Inomband hel duplex och dess prestanda för tr˚adlösa 5G sm˚acellsnätverk in-omhus har studerats med hjälp av datasimuleringar och jämförts med dynamisk och statisk tidsdelning. Utöver detta har prestanda för de olika duplexme-toderna med avseende p˚a tv˚a tekniker för störningsundertryckning, lobformning och störningseliminering, ocks˚a undersökts.

V˚ara resultat indikerar att för tr˚adlösa nätverk med högt radioresursutnyt-tjande bör inomband hel duplex kombineras med lobformning och st¨ orningse-liminering för att uppn˚a en prestandavinst jämfört med traditionella duplexme-toder. Bara d˚a kan inomband hel duplex anses vara fördelaktig oberoende av radioresursutnyttjande och andelen upp- och nedlänkstrafik.

Resultaten tyder ocks˚a p˚a att sändareffekten för radioaccesspunkterna bör vara jämförbar med den för mobilenheterna för att en prestandavinst med in-omband hel duplex ska kunna uppn˚as.

Nyckelord: Tr˚adlösa nätverk, Inomband hel duplex, Statisk tidsdelning, Dy-namisk tidsdelning, Störningsundertryckningsmetoder, Sm˚a celler, 5G, Millime-terv˚agsfrekvenser, Lobformning, Störningseliminering.

(8)

(9)

List of Figures

2.1 Interference in S-TDD DL time slots. . . 5

2.2 Interference in S-TDD UL time slots . . . 5

2.3 S-TDD time slots. . . 6

2.4 D-TDD time slots. . . 6

2.5 DL to UL interference in D-TDD . . . 6

2.6 IBFD time slots. . . 7

2.7 Interference in IBFD at MSs. . . 7

2.8 Interference in IBFD at APs . . . 7

2.9 Example on antennas placement. . . 8

4.1 2D sketch of the simulation environment. . . 13

4.2 CI and CIF Path losses with the distance. . . 15

4.3 A 2D layout of the simulation environment. . . 22

4.4 All possible MSs placements . . . 23

4.5 Association between APs and MSs . . . 23

4.6 DL SNR CDF with variable APs’ transmit power . . . 24

4.7 UL SNR CDF with variable MSs’ transmit power . . . 24

5.1 MS DL throughput with low network utilization. . . 28

5.2 MS DL throughput with high network utilization. . . 29

5.3 MS DL throughput with highly congested network. . . 29

5.4 MS DL throughput with beamforming only, under variable num-ber of antenna elements. . . 30

5.5 MS UL throughput with low network utilization. . . 30

5.6 MS UL throughput with high network utilization. . . 31

5.7 MS UL throughput with highly congested network. . . 32

5.8 MS UL throughput with beamforming only, under variable num-ber of antenna elements. . . 32

5.9 MS DL and UL throughputs with beamforming and interference cancellation, at a high network utilization and 1024 antenna ele-ments. . . 33

5.10 MS sum throughput with variable UL traffic demand. . . 34

5.11 MS DL throughput with variable APs’ transmit power. . . 34

5.12 MS UL throughput with variable APs’ transmit power. . . 35

5.13 Average hourly network energy consumption, with different com-binations of interference mitigation techniques. . . 36

(12)

(13)

List of Tables

4.1 Access points coordinates in the simulation environment. . . 14

4.2 List of heights and materials in the simulation environment. . . . 14

4.3 Parameters of the CI and the CIF models for indoor office envi-ronments. . . 15

4.4 The values of ω and ρ for different duplexing schemes. . . 18

4.5 Modified Shannon formula parameters. . . 19

4.6 Energy consumption model parameters. . . 21

5.1 Simulation parameters . . . 27

(14)

(15)

Acknowledgements

Firstly, I would like to express my truthful gratitude to my advisor Dr. Ki Won Sung for his continuous support, patience, and motivation, during the period of this thesis. His immense knowledge and guidance helped me during the time of research and writing of this thesis. I could not have imagined having a better advisor for this thesis. Dr. Sung provided me with the opportunity to access and use the research facilities at the RADIO SYSTEMS laboratory. It would not have been possible to conduct this research otherwise.

Besides my advisor, I would like to thank my examiner Dr. Anders V¨astberg and the PhD candidates under Dr. Sung supervision both Haris Celik and Yan-peng Yang, for their insightful comments that widen my research from various perspectives.

I thank my fellow labmate Adri´an De Miguel Herr´aiz, for the stimulating discussions and the company during the last five months.

I would like to thank my family, especially my parents who have always provided me with their unconditional love and support throughout my life.

This thesis has been produced during my scholarship period at KTH Royal Institute of Technology, thanks to a Swedish Institute scholarship.

(16)

(17)

List of Symbols

A The set of all access point in the system

Ad The subset of access points scheduled in downlink

Au The subset of access points scheduled in uplink

Dd Proportion of downlink time slots in static-time division duplexing

Du Proportion of uplink time slots in static-time division duplexing

GK,M Path gain between access point K and access point M

GK,j Path gain between access point K and mobile station j

Gi,j Path gain between mobile station i and mobile station j

Ha,b The channel gain matrix between node a and node b

L Total number of mobile stations positioning realizations

N FA The access points noise figure

N Fm The mobile stations noise figure

PA Access points transmit power

Pm Mobile stations transmit power

R Average mobile station sum throughput

Rd Average mobile station downlink throughput Ru _{Average mobile station uplink throughput}

Ri Mobile station i sum throughput

Rd

i Mobile station i downlink throughput

Ru

i Mobile station i uplink throughput

S The overhead scaling

T Total time slots budget

W Transmission bandwidth

max The maximum spectral efficiency

(18)

xvi

ηd The average access points downlink cell load

ηu The average access points uplink cell load

ηγ The signal-to-interference-plus-noise ratio efficiency

ηw The bandwidth efficiency

γd

j(t) Mobile station j instantaneous downlink signal-to-interference-plus-noise

ratio

γu

j(t) Mobile station j instantaneous uplink signal-to-interference-plus-noise

ratio

γmax The maximum signal-to-interference-plus-noise ratio that can be

de-tected

γmin The minimum signal-to-interference-plus-noise ratio required for a

suc-cessful transmission

Cm×n Set of complex valued m×n matrices

ψd The mobile stations downlink traffic demand to the total traffic demand

ψu The mobile stations uplink traffic demand to the total traffic demand

σ2

A The total noise power at the access points

σ2

m The total noise power at the mobile stations

ζ Self-interference cancellation capability factor

c The speed of light

fc Frequency of operation

gK,M The equivalent channel gain between access point K and access point

M

gK,j The equivalent channel gain between access point K and mobile station

j

gi,j The equivalent channel gain between mobile station i and mobile station

j

m The set of all of mobile stations in the system

md The subset of mobile stations scheduled in downlink

mu The subset of mobile stations scheduled in uplink

nlos The path loss exponent for line-of-sight propagation

nnlos The path loss exponent for non-line-of-sight propagation

rd

j(t) Mobile station j instantaneous downlink rate

(19)

xvii

sJ Access point J transmitted data symbol

sj Mobile station j transmitted data symbol

wr_J Access point J postcoding vector wt_J Access point J precoding vector

(20)

(21)

xix

Acronyms

IBFD In-Band Full Duplex

D-TDD Dynamic-Time Division Duplexing

S-TDD Static-Time Division Duplexing

AP Access Point

MS Mobile Station

DL Downlink

UL Uplink

SI Self-Interference

TDD Time Division Duplexing

FDD Frequency Division Duplexing

MAC Medium Access Control

LOS Line-of-Sight

NLOS Non Line-of-Sight

SINR Signal-to-Interference-Plus-Noise Ratio

SNR Signal-to-Noise Ratio

2D Two-Dimensional

Cell-DTX Cell Discontinues Transmission

Cell-DRX Cell Discontinues Reception

QoS Quality of Service

MIMO Multiple-Input Multiple-Output

(22)

(23)

Chapter 1

Introduction

In order to meet the future society’s demand for high data rate and capacity requirements, the next-generation of wireless networks must support one thou-sand times higher mobile data volume per area, ten to hundred times higher data rate per user, and ten to one hundred times higher number of connected devices at a very short latency [1]. Such challenging requirements necessitate more spectrum resources to be dedicated to mobile communications services.

The scarcity of the available spectrum resources is a fundamental challenge for the next generation of wireless networks. In-band full duplex is a promising technology that provides an insight into tackling this challenge [2].

It used to be a fundamental belief that radio units cannot transmit and receive concurrently at the same frequency resources, due to self-interference (SI) [3]. The duplexing of transmission and reception must be done by either frequency division duplex (FDD) or time division duplex (TDD). Both FDD and TDD have the disadvantage of wasting time and frequency resources [4]. However, recent advances in signal processing have made it possible to reduce the effect of SI, allowing in-band full duplex (IBFD) wireless communications [5][6].

A wireless system that operates in IBFD requires half the frequency and time resources required to operate in FDD and TDD, respectively. On the other hand, IBFD might not increase the spectral efficiency of wireless networks. That is especially true for small cell wireless networks with high utilization, where the number of mobile stations (MSs) is much more higher than the number of access points (APs), leading to severe interference originating from both APs and MSs at the same time. In such interference limited systems, using interference mitigation techniques for IBFD, possibly will lead to a relative performance gain over traditional duplexing schemes.

The performance of IBFD wireless networks should be investigated thor-oughly, due to the following reasons:

• IBFD’s performance is determined by number of factors, including but not limited to: the network utilization, the downlink (DL) to uplink (UL) traffic demand proportion, the area of the network, the APs’ and the MSs’ transmit powers, and the density of the APs [7]. Some of these parameters can be controlled, for instance, we can make a well-studied decision on the APs’ transmit power. Some other parameters are out

(24)

2 Chapter 1. Introduction

of control, however, such as the network utilization, and the DL to UL traffic demand proportion. Consequently, it is necessary to investigate the performance of IBFD with both types of parameters, before making any conclusion about the benefits that IBFD brings compared to other duplexing schemes.

• The performance of IBFD combined with interference mitigation tech-niques have not been thoroughly studied in the literature.

• IBFD requires a more sophisticated design of the transceiver. Therefore, unless IBFD brings major system performance improvement, it would not be beneficial to apply.

1.1 Research Questions

The aim of this thesis is to address the following questions:

• Will IBFD bring throughput improvement, in both links, to the existing duplexing schemes, under any network utilization?

• If the answer to the first question is no. Will implementing interference mitigation techniques with IBFD, bring throughput improvement, in both links, to the existing duplexing schemes, under any network utilization?

• Will IBFD adapt better to asymmetric traffic demands, in comparison to the existing duplexing schemes?

• IBFD causes interference between DL and UL, in general the APs’ trans-mit power is higher than the MSs’ transtrans-mit power. How should we select the APs’ transmit power, for a wireless network that utilizes IBFD, in a way that we limit the interference between DL and UL?

• Wireless networks that operate only the APs in IBFD are reported to have a poor energy efficiency performance, compared to half duplex (HD) schemes [8]. Will implementing interference mitigation techniques, help IBFD to become more energy efficient?

The fundamentals of the duplexing schemes that are considered in this thesis can be found in Section 2.1, however, they are listed as follows:

1. Static-time division duplexing (S-TDD).

2. Dynamic-time division duplexing (D-TDD).

3. In-band full-duplex (IBFD).

The interference mitigation techniques under study are given below. More explanation on the techniques is available in Section 2.2:

1. Maximal ratio transmit and receive beamforming.

(25)

1.2. Benefits, Sustainability, and Ethics 3

1.2 Benefits, Sustainability, and Ethics

IBFD provides an insight into tackling the challenge of spectrum resources scarcity, by utilizing the same resources for both DL and UL [2]. IBFD provides a better way to detect collisions in contention-based access protocols [9]. More-over, IBFD improves the secrecy performance of relay networks significantly [10].

IBFD has been proven to be less energy efficient than the HD schemes. On the contrary, IBFD provides throughput gain over the HD schemes [8]. So, there is a trade-off between throughput performance and energy efficiency performance when it comes to IBFD.

On the other hand, in our opinion, we see no ethical issues that can arise due to the implementation of IBFD radios.

1.3 Outline

The content of this thesis is organized as follows: Chapter 2 includes the theo-retical background; the methodology is presented in Chapter 3; the simulation models, algorithms, and parameters are described in Chapter 4; Chapter 5 is devoted for the simulation results; and in Chapter 6 we present our conclusions with suggestions for future work.

(26)

(27)

Chapter 2

Background

2.1 Overview of duplexing schemes

2.1.1 Static-Time Division Duplexing

For a radio node that operates in S-TDD, transmission and reception of data occurs using the same frequency band, through allocating distinct time slots for UL and DL. Additionally, the transmission bandwidth of each link is fixed depending on the average traffic load [11] [12]. At any time instant, all the APs are synchronized to operate in the same transmission mode, and each AP can serve one MS at most.

Consider a simple setup consisting of two APs as in Figure 2.1, AP1 is having two MSs within its coverage area, each MS is requesting a different link. AP2 has only one MS within its coverage area with a DL demand. Let us assume that each MS necessitate two time slots to complete downloading or uploading.

Figure 2.1: Interference in S-TDD DL time slots.

Figure 2.2: Interference in S-TDD UL time slots

AP1 will serve its MSs in an alternating manner, while AP2 will serve MS3 every other time slot; as a result AP2 is always wasting half the time resources. The situation is clearly depicted in Figure 2.3.

Nevertheless, S-TDD will always avoid the interference between DL and UL as shown in Figures 2.1 and 2.2, where we highlight the interference (dashed red line) that AP1’s cell is experiencing.

(28)

6 Chapter 2. Background

Figure 2.3: S-TDD time slots.

2.1.2 Dynamic-Time Division Duplexing

In D-TDD the nodes can transmit and receive data on the same frequency band, through allocating distinct time slots for UL and DL just like S-TDD. Conversely, with D-TDD the APs are not synchronized to be in the same trans-mission mode; besides, the DL and the UL bandwidths varies according to the instantaneous traffic load. D-TDD enables efficient and flexible asymmetric services, which improves the spectral efficiency of wireless networks [11].

For the same scenario introduced previously, with D-TDD, MS3 will take advantage of the situation and stay active until it finishes downloading. MS1 and MS2 are now downloading and uploading with variable bandwidths according to the instantaneous traffic load as depicted in Figure 2.4. As a result, MSs 1 and 3 finished downloading earlier with D-TDD. This is only true if the interference level in D-TDD is similar to the interference level in S-TDD.

Figure 2.4: D-TDD time slots.

Nevertheless, D-TDD allows interference between DL and UL as shown in Figure 2.5. Interference between DL and UL can be problematic if the transmit powers of the APs and the MSs are imbalanced, in that case, one of the links will cause massive interference to the other link, and the MSs will need more time finish downloading or uploading.

Figure 2.5: DL to UL interference in D-TDD

(29)

2.2. Interference Mitigation Techniques 7

2.1.3 In-Band Full Duplex

With IBFD, radio nodes transmit and receive on the same frequency and time resources, allowing more MSs allocations, on the expense of more interference. An AP operating in IBFD can serve two MSs at one time instant, one MS in each link. Operating in IBFD causes severe interference in the network, composed of inter-cell interference, intra-cell interference, as well as SI. Recent advances in signal processing made it possible to reduce the effect of SI [5] [6]. Nonetheless, reducing the effect of SI will not guarantee a better performance for IBFD in comparison to D-TDD and S-TDD.

Back again to our scenario, now all the MSs will stay active until they finish downloading or uploading as demonstrated in Figure 2.6. Figures 2.7 and 2.8 depict the interference the reference cell (i.e. AP1’s cell) is experiencing with IBFD. The extra interference imposed by operating in IBFD will cause the MSs to stay active for a longer time until they finish downloading and uploading. However, if we can reduce the level of the interference in IBFD we might achieve a performance gain over D-TDD and S-TDD.

Figure 2.6: IBFD time slots.

Figure 2.7: Interference in IBFD at MSs. Figure 2.8: Interference in IBFD at APs

2.2 Interference Mitigation Techniques

IBFD will have higher interference compared to S-TDD and D-TDD. The terference in IBFD is composed of SI, inter-cell interference, and intra-cell in-terference. In the next three subsections, we will go through possible ways to mitigate this interference.

2.2.1 Self-Interference Cancellation

There are three existing approaches for SI Cancellation:

(30)

1. Antenna Separation.

Antenna separation reduces the effect of SI by maximizing the path loss between the transmit and the receive antennas. The path loss between the antennas is controlled by three parameters:

• The distance between the antennas.

The more spacing between the antennas, the less SI received at the receive antenna, indeed the spacing between the antennas is limited to the physical size of the radio node itself [5].

• The type of the antennas.

Obviously, utilizing directional antennas can lower the power of SI at the receive antenna in comparison to the case with omnidirectional antennas [13].

• The antennas placement.

One example of using antennas placement as a mean to reduce the received SI is what the researchers in [9] proposed. That if we are uti-lizing two transmit antennas (TX1and TX2) and one receive antenna

(RX), with the spacing between TX1 and RX being d, adjusting the

position of TX2such that the distance between TX2and RX is d+λ/2

as in Figure 2.9, causes the signal from the two transmit antennas to add destructively at the receiver, leading to a significant attenuation in the SI at RX. This approach was able to achieve 20 dB reduction in the received SI in [9].

Figure 2.9: Example on antennas placement.

2. Analog Cancellation.

Analog cancellation is an active cancellation scheme performed in the ana-log domain before the received signal passes through the Anaana-log-to-Digital Converter, by subtracting an estimate of the transmitted signal from the received signal [9] [13].

3. Digital Cancellation.

Due to imperfections in the analog cancellation scheme, active digital can-cellation estimates the transmitted signal in the digital domain and sub-tracts this estimate from the received signal in the digital domain [5] [13].

2.2.2 Maximal Ratio Transmit and Receive Beamforming

Beamforming is well-known for being an excellent interference mitigation tech-nique, as it can combat fading, increase spectral efficiency, and reduce interfer-ence [14].

When the APs have multiple transmit and receive antennas, a multi-stream transmission and reception can maximize the throughput in DL and UL. One of the most famous beamforming techniques is the maximal ratio transmit and

(31)

2.3. Related Work and Research Gap 9

receive beamforming, which increases the reliability by exploiting channel vari-ations. Maximal ratio transmit and receive beamforming utilizes precoding and postcoding to weight information streams, which maximizes the desired signal power while keeping the interference signals at the same level [14][15].

2.2.3 Successive Interference Cancellation

As some of the received interfering signals may be strong enough to decode, successive interference cancellation is an interference mitigation technique that allows receivers to decode packets that arrive concurrently. The strongest signal can be decoded and subtracted from the collision, allowing the decoder to rec-ognize weaker interfering signals and decoding them as well. The procedure can be repeated iteratively, as long as the collided signals have different strengths [16] [17].

2.3 Related Work and Research Gap

With antenna separations of 20 cm and 40 cm in [5] it was possible to reduce the SI using analog and digital interference cancellation techniques by more than 78 dB. In [6] a 110 dB of SI cancellation was achieved in a dense indoor office environment, with 80MHz bandwidth. However, the analog cancellation circuit in [6] had the dimensions of (10cm × 10cm) which is not implementable on phones and other portable devices. Both [5] and [6] conducted their experiments with a center frequency of 2.4 GHz.

In [9] the researchers argue that IBFD provides a better way to detect col-lisions in contention-based access protocols. And in [10] it is concluded that IBFD can improve the secrecy of relay networks significantly.

Numerous investigations have been conducted on the performance of IBFD; in [18] the performance of IBFD in a dense small cell network is evaluated and compared against the conventional HD transmission. It was confirmed in [18] that IBFD provides 30 − 40% mean throughput gain over S-TDD for indoor scenarios. The 100% throughput gain was only noticed when the cells are isolated by extremely high wall loss figures. In [19] it was demonstrated that IBFD cannot double the throughput of HD transmission with ALOHA Medium Access Control (MAC) protocol, even with perfect SI cancellation, the actual throughput gain ranges from 0−33%, for the path loss exponent range (2,4]. The authors in [19] arrived to the conclusion that there is a strong need for a MAC protocol for IBFD wireless networks, that intelligently switches between IBFD and S-TDD based on different network configurations. Nevertheless, Both APs and MSs were assumed to be having IBFD capabilities in [18] and [19], which is far from being a realistic assumption, due to the size limitation mentioned earlier in this section. We also believe that unlike what have been done in [18] and [19] when it comes to IBFD the performances of the DL and the UL should be investigated separately, due to the fact that IBFD is known to enhance the performance of one of the links on the expense of the other, making the total network throughput a non intuitive performance metric.

Researchers in [20] proposed a hybrid scheduler that shifts between IBFD and S-TDD based on the best available circumstances. For a single cell simu-lation scenario where only APs are IBFD capable, with 85 dB SI cancelsimu-lation

(32)

capability, S-TDD throughput was improved by 69% in DL, and 81% in UL. Researchers in [8] have studied a multi-cell scenario where only APs are IBFD capable with intelligent scheduler, power allocation, and selection of MSs that maximizes the throughput. For an indoor simulation scenario, doubling the throughput was possible in both DL and UL, with perfect SI cancellation, how-ever, with a drawback in the energy efficiency. Researchers in [8] also concluded that the degradation in energy efficiency due to IBFD operation at the APs can be resolved by operating MSs in IBFD as well. On the other hand, the intelligent scheduler proposed in [20] and [8] might not be suitable for situations where the network is very highly utilized, as the search for the appropriate combination of MSs to serve will increase the latency of the network.

Researchers in [11], studied the feasibility of blind D-TDD for ultra-dense networks. It was observed in [11], that blind D-TDD outperforms S-TDD when the instantaneous traffic demands for DL and UL are highly asymmetric. How-ever, there has not been any research that compares the performance of D-TDD to IBFD.

Another advantage for D-TDD, that it does not require a highly compli-cated design for the transceivers, unlike IBFD. So, it is vital to compare the performances of IBFD to D-TDD, for the reason that, if IBFD cannot bring a high throughput gain over D-TDD it would not be sensible to implement IBFD in future wireless networks, where researchers are attempting to achieve any performance enhancements, at low system design complexity.

Also, to our best knowledge, there has been no research that investigates the performance of IBFD with variable and asymmetric traffic demands, which we believe are the most important parameters to determine the flexibility of a certain duplexing scheme, and its ability to adapt to various traffic demands.

2.4 Contribution

In this thesis, we will investigate the performance of the duplexing schemes S-TDD, D-S-TDD, and IBFD for 5G small cell wireless networks, under variable and asymmetric traffic demands, with the interference mitigation techniques beamforming and interference cancellation.

The performance metrics we will consider are the MS DL throughput, the MS UL throughput, the MS sum throughput, and the average hourly network energy consumption.

Through MATLAB computer simulations, we will focus our study into an indoor simulation environment. All simulation scenarios will be conducted with the assumption that the APs are IBFD capable with SI cancellation of 110 dB, while the MSs are HD.

(33)

Chapter 3

Methodology

We will be following an experimental research method in this thesis, as it is mostly suitable for evaluating a system’s performance and study relationships between variables [21]. Experiments are used as a data collection method, along-side with computational mathematics as a data analysis method.

Since using only mathematical methods to answer the research questions can be quite challenging, we will be combining the mathematical methods with a Monte Carlo snapshot-based computer simulations, implemented in MATLAB.

3.1 Experiments and Experiments Design

In this thesis we will conduct four key experiments, explained in the following subsections.

3.1.1 Impact of Interference Mitigation Techniques With

Variable Traffic Demand

In the first experiment, we will be measuring the MS DL and UL throughputs at variable number of active MSs in the system. We will repeat the same experi-ment for four different scenarios, each scenario represents a different combination of interference mitigation techniques. The four scenarios can be described as follows:

Scenario 1: A baseline scenario, without beamforming or interference cancel-lation being used.

Scenario 2: A scenario with only interference cancellation implemented.

Scenario 3: A scenario with only beamforming applied.

Scenario 4: A scenario with both beamforming and interference cancellation techniques functional.

All the experiments will be conducted with the assumption that the APs are able to cancel the SI by 110dB.

We will also study how the performance changes with different number of antenna elements, in the third scenario.

(34)

12 Chapter 3. Methodology

As will come later in Section 4.1.1, the total simulation environment area is 200 m2, containing five APs. Given that A stands for the number of APs, and m represent the number of mobile stations, we will highlight the achievable DL and UL throughputs and throughput gains for three specific network utilization states, described as follows:

Network Utilization State 1: m = 7, represents a network with low utiliza-tion (A ' m).

Network Utilization State 2: m = 20, representing a highly utilized net-work (A < m).

Network Utilization State 3: m = 200, representing a highly congested net-work (A m). It is highly unlikely to have 1 MS/m2 _{MSs’ density,}

however, this network utilization state will enable us to make a general conclusion about the performance of the duplexing schemes with variable traffic demand.

3.1.2 Impact of Interference Mitigation Techniques With

Asymmetric Traffic Demand

The second experiment aims at identifying the behavior of the duplexing schemes, with beamforming and interference cancellation, under variable DL to UL traf-fic demand proportions. The parameter to be varied in this experiment is ψu,

which stands for the proportion of the MSs UL demand to the total demand. ψu will be varied from 0 to 1, with ψu = 0 represent a network with all the

MSs requesting DLs, while at ψu = 1 all the MSs request ULs. For all other

experiments, we set ψu to 0.5.

To give clearer picture about the behavior of the duplexing schemes, we believe that presenting the MS sum throughput will be more intuitive in this experiment. Furthermore, the experiment will be conducted at the high network utilization state.

3.1.3 Impact of Access Points’ Transmit Power

In this experiment we will investigate the impact of variable APs’ transmit power, on the MS DL and UL throughputs, for the duplexing schemes. Same as the last two experiments, this experiment will be held with beamforming and interference cancellation being active, at the high network utilization state.

3.1.4 Impact of Interference Mitigation Techniques On

The Radio Access Network Energy Consumption

For the radio access network energy consumption study, will measure the total average hourly network energy consumption, with the four previously discussed scenarios. The experiment will be conducted at the high network utilization state.

(35)

Chapter 4

Simulation Models,

Algorithms, and

Parameters

4.1 Simulation Models

4.1.1 Simulation Environment

The simulation environment to be used in the numerical simulations is the sec-ond use case of METIS-II project [22]. This simulation environment represents one of the futuristic scenarios, where the nature of the indoor office work will require Internet services with high data rates and low latencies [23]. This simula-tion environment is suitable for investigating the performance of the duplexing schemes, because it is easy to manipulate different parameters related to the MSs’ traffic demand, and track the effect of these parameters on the perfor-mance. Also, this indoor simulation environment has a realistic environmental model geometry, including cubicle offices and tables, adding more credibility to the results obtained in this study.

Figure 4.1: 2D sketch of the simulation environment.

A two-dimensional (2D) sketch of the simulation environment is shown in

(36)

14 Chapter 4. Simulation Models, Algorithms, and Parameters

Figure 4.1. The simulation environment has the dimensions of (20m×10m×2.9m), where the APs are non-uniformly distributed, with the coordinates in Table 4.1. Furthermore, the heights and materials of the objects in the simulation envi-ronment are given in Table 4.2 [24]. The MSs are uniformly distributed within the simulation environment, at a fixed height of 0.75m.

Table 4.1: Access points coordinates in the simulation environment.

AP X[m] Y[m] Z[m] 1 2 2 2.85 2 2 8 2.85 3 18 8 2.85 4 18 2 2.85 5 10 5 2.85

Table 4.2: List of heights and materials in the simulation environment.

Height [m] Material Room 2.9 Concrete Cubicle 1.5 Wood

Table 0.7 Wood

4.1.2 Propagation Model

In order to satisfy the data rate and latency requirements expected in the future, it is necessary to deploy 5G wireless networks at frequencies in the mmWave bands. The mmWave bands are known to have wide bandwidths, also the very short wavelengths of the mmWave bands allow the implementation of beam-forming, with a huge number of antenna elements. In this thesis, we will utilize 70Ghz frequency of operation, at a bandwidth of 1GHz as suggested in [22]. Furthermore, the propagation models presented in [25], were recom-mended by METIS-II project deliverables, to model the propagation channel at the mmWave bands [22].

Accordingly, we will utilize the close-in (CI) free space reference distance path loss model, for the case of line-of-sight (LOS) propagation [25]. The CI path loss model is given by

P LCI(fc, d)[dB] = 10 × nlos× log10(d/1m) + LFS(fc, 1m) + Xσlos, (4.1)

where nlos refers to the path loss exponent for LOS transmission, d is the

separation distance between the transmitter and the receiver in meters, fc is

the frequency of operation in Hz which has to belong to the (0.5GHz - 100GHz) range of frequencies, Xσlos is a zero-mean Gaussian random variable with

stan-dard deviation σlos in dB representing the LOS shadow fading, and LFS(fc, 1m)

(37)

4.1. Simulation Models 15

LFS(fc, 1m)[dB] = 20log10(

4πfc

c ), (4.2)

with c being the speed of light.

On the other hand, the non-line-of-sight (NLOS) propagation path loss will be modeled with the close-in free space reference distance model with frequency-dependent path loss exponent (CIF) [25]. The CIF path loss model can be expressed as

P LCIF(fc, d)[dB] = 10 × nnlos× (1 + b(fc− f0)/f0)) × log10(d/1m)

+ LFS(fc, 1m) + Xσnlos,

(4.3)

where nnlosrepresents the path loss exponent for NLOS propagation, b is an

optimization parameter that models the frequency dependency of the path loss exponent, f0is a fixed reference frequency, and Xσnlos is a zero-mean Gaussian

random variable with standard deviation σnlos in dB representing the NLOS

shadow fading [25].

Table 4.3 summarizes the parameters of the CI and the CIF models, for an indoor office environment as suggested in [25], same parameters will be utilized in our simulations.

Table 4.3: Parameters of the CI and the CIF models for indoor office environments.

nlos σlos nnlos b f0 σnlos c

value 1.73 3.02 2.19 0.06 24.2 GHz 8.28 3 × 108_m/s

It is highly important to accurately determine the sight conditions between transmitters and receivers, since accordingly the path loss will be calculated. It is no surprise that a NLOS path will have higher loss compared to the LOS one. For example, according to the path loss models introduced here, together with the parameters in Table 4.3, at frequency 70GHz, with the shadow fading ignored, we can see the difference between the calculated CI and CIF path losses with the distance as in Figure 4.2.

(38)

We will also assume that the propagation occurs over a Rayleigh fading channel.

4.1.3 Mathematical Models

Signal-to-Interference-Plus-Noise Ratio Model

Let us start by analyzing the received signal in DL. If AP J is serving MS j in DL, MS j will be receiving interference from all APs in DL (Ad) except AP J ,

and all MSs in UL (mu). Then the received signal in DL at MS j (ydj) can be

formulated as yd_j = pPAGJ,jHJ,jwtJsJ | {z } Desired signal + X Ad6=J pPAGAd,jHAd,jw t AdsAd | {z }

Interference due to other APs in DL

+ ωX mu pPmGmu,jHmu,jsmu | {z } Interference due to MSs in UL +nm, (4.4)

where PA and Pm are the APs’ and the MSs’ transmit powers; GAd,j and

Gmu,j are the link gains from AP Ad and MS mu to MS j; HAd,j and Hmu,j

are the channel gain matrices from AP Ad and MS mu to MS j; wtAd denotes

the precoding vector of AP Ad; sAd and smu are the transmitted data symbols

from AP Ad and MS mu with

E[ksAdk

2_{] = E[ks} muk

2_{] = 1;} _(4.5)

nm represents the zero mean additive white Gaussian noise at the MSs

re-ceiver; and ω value depends on the duplexing scheme under study according to Table 4.4.

In Equation (4.4), HAd,j ∈ C

1×M_{, with M representing the number transmit}

antenna elements. As we are considering a Rayleigh fading channel, HAd,j can

be modeled as a complex Gaussian random vector [26], such that

HAd,j ∼ CN (0, I). (4.6)

On the other hand, Hmu,j ∈ C

1×1_{, and}

Hmu,j ∼ CN (0, 1) (4.7)

Moreover, wt Ad∈ C

M ×1_{denotes the precoding vector of AP A}

d. For maximal

ratio transmit beamforming, wt

Ad can be formulated as wtAd= H†_A d,md kHAd,mdk , (4.8)

with md denoting the MS which is being served by AP Ad[15]. It has to be

noticed that H† denotes the Hermitian of matrix H.

The zero mean additive white Gaussian noise at the MSs receiver is denoted by nm, which is

(39)

with σ2

m being the total noise power at the MSs. The value of σm2 is given

by

σ_m2 = N Fm× W × K × T0, (4.10)

with N Fm representing the MSs noise figure, W denotes the transmission

bandwidth in Hz, T0is the reference thermal noise temperature of 290 Kelvins,

and K is Boltzmann’s constant ≈ 1.38−23 Joule/Kelvin [27].

The instantaneous Signal-to-Interference-Plus-Noise Ratio (SINR) for MS j in DL can be calculated as γ_jd(t) = PAGJ,jgJ,j Id j + σ2m , (4.11) with Id

j indicating the total interference that MS j is experiencing in DL

expressed as I_jd= X Ad6=J PAGAd,jgAd,j+ ω X mu PmGmu,jgmu,j. (4.12)

In Equations (4.11) and (4.12), gJ,j, gAd,j, and gmu,j are the equivalent

channel gains such that

gJ,j∼ Γ(M, 1), (4.13)

while

gAd,j and gmu,j ∼ exp(1). (4.14)

Γ(M, 1) refers to the Gamma distribution with α = M and β = 1, and exp(1) refers to the Exponential distribution with λ = 1 [15].

For the case with no beamforming, all the equivalent channel gains become exponentially distributed, due to the fact that HAd,j, and HJ,jreduce to complex

Gaussian random variables with zero mean and unity variance [15] [26]. On the UL, however, the received signal at AP J can be written as

yu_j = (wr_J)†(pPAGj,JHj,Jsj | {z } Desired Signal + ωX Ad pPAGAd,JHAd,Jw t AdsAd | {z }

Interference due to APs in DL

+ X mu6=j pPmGmu,JHmu,Jsmu | {z } Interference due to UL MSs + ρpPAGJ,JHJ,JwtJ) | {z } Self-Interference +nA, (4.15) where Hj,J ∈ CM ×1and HAd,J ∈ C

M ×M _{are the channel gain matrices from}

MS j and AP Ad to AP J . Both Hj,J and HAd,J ∼ CN (0, I) [26].

In Equation (4.15), wr_J ∈ CM ×1_{denotes the postcoding vector of AP J . For}

(40)

wr_J= Hj,J kHj,Jk

. (4.16)

The value of ρ depends on the duplexing scheme under study according to Table 4.4.

The zero mean additive white Gaussian noise at the APs receiver is denoted by nA. nA∈ C1×1, with

nA∼ CN (0, σA2). (4.17)

In here, σ2

A stands for the total noise power at the APs, calculated as

σ2_A= N FA× W × K × T0, (4.18)

with N FAbeing the noise figure of the APs’ receiver.

Consequently, the instantaneous UL SINR at AP J is given by

γ_ju(t) = PmGj,Jgj,J Iu

j + σ2A

, (4.19)

with I_ju denoting the total interference at AP J formulated as Iju= ω X Ad PAGAd,JgAd,J + X mu6=j PmGmu,Jgmu,J + ρζPAgJ,J. (4.20) In Equations (4.19) and (4.20), gj,J ∼ Γ(M, 1), (4.21) while gAd,J, gmu,J, and gJ,J ∼ exp(1). (4.22)

With no beamforming applied gj,J, gAd,J, gmu,J, and gJ,J are exponentially

distributed [15].

The parameter ζ represents the SI cancellation capability factor, which ranges from 0 to 1, with 0 indicating a perfect SI cancellation. In this the-sis, we will consider ζ = 10−11, corresponding to a 110 dB SI cancellation.

Table 4.4: The values of ω and ρ for different duplexing schemes.

S-TDD D-TDD IBFD

ω 0 1 1

(41)

Capacity Evaluation Model

According to the instantaneous DL and UL SINRs, we can calculate the instan-taneous DL and UL rates of MS j, according to the modified Shannon capacity formula as in Equation (4.23) [28]. rj(t)[bps] =      0, γ(t) < γmin S.W.max, γ(t) > γmax S.W.ηw. log2(1 + γ(t).ηγ), o.w (4.23)

The modified Shannon formula takes into account the bandwidth efficiency (ηw), the SINR efficiency (ηγ), and the maximum spectral efficiency (max)

which depends on the modulation scheme. The modulation scheme is assumed to be 512 QAM, as high-order modulation schemes are expected to come into play for 5G wireless networks [29].

In Equation (4.23), γ(t) refers to the instantaneous SINR whether it was in DL or UL; γmin is the minimum SINR required for a successful transmission;

and S is the overhead scaling. The values of the parameters applied in the modified Shannon formula are listed in Table 4.5 [28].

Table 4.5: Modified Shannon formula parameters.

S γmin γmax max ηγ ηW

value 0.75 -7 dB 32 dB 9 bps/Hz 0,8 0,88

Over T time slots and L MSs position realizations, with m being the total number of MSs in the system, md the subset of MSs scheduled in DL, and mu

the subset of MSs scheduled in UL, we can evaluate the average MS experienced DL rate as Rd[bps] = PL l=1 PT t=1 P mdr d j(t, l) L × T × m , (4.24) the average MS experienced UL rate as

Ru[bps] = PL l=1 PT t=1 P mur u j(t, l) L × T × m , (4.25) and the average MS experienced sum rate as

R[bps] = PL l=1 PT t=1 P mdr d j(t, l) + P mur u j(t, l) L × T × m . (4.26) Henceforth, we will be referring to Rd_{, R}u_{, and R as the MS DL throughput,}

the MS UL throughput, and the MS sum throughput.

4.1.4 Radio Access Network Energy Consumption Model

For IBFD to be favorable over S-TDD and D-TDD, it has to provide not only throughput improvement, but also energy efficiency improvement, or at least to keep the energy efficiency at the same level, if it can provide some other

(42)

performance improvements. Here we introduce our model for the radio access network energy consumption.

Cell Discontinues Transmission (Cell-DTX) and Reception (Cell-DRX) is a new technology, that gives the APs the ability to go into a sleeping mode when they have no MSs to serve, so the APs’ circuit power consumption is reduced significantly in these idle time slots [30].

We assume that the APs have Cell-DTX and Cell-DRX capabilities, we assume the same traffic demand in the network all the time, and we will evaluate the average hourly network energy consumption, that is required to maintain a certain quality of service (QoS) requirement for the duplexing schemes, with different combinations of interference mitigation techniques.

For AP J to satisfy a certain QoS requirement in DL for the MSs within its coverage area, it has to stay in the active state for time slots TJ

d, out of all the

time slots (T × L). Therefore, the DL cell load of AP J can be formulated as

ηJ_d = T

J d

T × L, (4.27)

same goes for the UL, with the UL cell load of AP J given by

ηJu =

TJ u

T × L. (4.28)

The average DL cell load (ηd) and the average UL cell load (ηu) can be

calculated by averaging over all the APs DL and UL cell loads.

In DL, the average hourly network energy consumption in kilojoules (kJ) can be written as

Ed[kJ] = 3.6 × A(ηdνPA+ ηdP0+ (1 − ηd)P0δ), (4.29)

where A is the total number of APs in the network, ν represents the feeder losses and power amplifier consumption, P0is the APs’ transmitter circuit power

consumption in the active state, which reduces to δP0when the transmitter go

to sleeping mode, (1 − ηd) designate the proportion of time the transmitter is

sleeping, and δ is a factor representing the amount of the transmitter circuit power consumption saving due to Cell-DTX.

The average hourly network energy consumption in UL can be formulated as

Eu[kJ] = 3.6 × A(ηuP0+ (1 − ηu)δP0). (4.30)

We are assuming that the receiver’s circuit power is the same as the trans-mitter’s circuit power.

The total average hourly network energy consumption can be calculated as

E[kJ] = Ed+ Eu. (4.31)

In here, the QoS requirement will be set to 100 Mbps throughput, in each link, for all users 100% of the time. Furthermore, we will consider the parameters in Table 4.6 for the energy consumption model as proposed in [31] and [30].

One more assumption in our energy consumption study, that is the circuit power of IBFD transceiver, will approximately be the same as the one for S-TDD and D-TDD transceivers, as the extra components responsible for SI cancellation will not influence the circuit power consumption substantially.

(43)

4.2. Simulation Algorithms 21

Table 4.6: Energy consumption model parameters.

ν P0 δ

value 4.7 3.2 Watts 0.29

4.2 Simulation Algorithms

To answer the research questions, we will use a Monte Carlo snapshot-based computer simulation, implemented in MATLAB. The general simulation algo-rithm is briefly described in Algoalgo-rithm 1 at the end of this chapter. Algoalgo-rithm 1 is used for the first three experiments.

In every simulation run, we have L MSs positioning realizations, for every positioning realization we have T time slots budget.

MSs are distributed randomly; but they can only exist on grids 0.5 meters spaced in the x-y plane. The height of the MSs is fixed to 0.75 meters. MSs are associated to the closest AP.

MSs demand a DL with probability ψd, and UL with probability ψu. Indeed,

ψd + ψu =1. A full buffer traffic model is assumed, where the MSs are always

having data to transmit or receive [32].

The instantaneous S-TDD system transmission mode (i state), is decided randomly, such that the system goes to DL with probability Dd, and UL with

probability Du. Accordingly, each AP decides the MS to serve if there is any.

The selection is performed with equal probability among all the MSs with a traffic demand similar to the system’s transmission mode.

For simplicity, we will consider a blind D-TDD scheme in this thesis. In blind D-TDD each cell allocates resources in a completely uncoordinated man-ner whenever there is a need for transmission [11]. The selection in D-TDD is performed such that each AP selects one MS out of all the MSs within its cov-erage area. The MS is selected with equal probability, regardless of its demand. An IBFD AP selects two MSs within its coverage area to serve. The selected MSs have dissimilar link demand. In each link the MSs are selected with equal probability among the MSs with a similar link demand. Furthermore, in this thesis, we assume an SI cancellation capability of 110dB at the APs, which is the amount of SI cancellation achieved in [6]. We will not use the same frequency band the researchers in [6] used, but we assume that in the future, such SI cancellation capability will be possible at this frequency band.

For the beamforming, and as we assume low complexity of MSs, we will utilize the maximal ratio transmit and receive beamforming at the APs side only, with 256 antenna elements, as suggested in [33]. Maximal ratio transmit and receive beamforming is utilized in this thesis due to its simplicity [15].

A simplified successive interference cancellation technique will be imple-mented in this thesis, where we will use the assumption that both APs and MSs can cancel the strongest interferer. We will be referring to this technique as interference cancellation in this thesis.

Algorithm 2 at the end of this chapter, briefly describes the simulation al-gorithm we follow for the radio access network energy consumption study. Two more parameters are introduced here, Rd_QoS and Ru_QoS corresponding to the QoS requirement in each link. In each iteration, we evaluate the DL and UL

(44)

throughputs for each MS individually. As soon as a MS reaches the predefined QoS requirement,in one link, of a certain duplexing scheme, it will never be allocated again in that link and that duplexing scheme. For each duplexing scheme, in each iteration, we record which APs are serving MSs, and which APs are sleeping. We calculate the DL and UL cell loads for each AP, and we average over all the APs cell loads in DL and UL to get the average DL cell load (ηd) and the average UL cell load (ηu). Now, we can evaluate the average

hourly network energy consumption (E).

To determine the sight condition, we create a 2D layout for the simulation environment, where we model the cubicles by lines as shown in Figure 4.3. Then, we create a line between the transmitter and the receiver, after that we check if that line intersects with any of the walls. In case of intersection, we record a NLOS transmission and we calculate the path loss according to the CIF model. Otherwise, we record a LOS transmission, and the path loss will be according to the CI model.

Figure 4.3: A 2D layout of the simulation environment.

To reduce the simulation running time, we perform one very long simulation where we place a MS in every possible point in the x-y plane, then we calculate the number of walls between the MSs, and between the MSs and the APs, eventually we record the number of walls in lookup tables, which we then use for every simulation run, instead of calculating the number of walls in every run. This approach saves the simulation running time drastically. Nonetheless, it must be perceived that the AP to AP paths are always LOS, because the APs are at a height of 2.85m while the cubicles are at 1.5m height.

4.3 Simulation Parameters

Most of the simulation parameters were chosen according to METIS-I and METIS-II deliverables suggestions, however, we had the liberty to choose some of the simulation parameters, according to detailed studies as discussed in the following subsections.

(45)

4.3. Simulation Parameters 23

4.3.1 Time Slots and Positioning Realizations

The values of T and L were selected to be 500 each, after rigorous studies. The criteria was to have a L value that is as high as possible to cover enough points inside the simulation environment. While T value should be selected so that all possible selections of MSs are covered. However, both T and L values are directly affecting the processing time of the simulation, so a compromise between accuracy and processing time should be done.

4.3.2 Transmit Powers

As operating in D-TDD and IBFD causes interference between DL and UL. It is of significant importance to set the transmit powers of the APs and the MSs carefully. The transmit powers of the APs and the MSs needs to be comparable, otherwise one of the links will add destructive interference to the other.

The following steps can be followed to determine the transmit powers, keep-ing in mind that γmin = -7 dB, and γmax = 32 dB:

1. We place one MS in each possible location within the simulation environ-ment as in Figure 4.4. Then, we associate each MS to its closest AP as in Figure 4.5.

Figure 4.4: All possible MSs placements Figure 4.5: Association between APs and MSs .

2. We calculate the DL and the UL Signal-to-Noise ratios (SNRs), for each MS.

3. We plot the cumulative distribution function (CDF), of the calculated SNRs for variable PA and Pm as in Figures 4.6 and 4.7.

4. Accordingly, we should select PAand Pmsuch that not a high percentage

of the SNRs being outside the range of (-7 dB, 32 dB). Consequently, 30 dBm and 25 dBm are sensible decision for the APs’ and the MSs’ transmit powers respectively.

4.3.3 Quality of Service Requirements

As we mentioned before, for the radio access network energy consumption study, we set both Rd

QoSand RuQoSto 100Mbps. We made this choice by trial and error,

as we were aiming at QoS requirement that is achievable at every scenario, and for every duplexing scheme, with the high network utilization state.

(46)

Figure 4.6: DL SNR CDF with variable APs’ transmit power

(47)

4.3. Simulation Parameters 25

Algorithm 1 General simulation algorithm

1: Set the simulation environment, and simulation parameters 2: Determine which interference mitigation techniques to be active 3: Determine network utilization state

4: for l=1:L do 5: Randomly distribute MSs 6: Associate MSs to APs 7: Create MSs demand 8: for t=1:T do 9: Activate S-TDD 10: Generate i state 11: if i state= DL then 12: Select MSs in DL to serve. 13: Calculate rd i(t, l), ∀ i ∈ md 14: else 15: Select MSs in UL to serve. 16: Calculate ru_i(t, l), ∀ i ∈ mu 17: Activate D-TDD 18: Select MSs to serve. 19: Calculate rd i(t, l), ∀ i ∈ md 20: Calculate ru i(t, l), ∀ i ∈ mu 21: Activate IBFD 22: Select MSs to serve. 23: Calculate rd i(t, l), ∀ i ∈ md 24: Calculate ru i(t, l), ∀ i ∈ mu

(48)

Algorithm 2 Radio access network energy consumption evaluation Algorithm 1: Set the simulation environment

2: Determine which interference mitigation techniques to be active 3: Set ψu, Du, m, M , RdQoS, RuQoS, PA, and Pm

4: for l=1:L do 5: Randomly distribute MSs 6: Associate MSs to APs 7: Create MSs demand 8: for t=1:T do 9: Activate S-TDD 10: Generate i state 11: if i state= DL then 12: Select MSs in DL to serve. 13: Calculate Rd i, ∀ i ∈ m 14: else 15: Select MSs in UL to serve. 16: Calculate Ru i, ∀ i ∈ m 17: for i=1:m do 18: if Rd i ≥ RdQoS then 19: Silence MS i in S-TDD DL. 20: if Ru_i ≥ Ru QoSthen 21: Silence MS i in S-TDD UL. 22: Activate D-TDD 23: Select MSs to serve. 24: Calculate Rd i, ∀ i ∈ m 25: Calculate Ru i, ∀ i ∈ m 26: for i=1:m do 27: if Rd i ≥ RdQoS then 28: Silence MS i in D-TDD DL. 29: if Ru i ≥ RuQoSthen 30: Silence MS i in D-TDD UL. 31: Activate IBFD 32: Select MSs to serve. 33: Calculate Rd i, ∀ i ∈ m 34: Calculate Ru_i, ∀ i ∈ m 35: for i=1:m do 36: if Rd_i ≥ Rd QoS then 37: Silence MS i in IBFD DL. 38: if Rui ≥ RuQoSthen

39: Silence MS i in IBFD UL.

(49)

Chapter 5

Simulation Results

In this chapter we present the simulation results. The simulation parameters are listed Table 5.1, unless otherwise specified.

Table 5.1: Simulation parameters

Parameter Pm PA fc W T L Value 25dBm 30dBm 70GHz 1GHz 500 500 Parameter N Fm N FA Dd m ψu ζ Value 9dB 5dB 0.5 20 0.5 1 × 10−11 Parameter TR A M Value 290K 5 256

5.1 Impact of Interference Mitigation Techniques

with Variable Traffic Demand

First, we will present the results for the performance of the duplexing schemes with variable number of active MSs. We will also demonstrate the performance with different combination of interference mitigation techniques employed. In here the performance under the three previously discussed network utilization states is highlighted. The percentages on top of the bar graphs between paren-theses represent the throughput gains with respect to S-TDD in the same sce-nario, the other percentages are the throughput gains with respect to S-TDD in the first scenario.

5.1.1 Downlink

As the network starts to get lightly utilized, in the first scenario, the MS DL throughput attained with D-TDD and IBFD starts to be comparable to that achieved with S-TDD, this is clearly demonstrated in Figure 5.1. Interference cancellation will lead to a better DL performance enhancement for D-TDD and IBFD compared to S-TDD. Topping the duplexing schemes with beamforming will add a substantial DL performance gain to all the duplexing schemes, in this

(50)

28 Chapter 5. Simulation Results

Figure 5.1: MS DL throughput with low network utilization.

scenario D-TDD and IBFD are achieving about 44% and 73% DL throughput gains over S-TDD in the same scenario. In the fourth scenario, a very little performance gain is noticed with S-TDD and D-TDD compared to the third scenario, IBFD, nevertheless, experienced a substantial performance gain.

At a high network utilization, IBFD starts to fall behind D-TDD and S-TDD, whom are still having a comparable DL performance, this is clearly demon-strated in Figure 5.2. Interference cancellation at high network utilization is improving the performance of all the duplexing schemes somewhat. Activating beamforming now is vital, as the interference is very high. With beamform-ing DL performance gains of 13% and 63% are noticed for D-TDD and IBFD over S-TDD in the same scenario. Beamforming with interference cancellation, are leading IBFD to become extremely superior over D-TDD and S-TDD when the network utilization is high. In the fourth scenario, IBFD is having a MS DL throughput gain of 79%, while D-TDD could amount to only 13% MS DL throughput gain relative to S-TDD in the same scenario.

For highly congested networks, IBFD will always allocate twice as much MSs than S-TDD and D-TDD. Indeed, the interference is massive for such situation, but beamforming together with interference cancellation are still having excel-lent performance in reducing the effect of this interference, as can be seen in Figure 5.3.

Figure 5.4 shows the MS DL throughput, for a highly utilized network, in the third scenario, with variable number of antenna elements. The figure indicates that there is a room for performance enhancement by increasing the number of antenna elements, while keeping a practical size of the antenna. For instance, the MS DL throughput achieved with IBFD increased by 16% when the number of antenna elements increased from 256 to 1024. Figure 5.4 also reflects that 8 antenna elements is the minimum required number of antenna elements to reach

(51)

5.1. Impact of Interference Mitigation Techniques with Variable

Traffic Demand 29

Figure 5.2: MS DL throughput with high network utilization.

Figure 5.3: MS DL throughput with highly congested network.

for a positive throughput gain for IBFD with respect to S-TDD and D-TDD in the same scenario. An interesting finding here, that D-TDD will have a positive MS DL throughput gain with the reference being S-TDD in the same scenario for M ∈ [1, 1024], nevertheless, this throughput gain is very small, compared to that achieved with IBFD at a high number of antenna elements.

(52)

Figure 5.4: MS DL throughput with beamforming only, under variable number of antenna elements.

5.1.2 Uplink

Figure 5.5: MS UL throughput with low network utilization.

We can notice from Figure 5.5, that in UL, the moment the network gets lightly utilized, IBFD and D-TDD start to have way worse UL performance compared to S-TDD under the first and the second scenarios. Indeed, this is due to the interference originating from the DL APs, these DL APs are transmitting with a higher power than the MSs, also they have a LOS path to the UL APs. When beamforming comes into play, D-TDD and S-TDD will have comparable

(53)

5.1. Impact of Interference Mitigation Techniques with Variable

Traffic Demand 31

performance, while IBFD is obtaining an UL performance that is better than D-TDD and S-TDD by far. IBFD in the fourth scenario is obtaining a MS UL throughput gain of 59% and D-TDD is amounting to a 29% of MS UL throughput gain with respect to the MS UL throughput obtained with S-TDD in the same scenario.

In the UL, with high network utilization, under the first and the second scenarios, D-TDD’s and IBFD’s performances are still worse than S-TDD’s per-fromance. Beamforming is adding a slight performance gain for IBFD on top of S-TDD in the same scenario, and D-TDD is still out of the competition. Not even combining beamforming with interference cancellation will grant D-TDD any UL performance gain over S-TDD and IBFD in the same scenario. In the fourth scenario, IBFD accomplished a 35% MS UL throughput gain with the reference being S-TDD MS UL throughput in the same scenario.

Figure 5.6: MS UL throughput with high network utilization.

In the congested network utilization state, as presented in Figure 5.7, we can notice a trend that is similar to the situation of high network utilization. A 26 % MS UL throughput is all we can guarantee with IBFD over S-TDD in the fourth scenario.

Figure 5.8 shows the MS UL throughput, with variable number of antenna elements, for the three duplexing schemes, in the third scenario, with high net-work utilization. It is clear from the figure that the number of antenna elements should be at least 128 for IBFD to start having a similar performance to S-TDD in UL. Furthermore, increasing the number of antenna elements from 256 to 1024 will cause a 30% increase in the MS UL throughput of IBFD.

The joint effect of interference cancellation and beamforing with 1024 an-tenna elements, is demonstrated in Figure 5.9. IBFD is amounting to a total of 88% and 57% MS DL and UL throughput gains with respect to S-TDD under the same scenario.

(54)

Figure 5.7: MS UL throughput with highly congested network.

Figure 5.8: MS UL throughput with beamforming only, under variable number of antenna elements.

5.2 Impact of Interference Mitigation Techniques

with Asymmetric Traffic Demand

Now, we will be looking into the performance of the duplexing schemes, with beamforming and interference cancellation, under asymmetric traffic demand, for a highly utilized network.

As shown in Figure 5.10, S-TDD will have a poor performance when the traffic demand is highly asymmetric, as then plenty of the time slots will be

(55)

5.2. Impact of Interference Mitigation Techniques with

Asymmetric Traffic Demand 33

Figure 5.9: MS DL and UL throughputs with beamforming and interference cancellation, at a high network utilization and 1024 antenna elements.

wasted without allocations. At the absolute asymmetric traffic demand regions (i.e. ψu= 0 or ψu= 1), half of the time slots will be unexploited. The amount of

unused time slots will decline as the demand get symmetric, and the maximum achievable MS sum throughput will be noticed when ψu= 0.5.

D-TDD behaves excellently under extreme asymmetric traffic demand situ-ations. D-TDD will never waste a time slot as long as there is a MS to serve; thus, when the traffic demand is absolutely asymmetric D-TDD doubles the spectral efficiency of highly utilized S-TDD wireless networks. Conversely, as the demand starts to be symmetric, additional interference will be generated between DL and UL, this interference as shown previously will be accounted for with beamforming and interference cancellation, still the performance will decrease slightly in the symmetric traffic demand region. D-TDD is having a higher MS sum throughput compared to S-TDD at every DL to UL traffic proportion anyway.

IBFD will reduce to a D-TDD system when the traffic is absolutely asym-metric, as the APs can simultaneously serve two MSs only with different link demand. As a result, the more the demand is symmetric the more MSs are served, it is true that the DL to UL interference will also increase then, but doubling the number of allocated MSs in the symmetric traffic demand regions will improve IBFD performance considerably. Thus, IBFD overcomes S-TDD at every DL to UL traffic proportion. Besides, the gap in the performance be-tween D-TDD and IBFD turns out to be bigger as the demand becomes more symmetric.

(56)

Figure 5.10: MS sum throughput with variable UL traffic demand.

5.3 Impact of Access Points’ Transmit Powers

We will show the significance of choosing comparable APs’ and MSs’ transmit powers in this section. We will only show the results for the scenario with beamforming and interference cancellation applied, at high network utilization. The MSs’ transmit power is fixed at 25 dBm, while we vary the APs’ transmit power from 25dBm to 50dBm.

Figure 5.11: MS DL throughput with variable APs’ transmit power.

As we can see from Figures 5.11 and 5.12, S-TDD avoids DL to UL inter-ference, so, no change occurs on the performance for S-TDD in either links, regardless of the APs’ transmit power.

In Figure 5.11, we can see that, for IBFD and D-TDD a higher transmit power for the APs is directly translated into higher achievable MS DL

Performance of In-Band Full-Duplex for 5G Wireless Networks

IN

DEGREE PROJECT

ELECTRICAL ENGINEERING,

SECOND CYCLE, 30 CREDITS

,

STOCKHOLM SWEDEN 2016

Performance of In-Band

Full-Duplex for 5G Wireless

Networks

OSAMA AL-SAADEH

KTH ROYAL INSTITUTE OF TECHNOLOGY

Performance of In-Band Full-Duplex for

5G Wireless Networks

OSAMA AL-SAADEH

Master of Science Thesis performed at

the Communication Systems Department, KTH.

June 2016

Abstract

Sammanfattning

Contents

List of Figures

List of Tables

Acknowledgements

List of Symbols

Acronyms

Chapter 1

Introduction

1.1

Research Questions

1.2

Benefits, Sustainability, and Ethics

1.3

Outline

Chapter 2

Background

2.1

Overview of duplexing schemes

2.1.1

Static-Time Division Duplexing

2.1.2

Dynamic-Time Division Duplexing

2.1.3

In-Band Full Duplex

2.2

Interference Mitigation Techniques

2.2.1

Self-Interference Cancellation

2.2.2

Maximal Ratio Transmit and Receive Beamforming

2.2.3

Successive Interference Cancellation

2.3

Related Work and Research Gap

2.4

Contribution

Chapter 3

Methodology

3.1

Experiments and Experiments Design

3.1.1

Impact of Interference Mitigation Techniques With

Variable Traffic Demand

3.1.2

Impact of Interference Mitigation Techniques With

Asymmetric Traffic Demand

3.1.3

Impact of Access Points’ Transmit Power

3.1.4

Impact of Interference Mitigation Techniques On

The Radio Access Network Energy Consumption

Chapter 4

Simulation Models,

Algorithms, and

Parameters

4.1

Simulation Models

4.1.1

Simulation Environment

4.1.2