A Practical Analysis of the Eects of Opportunistic Nulling in LTE-based Systems

(1)

A Practical Analysis of the Eects of Opportunistic Nulling in LTE-based Systems

VANESSA BELEC

Master's Degree Project Stockholm, Sweden

XREESB 2012:009

(2)

(3)

A Practical Analysis of the Effects of Opportunistic Nulling in LTE-based

Systems

VANESSA BÉLEC

Master’s Thesis at Signal,Sensors and Systems Royal Institute of Technology (KTH)

Stockholm, Sweden

Supervisor and Examiner: Mats Bengtsson

(4)

(5)

iii

Abstract

Due to the Internet expansion over the last decades, the pressure for the telecom- munications companies to deliver a high performance broadband communication, especially wireless is imminent. Future wireless networks will need to support high data rates in order to meet the requirements of multimedia services. Furthermore, the user density will be much higher for every year and it will be an increase of amount of data communication between mobile devices. Consequently, a new generation network has been introduced, i.e. LTE also called 4G, promising better performance and speed. Nevertheless, it is not only the network that plays an important role in achieving speed and performance, but also the communication system including number of antennas used, antenna deployment, power, number of base station, and so on. Most research papers agree on one point and this is the fact that the performance and reliability may be improved when using multiple inputs and/or multiple outputs, i.e. MIMO.

This Master Thesis concerns the performance acquired by studying on one hand the capacity achieved by using multiple antennas in the receiver (1 x 4 MIMO) and on the other hand studying different methods used by the base stations for scheduling the transmission to users according to their channel quality. Furthermore, the evaluation has been done numerically using measured radio channels, obtained by Ericsson.

One of the methods used in this paper is opportunistic scheduling, which involves the tracking of each of the fading users’ channel fluctuation and scheduling transmission to these users when their instantaneous channel quality is close to their maximum. In order to improve the communication of those users with low channel quality, a well-known algorithm was added to the Opportunistic scheduling. This method considers previous capacity rates and it is called Proportional Fair scheduling.

Another effect of using the Opportunistic scheduling is the suppression of the inter-cell interference (ICI) generated by close-by base stations, called Opportunistic nulling. In this paper, Opportunistic nulling is analysed in order to find out whether a practical suppression is achieved by this scheduling and whether factors such as delayed channel information may affect the scheduling and prevent a reliable communication with minimum interference.

(6)

iv

Sammanfattning

Med anledning av Internets utbredning under de senaste decennier, är trycket på att få telekombolagen att leverera en hög standard bredbandskommunikation, särskilt trådlös, som allra starkast. De framtida trådlösa nätverken kommer nog att behöva stöda höga datahastigheter för att uppfylla de kraven som multimedia tjänsterna ställer på dem. Dessutom blir tätheten bland användarna högre för varje år så att mängden datakommunikation som överförs mellan mobila enheter ökar.

Därför har en ny generation av mobila nätverk introducerats, m.a.o LTE eller även s. k. 4G, som lovar bättre prestanda och hastighet. Det är dock inte bara nätet som spelar en stor roll för att uppnå höga hastigheter och prestanda, utan också kom- munikationssystemet är viktigt, inklusive antalet använda antenner, antennernas utplacering, effekt, antalet basstationer, etc.

De flesta forskningsuppsatser är överens om en sak och det är faktumet att prestanda och tillförlitlighet kan förbättras då multipla ingångar och utgångar införs, m.a.o MIMO. Denna uppsats avser å ena sidan den prestandan som man får av att studera kapaciteten som erhållits genom multipla antenner i mottagaren (1 x 4 Single Input Multiple Output, SIMO) och å andra sidan av att studera de olika metoder som används av basstationer för att skedulera överföringen till användarna enligt deras kanalegenskap. Dessutom har utvärderingen gjorts numeriskt genom att använda uppmätta radiokanaler från Ericsson.

En av de metoderna som används i denna uppsats är opportunisitsk skedulering, vilket innebär att kanalsvängningar spåras för de fädande kanalerna till varje användare och att överföringen skeduleras till de användare vars momentana kanalegenskap är nära sitt maximum. För att förbättra kommunikationen till de användare med sämre kanalkvalitet, har man stoppat in en välkänd algoritm i den Opportu- nistiska skeduleringen. Metoden kallas för Proportional Fair och den tar hänsyn till tidigare värden på datahastigheter i sin algoritm.

En annan effekt av att använda Opportunistisk skedulering är den undertryck- ningen som skapas på den störning som genereras av närliggande basstationer(ICI), den s.k. Opportunistisk störningsundertryckning (Opportunistic Nulling). I denna uppsats analyseras den opportunistiska störningsundertryckningen för att ta reda på ifall en praktisk undertryckning är uppnådd och ifall faktorer såsom försenad kanalinformation kan drabba skeduleringen och ev förhindra en tillförlitlig kommu- nikation med minsta interferens.

(7)

Acknowledgement

I would like to express my gratitude to Mats Bengtsson, my supervisor and examiner at KTH, for giving me the opportunity to complete this thesis, spending time with me when I needed it the most and for supporting and helping me to achieve my goal.

I would also like to thank with all my heart my husband Martin for his patience and full support during this study time.

Finally, I would like to dedicate this work to my boys, Christoffer and Nicklas, with all my love.

vii

(10)

(11)

Definitions

Abbreviations

LTE Long Term Evolution

OFDMA Orthogonal frequency division multiplexing Access TDMA Time division multiplexing Access

MIMO Multiple Input Multiple Output SIMO Single Input Multiple Output

SNR Signal to Noise ratio

SINR Signal to Interference plus Noise ratio 3GPP Third Generation Partnership Project ICI Inter-Celll Interference

CDM Code Domain

TDM Time Domain

FDM Frequency Domain

PF Proportional Fair

Symbols

σ² Variance of noise

n(t) Gaussian noise

x(t) Transmitting signal

y(t) Receiving signal

H(t) Channel Response

λ Memory Factor

P Power

ix

(12)

(13)

Chapter 1 Background

During the last decade, high speed wireless broadband is being preferred over wired communication due to its mobility, relatively low maintenance, and high practicality. The use of laptops, smart phones and Ipads has never been so popular. However, the speed and performance in wireless systems has not always been as expected and there is still plenty of work to do to meet these expectations.

In order to fulfil the need of the market, the Third Generation Partnership Project (3GPP), which is a collaboration that brings together several telecom- munications standard bodies in USA, Europe, Japan, South Korea and China, has released a considerable amount of specifications that provide great speed and capacity. One of these releases is the Long-term evolution (LTE), which introduces OFDMA technology in the downlink and DFTS-OFDMA for the uplink.

1.1 OFDMA

Orthogonal frequency division multiplexing Access (OFDMA) has as well as OFDM scheme multiple subcarriers in one channel. The subcarriers are divided into groups of sub-channels allocated to multiple users allowing simul- taneously data transmission in a time-frequency selective resource. In the downlink, each sub-channel may be intended for a different receiver. In the uplink, a transmitter may be assigned one or more sub-channels. Furthermore, in case of frequency multiplexing of OFDM signals from multiple users, it is critical that the transmission of each user arrives at the base station with time alignment of less that the length of the cyclic prefix to preserve orthogonality

1

(14)

2 CHAPTER 1. BACKGROUND

Figure 1.1: The LTE Physical layer

between subcarriers in order to avoid inter-channel interference.[11] The cyclic prefix may vary between normal or extended status depending on how it is configured. In a downlink slot with normal cyclic prefix there are six OFDM signals, whilst for an extended one there are instead seven OFDM signals.

The advantage of an extended cyclic prefix is to be able to cover large cell sizes with higher delay spread of the radio channel.

The structure of the OFDMA symbol consists of three types of subcarriers [6],[1]:

• data subcarriers for data transmission

• pilot subcarriers for estimation and synchronization purposes

• null subcarriers, where no signal is transmitted, used for guard bands and DC carriers.

OFDMA is normally used in the multiplexing scheme of LTE.

1.2 LTE

Long-Term Evolution, LTE (also called 4G) is the new generation mobile network offering higher data rates, lower latency and higher capacity.

(15)

1.2. LTE 3 LTE has a physical layer (LTE PHY) which is a highly efficient for con- veying both data and control information between an enhanced base station (eNodeB) and mobile user equipment (UE). The LTE PHY employs some advanced technologies that are new to cellular applications. These include Orthogonal Frequency Division Multiplexing (OFDMA) and Multiple Input Multiple Output (MIMO) data transmission.

The LTE physical layer supports different bandwidths from 1.4 MHz to 20 MHz. The smallest amount of resource that can be allocated in the uplink or downlink is called a resource block (RB). One RB is divided in 12 subcarriers of ∆f = 15kHz each, which makes a total of 180 kHz wide. In addition, each RB lasts for one time slot which is 0.5ms (see FIG. 1.1). In both cases, the subcarrier spacing is constant regardless of the channel bandwidth. [8]

Moreover, LTE supports more than 300Mbps in the downlink and 80Mbps in the uplink. Thanks to its introduction to OFDMA technology, LTE provides orthogonality between users within a cell contributing to an intra-cell interference free zone.[7]

One of the basic principles for LTE is the use of channel-dependent scheduling. Although the scheduling is not specified by 3GPP, the overall goal of scheduling is to take advantage of the channel variations between users and preferably schedule transmission to a user when the channel conditions are advantageous. The scheduling can be realised in both the downlink and uplink transmission.

The main difference between the uplink and the downlink scheduling is how the power resource is distributed. In the uplink scheduling, the power resource is generated among the users, whilst in the downlink scheduling, the resource is centralised within the base station. Though the power resource in the downlink scheduling is centralised in the base station, it may also need to be shared among multiple users within a cell depending on in which domain the downlink signal is transmitted. If the signal is divided in frequency (FDMA) or in code (CDMA), then the power resource needs to be shared within the cell. On the other hand if TDMA is used, no instantaneous sharing is needed because per definition, there is only one transmission at a time. Often, the downlink scheduling uses only the time domain so no cell sharing is needed and in combination with the centralised power resource, the link capacity is efficiently utilised.[3]

Another difference between the uplink and the downlink scheduling is that the transmission power of a mobile device for the uplink is significantly lower

(16)

4 CHAPTER 1. BACKGROUND

than the output power for the downlink from a base station to multiple users.

For this reason, the uplink scheduling uses not only the time domain but also the frequency and code domain to obtain better reliability.

For simplicity, we use in this thesis the time domain downlink scheduling for each frequency sub-channel because the transmission that sends from the base stations to the users varies in time.

1.3 MIMO

As previously discussed, the future communication requires an increase of data rates so higher data rates can be provided to multiple mobile users. However, the power and bandwidth allocation are limiting factors together with the interference which make this task even more complicated to fulfill.

A solution that may increase the performance without affecting the limiting factors is the use of Multiple-input, multiple-output (MIMO) wireless system.[4] In a MIMO system there are several transmitting and receiving antennas. This system which is multifading can actually achieve among others high capacity levels compare to those with only a single antenna. One of the reasons for that is that MIMO provides spatial diversity with multiple parallel spatial channels working independently. [9]:

1.4 Objective

The main goal of this project is to analyse the effects of opportunistic nulling.

For this task, channel data measured by Ericsson corresponding to LTE (OFDM) has been evaluated. This data is used to emulate a model where three base stations are used as transmitters and each has a cell of four users/receivers. Each transmitter/base station has one antenna and each receiver has four antennas.

One of the objectives of this project is to evaluate the opportunistic scheduling by calculating the maximal throughput for each user in a cell. In order to obtain more accurate results, the delay between the moment the scheduler receives the information and the moment when time slots are actually allocated among the users, is also considered when calculated the maximal throughput.

(17)

1.4. OBJECTIVE 5 Accordingly, the capacity is newly calculated considering this delay and is then compared to the originally scheduled max throughput without delay.

Another objective of this project is to investigate the proportional fair algorithm by calculating the throughput of the system and by evaluating the results.

The third objective of this project is to evaluate the maximal throughput scheduling without considering any interference from adjacent base stations.

The last objective of this project is to define a factor/parameter in the channel data that may affect the throughput of each user in relation to interference.

The channel is affected by a AWGN noise with variance σ² and E[n] = 0.

The results from each cell are averaged over both time and frequency for better accuracy.

(18)

(19)

Chapter 2 System Description

2.1 Channel Measurements

Here follows the specification for the data received from Ericsson. The data comprises 89 files and each file includes the channel data or frequency response H, which is a four dimension complex matrix which varies with time, frequency, base station and receive antenna. As seen in Table 2.1, H is divided in 432 frequencies and 2000 time slots. In this thesis, the results have been averaged over both time and frequency for better accuracy.

Table 2.1: LTE Drive Test Data Details

7

(20)

8 CHAPTER 2. SYSTEM DESCRIPTION

2.2 Emulated System Scenario

The system model is extracted from a test drive where the driving car was equipped with four receiving antennas mounted on the roof of the car. The received information came from three different base stations placed at a rea- sonable distance from each other and equipped with only one antenna each.

The channel frequency response was measured along the whole test drive so it varies with the distance and coverage from the base stations. The measure- ment was then stored in 89 files with a time interval of 10.66 seconds each, normalised for each combination of antennas and channel sample to be ready for analysis.

Figure 2.1: Three base stations linked with four users each.

In this thesis, 12 files were selected among the 89 Ericsson files to represent the twelve users in the system model. Each file contains the 4D complex channel data divided in 432 sub frequencies (also subcarriers), each having a bandwidth of 45MHz and 2000 time slots equivalent to 10,66s, i.e. 5,33 ms for each time slot.

In order to obtain a strong signal from the base station, the users were selected to be those files which were closest to a specific base station.

(21)

2.2. EMULATED SYSTEM SCENARIO 9 As shown in Figure 2.1, there are a total of 12 users receiving the signals from different base stations and each of the three base stations is associated with four users. The number written below each user is the number of the file selected among the 98 Ericsson files.

A picture of where the drive test took place and the location of each base station is depicted in Fig 2.2. The drive test is indicated as a blue lighted route and each base station is depicted as a green triangle. The stars indicate the location of the users and the colour of each star represents which base station the user belongs to. In addition, the number in each star represents the number of the user.

Figure 2.2: The location of the users in Kista

(22)

(23)

Chapter 3 Channel factors and scheduling methods

This chapter describes factors that affect the channel in different ways and scheduling methods that are used in this thesis.

3.1 Channel Data Rate

In order to analyse the complex data channel, the first step involves calculating the data rate R of each selected file/ user. It is assumed that the received signal is an additive white Gaussian Noise (AWGN ) with zero mean and a variance of σ².

Let’s begin with a channel model of a single link in the system.

Figure 3.1: A model of AWGN channel

11

(24)

12 CHAPTER 3. CHANNEL FACTORS AND SCHEDULING METHODS

In figure 3.1, x(t) is the transmitting signal, H(t) is the channel response, n(t) is the white noise and y(t) is the received signal with four antennas. This is a model where the narrow band subchannel or subcarrier is constant over time slots of length T samples and it is also assumed that the bandwidth of each sub-channel is narrow enough so the channel response is flat across the band of the sub-channel.

According to Fig.3.1, the receiving signal y(t) results in:

y1(t) = x(t)H1(t) + n(t) y₂(t) = x(t)H2(t) + n(t) y₃(t) = x(t)H3(t) + n(t)

y4(t) = x(t)H4(t) + n(t) (3.1) where it shows that the transmitting signal x(t) is multiplexed by the matrix H(t) resulting in four receiving signals r1(t), r2(t), r3(t), r4(t) aimed at four receiving antennas (NR = 4). In this model, only one transmitting unit is being considered (NT = 1).

Now we need to calculate the power. In this thesis, we assume that the transmit power P is constant at all times, i.e. E[|x(t)|²] = P T , where T is the length of each time slot.

During a T interval we have[13] :

E[|x1(t)|²] = P E[|x2(t)|²] = P

E[|x3(t)|²] = P (3.2)

where xi(t) is the transmitting signal from each base station i.

According to the general Shannon capacity theory [15], the rate is:

R = log2det(INR + P

σ² HH^H) bits/s/Hz (3.3) where I is an identity matrix, NR is number of receiving antennas, P is the power of the input signal, σ²is the variance of Gaussian noise and H ∈ C^N^R^×N^T and is the normalised channel matrix. In the special case, where NT = 1, the formula is then:

R = log2







(1 + P σ²

NR

X

i=1

|h_i|²)







bits/s/Hz (3.4)

(25)

3.1. CHANNEL DATA RATE 13

where i is the number of receiving antennas and NR= 4 .

In general, the Signal to Noise Ratio (SNR) is formulated as :

SN R= 10 log10

Psignal

P_noise

(3.5)

If considering Eq.3.5 and Eq.3.4, the SNR here results in ^{P |h}_σ2ⁱ^|².

As previously explained, the interference from close-by base stations should not be neglected and should therefore be considered into the calculations for the capacity. Accordingly, since we are not only dealing with white Gaussian noise but also with ICI, we need to replace the value of SNR with the SINR (Signal to Interference and Noise Ratio) value.

The SINR for base station 1 is :

SIN R1 = P h1HR⁻¹_w

1h1 (3.6)

where R⁻¹w1 is the inverted autocorrelation of the interference plus noise.

The same applies for SINR2 and SINR3.

Accordingly, the next step is to calculate the autocorrelation Rw1 [15]

which is a NR× N_R matrix, where NR is the number of receiving antennas in the system. In this thesis, NR = 4 which means that the autocorrelation of the interference Rw is a 4 × 4 matrix.

R_w₁(t) = E[W1(t)W1(t)^H] =

= E[(h2(t)x2(t) + h3(t)x3(t) + n(t))(h2(t)x2(t) + h3(t)x3(t) + n(t))] =

= h2(t)h2(t)^HE[x2(t)x2(t)] + h3(t)h3(t)^HE[x3(t)x3(t)] + E[n(t)²] =

=P = E[x2(t)x2(t)] and P = E[x3(t)x3(t)]=

= P h2(t)h2(t)^H + P h3(t)h3(t)^H + σ²I.

(3.7) The same procedure applies for Rw2 and Rw3.

In equation 3.7, we assumed that W(t) is the ICI plus noise as following:

(26)

W₁(t) = h2(t)x2(t)+ h3(t)x3(t) + n(t) W₂(t) = h1(t)x1(t)+ h3(t)x3(t) + n(t)

W₃(t) = h1(t)x1(t)+ h2(t)x2(t) + n(t) (3.8) In equation 3.8, the ICI plus noise Wi(t), the input signal xi(t) and the channel response hi(t) are represented for each base station i, where i=1,2,3.

Figure 3.2: The ICI generated by BS2 affecting a user with four receiving antennas served by BS1.

For a better understanding, the figure 3.2 shows how the ICI (h21, h₂₂, h₂₃, h₂₄), generated by base station 2, affects the receiving signal of a user with four re- ceiving antennas (h11, h₁₂, h₁₃, h₁₄) served by base station 1.

The last step when calculating the data rate for our model is to insert eq.3.6 into eq. 3.4 and we obtain the following:

R_1,k(t) = log2(1 + P h1(t)Rw⁻¹1(t)h1(t)^H) R_2,k(t) = log2(1 + P h2(t)Rw⁻¹2(t)h2(t)^H)

R_3,k(t) = log2(1 + P h3(t)Rw⁻¹3(t)h3(t)^H) (3.9)

The data rate is calculated for each base station i and user k at each time instant t, where Ri,k(t) is a scalar value, i=1...3 and k=1...4.

(27)

3.2. THERMAL NOISE VARIANCE 15

3.2 Thermal noise variance

The white noise added to the system may be considered as a thermal noise which is a noise generated by the random thermal motion of charge carriers inside an electrical conductor which happens regardless of any applied voltage.

The formula for the power of the thermal noise is:

σ² = kBT N_S∆f, where kB is the Boltzmann’s constant, T is the room’s tem- perature, NS is the Noise Figure, which is the noise sensibility factor for the receiver unit in particular and ∆f is the bandwidth for each subcarrier.

In this thesis the Noise Figure NS is the noise value for a mobile device in dB which in this case is assumed to be 7. The temperature is 273+20 K and the bandwidth is 45kHz.

3.3 Opportunistic scheduling

In a MIMO system where the channel conditions are time-varying due to fading from multiple spatial signals, different wireless users experience different channel conditions at a given time. In other words, the conditions within a radio cell vary independently for each user in that cell, and at each point in time there is a high probability that a user is having its channel quality near its peak [2]. In order to increase the total throughput of this system, only those users having a high channel quality are assigned time slots for transmission/reception. This method of enhancing the total system throughput is called opportunistic scheduling [12]. It is opportunistic because this method takes advantage of favourable channel conditions in sharing the radio channel. However, opportunistic scheduling requires a flexible and tolerant user in terms of delay in transmission/reception, especially when the radio conditions are not ideal. Therefore, the Opportunistic Scheduling may not be as useful for voice signals. [5]. Consequently, this method is widely used in data signals which allow a balance between efficiency and delay.

Another effect of the Opportunistic scheduling is the suppression of the inter-cell interference (ICI) produced by other close-by base stations, called Opportunistic nulling. In this thesis, Opportunistic nulling is analysed in order to find out whether a practical suppression is achieved by this scheduling and which factors may be of importance to be considered for obtaining a reliable communication with minimum interference in the near future.

(28)

3.4 Max Throughput Scheduling

Maximum Throughput scheduling is an opportunistic scheduling which cal- culates the maximum throughput among all users within a cell at a given time.

The opportunistic scheduling works as following:

• Each receiver tracks its own channel through a downlink single pilot signal and feeds back the instantaneous channel quality to the base station.

• The base station schedules transmission among the users by selecting the user that have the maximum capacity at a given time instant to assign any transmission/reception slots.

Mathematically, the Max Throughput or Rate Scheduler can be expressed as a selected user k* given by:

k^∗ = arg max

k R_k (3.10)

where Rk is the instantaneous data rate for user k.

However, this opportunistic scheduling may be far from fair. In real conditions, users are spread over a radio cell and some of those may have the base stations far from their locations. This type of scheduling will only favour those with high throughput e.g. closer to the base station and the rest of the users will have to struggle to be assigned any time slots.

In order to improve the communication of those users with temporarily low channel quality, another version of opportunistic scheduling may be used.

This version uses previous values of capacity rates and it is called Proportional Fair scheduling.

3.5 Proportional Fair Scheduling

Proportional Fair scheduling tries to satisfy different rate requirements of each user without missing out too much of the efficiency of multiuser diversity. In other words, the system can accommodate more users that are near their peak rates. The downside of such a scheduling is that many users may not reach their target/maximum rates and for those users the communication sys- tem will feel slower. For instance, if there are k users in a cell and Rk is an achievable rate for user k^∗ at the transmission interval t, which depends on the user’s current channel conditions, the scheduler keeps then track of the

(29)

3.5. PROPORTIONAL FAIR SCHEDULING 17

running average rate for every user.

According to [12], this general version of the proportional fair algorithm works as following:

Given a time slot t, a user k, the algorithm keeps track on:

• Users’ average throughputs T1(t), T2(t), ...Tk(t).

• Current requested data rates R1(t), R2(t), ...Rk(t) [14] transmit to the user k with the largest Rk(t), Tk(t).

Specifically, at each time slot t, the decision of the PF scheduler is to schedule the user k^∗ with the largest rate Rk(t)/Tk(t) among all active users in the system. In other words, user k^∗ is selected for transmission according to:

k^∗ = arg max

k

R_k

T_k (3.11)

where Rk is the rate (bit/s) achieved by the user and Tk(t) is the average rate calculated over a time window as a moving average. The average throughputs Tk(t) can be updated using a memory factor λ as following:

T_k(t + 1) =

( (1 − λ) Tk(t)+ λRk(t), k = k^∗

(1 − λ) Tk(t), k 6= k^∗ (3.12)

In this equation, the time window is defined by the memory factor λ which varies between 0 and 1. This memory factor is chosen to balance between the needs of estimating throughput which requires a small value of λ and the ability to track changes in the channel characteristics requiring a larger value of λ.

Hence, the algorithm allows the user with the statistically strongest channel to be chosen for transmission. The running average rates will however lower the total throughput over the maximum possible and instead will pro- vide more acceptable levels to users with poorer SNRs. In this case, the users do not compete for the resources based on their requested rates but only after normalization by their respective average throughputs.

(30)

Wide Band Channels

The communication system LTE provides a wide band channel divided in narrow band subchannels due to its OFDM signals. In such a wide band channel, it is important to consider not only the time selective fading but also the frequency selective fading. A simple model of such a system is used in this thesis which involves a set of l parallel narrow band subchannels or subcarriers, each of these channels has a invariant frequency level that fluctuates with time. The model is then consisting of 432 frequency divided subchannels, each with a bandwidth of 45 kHz making a total of 19.4 MHz, and transmitting a total transmitting power of 1W. Accordingly, the transmit power for each subchannel is very small (2.3mW) but it does not need to be higher since the noise level of each subchannel is even a smaller value. This issue will be discussed in a later chapter complemented with simulations regarding capacity values vs SNR values.

In order to be able to schedule the transmission to users within a cell, the users need to measure the SNR on each of the subchannels and feed back the requested rates or SNRs to the base station. The scheduler receives the information and allocates at each time a single user to transmit to for each of the narrow-band subchannels. If the proportional fair scheduler is used, the scheduler needs to keep track of the average throughputs Tk(t) across all narrow-band subchannels in a past window of length 1/λ for each user k. [12]

The algorithm is then modified from being used for the aforementioned pure time-selective channel to the following:

For each sub-channel l, the scheduler transmits to the user k^∗(l), where

k^∗(l) = arg max

k

R_kl(t)

T_k(t) (3.13)

where Rkl(t) is the requested rate of user k in channel l at time slot t. It is important to realise that Tk(t) is averaged over all the narrow-band sub- channels and not just the subchannel l. This is because the fairness criterion belongs to the total throughput of the users across the entire wide band channel.

3.6 Interference

Interference emerges as the key performance limiting factor, because it deter- mines how many users per area can be served at a certain data rate. If the number of users is high then each user will get just a small fraction of the

(31)

3.7. DELAY 19

overall resource. [10]

In order to avoid this to happen, the 4G/LTE network provides orthogo- nality between users within a cell contributing to no inter-channel interference between multiple users from a signal sent by the same base station. However, inter-cell interference (ICI ) can be generated if more adjacent base stations are sending to their users at the same time.

As explained above, MIMO offers another degree of freedom to avoid and mitigate interference. In such a multiuser network, the users within a cell compete for the available resources. This results in that the performance of some users can be increased at the cost of decreasing the performance of other users. The selected strategy is often a compromise between fairness and efficiency. Therefore, dynamic interference management and resources allocation are key factors to achieve a full exploitation of the available system capacity.

In other words, the available resource needs to be designed in a flexible way considering the current interference situation.[17]

In this thesis, the focus lies on whether different types of downlink scheduling such a Opportunisitc Nulling may mitigate the inter-cell interference pro- duced by three base stations in a LTE system with (1 x 4) MIMO.

3.7 Delay

In this report, delay refers to the delay caused by the latency between the moment the scheduler receives the information, as well as adapting users’ data rates to the instantaneous channel quality and the time when the time slots are actually allocated among the users. There is no doubt that this time delay affects the performance of the radio resources in a real time system. However, the question is how much such a delay may affect the users in such a system. As discussed earlier, one of the important issues when developing new systems is to give a better Quality of Service (QoS). If a user does not obtain the expected performance because the scheduler allocated time slots erroneously due to the above discussed delay, the delay is then affecting seriously the QoS.

When exploiting higher capacity in an opportunistic way, it is necessary to consider the trade-off problem between wireless resource efficiency and levels of satisfaction among users.

(32)

(33)

Chapter 4 Results and Analysis

In this chapter, the results from the simulations of the scenarios described in Chapter 3 are presented below. Moreover, an analysis of each figure is given when appropriate.

All simulations were performed by generating data following the system model introduced in chapter 2 and were implemented in MATLAB.

In short, the system model has three base stations linked with four users each. In addition, the channel has an additive white Gaussian noise and there is an inter-cell interference (ICI ) between the base stations.

In order to be able to analyse the performance of the opportunistic scheduling, the rate of each user from all three base stations has been calculated. For simplicity, the power used for each subcarrier has been chosen to be 2.3mW, which makes a total of 1W for the whole frequency band. Since the frequency band has 432 subcarriers, we have either randomly chosen the subcarrier in the position 200 or calculating the average of the results from all subcarriers, depending on what it was more appropriate at that time.

4.1 Data Rate

As explained in section 3.1, the data rate for each user at a time t is calculated according to the formula 3.9, where the ICI and white noise are considered in the calculations.

21

(34)

22 CHAPTER 4. RESULTS AND ANALYSIS

Figure 4.1: (a) Rate of four users vs Time for BS1 using four receiving An- tennas. (b) Rate of four users vs Time for BS2 using four receiving Antennas.

Figure 4.2: (a) Rate of four users vs Time for BS3 using four receiving Anten- nas. (b) Estimated Round Robin scheduling for total system vs Time using four receiving Antennas.

(35)

4.2. MAX THROUGHPUT SCHEDULING 23 In Fig. 4.1 and 4.2, the rate has been calculated for each user served by base station 1 (BS1), base station 2 (BS2) and base station 3 (BS3) .The time period selected here is set to 30 frames (0.16s) of a maximum of 2000 frames.

In these figures, four receiving antennas have been taken into account.

It can be seen from these figures that the rate varies severely among the users and base stations. Since each user has a different position in the map shown in Fig 2.2, it reveals that those users that are further away from their base station have a lower signal than other users that are closer to theirs. However, it is also clear from these results that the signal coming from BS3 is weaker than the other base stations. In BS3, the range in rates varies between 2 - 9 b/s/Hz, whilst in BS1 and BS2, the ranges vary between 8 -14 b/s/Hz and 5 -14 b/s/Hz respectively.

The second graph in Fig. 4.2 is depicting the estimated total throughput in the system considering an average user per time instant, i.e. ’Round Robin Scheduling’. This means that the time is shared equally by all four users and no consideration is taken to the instantaneous channel information when scheduling. [8]

4.2 Max Throughput Scheduling

As mentioned in Chapter 3, Opportunistic Scheduling involves assigning time slots for transmission/reception to those users with the highest channel quality, and Maximum Throughput scheduling assigns time slots to the user that has the maximum throughput among all the users within a cell at a given time.

In Figure 4.3, the Max Throughput scheduling is marked as a continuous red line with crosses in all graphs. For a clearer view, the time period is zoomed in to 0.2s for the three capacity/users graph.

As seen in the last graph in Fig. 4.4, the the total throughput is shown for the full 10.66 s (2000 time frames). Further, the Max Throughput scheduling has a higher total throughput in the system than the Round Robin sheduling.

This is comprehensible because the Max Throughput scheduling selects the users with the highest rate values within a cell instead of sharing the time equally without considering the instantaneous channel conditions, so the total throughput in the system per time instant is increased. In this case the increase varies between 5-7 b/s/Hz. However, the advantage of using Round Robin scheduling is that it gives more equal service quality in time between different users.

(36)

Figure 4.3: Max throughput scheduling/Rate vs Time for BS1 and BS2 using four receiving Antennas.

Figure 4.4: (a) Max Throughput scheduling/Rate vs Time for BS3 using four receiving Antennas. (b) Mean Max Troughput scheduling/Round Robin Scheduling vs Time using four receiving Antennas.

(37)

4.2. MAX THROUGHPUT SCHEDULING 25

Max Throughput scheduling without Interference

As explained in previous chapters, the rate has been calculated considering the presence of interference signals from adjacent base stations. In this case, this Max Throughput scheduling was computed to be free from inter-cell interference (ICI) so the performance of this scheduling could be analysed.

Figure 4.5: Max throughput scheduling with and without Interference vs Time for the total system using 1 or 4 receiving Antennas.

In Fig 4.5, each graph represents the Max Throughput with and without interference depending on a different number of receiving antennas. The Max Throughput was calculated over the whole system, which means adding the rate of the three base stations and obtaining an average over the 432 subcarriers in the frequency band.

As observed in these graphs in Fig. 4.5, the throughput is much higher when there is no ICI around and the difference in rate between the Max Throughput with and without interference decreases with the addition of antennas to the receiver.

Other figures that show the performance of the Max Throughput scheduling are Fig 4.6 and Fig 4.7.

(38)

Figure 4.6: Max throughput scheduling for a selected Bestuser with or without interference vs Time using 1 or 2 receiving Antennas.

Figure 4.7: Max throughput scheduling for a selected Bestuser with or without interference vs Time using 3 or 4 receiving Antennas.

(39)

4.3. PROPORTIONAL FAIR 27 In particular, these figures show how the Max Throughput scheduling (also called here Cmax) sometimes do not assign any transmission to the ’right’ best user in a ICI free environment (also called here ’bestuser without ICI’), i.e the highest rate user that would have been selected if no intercell interference was in the system.

As seen in the figures 4.6 and 4.7, the continuous blue line represents the Max Throughput scheduling with ICI, whilst the green line represents the Max throughput with ICI vs the best user without ICI.

Every time the best user with present interference selected by the Max Through- put scheduling does not coincide with the best user selected in a ICI free environment, it means that this Max Throughput schedules the transmission to another user in that base station at that time instant. Consequently, the best user without ICI is not assigned any transmission at that time instant. From the figures, this can be seen as sharp variations in rate (the rate for that best user turns to zero bringing a lower average for the whole system) from one instant to another.

For instance, Fig. 4.6 shows a green curve with drastic variations in some places when using one receiving antenna. With the addition of receiving antennas, the variations get less sharp but the number of variations is not necessarily reduced with the exception of the last graph in Fig. 4.7 using four receiving antennas. Accordingly, it seems like the reliability for a user is improved not only in achieving better performance or higher levels of throughput but also in accessibility, i.e being assigned time slots if the channel quality is better than for other users in that cell , when multiple receiving antennas are used.

4.3 Proportional Fair

Memory factor λ

In section 3.3.1, the Proportional Fair algorithm is discussed. In this algo- rithm, the average rates are updated using a memory factor λ. In general, λ should be chosen small enough so that it provides an acceptable measure of the throughput. In order to see how the memory factor influences the outcome of the algorithm, three different values have been then selected:

λ = 1/10, 1/50, 1/100, which means that the average rate is computed over the previous 10, 50 and 100 time slots.

(40)

Table 4.1: Total throughput in the system for PF; λ=1/10, 1/50, 1/100 for one selected subcarrier F1 and for Fmean.

In Table 4.1, the total throughput of the system when using the proportional fair algorithm mentioned in eq. 3.12 is evaluated. F1 in the table represents the subcarrier located in the position 200 of the frequency band and as well as all subcarriers it has a narrowband of 45kHz. The values shown in the second part of the table are first obtained for each subcarrier to be consequently averaged over all the subcarriers.

As seen in Fig 4.8, the staple graph represents the results from the second values of the Table 4.1, where the rate increases with the number of antennas.

However, there is no major improvement in throughput when increasing the value of λ. This figure also compares the throughput of the Proportional Fair scheduling with the Max Throughput scheduling, which shows that the latter achieves a higher throughput.

Fairness

As already discussed, the idea behind the Proportional Fair algorithm is to create fairness in the system. This means that users that have a relatively constant rate, but not higher than other users in the same cell, will be able to be assigned a time slot.

In Fig 4.9, the rate of four different users is shown and how the Max- and PF throughput methods schedule these users in base station 3. Fig 4.10 is depicting the total estimated rate that is requested from each user in all three base stations and the rate that is scheduled when using either the Max Throughput or PF Throughput scheduling. In both figures, the first graph is using only one receiving antenna whilst the second graph is using four receiving antennas. Again, the frequency values were averaged over the whole frequency band equivalent to 432 subcarriers.

(41)

4.3. PROPORTIONAL FAIR 29

Figure 4.8: Staple graph showing the total Max throughput and PF through- put for different values of λ and number of receiving antennas N.

As seen in the first graph in Fig 4.9 for a single receiving antenna, the Max Throughput scheduling prioritises the user #4 during this short period of time.

On the contrary, PF Throughput scheduling assigns more the transmission to other users than to user #4. Accordingly, it seems like the latter mentioned scheduling is fairer than Max Throughput as expected. The same is valid when using four receiving antennas shown in the second graph.

The estimated capacity for all three base stations is shown in Fig 4.10 for both a single receiving antenna and four receiving antennas. It is apparent that the PF Max Throughput and Max Throughput are at least fourfolded in capacity when using four antennas instead of a single one. Furthermore, the PF Throughput is lower than the Max throughput in the total system, which is a direct consequence of a fairer system.

(42)

Figure 4.9: PF Throughput with λ = 1/100/Max Throughput/Capacity vs Time for BS3 with 1 or 4 receiving Antennas.

Figure 4.10: Estimated PF Throughput with λ = 1/100 / Max Throughput scheduling/ Capacity vs Time for the total system using 1 or 4 receiving Antennas.

(43)

4.4. DELAYED SCHEDULING 31

4.4 Delayed Scheduling

Figure 4.11: Delayed Max Throughput/Max Throughput scheduling vs Time using 4 receiving Antennas

In this thesis the delay described in section 3.6 was set to be one time frame or 5.33ms which is the smallest amount of time measured by Ericsson.

In general, this sort of delay is considered to be less than 5ms and often no more than 1ms. [16]

In Fig 4.11, it is possible to see how the delay affects Max throughput in BS1 when four receiving antennas are used. During this short lapse of time, user #2 is here prioritised by the Max Throughput. However, when there is a delay in the scheduling, user#2 is missing out at certain points. For instance at t=1.13s, user #2 is not receiving any transmission slot when delay occurs even if this is a best user at this time instant.

This is better shown by Fig 4.12 and Fig 4.13. These figures represent how the Delayed Max Throughput differs from the Max Throughput in terms of choice of best user. When the best user chosen by the Max throughput does not coincide with the best user from the Delayed Max Throughput, it means that not all the users with the maximum rate were allocated transmisson slots and consequently less total throughput is achieved in the system as depicted by the figures.

(44)

Figure 4.12: Delayed Max Throughput/Max Throughput scheduling with ICI vs Time using 1 or 2 receiving Antennas.

Figure 4.13: Delayed Max Throughput/Max Throughput scheduling with ICI vs Time using 3 or 4 receiving Antennas.

(45)

4.4. DELAYED SCHEDULING 33 This decrease in capacity for the delayed scheduling is more evident when using more than one antenna.

Figure 4.14: Estimated Max Throughput/ Delayed Max Throughput scheduling with and without ICI for a selected Bestuser without ICI vs Time using 1 or 2 receiving Antennas

Other figures 4.14 and 4.15, show the difference in rate between the Max Throughput /Delayed Max Throughput with and without interference vs the selected best user that should be selected when no ICI is around (bestuser without ICI). It is clear from the figures that when no interference is present, the rate level is high. Furthermore, it is also clear that even a delay of 1 time frame(5.33ms) may contribute to a rate loss of up to 29 bps/Hz at a given time (at approx. 0.8 sec) if one receiving antenna is used and up to 10 bps/Hz (at approx.7 sec) when four receiving antennas are used and no ICI is around.

Even when the interference or ICI is considered in the calculations, the rate may decrease with up to 1-2 bps/Hz for such a small delay.

Another interesting matter is that the Max Throughput scheduling with ICI is rapidily approaching the curve for Max Throughput without ICI for every added receiving antenna. The same is valid for the delayed Max Throughput curve.

Another way of seeing how much throughput is lost because of a delay of one time frame is to calculate both the Max Throughput and the Delayed Max

(46)

Figure 4.15: Estimated Max Throughput/ Delayed Max Throughput scheduling with and without ICI for a selected Bestuser without ICI vs Time using 3 and 4 receiving Antennas.

Figure 4.16: Delayed Max Throughput/ MaxThroughput scheduling vs Time for a selected Bestuser without Interference in comparison with Max/ De- layed Throughput scheduling for a Bestuser with Interference and using 1 or 2 receiving Antennas.

(47)

4.4. DELAYED SCHEDULING 35

Figure 4.17: Delayed Max Throughput/ MaxThroughput scheduling vs Time for a selected Bestuser without Interference in comparison with Max/ De- layed Throughput scheduling for a Bestuser with Interference and using 3 or 4 receiving Antennas.

Throughput in relation to best users when the system is free from interference.

Those best users that were not assigned the transmission/reception according to the scheduling will give a null throughput making it easier to differentiate the performance between the two scheduling. Fig 4.16 and Fig 4.17 show how the Delayed Max Throughput has a lower total capacity for all number of antennas. Especially, it can be noted that the Delayed Max Throughput scheduling with ICI does not allocate as much time slots to the ’true’ best users (Bestuser without ICI) than the Max Throughput with ICI without delay.

In other words, the delayed Max Throughput provides less performance than without delay. This is specially notourious when using four receiving antennas.

Another interesting matter in the above figures is that the rate for Max Throughput scheduling without delay is increasing faster than when having delay for more than two receiving antennas. This is due to the mitigation of ICI achieved by the combination of Max Throughput scheduling with the increasing number of receiving antennas (SIMO). A staple graph in Fig. 4.18 shows the rate for all four scheduling vs number of antennas where the time and frequency has been averaged for simplicity.

In order to obtain some conclusions about how the delay decreases the

(48)

Figure 4.18: Staple graph showing a Delayed Max Throughput/ MaxThrough- put scheduling vs number of Antennas for a selected Bestuser without Interfer- ence in comparison with Max/Delayed Throughput scheduling for a Bestuser with Interference.

Figure 4.19: Delayed Max Throughput vs Delay using 1 or 4 receiving Anten- nas.

(49)

4.5. POWER 37

Figure 4.20: Delayed Max Throughput vs Delay using 1 to 4 receiving Anten- nas.

total throughput of the system, we have calculated the rate for each user with different values of delay, i.e. 1-10 time frames (approx. 5-53ms). The Max throughput has been taken on the delayed rates for each time instant to be finally averaged over time. In this case, only one subcarrier is chosen for simplicity. See Fig 4.19 for 1 and 2 receiving antennas and Fig 4.20 for 3 and 4 receiving antennas.

The abovementioned figures show that the total throughput has decreased five times more when using one receivng antenna than when using four receiving antennas. In other words, the more receiving antennas we have, the more rate is decreased in the system. However, from 4.20, it can be noted that the loss in rate is almost negligible when all different numbers of receiving antennas are illustrated in the same graph due to the difference in rate level.

4.5 Power

In earlier chapters, we have explained that the transmit power (Pt) used in most of the calculations in this thesis is Pt=2.3 mW per subcarrier, which is a very small value. However, it does not need to be higher since the noise

(50)

level of each subcarrier is 9.12e-16 W. In this case, whe are having a channel response for example of |Hi| of around 3.0e-5 for each base station and user.

In addition, if we apply eq. 3.5, this would correspond to a SNR of about 32dB. (see Fig. 4.21 and eq.4.1). It is also important to see how the Max Throughput varies with the Signal to Noise Ratio (SNR), which is propor- tional to power.

SN R= 10 log10(^P_P^signal_noise) If signal y(t) = x(t)H(t) then:

P_signal = E[|x(t)H(t)|²] = |H(t)|²E[|x(t)|²];

Since E[|x(t)|²] = Pt then Psignal = Pt|H(t)|²; We know also that Pnoise = σ² which in this case:

P_noise = 9.12 e-16 W;

Accordingly if for example|H(t)| = 3.0e-5 and Pt = 2.3e-3 W then:

SN R= 10 log10(^(3.0e−5)_9.12e−16²^∗2.3e−3) ' 32dB.

(4.1) In Fig 4.21, we can see the expected rate for different methods and number of antennas vs SNR. The highest rate when using one antenna is achieved at around 32dB (P=2.3mW) by Max Throughput scheduling without any delay or using a proportional fair algorithm. Furthermore, when using four antennas at the same time, the expected rate is increased nearly seven times compared with the use of only one antenna for the same amount of power. In addition, the rate does not saturate as it is the case when using one antenna. Instead, the rate increases with power during this specified SNR range. It is remarkable how the proportional fair method does not achieve more rate than 4 b/s/Hz when using one antenna and can be as high as 55 b/s/Hz when using four antennas. Another interesting finding is that the rate of the Max Throughput scheduling with a delay of 5.3ms is as high as the Max Throughput scheduling without delay when using at least four receiving antennas.

In Fig 4.22, we have the same graph as before but we have added the Max Throughput without interference. As seen here, the rate without interference can only be higher with the increase of the SNR. This is quite obvious because there is no limiting factor more than the (small-value) white noise to slow down

(51)

4.5. POWER 39

Figure 4.21: Expected Max Throughput/Delayed Max Throughput/ PF Max Throughput vs SNR for different numbers of receiving Antennas.

Figure 4.22: Expected Max Throughput/Delayed Max Throughput/ PF Max Throughput with and without Interference vs SNR for different numbers of receiving Antennas.

(52)

the power. For simplicity reasons, we only show the Max throughput when using one antenna or four antennas in this figure. Hereby, it can be seen that the Max throughput is only slightly affected by interference when a plurality of antennas is used.

4.6 Comparison between different schedulers

Figure 4.23: Cummulative Distribution Function vs different types of scheduling using one or four Antennas.

Figure 4.23 show that for all base stations the highest performance with respect to highest data rate is achieved by the Max Throughput scheduli- ing when using four receiving antennas. it can also be noted that the Max Throughput with 5 delayed time frames is slightly less reliable than with just one delayed time frame for all base stations. Regarding the Proportional Fair Max throughput scheduling, as for the others variations of scheduling, it increases in data rate with the addition of antennas and it gets closer in rate to the Max Throughput without ICI . This can be explained as how the fair system works, where even weak users have the opportunity to be assigned time slots though the quality of the channel is not at their maximum.

(53)

Chapter 5 Conclusions and Future Research

In this thesis, the main purpose has been to analyse how inter-cell interference (ICI ), latency from delayed scheduling and the level of fairness may affect a user in a LTE-based system.

According to the results from the simulations shown in previous chapters, it is clear that the inter-cell interference is mitigated during opportunistic scheduling, especially when using the Max Throughput Scheduling. As explained earlier, the Round Robin scheduling does offer an equal time sharing among the users but the performance in terms of throughput is much lower than when instantaneous channel information is considered.

It is also clear from all figures that an increase in receiving antennas does contribute to obtain a higher throughput in the system. In this case, we only have four receiving antennas but it would be interesting to see how much the interference can be mitigated by increasing the number of antennas. Obvi- ously, it would be of great interest to increase also the number of transmitting antennas, obtaining a MIMO system.

Regarding the latency from a delayed opportunistic scheduling, several delays have been considered in the simulations. As expected, the more the scheduled signal is delayed, lower throughput is achieved in the system. How- ever, for each antenna added to the receiver, the throughput increases.

In the terms of fairness, the analyses show that the opportunistic scheduling may benefit some users more than others due to its lack of consideration for users situated far from base stations and for users surrounded with several obstacles that are fading the signals. The simulations show that some users may have as much as sixty five percent of the transmission signal in average, whilst other less favourable users will only receive six percent of the transmission as an average during a period of time of 10 seconds. However,

41

(54)

42 CHAPTER 5. CONCLUSIONS AND FUTURE RESEARCH

the level of fairness is often increased by a few percent with every additional antenna. In order to test the fairness of one of the general proportional fair algorithms, simulations have been performed with different memory factors,λ.

It is evident from the figures that the higher the memory factor is, the higher throughput is in the system. However, this is a borderline because the ac- tual difference is almost negligible. This is probably because this algorithm focuses on the fairness of sending transmission to any user and not only to those users who have the maximum rate in the system at any time instant. In other words, the simulations show that the algorithm works well by achieving a fairer system for the users. When considering higher memory factors, the level of fairness was not increased as expected. In summary, this thesis has shown that the Opportunistic Nulling is a fact, i.e. the Opportunistic scheduling in combination with a SIMO system does mitigate the ICI.

(55)

Bibliography

[1] Physical channels and modulation (36.211) for LTE Overview. Technical report, 3GPP,E-UTRA, 2011.

[2] W. Ajib and D. Haccoun. An Overview of Scheduling Algorithms in MIMO-Based Fourth-Generation Wireless Systems. IEEE Network, p.43- 48, 2005.

[3] Furuskar A Jading Y Lindström M Astely D, Dahlman E and Parkvall S. LTE: the evolution of mobile broadband. Communications Magazine, IEEE, 47(4):44–51, April 2009.

[4] Helmut Bölcskei. Space-time wireless systems: from array processing to MIMO communications. Cambridge Press, 2006.

[5] Thomas Bonald and Lucas Muscariello. Opportunistic scheduling of voice and data traffic in wireless networks. In EuroFGI Workshop on IP QoS and Traffic Control, Portugal, 6-7 December 2007.

[6] C.Gessner. UMTS Long Term Evolution(LTE) Technology Introduction.

Technical report, Rohde & Schwarz, September 2008. Application notes 1MA111.

[7] K. Hassan and H. Haas. User Scheduling for Cellular Multi User Access OFDM System Using Opportunistic Beamforming. pages 26–30, Ham- burg, Germany, 2005.

[8] S. Parkvall J. Sköld, E. Dahlman and P. Beming. 3G Evolution: HSPA and LTE for Mobile Broadband. Academic Press, 2nd edition edition, 2008.

[9] Niklas Jaldén. Analysis and Modelling of Joint Channel Properties from Multi-site, Multi-Antenna Radio Measurements. PhD thesis, KTH, School of Electrical Engineeering, March 2010.

43

(56)

44 BIBLIOGRAPHY

[10] J. Zander L. Ahlin and Slimane. Principles of Wireless Communications.

Studentlitteratur AB, 2008.

[11] L.J Cimini P. Svedman, S.K. Wilson and Björn Ottersten. Opportunistic Beamforming and Scheduling for OFDMA systems. IEEE Transctions on Communications, 55(5), 2007.

[12] D.N. C. Tse P. Viswanath and R. Laroia. Opportunistic Beamforming using Dumb Antennas. IEEE Trans. on information theory, 48(6), 2002.

[13] John G. Proakis. Digital Communications. McGRAW-HILL, London, 4th. edition edition, 2001.

[14] Bilal Sadiq and Seung Jun Baek. Delay-optimal opportunistic schedul- ing and approximations: The log rule. In IEEE/ACM Trans. Network, number 19(2), pages 405–408, 2011.

[15] M. Sellathurai and S. Haykin. Space-Time Layered Information Process- ing for Wireless Communications. John Wiley & Sons Inc, 2009.

[16] Rohit Kapoor Siddharth Mohan and Bibhu Mohanty. Latency in HSPA Data Networks. Technical report, Qualcomm, 2011.

[17] Hui Zhang Xiaodong Xu and Qiang Wang. Inter-cell Interference Mit- igation for Mobile Communication System, Advances in Vehicular Net- working Technologies. Dr Miguel Almeida, 2011.

A Practical Analysis of the Eects of Opportunistic Nulling in LTE-based Systems