You are on page 1of 50

Institutionen fr systemteknik

Department of Electrical Engineering


Examensarbete

Full-Duplex Multiuser MIMO with Massive Arrays

Examensarbete utfrt i Kommunikationssystem


vid Tekniska hgskolan i Linkping
av
Hussain A. Wannas
LiTH-ISY-EX--14/4746--SE
Linkping 2014

Department of Electrical Engineering


Linkpings universitet
SE-581 83 Linkping, Sweden

Linkpings tekniska hgskola


Linkpings universitet
581 83 Linkping

Full-Duplex Multiuser MIMO with Massive Arrays

Examensarbete utfrt i Kommunikationssystem


vid Tekniska hgskolan i Linkping
av
Hussain A. Wannas
LiTH-ISY-EX--14/4746--SE

Handledare:

Hien Quoc Ngo


isy, Linkpings universitet

Examinator:

Danyo Danev
isy, Linkpings universitet

Linkping, 26 February, 2014

Avdelning, Institution
Division, Department

Datum
Date

Division of Communication Systems


Department of Electrical Engineering
Linkpings universitet
SE-581 83 Linkping, Sweden
Sprk
Language

Rapporttyp
Report category

ISBN

 Svenska/Swedish

 Licentiatavhandling

ISRN


 Engelska/English

 Examensarbete

 C-uppsats
 D-uppsats

 vrig rapport

2014-02-26

LiTH-ISY-EX--14/4746--SE
Serietitel och serienummer ISSN
Title of series, numbering


URL fr elektronisk version
http://www.commsys.isy.liu.se
http://urn.kb.se/resolve?urn=urn:nbn:se:liu:diva-ZZZZ

Titel
Title

Full-Duplex Multiuser MIMO with Massive Arrays

Frfattare Hussain A. Wannas


Author

Sammanfattning
Abstract
Half-Duplex Multiuser Multiple-Input Multiple-Output (HD MU-MIMO) systems
currently employed in communication systems are not experiencing the selfinterference (SI) problem but they are not optimal in terms of efficiency and in
terms of resources used (time and frequency resources). Ignoring the effect of largescale fading, we start by explaining the uplink (UL) and downlink (DL) parts of
the MU-MIMO system and how the sum-rate is calculated. We also introduce the
three linear receivers/precoders, Maximum-Ratio Combining (MRC)/MaximumRatio Transmission (MRT), Zero-Forcing (ZF), and Minimum Mean-Square Error
(MMSE) and which of the three types is going to be used in the study of FullDuplex Multiuser Multiple-input Multiple-output (FD MU-MIMO) system. Then
we introduce FD MU-MIMO system, and how the equation used to calculate the
sum-rate of the UL part changes when the SI occurs, and why SI problem is not
present in the DL part. Next, we introduce the spectral efficiency (SE), and how
to calculate it and why it is taken as a parameter to compare HD and FD systems.
Also the effect of SI on FD MU-MIMO system is presented through simulation
graphs, then we move to show how to reduce SI effect by increasing the number of
antennas in the base-station (BS). Lastly, we take the effect of large scale fading in
order to reach a simple statistical model in the form cumulative distribution function (CDF) graph for different values of SI and compare those of FD MU-MIMO
system to HD MU-MIMO. The results show that FD MU-MIMO together with
massive MIMO technology is very promising and would save time and frequency
resources which means an increase in the SE but SI must be below a certain level.

Nyckelord
Keywords
Full-Duplex, Massive MIMO, MIMO, MU-MIMO

Abstract
Half-Duplex Multiuser Multiple-Input Multiple-Output (HD MU-MIMO)
systems currently employed in communication systems are not experiencing the self-interference (SI) problem but they are not optimal in terms of
efficiency and in terms of resources used (time and frequency resources).
Ignoring the effect of large-scale fading, we start by explaining the uplink
(UL) and downlink (DL) parts of the MU-MIMO system and how the sumrate is calculated. We also introduce the three linear receivers/precoders,
Maximum-Ratio Combining (MRC)/Maximum-Ratio Transmission (MRT),
Zero-Forcing (ZF), and Minimum Mean-Square Error (MMSE) and which
of the three types is going to be used in the study of Full-Duplex Multiuser
Multiple-input Multiple-output (FD MU-MIMO) system. Then we introduce FD MU-MIMO system, and how the equation used to calculate the
sum-rate of the UL part changes when the SI occurs, and why SI problem
is not present in the DL part. Next, we introduce the spectral efficiency
(SE), and how to calculate it and why it is taken as a parameter to compare HD and FD systems. Also the effect of SI on FD MU-MIMO system is
presented through simulation graphs, then we move to show how to reduce
SI effect by increasing the number of antennas in the base-station (BS).
Lastly, we take the effect of large scale fading in order to reach a simple
statistical model in the form cumulative distribution function (CDF) graph
for different values of SI and compare those of FD MU-MIMO system to HD
MU-MIMO. The results show that FD MU-MIMO together with massive
MIMO technology is very promising and would save time and frequency resources which means an increase in the SE but SI must be below a certain
level.

Acknowledgments
I would like to start with my parents; they always kept me and my siblings
on check when it comes to education, I would like to thank them for being
so resolute in this regard because it paid us a lot being well educated. I
would like to thank my wife and my siblings for their morale support. I
can not forget my supervisor in this, so I would like to show my gratutude
to Mr. Hien Q. Ngo for being helpful, supportive and patient with an open
door whenever I needed him. My examiner, Associate Professsor Danyo
Danev, I would like to thank him for giving me the opportunity to pursue
my thesis work in the division of communication systems.

vii

Contents
1 Introduction

2 Multiuser-MIMO and Massive MIMO


2.1 Introduction . . . . . . . . . . . . . . .
2.2 Channel Model and Assumption . . .
2.2.1 Channel Assumption . . . . . .
2.2.2 Channel Estimation . . . . . .
2.3 Uplink System . . . . . . . . . . . . .
2.3.1 System Model . . . . . . . . . .
2.3.2 Linear Receivers . . . . . . . .
2.3.3 Achievable Rate . . . . . . . .
2.4 Downlink System . . . . . . . . . . . .
2.4.1 System Model . . . . . . . . . .
2.4.2 Linear Precoders . . . . . . . .
2.4.3 Achievable Rate . . . . . . . .

.
.
.
.
.
.
.
.
.
.
.
.

5
5
7
7
9
9
9
10
12
13
13
13
15

3 Full-Duplex Massive MIMO


3.1 System Model . . . . . . . . . . . . . . . . . . . . . . . . . .
3.2 Linear Receivers . . . . . . . . . . . . . . . . . . . . . . . .
3.3 Achievable Rate . . . . . . . . . . . . . . . . . . . . . . . . .

17
18
19
20

4 Simulation Results and Discussion


4.1 Uplink and Downlink Performance
4.2 Spectral Efficiency . . . . . . . . .
4.3 Fixed Large Scale fading . . . . . .
4.4 Random User Location . . . . . . .

21
21
22
22
24

.
.
.
.

.
.
.
.

Systems
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.

.
.
.
.

5 Conclusions

35

Bibliography

37

ix

List of Figures
2.1
2.2

MU-MIMO Uplink (UL) system model. . . . . . . . . . . .


MU-MIMO Downlink (DL) system model. . . . . . . . . . .

10
14

3.1

MU-MIMO Full-Duplex (FD) system model.

. . . . . . . .

18

4.1
4.2
4.3
4.4
4.5
4.6

Performance of UL linear receivers in Half-Duplex. . . . . .


Performance of UL linear receivers in Half-Duplex. . . . . .
Performance of DL linear receivers in Half-Duplex. . . . . .
Performance of DL linear receivers in Half-Duplex. . . . . .
Self interference versus spectral efficiency. . . . . . . . . . .
Spectral Efficiency versus Number of BS antennas (M ) for
SIM RC,ZF = 0dB. . . . . . . . . . . . . . . . . . . . . . . .
Spectral Efficiency versus Number of BS antennas (M ) for
SIM RC,ZF = 10dB. . . . . . . . . . . . . . . . . . . . . . . .
Spectral Efficiency versus Number of BS antennas (M ) for
SIM RC,ZF = 12dB. . . . . . . . . . . . . . . . . . . . . . . .
Spectral Efficiency versus Number of BS antennas (M ) for
SIM RC = 20dB and SIZF = 10dB. . . . . . . . . . . . . . .
Spectral Efficiency versus Number of BS antennas (M ) for
SIM RC = 25dB and SIZF = 10dB. . . . . . . . . . . . . . .
Spectral Efficiency versus Number of BS antennas (M ) for
SIM RC = 30dB and SIZF = 10dB. . . . . . . . . . . . . . .
Cumulative Distribution Function (CDF) for Zero-Forcing.
Cumulative Distribution Function (CDF) for Zero-Forcing
with SI equal to 5dB. . . . . . . . . . . . . . . . . . . . . .
Cumulative Distribution Function (CDF) for Zero-Forcing
with SI equal to 2dB. . . . . . . . . . . . . . . . . . . . . .
Cumulative Distribution Function (CDF) for Zero-Forcing
with SI equal to 1dB. . . . . . . . . . . . . . . . . . . . . .
Cumulative Distribution Function (CDF) for Zero-Forcing
with SI equal to 0dB. . . . . . . . . . . . . . . . . . . . . .

26
26
27
27
28

4.7
4.8
4.9
4.10
4.11
4.12
4.13
4.14
4.15
4.16

28
29
29
30
30
31
31
32
32
33
33

Chapter 1

Introduction
In mobile networks of today, multiple-input multiple-output (MIMO) is utilized in the shape of MU-MIMO (Multiuser-MIMO) in the downlink (DL)
and uplink (UL) communications link [14]. A MU-MIMO communications
system utilizes multiple antennas at the transmitter and receiver ends, and
uses spatial multiplexing to enable different signals to be forwarded on the
same frequency concurrently. MU-MIMO is very promising when the desire
is to increase data transfer rates but soon, the technology is reaching the
limits of its data transfer speed [14]. Size, cost and power consumption are
factors limiting the number of antennas that can be utilized on both ends
of the communication link.
In theory, to have the DL and UL working concurrently using the same
frequency would mean that time for transmission is lowered and the frequency resources needed will be less, and hence, an increase in the spectral
efficiency is obtained.
The Problem
MU-MIMO systems operate in half-duplex (HD), which means that
DL and UL do not share either the same time of operation or the same
frequency of transmission. When both DL and UL operate in the same
frequency at the same time, this is called full-duplex (FD), but this will
cause self-interference (SI) at the base-station (BS) between transmit and
receive antennas. Interference effect means that the self-interference (SI)
power compared to the power of the desired signal is increased and according to Shannons theorem, it means that the data rate will decrease.

Introduction

In order to reduce the effect of SI, one possible method is using a massive array of antennas in the order of hundreds in the BS. With massive
arrays, an increase in the power of the desired signal compared to that of
the introduced SI component in FD is achieved.
The Goal
In this work, we compare the system performance of FD and HD in
terms of spectral efficiency. We consider the effects of SI, number of antennas in BS and propagation environment.

Chapter 2

Multiuser-MIMO and
Massive MIMO Systems
2.1

Introduction

Multiple-input and multiple-output, MIMO for short, refers to the use of


multiple antennas at both ends of the communication link (transmit and
receive ends) and the purpose is to improve communication performance
and reliability. It is worth to mention that the terms input and output
refer to the radio channel carrying the signal and not the device where the
antennas are mounted.
When MIMO is used as a technique of communication among many terminals simultaneously, then we are speaking about multiuser MIMO and
for short, we call it MU-MIMO [10, 6].
The more antennas installed on transmitter/receiver, the more degrees
of freedom the propagation channel can provide, the better performance
(data rate) and reliability it can achieve. Communication channel varies
rapidly and continuously over time and frequency, the achievable rate scales
as
min(nt , nr )log2 (1 + SN R),
where nt is the number of transmit antennas, nr is the number of receive
antennas, and SN R is the signal-to-noise ratio. Multi-user systems offer
gains because such systems allow for simultaneous transmission where the
base-station can communicate with several users in the same time over the
same frequency resources. Benefits of high performance and reliability do
not come cheap, but they come on the expense of more complex hardware,
5

Multiuser-MIMO and Massive MIMO Systems

and also an increase in both, complexity of the signal processing and energy
consumption at transmit and receive ends of the communication system.
In MU-MIMO systems, transmitters are complicated because of the coding
techniques that are used in order to communicate simultaneously with more
than one user and keeping inter-user interference at a controlled level. More
antennas mean more physical space needed to host the increased number
of antennas which translates into real estate rental costs [16]. MU-MIMO
offers many advantages [10]:
Higher data rate: due to the large number of antennas installed in the
BS, the BS can transmit bit streams to many users simultaneously,
which simply means that BS can serve more users in the same time.
This is referred to as the multiplexing gain.
More reliable link: the more antennas communicating over a link
means more paths are created for the signal to propagate over between
transmit and receive ends of the link. This is referred to as diversity
gain.
More energy efficient: the BS is capable of directing its transmitted
beam towards where the users are located. This is called array gain.
Decreased interference: the BS can avert transmitting into directions
where effect of interference cannot be controlled.
The advantages listed above urged the industry to adopt MU-MIMO as
one of the wireless communication standards. The more antennas used in
BS, the better and more reliable communication is achieved [10].
In current LTE, a BS can have up to 8 antennas and serve several users
the same time, but when Massive MIMO is the subject of discussion, now
the talk is about a BS that utilizes a number of antennas in the range of
several hundreds that can serve tens of users simultaneously [10]. With
Massive MIMO, we want to make use of the advantages of MU-MIMO on
a larger scale. Advantages of Massive-MIMO include [10, 7]:
Massive MIMO may increase the capacity 10 times or more as well
as increasing the energy-efficiency of the radiated signal in the order
of 100 times. Increase of capacity is due to the aggressive spatial
multiplexing employed in Massive MIMO.
Massive MIMO can be constructed using cheap, low-power components.

2.2 Channel Model and Assumption

Massive MIMO significantly reduces the effect of fading.

For the above, Massive MIMO attracted attention and here in this work,
we try to investigate these advantages in terms of sum-rate and spectral
efficiency.

2.2
2.2.1

Channel Model and Assumption


Channel Assumption

It is not possible to exactly estimate the behavior of the wireless channel


due to the physical properties of the environment, but channel physics is
the major factor affecting the performance of a wireless communication
system. There are channel models proposed in order to consider the effect of channel when studying the performance of communication systems.
Rayleigh fading channel is a statistical model for the effect of a propagation
environment on a radio signal, such as that used by wireless devices. In
this model, the assumption is that the magnitude of a signal that passes
through a communications channel will alter at random, or fade according
to a Rayleigh distribution.
Rayleigh fading is considered as a possible model for signals propagating in the troposphere and ionosphere layers as well as the effect of heavily
built-up urban environments on radio signals. Rayleigh fading is most applicable when there is no dominant propagation along a line of sight between
the transmitter and receiver. If line of sight occurs, Rician fading may be
more applicable but here, we will stick to Rayleigh model.
A BS with M antennas and K single-antenna users will be considered,
which means we have a total of K antennas at the user end. Single-antenna
users are cheap, not complex in design and consume less power and users
still get high bit-rate [12].
We will assume that the effect of channel on the transmitted signal is
modeled by the matrix G which has dimensions of M K. Since we have
M antennas at BS and K single-antenna users at user end, then we will
have each one of M antennas transmitting to K users. Therefore,

Multiuser-MIMO and Massive MIMO Systems

..
.

g1K
g2K
..
.

gM 1 gM 2

gM K

g11

g21
G=
..
.

g12
g22
..
.

(2.1)

Each element of the channel matrix G, gmk , [G]mk , represents the


channel coefficient between the mth antenna of BS and the k th user. The
channel, gmk , can be expressed as [12]:
p

gmk = hmk k ,

m = 1, 2, ..., M,

k = 1, 2, ..., K.

(2.2)

In the expression above, hmk models the fast fading coefficient from the
k th user to the mth antenna of the BS and each single element, hmk , is
assumed to be a Gaussian. The geometric attenuation and shadow fading
which is assumed to be independent over m, also assumedto be not changing over many coherence time intervals and is modeled as k , and it is also
called the large-scale fading coefficient. We model k via k = zk /(rk /rh )v ,
where zk is a log-normal random variable with standard deviation (shadow ),
rk is the distance between the k th user and the BS, rh is the distance of
the closest user to the BS, and v is the path loss exponent [12]. The distances between the single-antenna users and the BS are much larger than
the distance between the antennas of the BS, and the value of k changes
very slowly with time. Then, we have
G = HD1/2

(2.3)

where H is the M K matrix of fast fading coefficients between the K


users and the BS, i.e., [H]mk = hmk [12], and since we assumed that we
have a Rayleigh fading channel, the sum of the multiple paths between BS
and users is statistically independent scattered and reflected paths with
random magnitudes is an independent and identically distributed (i.i.d)
Gaussian random variable with zero mean and 0.5 variance. The element
of the channel matrix can be written in a complex number form as
hmk = cmk + jdmk

(2.4)

2.3 Uplink System

1
1
where cmk N (0, ) and dmk N (0, ), [1], and D is a K K diagonal
2
2
matrix, where [D]kk = k .

2.2.2

Channel Estimation

In wireless communications, channel state information (CSI) refers to known


channel properties of a communication link. This information describes how
a signal propagates from the transmitter to the receiver and represents the
combined effect of, for example, scattering, fading, and power decay with
distance. The CSI makes it possible to adapt transmissions to current channel conditions, which is crucial for achieving reliable communication with
high data rates in multi-antenna systems. In order for the transmitting end
be able to estimate the channel matrix G, BS on the downlink sends pilot
sequence, then each user feedbacks this CSI back to the BS. The channel
estimation gets more complicated when the number of antennas in the BS
is increased and this costs time and higher energy consumption, so we assume that the channel is estimated in the BS using pilots sent in the uplink
by users [9, 18]. Throughout our work in this thesis, we will assume that
BS has perfect knowledge of channel state information, i.e., it knows G.

2.3
2.3.1

Uplink System
System Model

We assume K single-antenna users and a BS utilizing an array of M antennas as can be seen in Figure 2.1. The BS receives y, an M 1 vector
from K single-antenna users in the uplink

y=
=

pu

K
X

gk xk + n

(2.5)

k=1

pu Gx + n.

(2.6)

In equation (2.5), gk is the M 1 channel vector between the k th user

and the BS, pu xk is the transmitted signal from the k th user (the average
power transmitted by each user is pu ), and n CM 1 is a vector of additive white noise with zero-mean. We assume the noise variance to be equal
to 1 to minimize notation. In (2.5), G , [g1 , . . . , gk ] and x , [x1 , . . . , xk ]T

10

Multiuser-MIMO and Massive MIMO Systems

Figure 2.1. MU-MIMO Uplink (UL) system model.

[12, 11].
Using equation (2.3), equation (2.5) can be re-written as
y=

2.3.2

pu HD1/2 x + n.

(2.7)

Linear Receivers

In order for the BS to detect the received signal, it uses maximum-likelihood


detectors to gain ideal performance. Such receiver has a drawback; its complexity grows exponentially as the number of users, K, increases. That is
why and because in MU-MIMO we have M  K  1, under such circumstances, linear detectors perform pretty well and for that reason, linear
receivers (or detectors), Maximum Ratio Combining (MRC), Zero Forcing
(ZF), and Minimum Mean-Square Error (MMSE) are going to be considered. As we mentioned before that the BS has perfect knowledge of CSI,
i.e., it knows G [12, 11].
With MRC, the detector will maximize the SNR of the stream of inter-

2.3 Uplink System

11

est and neglects the effect of multiuser interference. With ZF, the detector
will deal with multiuser interference but neglects the effect of noise. With
MMSE, the detector is an estimator that estimates AH y and minimizes the
mean square error (MSE) between the estimate and the transmitted signal
x [12, 8].
Let A be an M K linear detector matrix which depends on the channel
matrix, G [12]:


1

H
G
G
G
A=

1

IK
G G G+

for MRC
for ZF

(2.8)

for MMSE

pu

By using the linear detector, the received signal in equation (2.5) is


multiplied by the conjugate-transpose of A or for short, AH (Hermitian
of A), in order to detect the signal sent by everyone of the K individual
single-antenna users

r = AH y.

(2.9)

Using linear detectors, and from equations (2.6) and (2.9), the received
vector is given by:
r=

pu AH Gx + AH n.

(2.10)

Assume that rk is the kth element of the K 1 vector r, and xk is the


kth element of the K 1 vector x. Then, for the BS to extract the signal
sent by the kth user, [12]

rk =

H
pu aH
k Gx + ak n =

pu aH
k gk x k +

pu

K
X

H
aH
k gi xi + ak n. (2.11)

i=1,i6=k

In equation (2.11), pu aH
desired signal received
k gk xk represents the
PK
th
H
at the BS transmitted by the k user and pu i=1,i6=k aH
k gi xi + ak n is

12

Multiuser-MIMO and Massive MIMO Systems

the interuser-interference plus noise component, and, ak and gk are the kth
columns of the matrices A and G, respectively [12].

2.3.3

Achievable Rate

Since elements of vector x are not correlated, so in (2.11), multiuserinterference plus noise can be modelled as additive Gaussian noise. In
general, Shannons theorem formulates the maximum rate that a channel
can deliver to be

R = log 2 1 +

S
N

(2.12)

where R is the achievable rate in bits/s/Hz, S is the power of the desired signal and N and is the power of unwanted signals such as multiuserinterference white Gaussian noise added by the receiver. The achievable
rate of the k th user is given by

L
RU
P,k

= log 2

2
pu |aH
k gk |
1 + PK
2
2
pu i=1,i6=k |aH
k gi | + kak k

(2.13)

L
The superscript in RU
P,k refers to the uplink phase and subscript,
P,
refers to perfect knowledge of G. In the channel matrix, G, gmk = hmk k ,
where hmk is the fast fading component so in order to calculate the achievable rate, we need to average the achievable rate over many values of hmk
in order to obtain the rate as accurate as possible, then

(
L
RU
P,k

= E log 2

2
pu |aH
k gk |
1 + PK
2
2
pu i=1,i6=k |aH
k gi | + kak k

!)

(2.14)

Equation (2.14) is the general expression to calculate the UL achievable


rate. In order to find the achievable rate for every one of the three linear
detectors, equation (2.8) is used to calculate the detector matrix and then
the result is substituted in equation (2.14).

2.4 Downlink System

2.4
2.4.1

13

Downlink System
System Model

The BS employs the CSI acquired from the training phase to process the
information signal before transmitting it over the channel to the K singleantenna users. Also, assume that the BS has a perfect knowledge of CSI,
and this can be assumed if the power of pilot signal in the uplink is high
or the coherence time is large. We assume a BS utilizing an array of M
antennas and K single-antenna users as shown in Figure 2.2. So far this
seems to be similar to the uplink model but signal processing is different.
Since the BS uses M antennas, then transmitting information signal from
M antennas with each antenna sending its signal with a power equal to Ptr
means that the BS will use M Ptr , and that means a lot of power consumption so the BS has to find a way to reduce the transmitted power while
maintaining an acceptable downlink bitrate at the user end. Precoding the
information comes here to solve this issue by scaling the power transmitted
by each antenna in the BS. Since it is a MU-MIMO system, two of the
many advantages of such layout are the diversity gain and the array gain.
This can be interpreted as, firstly, a user can still get many copies of the
same signal from different paths (diversity gain) and these copies can be
summed in order to get a stronger signal. Secondly, the BS can direct its
beam towards the desired user which means that the power is directed in
the desired direction instead of being transmitted in many directions (array
gain).

2.4.2

Linear Precoders

Assume that the information vector, x, which is a K 1 in dimensions, and


A CM K to be the precoding matrix [17]. Assume that
1
B.
(2.15)

For linear precoding, maximum ratio transmission (MRT), zero forcing


(ZF), and minimum mean-square error (MMSE) [5], matrix B is
A=

 T 1
G G G
B=
1


IK
G G G +

Ptr

for MRT
for ZF
for MMSE

(2.16)

14

Multiuser-MIMO and Massive MIMO Systems

Figure 2.2. MU-MIMO Downlink (DL) system model.

and is a normalization which guarantees that transmitted power is limited to a certain level.
To calculate the value of , the precoded vector is s = Ax and
E ksk2 = E kAxk2 and E kxk2 = 1





1
1 
E kAk2 = E k Bk2 = 2 E kBk2

1
tr(BBH ) = Ptr
2

so
s

tr(BBH )
Ptr

where G is the channel matrix and the transmitted vector is

(2.17)

2.4 Downlink System

15

s=

1
Bx.
2

(2.18)

The K single-antenna users receive the vector given by equation (2.19)

y = GT s + n
1
= GT Bx + n

(2.19)
(2.20)

and as in equation (2.11),

yk =

K
1 T
1
1 X
gk Bx + nk = gTk bk xk +
gT bi xi + nk .

i=1,i6=k k

(2.21)

The energy of the desired signal is


"
2 #
2
1 T

1


S = E gk bk xk = 2 gTk bk

(2.22)

and the energy of interference plus noise component is



2


K
K



1 X
1 X
T 2
T
N = E
gk bi xi + nk = 2
gk bi + 1.
i=1,i6=k
i=1,i6=k

2.4.3

(2.23)

Achievable Rate

Following the same discussion of (2.12), (2.13) and (2.14), the achievable
rate is

1 T 2
g bk
2 k

DL
.
RP,k = E log 2 1 +

T 2
1 PK

g bi + 1
2 i=1,i6=k k

(2.24)

16

Multiuser-MIMO and Massive MIMO Systems

The superscript in RDL


P,k refers to the downlink phase, and the subscript,
P , refers to perfect knowledge of G. Equation (2.24) is the general expression to calculate the achievable rate at each user in the downlink. In order
to find the achievable rate for every one of the three linear detectors, equations (2.16) and (2.17) are used to determine B and and the result is to
be substituted in equation (2.24).

Chapter 3

Full-Duplex Massive MIMO


With traditional MIMO, links are reliable and spectral efficiency is high
without the need to increase the transmit power, but these links are still
capable of bringing more gain in terms of data throughput and spectral
efficiency. The DL and UL of MU-MIMO systems employed in the current
cellular systems are either running on different time slots (Time Division
Duplex, TDD) or operate in two different frequencies (Frequency Division
Duplex, FDD) which means they are not sharing the same time and same
frequency of transmission and this is called the half duplex (HD) mode.
Therefore, systems are either losing time or frequency resources when links
operate in HD mode, which means that such systems do not achieve best
spectral efficiency possible [14, 13, 4, 3, 2, 16].
Advatages
What if the UL and DL run in the same time using the same frequency;
that means systems will save time and frequency resources, which in return
will increase the spectral efficiency, and longer time for both DL and UL
means more data to be transferred over the link. The term full duplex
mode (FD) is used when having both, the UL and DL, running at the same
time and the same frequency simultaneously [13, 14].
Disadvatages
Operating at the same time with the same frequency for both, the UL
and DL, will cause self-interference in the BS between the transmit and
receive antennas and the effect of this self-interference is what this work is
trying to determine [13, 14].
17

18

Full-Duplex Massive MIMO

Figure 3.1. MU-MIMO Full-Duplex (FD) system model.

3.1

System Model

The system model to be studied is a MU-MIMO relay system. Such systems can be used to extend the coverage of cell networks for instance. The
system is shown in figure 3.1 and it consists of two groups of users, each
with K Single-antenna users, one group in the UL end and the other is in
the DL end of the link. In the middle, a BS with 2M antennas, M antennas are used to transmit in the DL side and M antennas are employed for
the purpose of reception on the UL side. Further, it is assumed that the
two groups of users are far away from each other in terms of geography so
that the interference between users transmitting in the UL side and those
receiving in the DL side (co-channel interference) is ignored [14].
With the above system setup, it is clear that the M uplink antennas
will receive a self-interference component along with signal sent from the
UL users. The signal detected at the receive antennas of the BS, yU L , will
be similar to equation (2.6) with an extra SI component caused by the
downlink transmit antennas and it is given as
yU L =

pu GU L x + GSI WDL xDL + n

(3.1)

where GU L is the uplink channel matrix, x is a K 1 vector representing

3.2 Linear Receivers

19

the signal sent from K users in the unlink, n CM 1 is a vector of additive


white noise with zero-mean, GSI WDL xDL is the self-interference component from downlink that is added to the uplink received vector, yU L , GSI
is the self-interference channel matrix from downlink transmit antennas to
uplink receive antennas, and WDL is the downlink precoding matrix and
from equations (2.15), (2.16), and (2.17)
WDL =

1
B

(3.2)

and xDL is the signal vector sent over the downlink.


From the above assumptions, the users at the downlink end are not
affected and achievable rate calculation of the DL is the same as that introduced in section 2.4.

3.2

Linear Receivers

Assuming perfect CSI, which again means that the BS knows the channel
matrix GU L . Let AU L be an M K linear detector matrix which depends
on the channel GU L . By using the linear detector, the received signal is
separated into streams by multiplying it with AH
U L as follows
rU L = AH
U L yU L

(3.3)

We consider three conventional linear detectors MRC and ZF and MMSE,


such that

GU L


1

H
G
G
G
U
L
U
L
U
L
AU L =

1

IK
GU L GU L GU L +

pu

substituting (3.1) in (3.3), (3.3) becomes

for MRC
for ZF
for MMSE

(3.4)

20

Full-Duplex Massive MIMO

rU L =

3.3

H
H
p u AH
U L GU L x + AU L GSI WDL xDL + AU L n.

(3.5)

Achievable Rate

Let rU L,k and xk be the k th elements of the K 1 vectors rU L and x,


respectively. Then,

rU L,k =
=

H
H
pu aH
U L GU L x + aU L GSI WDL xDL + aU L n

(3.6)

K
X
H
H
pu aH
g
x
+
p
aH
u
UL UL k
U L,k gU L,i xi +aU L GSI WDL xDL +aU L n.
i=1,i6=k

(3.7)

Following the same discussion of sections 2.3.2 and 2.3.3, the achievable
rate in the uplink is

2
pu |aH
U L,kgU L,k |

(
L
RU
P,k =E

log 2 1+

pu

!)

PK

H
H
2
2
2
i=1,i6=k |aU L,k gU L,i | +kaU L,kGSIWDLk +kaU L,k k

(3.8)

Using equations (3.4) and (3.8), the achievable rate for each one of the
three linear receivers, MRC, ZF and MMSE is calculated.

Chapter 4

Simulation Results and


Discussion
4.1

Uplink and Downlink Performance

We will start our analysis of FD systems by studying the UL and DL of


a MU-MIMO system in order to determine the effect of transmit power
and number of antennas in the BS. The three linear detectors/precoders,
MRC/MRT, ZF and MMSE are considered to reduce decoding complexity.
Before we start, there are two observations to be noticed that are going to
serve us in reducing the amount of work required to do. For the uplink phase
of the FD system, from Figure 4.1 where M = 10, K = 10, and Figure 4.2
where M = 100, K = 10, we can see that ZF and MMSE linear receivers
perform almost identical to each other when transmit power increases or
when number of antennas in BS is increased and the same also applies
on the downlink phase as shown in Figure 4.3 (M = 10, K = 10) and
Figure 4.4 (M = 100, K = 10), and in this work, a massive MU-MIMO
system is considered where the BS is equipped with number of antennas
in the order of hundreds so it is very obvious that studying either ZF or
MMSE receiver would be sufficient. ZF linear receiver have been chosen
over MMSE because of the simplicity of signal processing which means it is
cheaper to implement. The graphs mentioned above were plotted assuming
the large scale fading (slow fading) to be equal to 1 so D1/2 = 1 and
equation (2.3) becomes G = H.
21

22

Simulation Results and Discussion

4.2

Spectral Efficiency

Spectral efficiency is a measure of how efficiently a limited spectrum is utilized .


In Half-Duplex (HD), UL and Dl are either work on different frequencies in the same time or same frequency but in different time slots. In
Full-Duplex (FD), both UL and DL work on same frequency in the same
time so the spectral efficiency (SE) for the HD is calculated as follows
1 UL
SEHD = (RHD
+ RDL )
2

(4.1)

and for the FD, the spectral efficiency is calculated as follows

L
SEF D = RFU D
+ RDL .

(4.2)

From equations (4.1) and (4.2), we can see the difference between the
1
two cases of HD and FD. The first difference is the ( ) component in
2
equation 4.1 which reflects to the fact that the spectrum is divided between
UL and DL phases so UL and DL are not utilizing the spectrum all the time
on the same frequency but either on different time using same frequency
and time is divided by the two phases, or two different frequency in the same
time which means doubling the frequency resources requirement. The other
U L and RU L , where these are calculated using equations
difference is RHD
FD
(2.14) and (3.8) respectively while RDL is calculated using equation (2.24)
because as discussed in chapter 3, the DL phase is not affected by BS
self-interference.

4.3

Fixed Large Scale fading

To start studying the effect of self-interference (SI) , we need to examine


the effect of self-interference (SI) on the spectral efficiency. The effect of
large scale fading (slow fading) was not considered when the spectral efficiency was calculated so D1/2 = 1 and G = H as explained in section
(4.1). Figure 4.5 demonstrates the effect of self-interference. It can be

4.3 Fixed Large Scale fading

23

noticed clearly that the spectral efficiency decreases when the SI level increases. From equations (4.1) and (4.2), we can expect that the FD mode
should perform better than HD mode when there is no SI effect or when
it is very low, but the spectral efficiency decreases with the increase of SI
level. The HD mode is not affected by SI because there isnt any so the
spectral efficiency is constant and it was added to the figure in order to
be able to specify the point where HD starts to perform better than FD.
Figure 4.5 shows the effect of SI for both, the MRC and ZF linear detectors
show different response. It can observed that while ZF has higher spectral
efficiency in normal cases under limited SI, it is not as robust to SI as the
MRC when we comparing the HD and FD modes of each type of detector.
The reason is that in MRC, the BS maximizes the SNR of received streams
and ignore the effect of interference.
Now, it is important to examine the advantage of massive MU-MIMO
where the BS is equipped with a large number of antennas in the order
of hundreds. Figure 4.6 displays the effect of increasing the number of BS
antennas on the spectral efficiency when the SI level is 0dB (=1 in leaner
scale). It is clear to see that FD performs better for all range of M (number
of BS antennas). In Figure 4.7, SI level is increased to 10dB, and it can noticed that FD in MRC is still performing much better than HD even when
M is below 100. In ZF, when M is low, HD is performing better than FD,
but FD starts to improve with the increase of M until they perform the
same when M is around 210 and after that, FD starts to perform better
with the increase of M (number of antennas in BS), so it can noticed that
increasing M will help reduce the effect of SI.
In Figure 4.8, the SI level was increased to 12dB and MRCs FD still
performing better than HD and FD performs better and better with the
increase of M increases. As for ZF, increasing the SI level meant that in
order to have FD performing better than HD, M should be increased to
around 500 so we can have FD and HD perform almost the same and for
FD to perform better than HD, M should be increased much higher to get
a reasonable increase in performance.
In Figure 4.9, SI level for ZF was kept at 10dB and for MRC, SI level was
increased to 20dB to find the point where HD starts to perform better than
FD but the almost performed the same as can be noticed. In Figure 4.10,
SI level for MRC was increased to 25dB and it was the same as explained
before and HD and FD performed similar to each other. In order to make

24

Simulation Results and Discussion

sure of the observations recorded in Figures 4.9 and 4.10, SI level for MRC,
was increased to 30dB and nothing changed as can be seen in Figure 4.11,
and that both FD and HD performed almost identical to each other which
meant that for MRC, HD can not perform better than FD regardless of the
amount of SI level. It is worth to mention that in all of the simulations of
this section, the transmitted power for both UL and DL was 10dB.

4.4

Random User Location

Random user location means that G = HD1/2 and D is not equal to 1 like
we assumed in the previous section. In order to take into account the effect
of changing the position of users, we refer to subsection 2.2.1 where the
calculation and parameters of equation (2.2) were explained. For our simulation, the values of parameters were: rk is a random number between 100m
and 1000m and it represents the distance of a user from the base-station,
rh = 100m, v = 3.8, shadow = 8dB and zk = 10(randnshadow /10) , where
randn is a MATLAB function that returns a scalar whose value changes
each time it is referenced.
For MRC, from Figures 4.9, 4.10, 4.11, it is clear that increasing SI level
along with increasing the number of BS antennas (M ) did not change the
value of spectral efficiency and that both FD and HD performed almost
identical to each other so we decided to only study ZFs HD and FD performance because it is clear to notice the effect of changing SI level along
with the number of BS antennas (M ).
In probability theory and statistics, the cumulative distribution function (CDF), or just distribution function, describes the probability that a
real-valued random variable X with a given probability distribution will
be found at a value less than or equal to x. In the case of a continuous
distribution, it gives the area under the probability density function from
minus infinity to x. So, our random variable X will be the Spectral Efficiency. We will plot CDF for three different values of SI level in order to
be able to compare the change in CDF values when SI level is changed.
The three values selected were 5dB, 10dB, and 15dB. The reason to chose
these three values is found in Figure 4.5. We want values when FD perform
better than HD, hence the 5dB value, when FD just starts to perform less
than HD, hence the 10dB value and when FD performs clearly less than HD
and hence the 15dB and the number of BS antennas (M ) was selected to
be 400 to take advantage of increased number of BS antennas to overcome

4.4 Random User Location

25

the SI effect and the number of single-antenna users (K) was selected to
be 10.
In Figure 4.12, it is easy to notice that HD performed better in the three
cases and this is in contrast with what we assumed and the reason for that
is the random user location that came into consideration and this factor
was not included when Figure 4.5 was plotted. Also, let us not forget the
parameters mentioned in the first paragraph in which we assigned values
for rk , rh , v, and shadow and these values change depending on transmission environment and they are not constant so the values we selected, the
effect was as shown in Figure 4.12.
The effect of changing the use location is big and it reduces the spectral
efficiency (SE) significantly. ZF detector became more affected by SI and
Figures 4.13, 4.14 and 4.15 show this clearly. In Figure 4.13, when SI is
5dB, HD is clearly the better performer, then when SI is 2dB, FD and HD
start to get closer in terms of performance and gets even closer when SI is
1dB and finally when SI is 0dB as shown in Figures 4.14, 4.15, and 4.16
respectively.

26

Simulation Results and Discussion

Figure 4.1. Performance of UL linear receivers in Half-Duplex.

Figure 4.2. Performance of UL linear receivers in Half-Duplex.

4.4 Random User Location

Figure 4.3. Performance of DL linear receivers in Half-Duplex.

Figure 4.4. Performance of DL linear receivers in Half-Duplex.

27

28

Simulation Results and Discussion

Figure 4.5. Self interference versus spectral efficiency.

Figure 4.6. Spectral Efficiency versus Number of BS antennas (M ) for


SIM RC,ZF = 0dB.

4.4 Random User Location

29

Figure 4.7. Spectral Efficiency versus Number of BS antennas (M ) for


SIM RC,ZF = 10dB.

Figure 4.8. Spectral Efficiency versus Number of BS antennas (M ) for


SIM RC,ZF = 12dB.

30

Simulation Results and Discussion

Figure 4.9. Spectral Efficiency versus Number of BS antennas (M ) for SIM RC =


20dB and SIZF = 10dB.

Figure 4.10. Spectral Efficiency versus Number of BS antennas (M ) for SIM RC =


25dB and SIZF = 10dB.

4.4 Random User Location

31

Figure 4.11. Spectral Efficiency versus Number of BS antennas (M ) for SIM RC =


30dB and SIZF = 10dB.

Figure 4.12. Cumulative Distribution Function (CDF) for Zero-Forcing.

32

Simulation Results and Discussion

Figure 4.13. Cumulative Distribution Function (CDF) for Zero-Forcing with SI


equal to 5dB.

Figure 4.14. Cumulative Distribution Function (CDF) for Zero-Forcing with SI


equal to 2dB.

4.4 Random User Location

33

Figure 4.15. Cumulative Distribution Function (CDF) for Zero-Forcing with SI


equal to 1dB.

Figure 4.16. Cumulative Distribution Function (CDF) for Zero-Forcing with SI


equal to 0dB.

Chapter 5

Conclusions
Massive MU-MIMO is a very promising in order to reach optimal utilization of communications link potentials but that depends on the environment where the approach is employed. Self-interference affects the spectralefficiency significantly. It can be easily said that self-interference (SI) is
what makes the full-duplex approach not the best option yet. Adopting
full-duplex means at least doubling the number of antennas (M ) that the
base-station is equipped with and this means more cost and physical space
needed to accommodate the big number of antennas and the cost the extra
property used to host the BS. The uplink sum-rate is affected a lot and
reduced hugely because of the SI component that is added to the equation
used to calculate the sum-rate while the downlink (DL) is not affected according to our assumption of having users of UL and DL geographically
away from each other so they dont affect each other. So in FD, UL is no
more performing the way it did with HD and depending on the type of
application this might be bad because what is the point of having a link
where the speed of transmission is governed by the slower part of the communications link.

35

Bibliography
[1] L. Ahlin. Principles of wireless communications. Studentlitratur AB,
2006.
[2] D. W. Bliss, P. A. Parker, and A. R. Margetts. Simultaneous transmission and reception for improved wireless network performance.
IEEE/SP 14th Workshop on Statistical Signal Processing, 2007.
[3] B P. Day, A. R. Margetts, D. W. Bliss, and P. Schniter. Full-duplex
MIMO relaying: Achievable rates under limited dynamic range. IEEE
JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, pages
15411553, Sept. 2012.
[4] Brian P. Day, Adam R. Margetts, Daniel W. Bliss, and Philip Schniter.
Full-Duplex bidirectional MIMO: Achievable rates under limited dynamic range. IEEE TRANSACTIONS ON SIGNAL PROCESSING,
pages 37023712, July 2012.
[5] X. Gao, O. Edfors, F. Rusek, and F. Tufvesson. Linear pre-coding
performance in measured very-large MIMO channels. Proc. IEEE Vehicular Technology Conf. (VTC), pages 15, Sept. 2011.
[6] David Gesbert, Marios Kountouris, Robert W. Heath Jr., ChanByoung Chae, and Thomas Salzer. Shifting the MIMO paradigm.
IEEE Signal Processing Magazine, Sept. 2007.
[7] J. Hoydis, S. Brink, and M. Debbah. Massive MIMO in the UL/DL
of cellular networks: How many antennas do we need? IEEE J. Sel.
Areas Commun., 2012.
[8] N. Kim and H. Park. Performance analysis of MIMO system with
linear MMSE receiver. IEEE Trans. Wireless Commun., vol. 7(no.
11):44744478, Nov. 2008.
37

38

Bibliography

[9] M. Kobayashi, N. Jindal, and G. Caire. Training and feedback optimization of multiuser MIMO downlink. IEEE Trans. Commun., vol.
59(no. 8):22282240, Aug. 2011.
[10] Erik G. Larsson, Fredrik Tufvesson, Ove Edfors, and Thomas L.
Marzetta. Massive MIMO for next generation wireless systems. IEEE
Communications Magazine, May 2013.
[11] Hien Quoc Ngo. Performance bounds for very large multiuser mimo
systems. Swedish Licentiate Thesis, Linkoping 2012.
[12] Hien Quoc Ngo, Erik G. Larsson, and Thomas L. Marzetta. Energy
and spectral efficiency of very large multiuser MIMO systems. IEEE
Transactions on Communications, 2013.
[13] Dan Nguyen, Le-Nam Tran, Pekka Pirinen, and Matti Latva-aho.
Transmission strategies for full duplex multiuser MIMO systems.
IEEE, pages 68256829, 2012.
[14] Dan Nguyen, Le-Nam Tran, Pekka Pirinen, and Matti Latva-aho. Precoding for full duplex multiuser MIMO systems: Spectral and energy
efficiency maximization. IEEE Transactions on Signal Processing, May
2013.
[15] Eakkamol Pakdeejit. Linear precoding performance of massive MUMIMO downlink system. Masters thesis, Linkoping University, 2013.
[16] F. Rusek, D. Persson, B. K. Lau, E. G. Larsson, T. L. Marzetta,
O. Edfors, and F. Tufvesson. Scaling up MIMO: Opportunities and
challenges with very large arrays. IEEE Signal Processing Magazine,
2013.
[17] V. Stankovic and M. Haardt. Generalized design of multiuser MIMO
precoding matrices. IEEE Trans. Wireless Commun., vol. 7:953961,
Mar. 2008.
[18] P. Viswanah and D. N. C. Tse. Sum capacity of the vector Gaussian broadcast channel and uplink-downlink duality. IEEE Trans. Inf.
Theory, vol. 49(no. 8):pp. 19121921, Aug. 2003.

You might also like