Thesisdoc Gizachew

Bahir Dar University
Bahir Dar Institute of Technology
School of Graduate Studies
Faculty of Electrical and Computer Engineering
Performance Analysis of Joint

Transmit Antenna Selection and
User Scheduling for Massive
MIMO Systems
By
Gizachew
Worku
A Thesis Submitted to the School of Graduate Studies of Bahir Dar

Institute of Technology in Partial fulfillment of the Requirements for the
Degree of
Master of Science
in Communication
Engineering
October 2020
Bahir Dar, Ethiopia
i
Bahir Dar University
Bahir Dar Institute of Technology
School of Graduate Studies
Faculty of Electrical and Computer Engineering
Performance analysis of Joint

Transmit Antenna Selection and
User Scheduling for Massive
MIMO Systems
By
Gizachew Worku
Adviser:
Dr.Fikreselam Gared (Assct.
professor)
Declaration
I declare that this thesis titled, ’Performance Analysis of Joint Transmit Antenna Selec-
tion and User Scheduling for Massive MIMO Systems’ and the work presented in it are my
own, and all sources of materials used for the thesis has been clearly stated and attributed.
Student Name: Gizachew Worku
Signature: Date of submission: 10/22/2020
This thesis has been submitted for examination with my approval as a university ad-
viser.
Advisor Name: Dr.Fikreselam Gared (Assct. Professor)
Advisor’s Signature: Date : 10/22/2020
ii
Abstrac
t
Massive multiple-input multiple-output (Ma-MIMO) system which uses a large number
of antennas to serve a relatively small number of users increases significantly the spectral
efficiency and energy efficiency of wireless communication . Hardware complexity ,energy
wastage and increased cost of deployment are the challenges faced by Ma-MIMO due to
the requirement of large number of radio frequency equipments. A joint antenna
selection and user scheduling algorithms are proposed to alleviate this problem which
reduce the computational complexity of optimal exhaustive search algorithm with
comparable performance. These algorithms remove the worst performance antenna
greedily and results best set of antennas and users at the same time.
However, all algorithms implement a semi-orthogonal user scheduling scheme to schedule

users and a zero forcing precoding schemes to mitigate inter-user interferences. A semi-
orthogonal user scheduling scheme generate a high computational complexity for massive
MIMO systems. Therefore, in this work with the objective of reducing the complexity of
SUS we implement two low complexity user scheduling in JASUS algorithm. These are
norm based user scheduling and random user scheduling. Also we apply MMSE and MRT
precoding techniques along with user scheduling schemes to examine their performance
when implemented in JASUS algorithm. Spectral efficiency performance evaluations and
computational complexity analysis of JASUS algorithm is done for each combinations of
user scheduling and precoding techniques. Compared with the originally implemented
SUS user scheduling technique, NUS shows a negligible spectral efficiency improvement
with a reduced complexity in all precoding types. Random user scheduling shows around
3 bits/s/Hz performance decrements with much less computational complexity as com-
pared to SUS in JASUS algorithm. Also performance comparison of precoding techniques
showed that MMSE results best spectral efficiency performance among the precoding
types in a system with a JASUS algorithm in most considered scenarios.
Key Words: Ma-MIMO, JASUS, User scheduling, Precoding, Computational

complexitry
Acknow ledgemen
t
First and foremost, praises and thanks to the God, the Almighty, for His showers of
blessings throughout my research work to complete the research successfully.
I would like to express my deep and sincere gratitude to my research advisor, Dr. Fikrese-
lam G. for providing invaluable guidance throughout this research. His dynamism, vision,
sincerity and motivation have deeply inspired me. He has taught me the methodology to
carry out the research and to present the research works as clearly as possible. It was a
great privilege and honor to work and study under his guidance. I am extremely grateful
for what he has offered me. I would also like to thank him for his empathy and great
sense of humor.
A heart-felt thanks to Mr. Amare K. (PHD candidate) for his guidance and inspiration
in the topic of this thesis during the early stage of my research work.
I am extremely grateful to my parents for their love, caring and sacrifices for educating
and preparing me for my future.
Finally, my thanks go to all the people who have supported me to complete the research
work directly or indirectly.
iv
Contents
Declaration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ii
Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iii
Acknowledgements . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . iv
List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viii
List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . x
Abbreviations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xi
Symbols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiii
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1.1 Large Scale (Massive) MIMO . . . . . . . . . . . . . . . . . . . . 4
1.2 Statement Of The Problem . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.3 Objective . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.3.1 General Objective . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.3.2 Specific Objective . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.4 Research Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.5 Thesis Contribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.6 Scope Of The Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.7 Thesis Organization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2 Literature Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3 System And Channel Model . . . . . . . . . . . . . . . . . . . . . . . . . . 23

3.1 System Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.2 Propagation Channel Model . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.3 Transmit And Received Signal Model . . . . . . . . . . . . . . . . . . . . 28
3.3.1 Transmit signal processing . . . . . . . . . . . . . . . . . . . . . . 28
3.3.2 The received signal model . . . . . . . . . . . . . . . . . . . . . . 29
3.4 Mathematical Model Of Linear precoding Techniques . . . . . . . . . . . 31
v
Contents vi
3.4.1 Zero-Forcing Beam Forming (ZFBF) . . . . . . . . . . . . . . . . 32

3.4.1.1 Water filling Power Allocation . . . . . . . . . . . . . . . 33
3.4.2 MMSE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.4.3 Maximum Ratio Transmission (MRT) . . . . . . . . . . . . . . . . 36
4 Joint Antenna Selection And User Scheduling . . . . . . . . . . . . . . 38

4.1 Joint Antenna And User Scheduling Algorithm and The problem Formu-
lation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
4.1.1 Problem formulation . . . . . . . . . . . . . . . . . . . . . . . . . 39
4.1.2 Joint antenna and user scheduling algorithm . . . . . . . . . . . 40
4.2 User Scheduling Schemes . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
4.2.1 Norm Based User Scheduling . . . . . . . . . . . . . . . . . . . . 42
4.2.2 Semi-orthogonal User Scheduling . . . . . . . . . . . . . . . . . . 43
4.2.3 Random User Scheduling . . . . . . . . . . . . . . . . . . . . . . . 44
4.3 Computational Complexity . . . . . . . . . . . . . . . . . . . . . . . . . . 44
4.3.1 Computaional complexity of user scheduling algorithms . . . . . . 45
4.3.2 Computational complexity of linear precoding techniques . . . . . 46
4.4 Computational complexity of JASUS algorithm . . . . . . . . . . . . . . 47
5 Simulation Results And Discussions . . . . . . . . . . . . . . . . . . . . . 48

5.1 Simulation setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
5.2 Simulation Result . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
5.2.1 Spectral efficiency performance evaluation at different values of SNR 51
5.2.2 Spectral efficiency performance evaluation for variable number of
antennas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
5.2.3 Spectral efficiency evaluations as a function of the number of se-
lected antennas . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
5.2.4 Number of available users versus spectral efficiency . . . . . . . . 61
5.3 Computational Complexity Numerical Evaluation . . . . . . . . . . . . . 64
5.3.1 Computational complexity evaluation of JASUS algorithm . . . . 65
5.3.1.1 Computational complexity of JASUS for the variations of
antenna number . . . . . . . . . . . . . . . . . . . . . . 65
5.3.1.2 Computational complexity of JASUS for the variations of
Scheduled number of users(N=K) . . . . . . . . . . . . . 66
5.3.2 Computational complexity comparison of NUS and SUS . . . . . 68
6 Conclusion and Future Work . . . . . . . . . . . . . . . . . . . . . . . . . 71

6.1 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
Contents vii
6.2 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
List of Figures
Figure 1.1 Block diagram of a multiple-input multiple-output system [1] . . 1

Figure 1.2 An illustration of a multi-user MIMO downlink [5] . . . . . . . . . 3
Figure 1.3 A schematic diagram of a massive MU-MIMO system equipped with
M transmit antennas serving K users [8] . . . . . . . . . . . . . . 5
Figure 1.4 Block diagram that shows the applied user scheduling and precod-
ing techniques in the JASUS algorithm with inputs and outputs . 10
Figure 1.5 Flow chart of JASUS algorithm . . . . . . . . . . . . . . . . . . . 11
Figure 1.6 Block diagram that shows flow of works . . . . . . . . . . . . . . . 12
Figure 3.1 System model of a single cell MU-MIMO system that selects N an-
tennas from M base station antennas and at the same time sched-
ules K ≤ N users from X total users . . . . . . . . . . . . . . . . 24
Figure 3.2 Multi-user MIMO system where K users are served by the M-
antenna BS in the same time-frequency resource [10]. . . . . . . . 25
Figure 3.3 Channel state information characterization [27] . . . . . . . . . . 26
Figure 3.4 Block diagram of the linear precoders at the BS [10] . . . . . . . . 28
Figure 5.1 Block diagram of simulation process . . . . . . . . . . . . . . . . . 49

Figure 5.2 Spectral efficiency performance comparison of user scheduling schemes
in JASUS algorithm using different precoding type for the case
M = X M = 10, X = 10,N = 3 and K = 3 . . . . . . . . . . . . . 52
Figure 5.3 Spectral efficiency performance comparison of Precoding types us-
ing different user scheduling schemes in JASUS algorithm for the
case M = X M = 10, X = 10,N = 3 and K = 3 . . . . . . . . . . 53
in JASUS algorithm using different precoding types for the case
M > X M = 15, X = 10,N = 3 and K = 3 . . . . . . . . . . . . . 55
Figure 5.5 Spectral efficiency performance comparison of user scheduling schemes in
JASUS algorithm when the available BS antenna is varied using
different precoding types, X = 10, N = 4, K = 4 and SNR = 10 dB 56
viii
List of Figures ix
Figure 5.6 Spectral efficiency performance comparison of precodings techniques

when the availlable BS antenna is varied using different scheduling,
X = 10, N = 4, K = 4 and SNR = 10 dB . . . . . . . . . . . . . 58
in JASUS algorithm with different number of selected antenna un-
der different precoding. M = 14, X = 14 and SNR = 10 dB . . . 59
Figure 5.8 Spectral efficiency performance comparison of precoding techniques
in JASUS algorithm with different number of selected antenna . M
= 14, X = 14 and SNR = 10 dB . . . . . . . . . . . . . . . . . . 60
Figure 5.9 Spectral efficiency performance comparison of user scheduling schemes in
JASUS algorithm with variable number of users with different
precoding M = 16, N=K=4 and SNR = 10 dB . . . . . . . . . . 62
Figure 5.10 Spectral efficiency performance comparison of precoding techniques when
the number of users is varied using different user scheduling techniques
M = 16, N=K=4 and SNR = 10 dB . . . . . . . . . . 63
Figure 5.11 Asymptotic computational complexity of JASUS algorithm with
different user scheduling and channel inversion precoding(MMSE
and ZF). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
different user scheduling and MRT precoding . . . . . . . . . . . . 67
different user scheduling and channel inversion precoding . . . . . 68
different user scheduling and MRT precoding . . . . . . . . . . . . 69
Figure 5.15 Asymptotic computational complexity of NUS and SUS without
JASUS algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
List of Tables
Table 4.1 Simplified computational complexity of user scheduling schemes . . 46

Table 4.2 Simplified computational complexity of precoding techniques . . . 46
Table 4.3 Approximate computational complexity of JASUS algorithm under
different user scheduling and precoding implementation . . . . . . 47
Table 5.1 System design assumption and considered parameter values . . . . 50

Table 5.2 Spectral efficiency numerical values at specified values of SNR for
the case M = X . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
Table 5.3 Spectral efficiency numerical values at specified values of SNR for
the case M > X . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
Table 5.4 Numerical values of spectral efficiency at specified number of total
antennas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
Table 5.5 Numerical values of spectral efficiency at specified number of total
antennas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
Table 5.6 Spectral efficiency values of JASUS at specified number of users for
implemented precoding and user scheduling schemes . . . . . . . . 64
x
Abbreviations
AWGN Additive White Gaussian Noise

BFS Brute Force Search
BS Base Station
CBSM Correlation Based Stochastic Model
CSI Channel State Information
CSIT Channel State Information at Transmitter
DL Down Link
DPC Dirty Paper Coding
EE Energy Efficiency
EESACUS Energy Efficient Successive Antenna Cancellation and
User Selection
FLOP Floating Point Operation
I.I.D Independent and Identical Distrbution
JASUS Joint Antenna Selection and User Scheduling
Ma-MIMO Massive Multiple Input Multiple Output
MATLAB Matrix Labratory
MF Matched Filter
MIMO Multiple Input Multiple Output
MISO Multiple Input Single Output
MMSE Minimum Mean Square Error
MRC Maximum Ratio Combinning
MRT Maximum Ratio Transmission
MU-MIMO Multi User Multiple Input Multiple Output
NUS Norm Based User Scheduling
RF Radio Frequency
RUS Random User Scheduling
xi
Abbreviations xii
RZF Regularized Zero Forcing

SE Spectral Efficiency
SINR Signal Interference Noise Ratio
SISO Single Input Single Output
SNR Signal Noise Ratio
SUS Semi-orthogonal User Scheduling
TDD Time Division Duplexing
TDMA Time Division Multiple Access
UE User Equipment
UL Up Link
WF Water Filling
ZF Zero Forcing
ZFBF Zero Forcing Beam Forming
Symbols
|.| Absolute value (cardinality of a set)

||.||2 2-norm
C An element of complex values
R An element of real values
β Large scale coefficient
σ Noise variance
N Normal distribution
tr(.) Trace operator
max {.} Maximization
−1
(.) Inverse operator
T
(.) Transpose operator
(.)H Hermatian operator
(.)∗ Conjugate operator
γk Effective channel gain of a user
µ Water level in waterfilling power allocation
π(.) User index
A A set of antennas
U A set of users
m Antenna index
k User index
S A set that holds selected users’ index
∅ Empty set
M Number of total antennas at BS
X Number of total users
K Number of Scheduled users
N Number of RF chains (Number of selected antennas)
xiii
Symbols xiv
iopt Index of selected antenna

T Computational complexity of user scheduling schemes
R Computational complexity of precoding techniques
Chapter 1
Introduction
1.1 Background
MIMO (multiple input multiple output) is a wireless communication technology that im-
plement multiple antennas at the transmitter and receiver. Multiple data streams which
are different or copies of the same signal are sent through multiple transmit antennas
where they are received through multiple or single antennas. Multiple antennas can be
available on either of the two, the transmitter or the receiver, as well as on both of the
two [1].
Antenna 1
Antenna 1
Antenna 2
Receiver
Transmitter Antenna 2
Antenna N r
Antenna N t
Figure 1.1: Block diagram of a multiple-input multiple-output system [1]
Figure 1.1 depicts a MIMO systems with Nt antennas at the transmitter and Nr antennas
at the receiver. At the transmitter, the data stream enters an encoder, whose outputs
are forwarded to Nt transmit antennas. From the antennas, the signal is sent
through
1
Chapter 1. Introduction 2
the wireless propagation channel, and received by N r antennas at the receiver. Each htr
represent channel fading coefficient from the t transmit antenna to r receive antenna[1]
. In MIMO systems there are two mechanisms for utilizing multiple antennas to
improve wireless system performance [2].These are
1. Spatial multiplexing
In spatial multiplexing an original high-rate data stream is multiplexed into several

parallel streams, each of which is sent from one transmit antenna element. The
channel “mixes up” these data streams, so that each of the receive antenna elements
sees a combination of them. If the channel is well behaved, the received signals
represent linearly independent combinations. This enables the MIMO systems to
have a gain from spatial multiplexing. The multiplexing gain results from the fact
that a MIMO channel can be decomposed into a number R of parallel independent
channels. By multiplexing independent data onto these independent channels, we
get an R-fold increase in data rate in comparison to a system with just one antenna
at the transmitter and receiver. This increased data rate is called the multiplexing
gain [1][2].
2. Spatial diversity
The same data stream is transmitted or received through multiple antennas to

increase the link reliability. If the antennas are placed sufficiently far apart, the
channel gains between different antenna pairs fade more or less independently,
and independent signal paths are created. Thus link reliability is increased by
selecting the best faded signal or by combining the independent signals [3].
Therefore, MIMO technology offers significant increment in data throughput and link
range without an additional increase in bandwidth or transmit power due to the multi-
plexing and diversity gain. Some of the technologies which rely on MIMO systems are
IEEE 802.11, Third Generation (3G) and Long Term Evolution (LTE) ones.
A single user mimo (point to point) is traditional way of using multiple antennas for
communication in which the base station communicate with a single user at a time. If
we consider a single user time division multiple acces (TDMA) MIMO communication
with M antennas at the base station and N antennas at each receiver, the capacity gain
is roughly min {M, N } times the capacity of single input single out put system (SISO).
Since the number of receive antennas are smaller than the number of transmit antennas,
it is clear that min {M, N } is equal to N in most situations. Therefore, the capacity
gain may not be significant in single user MIMO system. To solve this problem the
trend of single user MIMO communication is shifted to multi-user MIMO (MU-MIMO)
communication. In MU-MIMO communication the base station (BS) simultaneously
communicates with multiple user equipments (UEs) at the same bandwidth. It is due to
the fact that multiple users could simultaneously communicate over the same spectrum
improves the system performance [4]. MU-MIMO system is implemented in both uplink
and downlink communication. In the uplink a group of users transmits to the same
base station whereas in down link a base station with multiple antennas broadcast data
streams to multiple users at the same frequency and time. The UEs are equipped with
one or more antennas.
Figure 1.2: An illustration of a multi-user MIMO downlink [5]
Since users in MU-MIMO system communicate with the same bandwidth, each user re-
ceives both its own signal and other users signal as interference. So that, MU-MIMO
communications is limited by inter user or co-channel interference which greatly reduce
the capacity of the system. A down link MU-MIMO system is depicted in figure 1.2 where
the base station attempts to transmit over the same channel to two users, but there is
some inter-user interference for user 1 generated by the signal transmitted to user 2 and
vice verse. In downlink scenario to mitigate the interference a nonlinear coding technique
called dirty paper coding(DPC) is an optimal strategy. DPC is a multi-user encoding
strategy based on interference pre-subtraction. Although, it is an optimal strategy spe-
cially when the number of users increase, it is difficult to implement in practical systems
due to its high computational complexity [4]. To reduce the computational complexity
of DPC a sub-optimal linear pre-processing techniques commonly known as precoding or
beam forming is applied on the transmitted signal. Precoding strategies include a chan-
nel inversion variants and a simple conjugate of a channel are common in MU-MIMO
system. The knowledge of channel state information at the transmitter is required to
handle the interferences. In the downlink, in the absence of channel state information at
the transmitter (CSIT), user multiplexing is generally not possible, as the BS just does
not know in which direction to form spatial beams [4][5].
MU-MIMO system also exploits the benefits of multi-user diversity which increase the
system performance. Multi-user diversity takes advantage of the fact that, in a system
with many users whose channels fade independently, at any given time some users will
have better channels than others. By transmitting only to users with the best channels
at any given time, system resources are allocated to the users that can best exploit them,
which leads to improved system capacity. When the number of users is large as compared
to the number of base station antennas, the base station can schedule its transmission to
those users with favorable channel fading conditions to improve the system throughput.
Scheduling transmissions to users based on their channel conditions coupled with power
control, can significantly increase both uplink and downlink throughput as measured by
sum-rate capacity [2][4].
1.1.1 Large Scale (Massive)

MIMO
Massive MU-MIMO is a system where a BS equipped with many antennas that serve
several users simultaneously in the same frequency band. It is also known as large-scale
antenna systems and promises significant gains in terms of spectral efficiency (SE) and
energy efficiency (EE) to accommodate a large number of users at extraordinarily high
data rates with better reliability while consuming much less power. The BS with mas-
sive MIMO (Ma-MIMO) sends independent data streams to multiple UEs in the same
time-frequency resource, hence achieving ultra-high SE. Ma-MIMO is one of the exciting
technologies enabling the fifth generation (5G) and beyond fifth generation (B5G) wire-
less cellular networks [6][7].
Both theoretical and measurement results indicate that large scale MIMO (LS-MIMO) is
capable of significantly improving spectral efficiency. Also the use of large excess antenna
Figure 1.3: A schematic diagram of a massive MU-MIMO system equipped with M trans-
mit antennas serving K users [8]
at the base station compared to the available users allows to use simple linear precoding
at the transmitter. LS-MIMO can improve band width efficiency since it can achieve
large multiplexing gains when serving tens of UEs simultaneously. Multiple antenna ar-
rays focuses energy into narrow beams around the desired user which leads to energy
efficiency. Massive MIMO relies on spatial multiplexing, which in turn relies on the base
station having good enough channel knowledge, on both the uplink and the downlink [9].
• Massive MIMO can increase energy efficiency significantly. It is due to the fact that
the use of more antennas helps to focus energy with an extremely narrow beam on
small regions where the UEs are located [9].
• In Massive MIMO, simple low-complexity signal processing algorithms are capable

of approximating the performance achieved by optimal methods, such as Maximum-
Likelihood (ML) multi-user detection and Dirty Paper Coding (DPC). Linear pro-
cessing (linear combing schemes in the uplink and linear precoding schemes in the
downlink) is nearly optimal [9][10].
• Massive MIMO can increase the capacity 10 times or more. The capacity increase
results from the fact that massive MIMO can achieve large multiplexing gains when
serving tens of UEs simultaneously [10] [11].
However, the benefits of massive MIMO come with at the cost of some price. Using
large number of transmitter antennas at the base stations requires radio frequency (RF)
equipment for individual antennas. RF chains are made of different components
which contributes to the large consumption of power, hardware complexity and
increased cost of deployment. Antennas are usually cheap and easy to deploy, but RF
chains can be relatively expensive. Due to large array gains, massive MIMO is energy
efficient in terms of radiated transmit-power. However, energy consumption in hardware
can be quite high [12]. Therefore, when there are a large number of transmit antennas but
a limited number of RF chains, the system performance can be greatly improved by
selecting a subset of transmit antennas with good channel conditions.
When antenna selection is performed in massive MIMO systems the number of users to
be served is bounded by the number of selected antennas which is also determined by the
available number of RF chains. For this situation user selection must be performed to
find a set of users from the total number of users that achieve maximum performance.
The maximum number of users to be scheduled is limited by the number of RF chains
[13].
To alleviate these problem and optimize the performance of massive MIMO systems re-
searches have been conducted on joint antenna selection and user scheduling algorithms.
Different algorithms are proposed based on antenna selection and user scheduling to find
best performance on the selected performance metrics. A variants of greedy based joint
antenna selection and user scheduling algorithms which schedules antennas and users
jointly with the aim of maximizing the sum-rate performance of a massive MIMO system
is proposed in a few works with semi-orthogonal user selection and zero forcing precoding
techniques. We will present the review of these works in chapter 2 which is a literature
review part of this work.
In this thesis the performance of a greedy based JASUS algorithm is evaluated and com-
pared by implementing different user selection criterion and precoding techniques. Since
each user selection criteria and precoding techniques has its own performance advantage
and disadvantage depending on the performance metrics, it is important to incorporate
them in joint antenna selection and user scheduling algorithm and compare their perfor-
mance. This research work aims to implement low complexity user selection algorithms
into a greedy based sumrate maximizing joint antenna and user scheduling algorithm
which originally used a higher complexity semi-orthogonal user selection as user schedul-
ing technique. We scheduled users using norm based user selection (NUS) and random
user selection(RUS) which have lower computational complexity as compared to semi-
orthogonal user scheduling. Also in addition to zero forcing (ZF) precoding we implement
minimum mean square error (MMSE) and maximum ratio transmission (MRT) precoding.
Spectral efficiency and computational complexity of the algorithm will be the comparative
metrics. Different scenarios are considered for the evaluation of performances. Matlab
software is used for the simulation purpose.
1.2 Statement Of The Problem
A brute force search (BFS) algorithm which selects the best set of antennas and users from
all possible sets of antennas and users guarantee optimum performance. But it can’t be
used in practical scenario because of its high computational complexity. A greedy based
JASUS algorithm are proposed to reduce the computational complexity of BFS with a
close SE performance [13].
Although a few JASUS algorithms that select antennas greedily are proposed in a single
cell massive MIMO systems all algorithms implement SUS for user scheduling part which
contributes a higher computational complexity for massive MIMO systems. SUS has a
complexity of O(M 3 K) [14] which becomes computationally high for massive MIMO sys-
tems. The JASUS algorithm that implement this scheduling scheme will generate higher
computational complexity. SUS perform user selection task based on user separability
and maximum channel norm which are the reason for its higher computational
complexity[15].
At low SNR and limited number of RF chains MMSE is found to have better throughput
performance than ZF when a greedy antennna selection is performed [16]. MRT pre-
coding have a simplest computational complexity than ZF and MMSE. In spite of those
advantage of precoding techniques regarding to computational complexity and capacity
performance, all proposed JASUS algorithms implement only ZF precoding.
In line with these facts, in this work the performance of a greedy based JASUS algorithm is
evaluated and compared by implementing two low complexity user scheduling techniques.
These are norm based user scheduling which select users based on largest channel gain
only and a simple random user scheduling which select users randomly [17]. In addition to
ZF precoding, linear precoding techniques such as MRT and MMSE will be implemented
combined with user scheduling schemes in JASUS algorithm and their performance will
be evaluated.
1.3 Objective
1.3.1 General
Objective
The general objective of this work is to evaluate and compare the performance of greedy
based JASUS algorithm in a single cell massive MIMO system using low complexity user
selection methods and different linear precoding types.
1.3.2 Specific
Objective
• Implementing three user selection criterion in JASUS algorithm. These are norm
based user selection, random user selection and the originally implemented semi-
orthogonal user selection.
• Implementing three linear precoding types in the JASUS algorithm. These are ZF,
MMSE and MRT precoding techniques.
• Evaluating the JASUS algorithm SE performance for each combination of precoding

and user selection techniques for different scenarios using MATLAB as a simulation
tool
• Evaluating the JASUS algorithm computational complexity for each combination

of precoding and user selection techniques
• Comparing the performance and draw conclusion
1.4 Research Methodology
The greedy based JASUS algorithm attain a comparable SE performance with the op-
timal exhaustive search algorithm with lower computational complexity. Originally all
greedy based algorithms implement SUS user scheduling scheme which generate high
computational complexity for Ma-MIMO systems. Also ZF precoding is the only con-
sidered precoding technique in the previous literatures. But in this work we apply lower
complexity user scheduling schemes (NUS and RUS) in the greedy based algorithm. We
also evaluate and compare the performance of the three precoding techniques combining
with user scheduling schemes. Figure 1.4 shows how we apply these user scheduling and
precoding techniques in the JASUS algorithm. At a given time one user scheduling tech-
nique is used to carry out the user selection task and one precoding technique is applied
to calculate the sum-rates in a JASUS algorithm. The yellow highlighted items in this
figure shows the user scheduling and precoding techniques that we apply in the algorithm.
The JASUS algorithm use the channel matrix H as input and it use the rows and columns
as a set of antennas and users respectively.
Figure 1.4: Block diagram that shows the applied user scheduling and precoding tech-
niques in the JASUS algorithm with inputs and outputs
The flow chart in figure 1.5 shows how JASUS algorithm works. The JASUS algorithm
selects a subset of antennas greedily with the help of inner and outer loops. The blocks
in blue are sections of the algorithm where we apply modifications. At each inner loop
iteration the JASUS algorithm removes one antenna from the antenna set A.
Figure 1.5: Flow chart of JASUS algorithm
Then a user is selected from a user set U by using antenna set deprived of one antenna.
The user selection is done using one of the three user selection criterion. By applying
one of the three precoding technique, sum-rate is calculated using the antenna set de-
prived of one antenna and the selected user. The outer loop of the JASUS algorithm
removes an antenna whose removal maximize the sumrate of the system. The outer loop
is executed until the number of antennas becomes equal to the number of RF chains N .
Finally, the algorithm returns the index of the selected antennas and users. From the
simulation point of view, the SE that can be achieved by the selected set of antennas
and users is calculated for each combination of user scheduling schemes and precoding
techniques. Also the computational complexity of the JASUS algorithm resulting from
the implementation of precoding techniques and user scheduling technique is evaluated.
Reviewing the literatures on the recent and background works helps us to find the research
gap on the on the area of antenna selection and user scheduling in massive MIMO systems.
Also we reviewed literatures to grasp fundamental concepts on the principles of multi-user
MIMO communication, precoding techniques and massive mimo.
The rest part of the steps we followed to accomplish this work can be classified into four
parts as shown in figure 1.6.
System Design
Specifications
• Determining the channel model and signal model
• Defining the precoding techniques and the respective power allocation strategies
• Formulating the SINR that results from using each precoding techniques
• Developing the capacity of MU-MIMO for each precoding case
• Determining the computational complexity of each precoding type
Analysis Of
Algorithms
• Analysis of the mathematical problem formulation of JASUS
• Modifying the problem formulation based on the precoding types we used
• Analysis of the user selection techniques
• Determining the computational complexity of each user selection techniques
• Implementing the combination of each user selection and precoding techniques into
the JASUS algorithm
• Determining the computational complexity that arise from JASUS algorithm by

implementing each combination of precoding and user selection techniques
Figure 1.6: Block diagram that shows flow of works
Simulation
• Identifying the variable and fixed parameters for a given simulation scenario
• Identifying which precoding technique and user selection scheme is applied at a

given time
• Generating channel matrix (channel realization) based on the channel model
• Applying JASUS algorithm using the generated single channel realization. A given
precoding and user selection technique are used based on a given simulation scenario.
• The system capacity as a sum-rate is calculated using the selected antennas and
users which are the output of JASUS algorithm.
• Averaging the sum-rates that result from all channel realization
Result comparison and interpretation
• By fixing a precoding technique, SE that results from using different user selection
criterion into the JASUS is evaluated and performance comparision is done
• By fixing a user selection technique, SE that results from using different precoding
techniques into the JASUS algorithm is evaluated and performance comparison is
done
• Asymptotic computational complexity comparison of JASUS algorithm for different

combination of user selection and precoding technique is done
1.5 Thesis Contribution
The contribution of this work can be summarized as follows
• When algorithms are formulated to maximize the SE in massive MIMO systems

the level of computational complexity it generate should be considered. So that, in
formulation of algorithms there exist a trade-off between SE performance and com-
putational complexity. Stressing with this trade off, in this work we implement two
low complexity user scheduling schemes and two precoding techniques with different
SE performance in JASUS algorithm. We examine the SE performance achieved
by each combination of user scheduling schemes and precoding techniques. The
computational complexity generated by JASUS algorithm in implementing these
combinations is evaluated and compared with the previous works that implement
only SUS-ZF. Norm based user scheduling shows better spectral efficiency perfor-
mance than SUS with a reduced complexity for all precoding techniques. With a
scarification of some SE values RUS makes the complexity of JASUS simple.
• It is clear that in massive MIMO systems MMSE or MRT precoding techniques can
be applied to reduce the effect of inter-user interference. Due to this, we formulate
JASUS algorithm for systems that use MRT or MMSE precoding techniques by
modifying the original JASUS that apply only SUS-ZF. Then performance compar-
ison is done with the original algorithm that use SUS-ZF.
• Since JASUS algorithm maximize SE by exploiting the benefits of spatial selec-

tivity and multi-user diversity, in this work we showed that which user
scheduling and precoding technique is more benefited from spatial selectivity and
multi-user diversity when applied in the JASUS algorithm.
• Generally, this work is more comprehensive than the existing literatures on the area
of JASUS in testing and showing the performance of different user scheduling and
precoding techniques. For one who considers trade off between SE and computa-
tional complexity for the greedy JASUS algorithm, the results in this work can be
used as a reference tool to select the appropriate user scheduling and precoding
techniques.
1.6 Scope Of The Study
The over all system depends on the availability of perfect channel state information at the
base station. This means we assume that the base station knows the channel perfectly
and no channel estimation technique is employed. In this thesis we are constrained with a
single cell centralized system where all the antennas are located on a single base station.
With the assumption of i.i.d Rayleigh fading channel no practical antenna configuration
is considered in this thesis. Performance evaluations are done using software simulations.
1.7 Thesis Organization
The rest of this document is organized as follows
• Chapter 2 After clearly discussing about massive MIMO benefits and the chal-
lenges it faced, we present the background and recent works related to antenna
selection, user scheduling as well as joint antenna selection and user scheduling
which aimed to solve the short coming of massive MIMO.
• Chapter 3 The system, channel and signal model is presented. Also in this chapter
the linear precoding techniques are described both mathematically and theoretically.
• Chapter 4 The greedy based JASUS algorithm with its problem formulation is
justified. Modification on the problem formulation is made to meet the objective
of this work. Three user scheduling schemes are also presented in the form of
pseduo-code. Finally we describe the mathematical expressions of computational
complexity generated by JASUS algorithm when different user scheduling and pre-
coding techniques are applied.
• Chapter 5 First we provide procedures which describes how numerical simulation

is performed. Then simulation results which shows the SE of JASUS with different
user scheduling and precoding techniques under various scenarios are presented.
Discussions and interpretation are made on each output of SE evaluations. Finally

plots which shows how the computational complexity of JASUS varies with different
parameters are provided.
• Chapter 6 This chapter consists the overall summary and conclusion of this the-
sis. Also we put forward recommendations which leads future work and further
investigation on this work.
Chapter 2
Literature Review
In a condition where there is large number of antennas at the base station using all the
available antennas will results hardware complexity, energy inefficiency and high cost of
deployment. Antenna selection is one of the proposed solution to solve this difficulty
in large scale MIMO systems. And at the same time the number of selected antennas
limit the number of users to be served which forces the base station to schedule a portion
of users from the available. To alleviate these problems efforts are done to provide an
algorithm that selects antennas at the base station and schedule the optimum number of
users jointly, so that the system operates at optimum performance.
In this chapter we provide the literature review of some of the recent and back ground
works on on antenna selection, user scheduling and JASUS algorithms.
In [4] A. Goldsmith et al. proposed a semi-orthogonal user scheduling algorithm for MU-
MIMO systems and its performance is studied in combination with ZF precoding. The
authors studied the asymptotic optimality of zero forcing beam forming as the number of
users goes to infinity in MU-MIMO system. In this paper the proposed semi-orthogonal
user scheduling scheme combined with zero-forcing beam forming showed a close per-
formance with the optimal DPC strategy as the number of users goes to infinity. The
proposed SUS algorithm is lower in complexity as compared with DPC. The authors
reasoned out the asymptotic optimality of ZFBF-SUS with large number of users. They
stated that with a large number of users the transmitter can choose user channels that
are nearly orthogonal to one another.
16
Chapter 2.Literature 17
Although the proposed user scheduling algorithm combined with ZF beam forming in [4] is
asymptotically optimal and lower in complexity than DPC, its computational complexity
for Ma-MIMO systems is high.
In [17] the average beam forming performance of the four user selection algorithms are
compared for MIMO broadcast channel. These user selections are semi-orthogonal user
selection, angle based user selection , norm based user selection and random user selection.
The authors studied the properties of four user selection algorithms in conjunction with
beam-forming that guarantee certain SINR requirements under transmit power minimiza-
tion. Based on Performance evaluations NUS showed asymptotically close performance
with the optimal exhaustive search and SUS when the number of antennas increase. RUS
showed the lowest performance. Xi Zhang et al. concluded that in scenarios with large
number of transmit antennas, the simple NUS scheme is sufficient in terms of average
transmit power needed to support certain SINR requirements.
The results found in [17] is important for massive MIMO systems where the number
of antennas are large. The simplicity of NUS and its close optimal performance as the
number of antennas increases is a good feature for massive MIMO systems. But in this
paper the sum-rate performance of user scheduling scheme is not evaluated. Also the
performance evaluation is done without antenna selection.
In [16] [18, 19] the authors works on the performance of transmit antenna selection. In
[16] Pu-hsuan et al. investigates the performance of transmit antenna selection with ZF
and MMSE precoding techniques. The authors showed that the computational complex-
ity of exhaustive search can be reduced by proposed antenna selection algorithm. They
also showed that when the SNR is low and the number of RF units is small, the sum
throughput of the MMSE precoder outperforms the sum throughput of the ZF precoder
whereas the performance gap decreases as the SNR value or the number of total RF
units increases. In [18] the authors analyze the capacity performance antenna selection
in real measured propagation channel. In their investigation Gao et al. showed that with
more available antennas than the RF chains, the antenna selection can greatly improve
the system performance, by exploiting the spatial selectivity in the propagation channels,
with more available antennas than the RF chains. In [19] Meroun et al. derived a good
approximation of the distribution of the mutual information with selecting any number
of antennas for Ma-MIMO systems. They also modeled the power consumption as the
addition of the transmit power and the circuit power consumption and proposed antenna
selection algorithm. The authors in this paper concluded that antenna selection can im-
prove the energy efficiency in large scale multiple antenna system.
These works in [18]-[19] proved that antenna selection can greatly improve the spectral
and energy efficiency when there is more available antennas than RF units at the base
stations. In [16] the MMSE precoder with antenna selection showed better performance
than ZF precoder at low SNR in multi-user MIMO systems. However, these works doesn’t
consider the problem of user scheduling by assuming the number of user are lower than
the number of antennas.
In [20] the authors proposed a joint transmit antenna selection and user scheduling for
multi-user MIMO system using block diagonalization precoding scheme. The system
model is based on a downlink transmission scheme where both the base station and
the each user equipped with multiple antennas. The authors proposed a joint transmit
antenna selection and multi-user scheduling based on the criterion maximizing averaged
eigenvalues (MAE) or maximizing minimum eigen value (MME). They used two scenarios
to evaluate the performance of their algorithms.
The authors compared their proposed algorithm with one another they do not use the
upper bound reference mainly the exhaustive search. They also do not consider the com-
putational complexity of their algorithm which matters in massive MIMO systems.
In [21] the authors propose a joint antenna selection and user scheduling scheme for the
downlink distributed massive MIMO systems employing ZFBF. In this scheme, the joint
antenna selection and user scheduling is formulated as a sum-rate maximization problem
under backhaul capacity constraint. The back-haul capacity determines the number of
active antennas. Three iterative algorithms are proposed to solve the resulting joint an-
tenna selection and user scheduling problem. They are either based on greedy fashion,
Frobenius-norm criteria or their combination. In norm based JASUS, antenna selection
and user scheduling is done simply based on the F-norm criteria which would sacri-
fice sum rate capacity. To improve the sum-rate performance over norm-Based JASUS,
sub-optimal antenna selection and user scheduling scheme called Greedy-Based JASUS
algorithm is proposed. Greedy-Based JASUS maximize the sum rate capacity at each
step. An algorithm which is called TCB JASUS is also proposed to balance throughput
and complexity.
In [22] Liu et al. investigated the antenna selection and user scheduling problems in
massive MIMO non-orthogonal multiple access (NOMA) system. The authors proposed
an antenna selection and user scheduling algorithms for two scenarios. The first is a
search algorithm for antenna selection in a simple single-band two-user scenario. In this
scenario the desired antennas are searched from limited candidate antennas with high
channel gain which are beneficial to the relevant users. The second proposed algorithm
is a joint AU contribution algorithm for joint antenna selection and user scheduling in
multi-band multi-user scenario. This algorithm jointly takes account of the contribution
of each antenna’s and user’s channel gain to total channel gain.
However in [21] the authors considers a distributed massive MIMO systems where the
base station antennas are scattered throughout the cell and for the problem formulation
they use backhaul capacity constraint. And in [22] the paper is on massive MIMO-NOMA
transmission scheme where a successive interference scheme is implemented to solve inter
user interference problem. But our focus is on a centralized massive MIMO systems for a
single cell systems which employ linear precoding techniques to deal with interferences.
M.Olyaee et al. in [23] developed a joint antenna and user selection algorithm for single
cell downlink multi-user massive MIMO system with an aim of maximizing energy effi-
ciency. The authors developed an algorithm which they called energy-efficient successive
antenna cancellation and user selection (EESACUS) that successively removes antenna
that have lesser contribution to sum-rate and selecting the best subset of users for the
remaining antennas, the proposed algorithm tries to maximize EE. In this paper the pro-
posed algorithm energy efficiency is evaluated using different scenarios. Using the number
of users as a scenario the authors showed that their proposed algorithm results almost
the same EE performance with exhaustive search algorithm .
The authors implement SUS user scheduling and did not provide the complexity and
spectral efficiency performance of their algorithm.
In [13] Benmimoune et al. proposed a joint transmit antenna selection and user schedul-
ing algorithm (JASUS) for single cell massive MIMO systems. The proposed algorithm is
a jointly strategy which performs antenna selection and user scheduling. This algorithm
competently resolves all difficulties of combining antenna selection and user scheduling
in a single cell multi-user massive MIMO system. A single base station with multiple
antennas that broadcast data to multi-users (one antenna per user) scenario is consid-
ered. The authors used zero-forcing beam forming as a precoding scheme. With the
objective of maximizing the system sum rate the authors proposed a two-step JASUS
algorithm that select a set of antennas and at the same time schedule users. The JASUS
algorithm successively eliminates an antenna and user that have a minimum contribution
to the system capacity performance. The users are scheduled based on semi-orthogonal
user scheduling scheme that tries to find the best possible orthogonal set of users. The
performance of this algorithm is evaluated using different scenarios. The authors used an
optimal brute force search (BFS) algorithm for comparison with the proposed method.
The numerical results show that the proposed joint antenna selection and user schedul-
ing algorithm performs close to the optimal algorithm. The paper discusses that the
complexity of the algorithm comes from the channel inversion of ZF precoding during the
sum-rate calculation at each iteration. Based on this assumption, the proposed algorithm
results lower complexity than BFS algorithm.
Even though this algorithm is good in reaching the capacity attained by the optimal BFS
algorithm in all scenarios, the SUS user scheduling algorithm combined with greedy an-
tenna selection will have a higher complexity. The authors did not consider its complexity
during the evaluation of computational complexity.
In [24] with the aim of maximizing users sum-rate and reduce complexity, a joint antenna
selection and user scheduling algorithm in downlink multi-user MIMO systems with a zero
forcing precoding is proposed. The problem formulation is based on sum rate maximiza-
tion of a single cell multi-user system. The authors proposed a user scheduling scheme
based on matrix Gauss elimination method, Combined with maximum norm antenna
selection algorithm. A matrix Gauss elimination based user scheduling algorithm is pro-
posed to improve the capacity performance of SUS . The authors showed that compared
with exhaustive search the proposed user scheduling algorithm have lower complexity
and approximately equal capacity performance. Compared with SUS, the proposed user
scheduling algorithm showed higher complexity and better performance than SUS.
However the authors compare the performance of proposed user scheduling with SUS in
terms of computational complexity and spectral efficiency. They did not consider the
performance comparison of the JASUS algorthm. The results showed that their proposed
user scheduling scheme have higher computational complexity than SUS. The considered
scenarios are limited to the performance of spectral efficiency with variable number of
users and signal to noise ratio.
In [25] a joint antenna selection and user Scheduling for Massive MIMO System is pro-
posed. The authors used the uplink single cell system sum rate performance maximization
for problem formulation. A zero-forcing receiver is implemented at the base station to
detect user signal. The authors proposed a two-step JASUS algorithm which consist a sim-
plified downward branch and bound (SDBAB) method and a swapping based algorithm
to further increase the sum rate performance. In its first step, a simplified downward
branch and bound based JASUS is used to find a near-optimal antenna and user sets
whose channel matrix has the near-largest MSV. In its second step, a swapping-based al-
gorithm is proposed to find a better solution by swapping antennas and users between the
selected and the discarded. The user scheduling part is done based on semi-orthogonal
user selection criteria which is the same with [13]. The authors on their work justify the
complexity of the work in [13] and modify the antenna selection technique by setting a
trade off number that results lower complexity and better sum rate performance. Before
the greedy algorithm a set of candidate antennas will be selected based on the maximum
channel gain of antennas. The trade off number is in between the number of available
antennas and the number of selected antennas. The lower computational complexity of
this algorithm mainly comes from using a trade-off number which is selected before the
first step of algorithm. The authors showed that based on a given trade off number their
proposed algorithm is better in spectral efficiency than [13] by 0.5 b/s/Hz. Moreover the
proposed algorithm shows a lower complexity for a given trade off number.
But the authors doesn’t state implicitly or explicity how to decide the trade of number
which contributes for the low complexity of their algorithm. Also they consider max-
imization of uplink capacity performance using zero forcing receiver. The considered
scenarios are limited to the variations of SNR and antenna front ends. In this paper SUS
is implemented to scheduling users and the authors did not consider the computational
complexity this user scheduling scheme. Also the authors didn’t consider the computa-
tional complexity of the swapping operation.
The papers in [24]-[25] aims to improve the complexity and SE performance of JASUS
algorithm in [13]. But all algorithms implement the complex SUS for user scheduling and
zero-forcing for linear precoding. The performance of other precoding techniques other
than zero-forcing is not evaluated in these papers including [13].
In [17] NUS which schedules users based on largest channel magnitude shows asymp-
totically close performance with SUS and exhaustive search as the number of antennas
increases which is a desirable property in massive MIMO systems.
In line with the above facts and due to the complexity problem of SUS proposed in [4]
for massive MIMO systems we will implement the two low complexity user scheduling
schemes in JASUS algorithm proposed in [13]. These are norm based user scheduling
(NUS) and random user scheduling (RUS). Although random user scheduling showed
lowest performance in literatures, it have lowest computational complexity. Due to its
simple complexity and for the purpose of performance comparison purpose, we will eval-
uate its performance in the JASUS algorithm.
Also in [16] MMSE showed better performance with antenna selection in low SNR values
than ZF precoding.
Based on this we will implement MMSE precoding in the original JASUS algorithm pro-
posed in [13]. Also due to its simplicity we will evaluate the performance of MRT in
JASUS algorithm.
We select the algorithm proposed in [13] to evaluate and compare its performance with
low complexity user scheduling schemes and different precoding. Because the algorithm
is optimized for centralized massive MIMO systems in contrast to the algorithm in [21]
which is for distributed massive MIMO. It implement linear precoding for interference
cancellation rather than SIC which is applied for MIMO-NOMA systems in[22]. Although
the algorithm in [25] improved the SE performance of the algorithm proposed in [13]
by a fraction of SE values, its performance is evaluated with limited scenarios. The
algorithm in [13] is evaluated with diverse scenarios and its performance is compared
with the optimal BFS algorithm for each scenario. This reason motivated us to select
this algorithm and analyze its performance.
Chapter 3
System And Channel
Model
In this chapter we will first provide the system, channel and signal model. Then the
mathematical expression of signal to interference plus noise ratio and the broadcast sum-
rate of the system will be derived and at last the necessary mathematical expression for
all precoding types which are used for the problem formulation of JASUS algorithm will
be presented.
3.1 System Model
It is clear that multi-user massive MIMO systems has a great advantage in increasing
spatial multiplexing gain by sending several parallel streams of data at the same time
to tens of users through large number of antennas. However using all the antennas at
the same time results wastage of energy, increased system complexity and cost of deploy-
ment [26]. When there are a large number of transmit antennas but a limited number
of RF chains, the system performance can be greatly improved by selecting a subset of
transmit antennas with good channel conditions. On the other hand, the number of users
that can be simultaneously supported is limited by the number of transmit antennas. So
that, when antenna selection is performed at the base station user selection should be
done at the same time. Joint antenna selection and user scheduling algorithms alleviate
these problems by a joint strategy which performs antenna selection and at the same
time schedule users based on maximizing the required system performance metrics. In
the crowded sub-6 GHz bands sum-rate maximization is one of the goal of antenna selec-
tion and user scheduling algorithms with reduced computational complexity to achieve
23
Chapter 3. System and Channel 24
user 1 user 1 1
RF 1
RF chain C
p
s h User 1
user 2 user 2 r
w a
Joint e 2
c RF chain
i 2 n
antenna User 2
t n
selection o
& user c e
d
scheduling h l
e
user X user K r N
RF chain M
User K
N RF chains M BS
antennas
Activate N selected switches out of M
Figure 3.1: System model of a single cell MU-MIMO system that selects N antennas from
M base station antennas and at the same time schedules K ≤ N users from X total users
.
an improved spectral efficiency [13][15].
As depicted in figure 3.1 we consider a single cell multi-user MIMO system which consists
X total users each with a single antenna. A base station is equipped with M antennas
and N RF chains. A JASUS algorithm at the transmitter selects K(≤ N ) users and N
antennas with the objective of maximizing broadcast sum-rate. The selected N antennas
are then connected to the N RF chains through the RF switch [13][18]. In other words,
the transmitter chooses two sets A and U, defined as sets of selected transmit antennas
and scheduled users, respectively. The scheduled users’ data is then processed by the
precoder and transmitted by the N selected transmit antennas through the channel [13].
In the next section we will develop the propagation channel and signal model. This section
provides mathematical model of broadcast sum-rate which is used for problem formulation
of JASUS algorithm in chapter 4. Also the mathematical and theoretical model of the
three linear precoding techniques will be described in the subsequent sections.
3.2 Propagation Channel Model
In this work we consider a single cell MU-MIMO system, where a base station is equipped
with a large number of antennas and serves many terminals each with a single receive
antenna in the same time-frequency resource as depicted in figure 3.2. More Consistently,
M/K ≥ 1 , which restricts the total number of users from exceeding the total number
of available antennas. We assume that the base station has a full knowledge of channel
state information or a perfect CSI is available at the transmitter. Massive MIMO is
known to operate in time division duplexing (TDD) mode of operation in which the
channel estimation overhead doesn’t increase with the number of antennas. In TDD
mode, uplink and downlink transmissions use the same frequency but separated in time
[10]. Therefore, we assume a TDD mode of operation where the uplink and downlink
transmissions are
separated in time.
Figure 3.2: Multi-user MIMO system where K users are served by the M-antenna BS in
the same time-frequency resource [10].
The wireless transmission channel is modeled based on a non-dispersive i.i.d rayleigh

channel model which is under the category of correlation-based stochastic channel model
(CBSM). In a non-dispersive i.i.d rayleigh channel model of CBSM, it is assumed that no
correlation exists between the transmit and receive antennas. Hence, the elements of the
small scale fading matrix are i.i.d Gaussian random variables [9].
Channel Sate Information: In order to take advantage of the MU-MIMO system, the
spatial channel between antenna elements and user terminals needs to be characterized
and should be known at the transmitter . This response is generally referred to as channel
state information (CSI). This CSI is effectively a collection of the spatial transfer functions
between each antenna and each user terminal. This spatial information is gathered in a
matrix (H) [27], as shown in figure 3.3.
Figure 3.3: Channel state information characterization [27]
From figure 3.3 above the fading of each signal from user ’k’ to antenna ’m’ can be
modeled as
1
hmk = gmk β mk
2
(3.1)
hmk ∈ C1 and βmk ∈ R1
where
gmk is a complex small scale fading coefficient between the mth antenna and the kth user
βmk is a rel valued large scale fading coefficient between the mth antenna and the kth user
From the small scale fading perspective the channel is assumed to be Rayleigh fading
channel so that |hmk | is a Rayleigh distributed random variable.
Since the spacing between the antennas is small enough the large scale fading is assumed
to be the same for all antennas. But the users are spatially distributed throughout the
coverage of the base station and their location difference is large enough to have different
large scale fading coefficients [6][28].
βmk ≈ βk
The above model is based on each user channel. Now let’s develop a channel matrix that
encompasses all users and antennas in the system.
The total channel matrix can be expressed as
1
H = GD 2 (3.2)
G consists of small scale fading coefficients and D is a diagonal matrix where the diagonal
elements are the large scale fading coefficient of each user (βk ).
Equation 3.2 can be written in a matrix form as
G ∈ CM XK H ∈ CM XK D ∈ RKXK
 
 g11 g12 g13 ··· g1K 

 
 g21 g22 g23 ··· g1K 

G=  .
 . . .  (3.3)
 .
 . . . 
gM1 gM 2 gM3 · · · gMK
The entries of G are a complex small scale fading coefficient and are independent and
identically distributed(i.i.d).
 
 h11 h12 h13 ··· h1K 

 

 h21 h22 h23 ··· h1K 
H=  .
 . . . 
 (3.4)
 .
 . . . 

hM1 hM2 hM3 · · · hM K
H = [h1 , h2 , ...., hK ]
where hk is each users total fading channel vector

 
 β1 0 0 0

 
0 β2 0 0 

D =
 .. 
 (3.5)
0
 0 . 0
0 0 0 βK
Since we are considering Rayleigh fading channel model and each path fading is as-
sumed to be independent, each entries of G (gmk ) are generated from normal distribution
(N (0, 1)).
hmk ∈ N (0,1)
In this thesis the effect of small scale fading is considered and the large scale fading
coefficients are set to unity, |βk | = 1 and H = G.
3.3 Transmit And Received Signal Model
3.3.1 Transmit signal

processing
In MU Ma-MIMO system a large number of antennas serve a relatively small number

of users at the same time and frequency resources. Communicating with the same time
and frequency results interference among the multiple users which greatly degraded the
performance of the system. A nonlinear technique such as maximum likely hood detection
at the uplink and dirty paper coding in the downlink can achieve optimal capacity. But
they give rise high complexity which prohibits implementing them in reality. When the
number of antennas is large as compared to the number of users, linear precoding system
in the downlink and linear detection in the uplink results in nearly optimal performances
[10].
Figure 3.4: Block diagram of the linear precoders at the BS [10]
Figure 3.4 depicts a process of linear precoding. Each user symbol sk is multiplied by a
weight vector wk then the weighted symbols are added together to form a transmitted
vector X. In another words, The symbol streams are coded independently and multiplied
by a linear precoder W (accounting for the beamforming weight and power) for trans-
mission through multiple antennas. Careful selection of the precoders can reduce (or
eliminate) interference among different streams by taking advantage of spatial separation
between users and thereby support multiple users simultaneously [29]. Let a set S and
W consists of each user symbol and precoding vector [7].
X
K
X= sk wk (3.6)
k=1
Where, X ∈ CM X1
Equation 3.6 can be re written as

X = WS (3.7)
S = [s1 , s2 , ....., sK ]T
n o
2
Each users symbol sk is a complex i.i.d random variable with E |sk | =1
W = [w1 , w2 , .....wK ]
W ∈ CM XK and wk ∈ CM X1
W is a normalized precoding Matrix to full fill the total transmission power constraint
,P, according the type of precoding employed.
tr(WH W) = P (3.8)
where tr(∗) is a trace function
The power allocation and precoding matrix normalization will be justified in section 3.4.
3.3.2 The received signal model
After a signal is transmitted through the available antennas it is encountered by fading

and the received signal vector y [7] is expressed as
y = HT X + n (3.9)
n ∈ N 0, σ2 ,
y ∈ CKX1
Where X is transmitted signal vector

n is additive white Gaussian noise vector with mean 0 and variance σ 2
Each users received vector can be obtained from (3.9)
yk = hkT X + n (3.10)
X
K
yk = hkT sk wk + n (3.11)
k=1
X
yk = hkT wk + hkT wi si + n (3.12)
i=k
The second term in (3.12) represents inter-user interference. The noise power σ 2 is as-
sumed to be the same for all users.
From (3.12) we can derive the SINR of each user as follows:
2
hk T w k
SIN Rk = P T 2 (3.13)
i=kh k w
i + σ
2
The type of w differs with each other depending on the precoding type implemented
at the transmitter. The average broadcast sum-rate which is averaged over all channel
realization is given by [24]
 
X
Rsum (A, U) = E  log2 (1 + SIN Rk (A, U))
(3.14)
k∈U
Where E[∗] is the expectation operator
Rsum (A, U) implies the broadcast sum rate achieved by the users in the user set U simul-
taneously served by antennas in the set A and SIN Rk (A, U) is the signal to interference
plus noise ratio of each users in the set U served by antennas in the set A.
JASUS algorithm selects transmit antennas and users simultaneously with the aim of
maximizing (3.14). The number of selected antennas are limited by the available antennas
where as the maximum number of scheduled user are limited by the number of selected
antennas. In chapter 4 the problem formulation and the constraints for the JASUS
algorithm will be presented in detail.
3.4 Mathematical Model Of Linear precoding Tech-

niques
With practical large number of antennas at the BS, nonlinear precoders are almost im-
practical in massive MIMO systems. By contrast, linear precoders, such as MF and
ZF, have been proven to be near-optimal or as effective in most cases. Thus, it is more
desirable to use linear precoding techniques in massive MIMO systems ,which have sig-
nificantly lower implementation complexity in practice [6].
Since our system model depends on perfect CSI at the transmitter the linear precoder ap-
ply linear processing on the total channel matrix so that inter-user interference becomes
negligible or totally canceled in the desired spatial direction.
Three conventional linear precoders we used are maximum-ratio transmission (MRT)

(also called conjugate beamforming), ZF, and MMSE precoders .
These are
1. Zero-forcing beam forming (ZFBF)

Zero-forcing precoding attempts to address the inter-user interference problem by
designing the optimization criteria to minimize for it. The precoding matrix is the
pseudo-inverse of the CSI matrix. Calculating the pseudo-inverse is more compu-
tationally expensive than the complex conjugate as in the MRT case.
2. Minimum mean square error (MMSE)

MMSE tries to strike a balance between getting the most signal amplification and
reducing the interference. The MMSE approach introduces a regularization term to
the optimization that allows for a balance to be found between the noise covariance
and the transmit power. It is sometimes also referred to in literature as regularized
zero-forcing (RZF).
3. Maximum ratio transmission(MRT)

Maximum ratio transmission, as the name indicates, aims to maximize the signal-
to-noise ratio (SNR). It is the simplest approach from a signal processing viewpoint,
as the precoding matrix is just the conjugate of the CSI matrix, H. The big downside
of this method is that inter-user interference is ignored [27].
In the next section we will develop the mathematical model of each type and the resulting
SINR which will be used in the JASUS algorithm in chapter 4 .
3.4.1 Zero-Forcing Beam Forming

(ZFBF)
Zero-forcing beam forming vector is found by using the pseudo inverse of the channel ma-
trix. It essentially relies on the idea of pre-inverting the channel matrix at the transmitter
so as to completely remove the interference at all users. The channel matrix formed by
users in the user set |U| can be re-written From (3.4) as
H = [h1 , h2 , ...., h|U| ] (3.15)
Where |U| is the cardinality of set U
The un-normalized ZFBF matrix W can be calculated from the pseudo inverse of H [30]
W = H(HH H)−1 (3.16)
W can be expressed with respect to each users precoding vectors

W = [F1 , F2 , ........, F|U| ]T (3.17)
Fk is a precoding vector for each user .
W is then normalized so that the 2-norm of Fk is unity. In ZFBF precoding all the
interferences in the desired direction is effectively cancel out. Therefore there will be |U|
parallel channel in the system. For such system water filling power allocation results an
optimum performance [2]. We used this power allocation for ZFBF precoding.
Let pk is the power allocated for user k using WF power allocation then the ZFBF
precoding matrix is given as  
√ √ √
pF pF p F
| U| | U
WZF =  1 1 , 2 2 , ....,  (3.18)
||F1 || ||F2 || 2
|
2
F|U| 2
WZF = [w1 , w2 , ......, w|U| ] (3.19)

√
pk Fk
wk = (3.20)
||Fk ||2
From (3.12) the second term will be zero for ZFBF case so the the received signal and
SINR for each user can be expressed as
zf
yk = hkT wk sk + n (3.21)
2
zf hTk wk
SIN Rk = (3.22)
σ2
Substituting (3.20) into eqn. 3.22 we obtain a simplified expresion as
2
zf hTk wk
SN Rk = (3.23)
σ2
SINR in (3.22) reduces to SNR (3.23) as the the inter-user interference becomes zero.
Zero-forcing scheme completely eliminate interference among users, which create inter-
frence free communication. However , this results signal strength degradation at the
receiver. The main reason for this degradation is essentially due to the power boosting
effect, which occurs in the pseudo-inverse computation of ill-conditioned channel matri-
ces. This translates into high power consumption which inevitably reduces the SNRs at
the users with ensuing degradation of the system performance at low SNRs [31].
We can derive the average broadcast sum-rate that can be achieved by users when zero-
forcing precoding is applied from (3.14).
 
X
ZF (A, U) = E  log2 (1 + SIN R 
Rsum
(3.24)
ZFk (A, U))
k∈U
3.4.1.1 Water filling Power Allocation
From equation 3.23 we can observe that each user is interference free and the total system
can be considered as a composite of |U| SISO channel where each user received its own
signal corrupted by AWGN noise. In an other words a |U| parallel sub-channels are
formed. The sum-rate capacity achieved by users in the set U is found by the sum of the
capacities achieved by each parallel channels. Mathematically
ZF X zf
Rsum = log2 1 + SN Rk (3.25)
k∈U
With the objective of maximizing (3.25) and under the total transmit power constraint P
at the base station, the optimal power allocation strategy is achieved by means of water-
filling algorithm [31]. In this power allocation strategy the total power P is allocated to
each user according to
+
pk = [µγk − 1] (3.26)
where γk represents the effective channel gain at the k th user and it is given by
1
γk = (3.27)
[(HH )−1 ]kk
H
µ is a water level and it is chosen to satisfy following equation

" #+
X
|U|
1
µ− =P (3.28)
k=1 γk


+ [x] =x if x > 0
Where the function [x] = max(x, 0) = +
+
[x] = 0 if x < 0
In this thesis we implement user selection to form a subset of users from a user set U
and power is allocated to the users in the subset. All users in the selected subset are
allocated a non zero power. In this scenario [32] a closed form solution for the water level
µ is given by  
1  X
µ= P + γk−1  (3.29)
|U| k∈U
After the value of µ is found in (3.29), pk which is the power allocated for each user can
be solved using (3.26).
A water-filling power allocation policy allocate more power to users with a stronger chan-
nel and less power is allocated to users with a weaker channel strength.
3.4.2 MMSE
To compensate the disadvantage of ZF precoding scheme at low SNR regularized

zero- forcing (RZF) is proposed [33].
WRZF = H(HH H + αI)−1 (3.30)

where α is the regularizing factor
The idea behind RZF is to add a multiple of identity matrix before inversion which allows
some interference between users. The amount interference is determined by the value of
α. The MMSE is derived From (3.30) such that the regularizing factor α is derived with
the objective of maximizing SINR [33].
| U| σ2
αM M SE = (3.31)
P
Where |U| is the number of users

P is the total transmission power constraint
σ2 is the noise variance
Therefore the un-normalized MMSE precoding matrix can be obtained from (3.30) and
(3.31)
2
|U| σ
Wf = H(H H + H
I)−1
(3.32)
P
Then
is normalized to satisfy the total base station power constraint P .
Wf
f
W
WM M SE = (3.33)
η
Equation 3.8 can be written for MMSE case, as

H
tr(WMMSE WMMSE ) = P (3.34)
where tr(∗) is a trace function The normalization factor is derived with the aim of satis-
fying (3.34), and it is given by
v
u
u
t tr f
WWfH
η= (3.35)
P
n o
WMMSE = w1 , w2 , .....w|U| (3.36)
wk represents each user precoding vector.

Each user received signal contains some interference from other users and noise and it
can be expressed as
X
ykM M SE = hkT wk + T
si hk wi + n (3.37)
i=k
The SINR mathematical model for MMSE precoding can be derived from the received
signal model.
T 2
h
k w
k
SIN RM M SE
k =P (3.38)
h Tw 2 + σ 2
i=k k i
The average sum-rate achieved by users in the set U is given by

 
X
MM SE (A, U) = E  log2 (1 + SIN RMkM SE (A, U))
Rsum
(3.39)
k∈U
3.4.3 Maximum Ratio Transmission

(MRT)
MRT is the downlink version of maximum ratio combining of the uplink. MRT is a
pre-processing technique at the transmitter whereas MRC is a receiving technique at
the receiver. Both techniques focus on magnifying the SNR while it doesn’t consider
inter-user interference minimization. Due to this reason the signal processed with MRT
precoding and received at the user suffers with inter-user interference. But MRT is a low
complexity technique for implementation because it doesn’t need any channel inversion
like that of zero-forcing and MMSE [10].
The normalized MRT beam forming matrix WMRT can be obtained by the complex
conjugate of the CSI matrix W and normalization factor η [10].
W = H∗ (3.40)
The normalization factor is obtained by using W
s
tr (WWH )
η= (3.41)
P
1 ∗
WMRT = H (3.42)
η
The columns of (3.41) represents the precoding vector for each users. And it can be
rearranged as " #
1 1 ∗ 1
WMRT = h∗1 , h2 , ..... h|U| ∗ (3.43)
η η η
wk = hk∗ (3.44)
The SINR for each users can be derived from equations (3.12) ,(3.13) and (3.43) for MRT
precoding case.
1
η2
|hk |4
SIN RM RT
k = P 2 (3.45)
1
h∗ + σ2
η2 i=k hk T i
In a MISO antenna configuration, the matched filter or maximum ratio transmission

(MRT) precoding maximizes the signal power at the intended users. This is performed
by projecting the data symbol onto the beam-forming vector given by the spatial direction
of the intended channel as shown in equation [15]. Due to this, the spectral efficiency of
MRT is better than ZF in noise limited transmission.
The average sum-rate for MRT precoding is given by

 
X
M RT (A, U) = E  log2 (1 + SIN RMkRT (A, U))
Rsum
(3.46)
k∈U
In the next chapter we apply the broadcast sum-rate mathematical models of the three
precoding techniques in the optimization problem of the JASUS algorithm.
Chapter 4
Joint Antenna Selection And User
Scheduling
A joint antenna selection and user scheduling (JASUS) is a joint strategy to select a set
of serving antennas from the available antennas at the base station and schedule a set
of users to be served within a specified time slot. The necessity of antenna selection for
massive MIMO systems comes from its hardware complexity and energy consumption
due to mass number of antennas at the transmitter. The selection of a limited number of
antennas forces the massive MIMO system to select and schedule a set of users from the
available. The strategy of selecting antennas and schedule users depends on the system
performance metrics to be optimized. Since band width is a scarce resource in wireless
communications, spectral efficiency is one of the performance metrics to be optimized in
massive MIMO systems. In line with this fact, the goal of most JASUS algorithms is to
maximize the spectral efficiency as much as possible with low computational complexity.
In this chapter we will justify the JASUS algorithm that we are going to analyze its
performance by modifying its user scheduling part with conventional low computational
complexity scheduling scheme.
This chapter is divided into three sections. The first section will present and Justify
how the JASUS algorithm works to maximize the SE performance. In the next section
we will describe the two user scheduling schemes that we will implement and the
SUS user scheduling scheme which is originally used. The last section is about the
computational complexity of the user scheduling schemes, precoding techniques and
the JASUS algorithm.
38
Chapter 4. Joint Antenna Selection And User 39
4.1 Joint Antenna Selection And User Scheduling

Algorithm and The problem Formulation
4.1.1 Problem formulation
The joint antenna selection and user scheduling optimization problem can be written as
[13].
 
X
max.Rsum (A, U) = max.E  log2 (1 + SIN Rk (A, U))
(4.1)
A,U A,U
k∈U
Subject to
|U| ≤ |A| ≤ N (4.2)

X X
pk = P (4.3)
m∈A k∈U
For the given applied precoding type Rsum (A, U) expression is substituted by one of
the respective broadcast sum-rate equation presented in chapter 3 (equations 3.24,3.39
and
3.46).

RZF (A, U) for ZF precoding
 sum
Rsum (A, U) = Rsum
MM SE (A, U) for MMSE precoding


sum (A, U)
R MR for MRT precoding
T
The joint antenna selection and user scheduling algorithm tries to select a subset of
antennas A and schedule a subset of users U that maximize the system capacity (sum-
rate) as described by (4.1). Constraint in (4.2) controls that the number of selected
antenna will not exceed the available number of RF chains. Also it makes sure that the
maximum number of scheduled users will not exceed the number of selected antennas.
The constraint described by (4.3) ensures that the sum of power allocated to each user
in the subset U will not be more than the total available power at the BS.
4.1.2 Joint antenna and user scheduling algorithm
Algorithm 1 describes the steps of JASUS algorithm. The highlighted parts (line 4 and
5 of Algorithm 1) shows the sections of algorithm where we apply low complexity user
scheduling schemes (line 4) and two different precoding techniques (line 5).
Algorithm 1: Steps of JASUS [13]

Input:
Channel coefficients H;
Number of RF chains N
Initialization:
t← 1
A ← {1, ...., M}
1 while t < N do
2 maxRate ← 0;
3 foreach m ∈ A do
4 Ut ←a set of N users using USS(A\ {m} , N );
5 R m = Rsum (A\ {m} , Ut );
6 if maxRate > R m then
7 maxRate ← R m;
8 mbad = m;
9 U = Ut
10 end
11 end
12 A ← A\ {mbad };
13 t ← t + 1;
14 end
Output: The set of antennas given by A and the set of users given by U
At each iteration the algorithm removes the worst performance antenna and schedule users
using user scheduling schemes (USS). By exploiting the spatial selectivity and multi-user
diversity gain offered by the antenna selection and user scheduling respectively, the al-
gorithm tries to achieve the maximum spectral efficiency. The JASUS algorithm takes
as input the channel coefficients of all users as well as the number of RF chains. The
algorithm seeks to construct two sets, namely the set of activated antennas A and
the
set of scheduled users U [13].
It starts by initializing A with the set of all antennas. In each new iteration, the algo-
rithm looks for the antenna to be removed from A. To this end, it iterates over all the
elements in this set in order to find the worst antenna, i.e. the one without which the
system can provide the maximum sum-rate[13]. The sum-rates ( line 5 of Algorithm 1)
are computed based on one of the three user scheduling schemes (USS) taking as input
the set A deprived of one antenna and the set of all users U. The JASUS algorithm
selects the user set (Ut ) (line 4) by using a set of antennas deprived of one antenna and
number of RF chains. Line 12 shows the removal of the worest performance antenna
designated by mbad from the current antenna set A. Also to calculate the sum-rates the
channel matrix formed by the selected user set (Ut ) is precoded using one of the three
precoding types (MMSE,MRT,ZF).
By the end of the first foreach loop (lines 3-11 in algorithm 1), JASUS removes the worest
antenna from set A and then proceeds to the second iteration of the whileloop (lines 1-
14) and so on. The algorithm terminates when it performs exactly M − N iterations
producing a set of exactly N antennas and a set of at most N users [13].
4.2 User Scheduling Schemes
In conventional MU-MIMO systems the number of users are large compared to the number
of total antennas at the base station in many cases. To achieve optimal capacity by
using a simple linear precoding techniques it is necessary to schedule users by selecting
a specified number of users less than the available antennas at the BS. As the number
of users increase the capacity achieved by precoding techniques combined with user
selection increases due to diversity gain [4]. However, in massive MIMO systems most of
the time the number of user is much less than the number of antennas and user
scheduling may not be important as conventional MU-MIMO systems. But there are
conditions that forces a system to select and schedule users . In our case antenna selection
is the factor that makes user scheduling important. A semi-orthogonal user scheduling
which is widely used in conventional MU-MIMO system generates a cubic scale of
computational complexity as the number of antennas increase. It is obvious that JASUS
algorithm which employs SUS will suffer with computational burden. In this section we
will provide the description of the three user scheduling schemes which will be
implemented in JASUS algorithm. The
two user scheduling schemes are norm based and random user scheduling which have low
computational complexity and the third is a semi-orthogonal user scheduling.
4.2.1 Norm Based User

Scheduling
This user scheduling scheme is based on the norm (channel gain) of users’ channel. And
it is simple as it requires a single iteration. The frobenius norm of each user channel
vector is first calculated and sorted in descending order. Then the first largest channel
gain users are selected. The number of users to be scheduled is determined by the system
requirement. In the JASUS algorithm the number of RF chains limits the maximum
number of users to be selected. The appendix part provides how the frobenius norm of a
matrix and a vector is calculated.
Algorithm 2 shows the pseduo code of norm based user scheduling. NUS ( line 1 of algo-
rithm 2) calculates the norm of each user channel vector from the channel matrix formed
by antenna set A and user set U and sorts in descending order based on the magnitude
of the channel vectors. Then N users with largest channel gain is selected (line 2 of
algorithm 2).
Algorithm 2: Steps of norm based user scheduling [34]

Input:
Set of available antennas A
Number of available RF chains N
Initialization:
U = {1, 2....., K};
S =∅
1 For k = 1, 2, ....., K,sort all ||hk,A ||2 ,such that;
||hπ(1) ≥ hπ(2) ≥, ..... ≥ hπ(K) ||
2 S = {π(1), π(2), ....., π(N )}

4.2.2 Semi-orthogonal User

Scheduling
Semi-orthogonal user scheduling scheme select users whose channels are orthogonal as
possible by maintaining as large channel gains as possible. More exactly, it selects users
one at a time, and each time it tries to maximize the channel projection to the orthogonal
subspace spanned by the channels of all the users already selected. Steps of SUS is shown
in algorithm 3.
Algorithm 3: Steps of SUS [4]

Input:
Set of Available antennas A;
Number of RF chains N
Initialization:
U ←− {1, 2....., K};
S ←− ∅
i ←− 1
1 while i < N do
2 for k ∈ U do
!
P ˜gj ˜g
i−1 H
3 gk,A = I − j
||˜gj || hk,A
j=1
2
4 end
5 iopt = arg max ||gk,A ||2 ;
∈
 k U
6 S =S iopt ;
7 U = U\iopt ;
8 ˜gi = giopt ;
9 i ←− i + 1
10 end
Output: The set of users is given by S
In SUS algorithm, first of all, for every user in the set U it calculates, gk,A the component
of the channel matrix corresponding to the k th user, which is orthogonal to the subspace
spanned by {˜g1 ....˜gi−1 }. The user that maximizes norm of this orthogonal
subspace(iopt , the index of selected user) is selected. As results, this procedure make
sure that the
selected user has a good level of orthogonality with other user. The sets S and U are up-
dated(line 6 and 7 of algorithm 3) by adding and removing the selected user respectively.
The procedure is repeated until the number of selected user is N [13][25].
4.2.3 Random User

Scheduling
RUS is the simplest user selection method and select users randomly independent of their
channel realizations. Algorithm 4 describes the steps of RUS. It simply picks a random N
user index from the available K user index and the set S consists the selected N users [34].
Algorithm 4: Steps of random user scheduling [34]

Input:
Set of available antennas A
Number of available RF chains N
Initialization:
U = {1, 2....., K};
S =∅
1 Let S = {π(1), π(2), ....., π(N )} be a set of any N random indices from
{1, 2, ....., K} ;
2 S = {π(1), π(2), ....., π(N )}
4.3 Computational Complexity
Computational complexity refers to the processing time required to execute a certain

algorithm at the transmitter [15]. When computers carry out addition, subtraction, mul-
tiplication, division, or other arithmetic operations on numbers represented in floating
point format, the result is rounded to the nearest floating point number. These oper-
ations are called floating point operations. A very rough estimate of the time required
to carry out some computation, such as channel inversion can be found by counting the
total number of floating point operations, or FLOPs. The complexity of an operation is
the number of flops required to carry it out, as a function of the size or sizes of the input
to the operation. The number of flops required to execute an algorithm is termed the
flop count of the algorithm [35][36].
In MU-MIMO systems data processing involves matrix computation. For instance, in

linear precoding techniques channel matrix inversion and multiplication is required. Also
user scheduling and antenna selection algorithms perform matrix computation at the base
station. Due to this, time consuming floating point operations are performed at the base
station. Therefore, a low complex algorithms with low execution time that results opti-
mum performance on the selected performance metrics is preferred.
In this thesis we use a simplified flop count method to compare the computational com-
plexity performance of the JASUS algorithms with different user scheduling and precoding
techniques.
The “Big-O” (O(.)) notation will be used to simplify the flop count of operations. In this
notation the algorithm is considered to have a flop count of O(nk ) if the dominant term
in the flop count is Cnk , where n is the size of the input to the computation and C is a
constant. For large n, the other terms are negligible in comparison to the term nk [36].
In this section we will first justify the computational complexity of user scheduling al-
gorithms and precoding techniques. In the subsequent section we will evaluate the com-
putational complexity of the JASUS algorithm by implementing each scheduling and
precoding techniques.
4.3.1 Computaional complexity of user scheduling

algorithms
Let the size of a channel matrix H is M XS, where M is the number of total antenna at
the base station and S is the total number of users.
1. Semi-orthogonal user scheduling(SUS)

The asymptotic computational complexity of SUS for M antennas and and S users
is O(M 3 S) [4] [14]. In this scheduling algorithm the computational complexity in-
creases in a cubic factor with the number of antennas.
2. Norm based user scheduling(NUS)

In this scheduling scheme the users are selected based on their channel magnitude.
So that the computational complexity of NUS comes mainly from the computation
of the frobenius norm of each users channel vector which involves multiplication
and addition floating point operations. The computational complexity of calcu-
lating the frobenius norm of a channel matrix of size M XS is given by O(M S)
[35]. Calculating the norm of S users with M X1 channel vector is the same as
calculating the frobenius norm of a channel matrix of size M XS. Therefore, the
computational complexity of NUS is defined as O(M S). The complexity of NUS
rises linearly with the number of antennas. As compared to SUS, NUS has a low
computational complexity.
3. Random user scheduling

Random user scheduling select users to be served by randomly selecting users. No
floating point operation is performed on the channel matrix. Therefore its com-
putational complexity is given by O(1) [37]. RUS is the simplest form of user
scheduling.
Table 4.1: Simplified computational complexity of user scheduling schemes
User scheduling schemes Computational complexity

SUS O(M 3 S)
NUS O(M S)
RUS O(1)
4.3.2 Computational complexity of linear precoding

techniques
A computational complexity of ZF and MMSE is dominated by the channel inversion

computation. Pseduo-inversion computation of SXS matrix is approximated as O(S 3 )
[16][38]. Since we are considering the asymptotic complexity, the floating point operation
of addition term before inversion in MMSE can be ignored. As a matched filtering
operation the computational complexity of a MRT for MXS matrix is given by O(M S).
Table 4.2: Simplified computational complexity of precoding techniques
Precoding techniques Computational comlexity

ZF and MMSE O(S 3 )
MRT O(M S)
4.4 Computational complexity of JASUS algorithm
We will use algorithm 1 to calculate the computation complexity of JASUS algorithm

with different combination of user scheduling and precoding.
From algorithm 1 the outer loop consists M −N iterations. For each outer loop iteration
,i, the inner loop executes M −i+1 iterations as one antenna is removed for the ith
completed outer loop. At each inner loop iteration the JASUS algorithm perform
scheduling and precoding . Let’s denote T as the computational complexity of user
scheduling schemes and R as the computational complexity of precoding techniques.
The computational complexity of JASUS algorithm accounting both scheduling and pre-
coding techniques can be given as
MX
−N
M − i + 1(T + R) (4.4)
i=1
After the user scheduling selects at most N users, precoding is performed on N column
channel matrix to calculate sum-rate (line 5 of algorithm 1). Therefore, we approximate
the computational complexity of MMSE and ZF to matix channel inversion complexity (
O(N 3 ) ) for both techniques. Since one antenna is removed at each outer loop iteration
and N user are selected in the inner loop the channel matrix for scheduling and precoding
in the inner loop will have a dimension of (M − i + 1)XS and (M − i + 1)XN respec-
tively. Now we can derive the computational complexity of JASUS algorithm for each
combination of precoding type and scheduling schemes. Table 4.3 shows the asymptotic
computational complexity of JASUS algorithm for each combination of user scheduling
and precoding techniques.
Table 4.3: Approximate computational complexity of JASUS algorithm under different
user scheduling and precoding implementation
Combination of precoding and scheduling in JASUS Asymptotic computational complexity

MP
−N
SUS-channel inversion M − i + 1((M − i + 1)3 S + N 3 )
i=1
MP−N
SUS-MRT M − i + 1((M − i + 1)3 S + (M − i + 1)N
i=1
MP−N )
NUS-channel inversion M − i + 1((M − i + 1)S + N 3 )
i=1
MP−N
NUS-MRT M − i + 1((M − i + 1)(S + N ))
i=1
MP−N
RUS-channel inversion M − i + 1(N 3 )
i=1
MP−N
RUS-MRT M − i + 1((M − i + 1)N )
i=1
Chapter 5
Simulation Results And
Discussions
In this chapter we will present the results obtained from simulation in the form of plots
and tables. A detailed comparative discussion will be made based on the results. The
chapter is divided into three sections where the first section provide the steps how the
simulation is done and the next section will present the spectral efficiency performance
evaluations of JASUS algorithm for different scenarios. The last section will provide the
computational complexity of JASUS algorithm with different user scheduling schemes
and precoding techniques.
5.1 Simulation setup
For the simulation purpose we used MATLAB software. Four scenarios are
considered to compare the performance of user scheduling and precoding techniques in
the JASUS algorithm. These are
1. Spectral efficiency comparison of user scheduling and precoding technique at differ-

ent values of transmit SNR

ent number of available antenna(M)

ent number of selected antenna(N)

ent number of total users(X)
48
Chapter 5.Simulation Results and Discussions 49
Figure 5.1 illustrates the work flow how the system is simulated to generate results.
Start
pre-allocate
variables
M,X
Based on the scenario

Generate channel set variable values
realizations
N,K
JASUS
Select precoding Do antenna selection select user scheduling
types types
and user scheduling
calculate sum-rate
for each realization
Average the sum-rates

of all realizations
Repeat the process

by changing variable
values
End
Figure 5.1: Block diagram of simulation process
For each single channel realization a channel matrix is generated based on perfect CSI
assumption. The matrix size depends on M and X whose value is fixed or variable based
Table 5.1: System design assumption and considered parameter values
System design asumptions/paramerter values considerations/values

Number of cells Single cell
Channel state information(CSI) Perfect CSI at the BS
Channel model CBSM/uncorrelated/rayleigh i.i.d fading [9]
total power at B S
Transmit SNR noisepower
= σP2 [39]
2
Noise power σ =1
Transmit SNR value 10 dB for all fixed SNR scenarios
Number antennas and users Based on [13]
Number of channel realizations 1000
on the scenarios stated above. The JASUS algorithm uses the generated channel matrix to
select antennas and schedule users by implementing a given precoding technique and user
scheduling scheme. The JASUS algorithm uses N and K as input variables. The next step
is calculating the sum-rates that could be achieved by the scheduled users transmitting
through the selected antenna set for each realizations. This steps will continue for all
channel realizations. Finally the average sum-rates is calculated using the sum-rates
found by each single realizations. Each scenario is simulated by varying the values
of the parameters(M,N,X,K). To remove a user or antenna during the execution of
JASUS algorithm, matlab performs the following operations on the channel matrix.
• Deleting a row means removing a single antenna
• Deleting a column means removing a user
For each scenario the performance comparison is classified into sub parts. The first
is comparison of user scheduling scheme in JASUS algorithm by fixing the precoding
technique. The next part is comparison of the performance of precoding techniques when
implemented in JASUS algorithm for a fixed user scheduling schemes.
5.2 Simulation Result
The results are generated using the above assumptions. Additional parameters and values
used for the simulation are briefed with the respective plots. The purpose of the plots
is to compare the performance of user scheduling types and precodings when they are
implemented in JASUS algorithm. All plots are based on the results generated from
JASUS algorithm.
5.2.1 Spectral efficiency performance evaluation at

different values of SNR
In this section the spectral efficiency performance for different values of SNR in ( dB) is
provided for two cases. First we will evaluate the spectal efficiency for the case M = X.
Then Evaluations for the case M > X will be presented. The transmit SNR is varied
between 0 dB and 30 dB and the results are plotted. The total number of antennas(M)
and and the available number users(X) are M = 10 and X = 10 for the first case and
M=15 and X=10 for the second case. Where as, the selected number of antennas and
users are N = 3 and K = 3 respectively for both cases.
Spectral efficiency performance evaluation at different values of SNR for the

case M=X
From figure 5.2 we can observe that spectral efficiency of JASUS increases as SNR in-
creases for all types of user scheduling. NUS shows a slightly higher performance than
SUS in all SNR region. The performance gap increases at low rate as SNR increase for all
precoding. At 0 dB NUS and SUS have a performance gap of 0.07 b/s/Hz but at 25 dB a
0.43 b/s/Hz performance gap is observed in ZF precoding. This shows the performance
gap is increasing for higher values of SNR.
Table 5.2: Spectral efficiency numerical values at specified values of SNR for the case
M = X
User scheduling combined Spectral Efficiency

with SNR=0 dB SNR=15 dB SNR=25 dB SNR=30 dB
precoding schemes
SUS-ZF 3.66 15.73 25.54 30.53
SUS-MMSE 4.00 15.57 25.34 30.30
SUS-MRT 2.58 5.16 5.45 5.50
NUS-ZF 3.73 16.09 25.97 30.93
NUS-MMSE 4.12 16.02 25.81 30.76
NUS-MRT 2.62 5.38 5.65 5.68
RUS-ZF - 12.45 22.13 27.08
RUS-MMSE 2.94 12.24 21.65 26.55
RUS-MRT 1.90 4.31 4.62 4.67
For MRT at 0 dB and 25 dB of SNR values NUS outperform SUS by 0.04 b/s/Hz and
0.2 b/s/Hz respectively.
35 35
NUS
ZF
RUS NUS
ZF MMSE
30 30
SUS ZF RUS
spectral efficiency(bits/s/Hz)
25 25
20 20
15 15
10 10
5 5
0 0
0 5 10 15 20 25 0 5 10 15 20 25
30 30
SNR SNR
(a) zero-forcing (b) MMSE

6
5.5
5
4.5
4
NUS
3.5 MRT
RUS
3
2.5
2
1.5
0 5 10 15 20 25
30
SNR
(c) MRT
Figure 5.2: Spectral efficiency performance comparison of user scheduling schemes in

JASUS algorithm using different precoding type for the case M = X M = 10, X = 10,N =
3 and K = 3
As shown in figure 5.2(a) the minimum transmit SNR value we used for the combination
of RUS and zero forcing precoding is 5 dB. Since RUS selects users randomly, for some
channel realization it select users with very bad channel conditions. For such user channel
conditions the water-filling power allocation strategy tries to allocate zero power for deep
fade channels in low transmit power regions. This leads wrong power allocation and a
wrong performance result as we maintain the water level in a way that no user is allocated
zero power. But a large transmit SNR increases the chance that all of the users obtain
nonzero power, in which case a right power allocation will be done. This is why we used
5 dB as a minimum transmit SNR for RUS-ZF case. For MRT and MMSE precodings
we implement equal power allocation strategy which allocate power fairly, so we used
0 dB as a minimum transmit power.
NUS selects users that possess higher values of channel magnitude that results a set of
users larger in SNR values than the unselected users. In addition to this the removal of
the worst performance antenna by the JASUS algorithm helps users selected by NUS to
have a minimum inter-user interference. These combined effects enabled NUS to have a
better performance than SUS with a fraction of SE values.
35 35
30 30 NUS
MMSE
SUS NUS ZF
25 25
MMSE NUS MRT

SUS ZF
efficiency(bits/s/Hz)
20 20
15 15
spectral
10 10
5 5
0 0
0 5 10 15 20 25 0 5 10 15 20 25
30 30
SNR SNR
(a) SUS (b) NUS

30
RUS
MMSE
RUS ZF
25 RUS MRT
20
15
10
0
5 10 15 20 25
30
SNR
(c) RUS
Figure 5.3: Spectral efficiency performance comparison of Precoding types using different
user scheduling schemes in JASUS algorithm for the case M = X M = 10, X = 10,N = 3
and K = 3
RUS results lowest performance for all precoding types. NUS select users based on max-
imum channel gain that can result higher SNR users than the unselected and SUS select
users that are orthogonal as much as possible which mnimize inter-user interference but
RUS doesn’t use any techniques to group users which is the reason for its lowest SE
performance in JASUS algorithm. A spectral efficiency gap that varies between 3 and
3.5 b/s/Hz is observed between SUS and RUS in ZF precoding at higher SNR values.
The spectral efficiency of JASUS becomes flattened for all user scheduling types in MRT
precoding case. This is due to the spectral efficiency property of MRT precoding as it
approaches constant value for higher SNR values.
Figure 5.3 shows the spectral efficiency performance comparison of precoding types using
different scheduling schemes in JASUS algorithm for the case M = X. From these figures
we can see that for all scheduling schemes MMSE shows better performance than ZF in
low SNR regions. But, at higher SNR regions ZF outperform MMSE slightly due to the
optimal power allocation (water-filling) used for ZF precoding. MRT shows the lowest
performance among all precoding types because MRT which magnifies the strength of the
users received signal doesn’t consider minimization of inter-user interference which is the
reason for its lowest performance in comparison with ZF and MMSE precodings. Since
the values of the SNR we used is 0 dB and above, this SNR values are enough for ZF to
outperform MRT.
Spectral efficiency performance evaluation at different values of SNR for the

case M > X
Because for most of cases the number of antennas is large as compared to the number
of users in massive MIMO systems in this section we will present the spectral efficiency
evaluations of JASUS for the case M > X. The value of the parameters is M = 15, X
=
10 and N = K = 3.
Figure 5.4 shows that JASUS with NUS still outperforms JASUS with SUS and RUS
shows the least performance for the case M > X. And from the table 5.3 we can see
that MMSE outperform better than ZF at low SNR. But at high SNR values ZF is better
than MMSE in spectral efficiency values.
Table 5.3: Spectral efficiency numerical values at specified values of SNR for the case
M > X
User scheduling combined Spectral Efficiency

with SNR=0 dB SNR=15 dB SNR=25 dB SNR=30 dB
precoding schemes
SUS-ZF 3.94 16.33 26.06 31.01
SUS-MMSE 4.23 16.06 25.84 30.81
SUS-MRT 2.79 5.49 5.76 5.75
NUS-ZF 3.98 16.61 26.49 31.41
NUS-MMSE 4.31 16.54 26.40 31.27
NUS-MRT 2.83 5.65 6.00 6.03
RUS-ZF - 12.64 22.30 27.21
RUS-MMSE 3.00 12.38 21.83 26.69
RUS-MRT 1.97 4.36 4.71 4.69
35
35
NUS ZF
30 NUS
RUS ZF
MMSE
SUS ZF 30
RUS
MMSE
SUS
MMSE
25
25
20
20
15
15
10
10
5
5
0
0 5 10 15 20 25 0
30 0 5 10 15 20 25
30
SNR
SNR
(a) zero-forcing (b) MMSE
NUS
RU
S
SUS
0 5 10 15 20 25
30
SNR
(c) MRT

JASUS algorithm using different precoding types for the case M > X M = 15, X = 10,N
= 3 and K = 3
The spectral efficiency performance gap of the JASUS algorithm with NUS and SUS user
scheduling algorithm is still in the order of fractions . For instance from table 5.3 the
performance difference is 0.04 b/s/Hz at SNR=0 dB with ZF precoding but it rises to
0.4 b/s/Hz at SNR=30 dB. These numerical values shows the spectral performance gap
between NUS and SUS is increasing favoring NUS as SNR goes to high. But at low SNR
values ignoring the negligible performance we can conclude the performance of NUS is
the same with SUS. The performance of RUS is still the least as compared to the two
user scheduling schemes. The spectral efficiency values for the case M > X greater than
the spectral efficiency values in the case M = X for each SNR values. This is because
the additional antennas for the case M > X increase the possibility of finding best
performance antenna. This means it increases the spatial diversity gain of the system.
But increasing small number of antennas favors both scheduling schemes in the same
manner so that the performance gap between NUS and SUS remains more or less the
same for the two cases.
5.2.2 Spectral efficiency performance evaluation for variable

number of antennas
For the spectral evaluation in this section we vary the number of total antennas by
fixing the values of SNR, total number of users, the number of selected antennas and
scheduled users. The number of total antennas is varied between 10 and 30. The
available number of users is X = 10 and the selected number of antennas and users are
N = 4 and K = 4 respectively.
15 15
NUS
14.5 MMSE
RUS
14
14
13.5
13
NUS 13
ZF
RUS 12.5
ZF
12 SUS ZF
12
11
11.5
11
10
10.5
9 10
10 12 14 16 18 20 22 24 26 28 10 12 14 16 18 20 22 24 26 28
30 30
Availabel antenna number (M) Availabel antenna number (M)
(a) ZF (b) MMSE

6
NUS
MRT
5.5 RUS
MRT SUS
MRT
4.5
3.5
10 12 14 16 18 20 22 24 26 28
30
Availabel antenna number (M)
(c) MRT

JASUS algorithm when the available BS antenna is varied using different precoding types,
X = 10, N = 4, K = 4 and SNR = 10 dB
The results from figure 5.5 showed that the spectral efficiency performance of the JASUS
algorithm increases slowly with the increased number of total antennas. NUS shows a
better performance than SUS throughout all regions of total number of antennas in all
scenarios. Increasing the number of antennas results lower interference between users
by decreasing the channel correlation between users and making the user channels more
orthogonal. Therefore, the JASUS algorithm will have an opportunity to select antennas
that result a good level of orthogonality between users. When NUS is implemented
the best SNR set of users are selected since the selected users are better in channel
magnitude than the unscheduled users. So that, user scheduling using NUS results user
set with better SINR when combined with antenna selection which exploits the benefits of
Table 5.4: Numerical values of spectral efficiency at specified number of total antennas
User scheduling combined with Spectral Efficiency

precoding schemes M=10 M=20 M=25 M=30
SUS-ZF 12.88 14.11 14.40 14.66
SUS-MMSE 13.11 14.20 14.46 14.65
SUS-MRT 4.70 5.41 5.59 5.71
NUS-ZF 13.15 14.32 14.62 14.84
NUS-MMSE 13.35 14.39 14.68 14.87
NUS-MRT 4.75 5.52 5.68 5.80
RUS-ZF 9.74 10.09 10.18 10.23
RUS-MMSE 10.26 10.52 10.58 10.67
RUS-MRT 3.89 4.09 4.14 4.17
spatial selectivity. SUS which tries to select a subset of users to minimize interference will
be dominated by NUS performance because already the channels are going to be more
orthogonal as the number of antennas goes to large. The large number of antennas at the
base station also enables a simple user grouping schemes like NUS to have a comparable
SE performance with user scheduling schemes that relies on user separability. Because
as the number of antennas increase the user channels becomes more orthogonal which
greatly minimize inter-user interference. Due to this NUS shows a better performance
than SUS throughout all regions as shown in figure 5.5. At M=10 NUS and SUS have
a performance gap of 0.27 b/s/Hz gap for ZF precoding and a 0.24 b/s/Hz for MMSE
precoding. RUS shows the lowest performance in comparison with NUS and SUS and it
shows a performance gap of 3.41 b/s/Hz with NUS at M=10.
Figure 5.6 shows that the SE performance of JASUS in MMSE is better than ZF for entire
number of antennas. ZF and MMSE shows a close performance in SUS and NUS user
scheduling scheme. The numerical values in table 5.4 shows at M=25 MMSE performance
is greater than ZF by 0.06 b/s/Hz SE value for NUS which is a negligible performance
16 16
14 14
12 12
SUS ZF
NUS ZF
SUS MRT NUS MRT
SUS MMSE NUS
10 10 MMSE
8 8
6 6
4 4
10 12 14 16 18 20 22 24 26 28 10 12 14 16 18 20 22 24 26 28 30
30
Availabel antenna number (M) Availabel antenna number (M)
(a) With SUS (b) With NUS

11
10
9
RUS ZF
8
RUS MRT
RUS
7
3
10 12 14 16 18 20 22 24 26 28
30
Availabel antenna number (M)
(c) with RUS
Figure 5.6: Spectral efficiency performance comparison of precodings techniques when the
availlable BS antenna is varied using different scheduling, X = 10, N = 4, K = 4 and SNR
= 10 dB
distance. For RUS case MMSE results better performance with almost equal performance
gap in all regions.
Due to the spatial selectivity gain, we can see that the SE performance is increasing for
all precoding types with the number of antennas as shown in figure 5.6. But the gain is
higher for channel inversion type precoding (ZF and MMSE) than MRT. From this, one
can conclude that, MRT is poor in exploiting the spatial selectivity gain as compared to
ZF and MMSE.
5.2.3 Spectral efficiency evaluations as a function of the

number of selected antennas
In this section the spectral efficiency performance of JASUS algorithm is evaluated by

varying the number of selected antenna. The total number of antennas and users are
fixed. The number of scheduled users varies with the number of selected antenna with
constraint N=K. The number of selected antenna is equal with the number of RF chains
at the BS.
Figure 5.7 illustrates the spectral efficiency performance comparison of different user
scheduling techniques when implemented in JASUS. The three plots are for the three
20 24
NUS ZF
RUS ZF 22
18
SUS ZF
20 NUS
16 MMSE
RUS
18
14
16
12
14
10
12
8 10
4 5 6 7 8 9 4 5 6 7 8 9 10
10
no. of selected antenna no. of selected antenna (N)
(a) ZF (b) MMSE

6
NUS
MRT
5.5 RUS
MRT
SUS
MRT
4.5
3.5
4 5 6 7 8 9 10
no. of selected antenna
(c) MRT

JASUS algorithm with different number of selected antenna under different precoding. M
= 14, X = 14 and SNR = 10 dB
precoding techniques. The results shows that the spectral efficiency of JASUS is increas-
ing as the number of selected users increasing for ZF and MMSE case and a decreasing
performance for MRT. Because the number of selected users is increased with the se-
lected antenna, each user is suffered with inter-user interference. In contrast to MRT,
ZF and MMSE can eliminate inter-user interference and their performance increase as
the selected antenna increase. NUS shows the highest performance than the other user
scheduling schemes. From table 5.5 at N=10 selected antenna number NUS and SUS
have a performance gap of 0.11 b/s/Hz with ZF precoding.
25 25
SUS MRT
NUS MRT
SUS NUS MMSE
MMSE
20 NUS ZF
20 SUS ZF
15 15
10 10
5 5
0 0
4 5 6 7 8 9 4 5 6 7 8 9 10
10
no. of selected antenna (N) no. of selected antenna (N)
(a) SUS (b) NUS

22
20 RUS MRT
RUS
MMSE
18 RUS ZF
16
14
12
10
2
4 5 6 7 8 9 10
no. of selected antenna (N)
(c) RUS
Figure 5.8: Spectral efficiency performance comparison of precoding techniques in JASUS

algorithm with different number of selected antenna . M = 14, X = 14 and SNR = 10 dB
Figure 5.8 shows the performance comparison of the three precoding types implemented
in JASUS with a given user scheduling. From the results we can observe that MMSE has
the highest performance than ZF and MRT. Specially, as the number of selected antenna
increases, the performance gap goes to high. From table 5.5 the performance gap between
ZF and MMSE in SUS use scheduling scheme and at N=7 is 0.47 b/s/Hz, but at N=10
the gap becomes 6.08 b/s/Hz which confirmed MMSE performance superiority as the
number of selected antenna increases.
As we noted earlier, the number of selected user is equal to the number of scheduled
users. So that, the transmitted power is shared by the newly added user which degrades
the SNR of each users. Due to this MMSE which has a property of a stable increasing
performance in low and high SNR regions has highest performance in all scheduling types.
Table 5.5: Numerical values of spectral efficiency at specified number of total antennas

precoding schemes N=4 N=7 N=9 N=10
SUS-ZF 13.7 18.01 19.48 19.71
SUS-MMSE 13.87 19.48 22.49 23.79
SUS-MRT 5.01 4.74 4.27 3.98
NUS-ZF 14.01 18.20 19.59 19.82
NUS-MMSE 14.19 19.58 22.55 23.82
NUS-MRT 5.1 4.77 4.27 3.99
RUS-ZF 9.72 13.61 15.70 16.65
RUS-MMSE 10.24 16.19 20.05 21.88
RUS-MRT 3.85 4.03 3.95 3.81
5.2.4 Number of available users versus spectral

efficiency
In this subsection the results found from the spectral evaluation of JASUS algorithm with
variable number of total users(X) will be presented. The fixed parameters are the total
transmit antenna (M), the number of selected antenna (N) and the number of scheduled
users(K). In the first part of the plot the performance of the user scheduling schemes is
compared when implemented in JASUS using a given precoding techniques. In the next
plot the performance of precoding techniques is compared in JASUS algorithm for a given
user scheduling.
It is clear that as the number of users increases the spatial orthogonality between users
diminishes. This increases the inter-user interference between users by increasing the
channel correlation. MRT which doesn’t consider elimination of co-channel interference
is prone to inter-user interference in this scenario. From figure 5.9 we can see that the
SE performance of JASUS is increasing at slow rate for NUS and SUS with channel
inversion type precodings. As the user number increase, the probability of getting users
with good channel conditions will increase for NUS and SUS. With the support of channel
inversions which minimize interference in the desired direction SINR will increase as the
total available users increase for SUS and NUS. So that the SE performance of JASUS
increases with the number of users. But RUS shows decreasing performance when the
number of users increase as it selects users randomly without any criteria.
15
14
14
13 RUS ZF RUS
SUS ZF 13 MMSE
SUS
MMSE
12
12
11
10
11
9
6 7 8 9 10 11 12 13 14 15
16 10
6 7 8 9 10 11 12 13 14 15 16
number of users(X)
number of users (X)
(a) zero-forcing precoding
(b) MMSE
6
NUS
MRT
RUS
3
6 7 8 9 10 11 12 13 14 15
16
number of users (X)
(c) MRT

JASUS algorithm with variable number of users with different precoding M = 16, N=K=4
and SNR = 10 dB
The performance gap between NUS and SUS in channel inversion type precoding is larger
than the performance gap observed in matched filtering precoding. To support this in
numerical value we can refer table 5.6. At X=15 NUS has a performance distance of 0.31
b/s/Hz from SUS under ZF precoding where as the performance gap decreases to 0.11
under MRT precoding.
15 15
14 14
13 SUS ZF 13
NUS MRT
SUS MRT
NUS
12 SUS MMSE 12 MMSE
11 11
10 10
9 9
8 8
7 7
6 6
5 5
6 7 8 9 10 11 12 13 14 15 6 7 8 9 10 11 12 13 14 15
16 16
number of users (X) number of users (X)
(a) SUS (b) NUS

12
RUS ZF
11 RUS MRT
RUS
10
3
6 7 8 9 10 11 12 13 14 15
16
number of users (X)
(c) RUS
Figure 5.10: Spectral efficiency performance comparison of precoding techniques when

the number of users is varied using different user scheduling techniques M = 16, N=K=4
and SNR = 10 dB
This is because SUS can minimize interference and approach the performance of NUS in
MRT precoding where interference is sever. But for channel inversion case the precoding
minimize or eliminate interference for each user and NUS outperform SUS more as the
number of users increase.
Figure 5.10 shows performance comparison of precoding techniques along with a given
user scheduling scheme when the number of users varied. With NUS and SUS a slow
performance increment is observed as the number of users increase. But MRT shows a
decreased performance with the number of users. With a fraction of SE gap MMSE shows
a higher performance gap than ZF with SUS and NUS. Table 5.6 reveals that MMSE is
higher than ZF by 0.11 b/s/Hz at X=15 with NUS. The performance gap between MMSE
and ZF is higher when RUS is implemented as compared to SUS and NUS . To support
this in numerical values, at X=15 the SE gap is 0.47 b/s/Hz in RUS. This is expected
because in RUS the probability of selecting poor channel condition is higher than other
scheduling schemes which degrade the SINR of each user. It is known that for poor
channel condition MMSE have a better performance than ZF. From the figures 5.9 and
Table 5.6: Spectral efficiency values of JASUS at specified number of users for
implemented precoding and user scheduling schemes

precoding schemes X=6 X=9 X=15 X=16
SUS-ZF 13.53 13.73 13.95 13.94
SUS-MMSE 13.60 13.84 14.08 14.08
SUS-MRT 5.41 5.26 5.11 5.10
NUS-ZF 13.63 13.91 14.26 14.31
NUS-MMSE 13.69 14.04 14.40 14.42
NUS-MRT 5.45 5.34 5.22 5.21
RUS-ZF 10.83 10.06 9.72 9.69
RUS-MMSE 11.15 10.54 10.25 10.21
RUS-MRT 4.58 4.12 3.86 3.83
5.10 a slow increasing property is observed for the combinations of NUS and SUS with
channel inversion variant precodings. This increasing property comes from multi-user
diversity gain which increase with the number of total users increase. But multi-user
diversity doesn’t favor MRT precoding and RUS from the user scheduling schemes. They
show a decreasing pattern as the number of users increase.
5.3 Computational Complexity Numerical Evaluation
One objective of this work is to implement the low complexity scheduling algorithm along
with variants of precoding types in the JASUS algorithm and compare their computa-
tional complexity performance in addition to spectral efficiency. In this section we will
provide the asymptotic complexity to assess how the algorithm responds to input size
changes expressed in terms of M, N, X and K. All the numerical results are expressed in
number of floating point operations (flops).
This section is divided into two parts. The first part will provide the computational com-
plexity comparison of JASUS algorithm for all combination of scheduling and precoding.
The next part will present only the comparison of the computational complexity of the
two scheduling schemes (SUS and NUS) for variable number of antennas without the
JASUS algorithm. The results will be presented in plots and tables.
5.3.1 Computational complexity evaluation of JASUS

algorithm
One of the objective of this thesis is to evaluate the computational complexity of JA-
SUS algorithm by implementing low complexity scheduling techniques. In line with this
objective, in this section we will provide the asymptotic computational complexity of
JASUS algorithm with different user scheduling and precoding techniques. We classify
the evaluation into two based on the the input to be varied. These are evaluations which
cover how the computational complexity of JASUS algorithm responds to the variation
of antenna number and number of scheduled users.
5.3.1.1 Computational complexity of JASUS for the variations of antenna

number
The numerical values are found using table 4.3 by setting S = X. To plot the computa-
tional complexity of JASUS we varied the number of antennas between 10 to 80 where
as the other parameters are constant with values X = 20, N = k = 6.
Figure 5.11 shows the complexity of JASUS algorithm with the combination of user
scheduling and precoding types which perform channel inversion. From this figure we
observe the curved scaling for NUS and SUS. This is the result of M n term of equa-
tions (table 4.3) for the two scheduling schemes . Where as JASUS in RUS shows a
linear increment scale because of M N 3 term with channel inversion precoding. N is not
3 3
varying so that N is constant making M N linear. JASUS with RUS generates
lowest computational complexity with a slow increment as the number of antennas
increase.
Even though JASUS in NUS and SUS shows a curved scaling, the computational com-
plexity in SUS is rapidly increasing as the the number of antennas increase as compared
to NUS. The computational complexity of NUS is in mega flops while that of SUS is in
giga flops. This confirms that JASUS algorithm with NUS results lower computational
complexity than JASUS with SUS.
6
10
4
NUS-inversion
RUS-inversion
3
flops
0
10 20 30 40 50 60 70 80
number of antennas
(M)
10
15
NUS-inversion
RUS-inversion
10 SUS-inversion
flops
0
10 20 30 40 50 60 70 80
number of
antennas
Figure 5.11: Asymptotic computational complexity of JASUS algorithm with different

user scheduling and channel inversion precoding(MMSE and ZF).
Figure 5.12 compares the complexity of JASUS algorithm under the implementation of
MRT precoding with the three user scheduling schemes. Since the number of antennas
(M) is the variable parameters by assuming other parameters constant, the computational
complexity scaling is dominated by user scheduling complexity. This is why the two
figures shows the same increment pattern for the channel inversion and MRT precoding.
5.3.1.2 Computational complexity of JASUS for the variations of Scheduled

number of users(N=K)
For the purpose of evaluation and comparison of the computational complexity of JASUS
algorithm for different number of scheduled users(K) we use the following parameter
values. K ranges from 10 to 20 along with N and X=25 ,M=64.
From figure 5.13 computational complexity plots of JASUS in RUS and NUS are a flat-
tened curve due to N 3 term in table (table 4.3). Before K = 17 RUS is in mega flops
where as NUS is in the range of 107 flops. But both plots fall into the range of 107
flops after K=17 with NUS dominating RUS. JASUS with SUS seems to be constant
10
5
4 NUS-MRT
RUS-MRT
3
flops
0
10 20 30 40 50 60 70 80
number of antennas
10
15
NUS-MRT
RUS-MRT
10
SUS-MRT
flops
0
10 20 30 40 50 60 70 80
number of antennas

user scheduling and MRT precoding
because as compared to the scheduling complexity the rising N 3 term is very small and
doesn’t contribute a significant flop. SUS generate a much higher complexity than the
two scheduling schemes.
Figure 5.14 shows how the computational complexity of JASUS algorithm varies with
different number of scheduled users when MRT precoding is implemented. The top plot
shows when MRT is used with NUS and RUS scheduling types. The computational com-
plexity for two user scheduling is rising linearly at slow rate with number of scheduled
users. It is clear that RUS shows the least computational complexity. The bottom plot
compares the computational complexity of the three scheduling schemes with MRT pre-
coding in JASUS algorithm. The SUS shows almost constant computational complexity
throughout all values of K. This is due to the huge value of SUS computational com-
plexity and the varying number of scheduled users doesn’t contribute significant
amount complexity relative to SUS complexity to affect the plot.
Generally, we can conclude that the complexity of JASUS algorithm becomes huge under
10 7
2
NUS-inversion
RUS-inversion
1.5
flops
0.5
0
10 11 12 13 14 15 16 17 18 19 20
number of scheduled user(N=K)
10 9
6
4
NUS-inversion
flops
RUS-inversion
SUS-inversion
2
0
10 11 12 13 14 15 16 17 18 19 20
number of scheduled users (N=K)

user scheduling and channel inversion precoding
the implementation of SUS. In contrary to this RUS results least computational com-
plexity. NUS is much better in reducing the computational complexity of the JASUS
algorithm.
5.3.2 Computational complexity comparison of NUS and

SUS
Based on table 4.1 we plot the complexity of the two user scheduling schemes (NUS and
SUS) for variable number of antennas to show how the computational complexity of the
these user scheduling increases as the number of antennas increase. Since computational
complexity of RUS is independent of M, it will not show increasing property. For
this reason it is not included in the following plots. The number of user is X=2 and
the number of antennas varied between 10 to 60.
10
4
3 NUS-MRT
RUS-
flops
MRT
1
10 11 12 13 14 15 16 17 18 19 20
number of scheduled
users(K=N)
109
6
4 NUS-MRT
RUS-
flops
MRT
SUS-
MRT
0
10 11 12 13 14 15 16 17 18 19 20
number of scheduled
users(K=N)

user scheduling and MRT precoding
From figure 5.15 the top subplot shows how the computational complexity of SUS in-
creases with the number of antennas. It is apparent that it is increasing rapidly in
thousands scale as antenna number increase. The middle plot shows how the computa-
tional complexity NUS is increasing with the number of antennas. From the plots we can
observe the computational complexity of NUS is increasing in tens of scale. It is rising lin-
early at very low rate as compared to SUS. This result confirmed that the computational
complexity of SUS is huge for massive MIMO systems as compared to NUS.
5
10
4
flops
2
SUS
0
10 15 20 25 30 35 40 45 50 55 60
number of antennas(M)
100
flops
NUS
50
0
10 15 20 25 30 35 40 45 50 55 60
5
10
4
flops
2
NUS
SUS
0
10 15 20 25 30 35 40 45 50 55 60
Figure 5.15: Asymptotic computational complexity of NUS and SUS without JASUS
algorithm
Chapter 6
Conclusion and Future
Work
In this thesis we present the recent and back ground works in the area of joint antenna
selection and user scheduling for massive MIMO systems in chapter two. In chapter three
we described the cellular, channel and signal models we apply. Also in this chapter we
present the general overview how JASUS works to find optimal set of antennas and users.
In chapter four we made modifications and justifications of algorithms. Plotting set up
and simulation results with discussion and interpretation are described in chapter five.
In this chapter we present the conclusion we made about this work and the research gaps
that should be investigated further to the future.
6.1 Conclusion
Due to the huge computational complexity generated by semi-orthogonal user scheduling

when implemented in the JASUS algorithm, it is necessary to apply lower complexity user
scheduling schemes in massive MIMO systems. In this research work we apply two lower
complexity user scheduling schemes. The first is norm based user scheduling which select
user based on the user channel vector 2-norm.The second is the random user scheduling
which picks user randomly without considering users channel condition. Their spectral
efficiency performance is compared with the performance of SUS in the JASUS algorithm.
Simulation results reveals that norm based user scheduling shows a higher spectral effi-
ciency performance than semi-orthogonal user scheduling when used in JASUS algorithm
under all scenarios and all precoding techniques. The spectral efficiency performance gap
between NUS and SUS is in the order of fractions through out all SNR values. In terms
71
Chapter 6. Conclusion and Future 72
of computational complexity the JASUS algorithm with SUS results a much higher com-
putational complexity than JASUS with NUS. NUS in JASUS algorithm results better
spectral efficiency with much lower complexity than SUS. RUS with its least complexity
in the JASUS algorithm shows a spectral efficiency performance decrement than SUS
and NUS. It shows around 2 and 3 bits/s/Hz performance decrement than the other
scheduling schemes. Unlike NUS and SUS, RUS doesn’t exploit the benefits of multi-user
diversity.
Therefore, we can conclude that in massive MIMO systems selecting users based on their
channel magnitude results a significant reduction in computational complexity with a
negligible spectral efficiency improvement when applied in a greedy based antenna selec-
tion in comparison with a semi-orthogonal user scheduling .
We also made a performance evaluation of the three precoding types when they are im-
plemented in JASUS algorithm. With a negligible performance gap MMSE out performs
ZF precoding in low SNR values while at higher SNR values ZF shows better performance
than MMSE. The better performance of ZF at higher values is a result of water filling
power allocation. With the moderate transmit SNR value we used, MMSE outperform
ZF in most considered scenarios proving that MMSE is a suitable precoding techniques
in a poor channel conditions. But spectral efficiency evaluation as a function of antenna
number shows that ZF approaches the spectral efficiency performance of MMSE when
the antenna number becomes large as compared to the user number. This is due to the
fact that simple linear processing techniques become optimal as the number of antennas
goes to large. MRT shows the lowest spectral efficiency among all precoding types. Also
the benefits from spatial selectivity and multi-user diversity is poor in MRT precoding.
6.2 Future
Work
In this work we investigate the performance of antenna and user selection algorithm which
select antennas greedily and schedule users at the same time using different user schedul-
ing and precoding techniques. Although performance evaluation is done by considering
different scenarios and assumptions, there are still cases and considerations that should
be investigated further. Here we present some of the research gaps and directions that
needs future investigation.
Chapter 6. Conclusion and Future 73
• In this thesis we prove that the performance of NUS is better than SUS when im-
plemented with a greedy based antenna selection with a limited number of antennas
at the base station. In fact the users channel correlation decreases as the number
of antennas increases making the low complexity user selection strategies like RUS
optimal. Fortunately, massive MIMO concept allows the usage of extended number
of antennas as compared to the number of users it serves at the same time. There-
fore, future works should be done on this work by increasing the antenna number
at greater scale to assess the performance of this low complexity user scheduling
schemes.
• Also our work depends on i.i.d rayleigh fading channel model by assuming antennas
at the base station are uncorrelated. In fact most of the time this may not be the
scenario in real world implementation. Thereby, a more realistic channel models
which take into account antenna correlation and pathloss should be considered in
future works.
• We made performance evaluation by considering the number of scheduled user

is equal to the number of selected antenna in this thesis. But further
investigation can be done by scheduling less number of users than the selected
antenna number.
• We implement an optimal and more complex a water filling power allocation

for zero-forcing precoding. When antenna number goes to large user channel
becomes more orthogonal and zero forcing can easily directs user signal to the
intended user without costing signal power degradation. Therefore, we recommend
performance evaluation and comparison of zero-forcing and MMSE precodings with
equal power allocation policy in the JASUS algorithm using larger number of
antennas.
Bibliography
[1] Andreas F Molisch. Wireless communications, volume 34. John Wiley & Sons, 2012.
[2] Andrea Goldsmith. Wireless communications. Cambridge university press, 2005.
[3] David Tse and Pramod Viswanath. Fundamentals of wireless communication.

Cam- bridge university press, 2005.
[4] Taesang Yoo and Andrea Goldsmith. On the optimality of multiantenna broad-
cast scheduling using zero-forcing beamforming. IEEE Journal on selected areas in
communications, 24(3):528–541, 2006.
[5] Quentin H Spencer, Christian B Peel, A Lee Swindlehurst, and Martin Haardt. An
introduction to the multi-user mimo downlink. IEEE communications Magazine, 42
(10):60–67, 2004.
[6] Muhammad RA Khandaker and Kai-Kit Wong. Signal processing for massive mimo
communications. In Academic Press Library in Signal Processing, Volume 7, pages
367–401. Elsevier, 2018.
[7] Hien Quoc Ngo, Erik G Larsson, and Thomas L Marzetta. Massive mu-mimo down-
link tdd systems with linear precoding and downlink pilots. In 2013 51st Annual
Allerton conference on communication, control, and computing (Allerton),
pages
293–298. IEEE, 2013.
[8] Weiqiang Tan, Wei Huang, Xi Yang, Zheng Shi, Wen Liu, and Lisheng Fan.
Multiuser precoding scheme and achievable rate analysis for massive mimo system.
EURASIP Journal on Wireless Communications and Networking, 2018(1):210,
2018.
74
Bibliography 75
[9] Kan Zheng, Long Zhao, Jie Mei, Bin Shao, Wei Xiang, and Lajos Hanzo. Survey
of large-scale mimo systems. IEEE Communications Surveys & Tutorials, 17(3):
1738–1760, 2015.
[10] Hien Quoc Ngo. Massive MIMO: Fundamentals and system designs, volume 1642.
Linkoping University Electronic Press, 2015.
[11] Erik G Larsson, Ove Edfors, Fredrik Tufvesson, and Thomas L Marzetta. Massive
mimo for next generation wireless systems. IEEE communications magazine, 52(2):
186–195, 2014.
[12] Xiang Gao. Doctoral thesis: Massive MIMO in real propagation environments. PhD
thesis, Lund University, 2016.
[13] Mouncef Benmimoune, Elmahdi Driouch, Wessam Ajib, and Daniel Massicotte. Joint
transmit antenna selection and user scheduling for massive mimo systems. In
2015
IEEE Wireless Communications and Networking Conference (WCNC), pages 381–
386. IEEE, 2015.
[14] Junling Mao, Jinchun Gao, Yuanan Liu, and Gang Xie. Simplified semi-orthogonal
user selection for mu-mimo systems with zfbf. IEEE Wireless Communications
Let- ters, 1(1):42–45, 2012.
[15] Eduardo Castaneda, Adao Silva, Atilio Gameiro, and Marios Kountouris. An
overview on resource allocation techniques for multi-user mimo systems. IEEE Com-
munications Surveys & Tutorials, 19(1):239–284, 2016.
[16] Pu-Hsuan Lin and Shang-Ho Tsai. Performance analysis and algorithm designs for
transmit antenna selection in linearly precoded multiuser mimo systems. IEEE
transactions on Vehicular Technology, 61(4):1698–1708, 2012.
[17] Xi Zhang, Eduard A Jorswieck, Bjorn Ottersten, and Arogyaswami Paulraj. User
selection schemes in multiple antenna broadcast channels with guaranteed perfor-
mance. In 2007 IEEE 8th Workshop on Signal Processing Advances in
Wireless Communications, pages 1–5. IEEE, 2007.
[18] Xiang Gao, Ove Edfors, Jianan Liu, and Fredrik Tufvesson. Antenna selection in
measured massive mimo channels using convex optimization. In 2013 IEEE
globecom workshops (GC Wkshps), pages 129–134. IEEE, 2013.
Bibliography 76
[19] Hui Li, Lingyang Song, and Mérouane Debbah. Energy efficiency of large-scale
multiple antenna systems with transmit antenna selection. IEEE Transactions on
Communications, 62(2):638–647, 2014.
[20] Feng Wang and Marek E Bialkowski. A joint design of transmit antenna selection
and multiuser scheduling for multiuser mimo systems employing block
diagonalization precoding scheme. In 2011 International Conference on Wireless
Communications and Signal Processing (WCSP), pages 1–5. IEEE, 2011.
[21] Guozhen Xu, An Liu, Wei Jiang, Haige Xiang, and Wu Luo. Joint user scheduling
and antenna selection in distributed massive mimo systems with limited backhaul
capacity. China communications, 11(5):17–30, 2014.
[22] Xin Liu and Xianbin Wang. Efficient antenna selection and user scheduling in 5g
massive mimo-noma system. In 2016 IEEE 83rd Vehicular Technology
Conference (VTC Spring), pages 1–5. IEEE, 2016.
[23] Maryam Olyaee, Mohsen Eslami, and Javad Haghighat. An energy-efficient joint
antenna and user selection algorithm for multi-user massive mimo downlink. IET
Communications, 12(3):255–260, 2017.
[24] Haojin Li, Haijun Zhang, Dong Li, Yupei Liu, and Arumugam Nallanathan. Joint
antenna selection and user scheduling in downlink multi-user mimo systems. In
2018 IEEE 4th International Conference on Computer and Communications (ICCC),
pages 1072–1076. IEEE, 2018.
[25] Yuhan Dong, Yuanyuan Tang, and Kai Zhang Shenzhen. Improved joint
antenna selection and user scheduling for massive mimo systems. In 2017
IEEE/ACIS 16th International Conference on Computer and Information Science
(ICIS), pages 69–74. IEEE, 2017.
[26] Tasher Ali Sheikh, Joyatri Bora, and Anwar Hussain. A survey of antenna and user
scheduling techniques for massive mimo-5g wireless system. In 2017 International
Conference on Current Trends in Computer, Electrical, Electronics and
Communi- cation (CTCEEC), pages 578–583. IEEE, 2017.
[27] Claire Masterson. Massive mimo and beamforming: the signal processing behind the
5g buzzwords. 10 Massive MIMO and Beamforming: The Signal Processing
Behind the 5G Buzzwords, page 10, 2017.
[28] Zheng Kan, Ou Suling, and X Yin. Massive mimo channel models: a survey. Int. J.
Antennas Propag, 2014(1), 2014.
Bibliography 77
[29] Bruno Clerckx and Claude Oestges. MIMO wireless networks: channels, techniques
and standards for multi-antenna, multi-user and multi-cell systems. Academic Press,
2013.
[30] Emil Bjornson, Luca Sanguinetti, Jakob Hoydis, and Mérouane Debbah.
Optimal design of energy-efficient multi-user mimo systems: Is massive mimo the
answer? IEEE Transactions on Wireless Communications, 14(6):3059–3075, 2015.
[31] Luca Sanguinetti and H Vincent Poor. Fundamentals of multi-user mimo commu-
nications. In New Directions in Wireless Communications Research, pages 139–173.
Springer, 2009.
[32] Jianqi Wang, David J Love, and Michael D Zoltowski. User selection with zero-
forcing beamforming achieves the asymptotically optimal sum rate. IEEE
Transac- tions on Signal Processing, 56(8):3713–3726, 2008.
[33] Christian B Peel, Bertrand M Hochwald, and A Lee Swindlehurst. A vector-

perturbation technique for near-capacity multiantenna multiuser communication-
part i: channel inversion and regularization. IEEE Transactions on
Communications,
53(1):195–202, 2005.
[34] Xi Zhang, Eduard A Jorswieck, Bjorn Ottersten, and Arogyasvsami Paulraj. On

the asymptotic optimality of opportunistic norm-based user selection with hard sinr
constraint. EURASIP Journal on Advances in Signal Processing, 2009:1–12, 2009.
[35] Stephen Boyd and Lieven Vandenberghe. Introduction to applied linear

algebra:
vectors, matrices, and least squares. Cambridge university press, 2018.
[36] William Ford. Numerical linear algebra with applications: Using MATLAB.
Aca- demic Press, 2014.
[37] Haijing Liu, Hui Gao, Shaoshi Yang, and Tiejun Lv. Low-complexity downlink user
selection for massive mimo systems. IEEE Systems Journal, 11(2):1072–1083, 2015.
[38] Hemanth Prabhu, Joachim Rodrigues, Ove Edfors, and Fredrik Rusek. Approxi-
mative matrix inverse computations for very-large mimo and applications to linear
pre-coding systems. In 2013 IEEE Wireless Communications and Networking
Con- ference (WCNC), pages 2710–2715. IEEE, 2013.
Bibliography 78
[39] Sebastian Wagner, Romain Couillet, Mérouane Debbah, and Dirk TM Slock.
Opti- mal training in large tdd multi-user downlink systems under zero-forcing and
regularized zero-forcing precoding. In 2010 IEEE Global Telecommunications
Conference
GLOBECOM 2010, pages 1–5. IEEE, 2010.
Appendix
Matrix Norm
The norm of mXn matrix A, denoted ||A||, is the square root of the sum of the squares
of its entries, v
u
uXm X
n
||A|| = t (1)
A2 ij
i=1 j=1
Where Aij is entries at the ith row and j th column of matrix A.
the norm in equation 1 is called the Frobenius norm, after the mathematician Ferdinand
Georg Frobenius, and is usually denoted with a subscript, as ||A||F . The vector norm
(frobenius norm) is the special case of equation 1 when n = 1. Let a1 , a2 , ......an are the
column vectors of of matrix A, the frobenius norm of A can be calculated as
q
||A|| = ||a21 || + ..... + ||an2 || (2)
79

Thesisdoc Gizachew

Uploaded by

Copyright:

Available Formats

Thesisdoc Gizachew

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Thesisdoc Gizachew

Uploaded by

Copyright:

Available Formats

Bahir Dar University

Bahir Dar Institute of Technology

School of Graduate Studies

Faculty of Electrical and Computer Engineering

Performance Analysis of Joint

A Thesis Submitted to the School of Graduate Studies of Bahir Dar

Bahir Dar University

Bahir Dar Institute of Technology

School of Graduate Studies

Faculty of Electrical and Computer Engineering

Performance analysis of Joint

Student Name: Gizachew Worku

Signature: Date of submission: 10/22/2020

Advisor Name: Dr.Fikreselam Gared (Assct. Professor)

Advisor’s Signature: Date : 10/22/2020

However, all algorithms implement a semi-orthogonal user scheduling scheme to schedule

Key Words: Ma-MIMO, JASUS, User scheduling, Precoding, Computational

List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viii

3 System And Channel Model . . . . . . . . . . . . . . . . . . . . . . . . . . 23

3.4.1 Zero-Forcing Beam Forming (ZFBF) . . . . . . . . . . . . . . . . 32

4 Joint Antenna Selection And User Scheduling . . . . . . . . . . . . . . 38

5 Simulation Results And Discussions . . . . . . . . . . . . . . . . . . . . . 48

6 Conclusion and Future Work . . . . . . . . . . . . . . . . . . . . . . . . . 71

6.2 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72

Figure 1.1 Block diagram of a multiple-input multiple-output system [1] . . 1

Figure 5.1 Block diagram of simulation process . . . . . . . . . . . . . . . . . 49

Figure 5.6 Spectral efficiency performance comparison of precodings techniques

Table 4.1 Simplified computational complexity of user scheduling schemes . . 46

Table 5.1 System design assumption and considered parameter values . . . . 50

AWGN Additive White Gaussian Noise

RZF Regularized Zero Forcing

|.| Absolute value (cardinality of a set)

iopt Index of selected antenna

In spatial multiplexing an original high-rate data stream is multiplexed into several

The same data stream is transmitted or received through multiple antennas to

Figure 1.2: An illustration of a multi-user MIMO downlink [5]

1.1.1 Large Scale (Massive)

• In Massive MIMO, simple low-complexity signal processing algorithms are capable

1.2 Statement Of The Problem

• Evaluating the JASUS algorithm SE performance for each combination of precoding

• Evaluating the JASUS algorithm computational complexity for each combination

• Comparing the performance and draw conclusion

1.4 Research Methodology

Figure 1.5: Flow chart of JASUS algorithm

• Determining the channel model and signal model

• Developing the capacity of MU-MIMO for each precoding case

• Determining the computational complexity of each precoding type

• Analysis of the mathematical problem formulation of JASUS

• Modifying the problem formulation based on the precoding types we used

• Analysis of the user selection techniques

• Determining the computational complexity of each user selection techniques

• Determining the computational complexity that arise from JASUS algorithm by

Figure 1.6: Block diagram that shows flow of works

• Identifying which precoding technique and user selection scheme is applied at a

• Generating channel matrix (channel realization) based on the channel model

• Averaging the sum-rates that result from all channel realization

Result comparison and interpretation

• Asymptotic computational complexity comparison of JASUS algorithm for different

1.5 Thesis Contribution