Thesis Title - Department of Electrical Engineering
-
Upload
khangminh22 -
Category
Documents
-
view
1 -
download
0
Transcript of Thesis Title - Department of Electrical Engineering
CITY UNIVERSITY OF HONG KONG
妆䉬㢢喽㦌
Maximum Sum Rate of Aloha Networks
磾胟Aloha䙝䖲䀛斁㦌䏚嚄䧇䀛惯䛄
Submitted to
Department of Electronic Engineering
秵敦䋳㢧奘
in Partial Fulfillment of the Requirements
for the Degree of Doctor of Philosophy
揳㟆喬墠
by
LI Yitong
䡯炍埕
June 2016
䄓䤢憘䤸兣䤸挛
Abstract
Aloha, one of the most representative random-access protocols, has
elicited much attention over the past four decades. Aloha is applied
extensively in satellite communications, cellular systems, wireless ad-
hoc networks, and machine-to-machine (M2M) networks because of its
simplicity and distributed nature.
Numerous studies have analyzed the performance of Aloha network-
s. Most of them focused on network throughput performance, which
measures the time-average of the number of successfully decoded pack-
ets. However, in reality, data rate performance should be given more
concern than network throughput performance. Although efforts have
been exerted to explore the information-theoretic limit of Aloha, how
to characterize the maximum sum rate and properly tune the system
parameters to achieve such a limit remains largely unknown.
This thesis is devoted to the characterization of the maximum sum
rate of Aloha networks under various assumptions on receiver struc-
tures. The study begins with the capture model, with which each pack-
et is decoded independently by regarding others as background noise.
A packet can be successfully decoded as long as its received signal-to-
interference-plus-noise ratio (SINR) is above a certain threshold. By
assuming that the nodes are unaware of the instantaneous realization of
channel gains and thus have a uniform information encoding rate, the
network steady-state point in saturated conditions is derived as a func-
tion of the SINR threshold, which determines a fundamental tradeoff
between the information encoding rate and the network throughput.
The maximum sum rate is further obtained as an explicit function of
the mean received signal-to-noise ratio (SNR), which is found to be
much smaller than the ergodic sum capacity of multiple access fading
channels at the high-SNR region.
iii
The capture model is essentially a single-user detector. The anal-
ysis is further extended to incorporate the capacity-achieving receiver
structure, successive interference cancellation (SIC), to identify if the
rate loss is caused by the random access nature or sub-optimality of the
receiver. Two representative SIC receivers are considered. These two
are ordered SIC in which packets are decoded in a descending order of
their received power and unordered SIC in which packets are decoded
in a random order. The analysis shows that compared with the cap-
ture model, the rate gains are significant only with the ordered SIC at
moderate values of the mean received SNR. The rate gap diminishes
at the high-SNR region, and they all have the same high-SNR slope of
e−1, which is below that of the ergodic sum capacity of fading channels.
This condition indicates that the rate loss is significant because of the
uncoordinated random transmissions of nodes. The effects of key fac-
tors, including backoff, power control, multipacket reception (MPR),
and channel fading, on the sum rate performance of Aloha networks
are also discussed to shed light on practical network design.
Acknowledgements
I would like to express my sincere gratitude to all those who
helped me during good and bad moments. My deepest gratitude goes to
my supervisor, Prof. Lin Dai, for guiding me in the area of random-
access networks. She was supportive and dedicated whenever I encoun-
tered difficulties in my research. She inspired me with her rigorous
attitude toward research, her creative ways to solve difficult problems,
and her correct outlook on life. Under her tutelage, I realized that peo-
ple should not resign themselves to fate whenever they experience bad
luck. Misfortune may be a blessing. The important thing is to continue
moving on. It is a great honor to be her student. I will strive to be a
good teacher and pass her knowledge and outlook to my future students.
I also thank my qualifying panel members, Prof. Ping Li and Prof.
Moshe Zukerman, for their valuable comments on my annual reports
and presentation skills. I thank Prof. Moshe Zukerman, Prof. Guan-
rong Chen, and Prof. Man Cho So for their outstanding teaching. I
thank my groupmates Xinghua Sun, Zhiyang Liu, Yayu Gao, Junyuan
Wang, Yue Zhang, and Wen Zhan for their encouragement and gen-
erous help. I also thank Jingjin Wu, Xueqing Gong, Jing Fu, Minrui
Cheng, Wu Yu, Chang Liu, Yi Jiang, Hongbo Yan, Jingwei Zhang, and
Weiping Wu, who helped me stay sane through these difficult years.
Lastly, I thank my parents, brothers, and old friends for supporting
me spiritually in the past eight years.
Contents
Contents i
List of Figures v
Abbreviations vii
Notations ix
1 Introduction 1
1.1 Random Access: Aloha and CSMA . . . . . . . . . . . . . . . . . . 2
1.2 Performance Analysis of Slotted Aloha . . . . . . . . . . . . . . . . 4
1.2.1 Network Throughput . . . . . . . . . . . . . . . . . . . . . . 4
1.2.2 Sum Rate . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.3 Thesis Contributions and Outline . . . . . . . . . . . . . . . . . . . 7
2 System Model 11
2.1 Channel Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.2 Transmitter Model . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.3 Receiver Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.4 Network Throughput and Sum Rate . . . . . . . . . . . . . . . . . . 15
3 Maximum Sum Rate with Capture Model 17
3.1 Previous Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.2 Maximum Network Throughput . . . . . . . . . . . . . . . . . . . . 19
3.2.1 Steady-state Point in Saturated Conditions . . . . . . . . . . 20
3.2.2 Maximum Network Throughput for Given µ and ρ . . . . . 22
3.2.3 Simulation Results . . . . . . . . . . . . . . . . . . . . . . . 23
3.3 Maximum Sum Rate . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.3.1 Optimal SINR Threshold µ∗ . . . . . . . . . . . . . . . . . . 26
3.3.2 Maximum Sum Rate C at High SNR Region . . . . . . . . . 28
3.3.3 Maximum Sum Rate C at Low SNR Region . . . . . . . . . 29
3.3.4 Simulation Results . . . . . . . . . . . . . . . . . . . . . . . 30
3.4 Discussions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.4.1 Effect of Adaptive Backoff . . . . . . . . . . . . . . . . . . . 31
3.4.2 Effect of Power Control . . . . . . . . . . . . . . . . . . . . . 32
3.4.3 Effect of Fading . . . . . . . . . . . . . . . . . . . . . . . . . 36
i
ii Contents
3.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
4 Maximum Sum Rate with Successive Interference Cancellation(SIC) 41
4.1 Previous Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
4.2 Steady-State Point in Saturated Conditions . . . . . . . . . . . . . 43
4.2.1 Conditional Probability of Successful Transmission ri . . . . 44
4.2.1.1 Ordered SIC . . . . . . . . . . . . . . . . . . . . . 45
4.2.1.2 Unordered SIC . . . . . . . . . . . . . . . . . . . . 47
4.2.1.3 Comparison . . . . . . . . . . . . . . . . . . . . . . 48
4.2.2 Steady-State Point in Saturated Conditions . . . . . . . . . 49
4.2.3 Simulation Results . . . . . . . . . . . . . . . . . . . . . . . 50
4.3 Maximum Sum Rate . . . . . . . . . . . . . . . . . . . . . . . . . . 51
4.3.1 Maximum Network Throughput . . . . . . . . . . . . . . . . 52
4.3.2 Maximum Sum Rate . . . . . . . . . . . . . . . . . . . . . . 53
4.3.3 Simulation Results . . . . . . . . . . . . . . . . . . . . . . . 56
4.4 Discussions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
4.4.1 Effect of MPR . . . . . . . . . . . . . . . . . . . . . . . . . . 57
4.4.2 Rate Loss Due to Random Access . . . . . . . . . . . . . . . 60
4.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
5 Conclusion and Future Work 63
5.1 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
5.2 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
A Proof of Theorem 3.1 67
B Proof of Theorem 3.2 69
C Proof of Theorem 3.3 73
D Proof of Corollary 3.4 77
E Proof of Corollary 3.5 79
F Derivation of (3.38-3.40) 81
G Derivation of (4.7)-(4.8) 83
H Derivation of (4.13) 87
I Derivation of (4.14) 89
J Derivation of (4.20)-(4.23) 91
K Derivation of (4.27)-(4.28) 93
List of Figures
1.1 Graphic illustration of successful transmission and collision in slot-ted Aloha networks. . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Graphic illustration of successful transmission and collision in CS-MA networks. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.1 Graphic illustration of an n-node slotted Aloha network. . . . . . . 12
2.2 State transition diagram of an individual HOL packet in slottedAloha networks. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.1 Maximum network throughput of slotted Aloha networks with cap-ture model λmax versus (a) SINR threshold µ and (b) mean receivedSNR ρ. n = 50. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.2 Steady-state point of slotted Aloha networks with capture modelpA versus initial transmission probability q0. (a) n = 50. µ = 1 andρ = 10dB. (b) K = 0. µ = 0.01 and ρ = 0dB. . . . . . . . . . . . . . 24
3.3 Network throughput of slotted Aloha networks with capture modelλout versus initial transmission probability q0. (a) n = 50. µ = 1and ρ = 10dB. (b) K = 0. µ = 0.01 and ρ = 0dB. . . . . . . . . . . 25
3.4 (a) Optimal SINR threshold µ∗ and (b) maximum network through-put λµ=µ
∗max of slotted Aloha networks with capture model versus
mean received SNR ρ. . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.5 Maximum sum rate performance of slotted Aloha networks withcapture model. (a) Maximum sum rate C at the high SNR region.(b) Maximum sum rate C at the low SNR region. . . . . . . . . . . 28
3.6 Sum rate of slotted Aloha networks with capture model Rs versusSINR threshold µ under different values of mean received SNR ρ.n = 50. K = 0 and q0 = q∗0. . . . . . . . . . . . . . . . . . . . . . . 30
3.7 Maximum sum rate of slotted Aloha networks with capture modelversus ρ1/ρ2 for a two-group slotted Aloha network. n1 = n2 = 25.K = 0. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.8 Network performance of slotted Aloha with the capture model overAWGN and fading channels. n = 50. (a) Maximum networkthroughput versus mean received SNR. (b) Maximum sum rate ver-sus mean received SNR. . . . . . . . . . . . . . . . . . . . . . . . . 38
v
vi List of Figures
4.1 Conditional probability of successful transmission rn−1 given thatthere are n−1 concurrent packet transmissions versus SINR thresh-old µ in slotted Aloha networks with SIC receivers. n = 20 andρ = 20dB. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
4.2 Steady-state point of slotted Aloha networks with SIC receivers pAversus SINR threshold µ. n = 20. q0 = 0.5 and ρ = 20dB. . . . . . . 50
4.3 Conditional probability of successful transmission ri in slotted Alo-ha networks with SIC receivers versus number of concurrent packettransmissions i. µ = 1. (a) ρ = 0dB. (b) ρ = 20dB. . . . . . . . . . 51
4.4 Steady-state point of slotted Aloha networks with SIC receivers pAversus transmission probability q0. n = 20 and µ = 1. (a) ρ = 0dB.(b) ρ = 20dB. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
4.5 Maximum network throughput of slotted Aloha networks with cap-ture model and SIC receivers λmax versus SINR threshold µ. n = 20and ρ = 20dB. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
4.6 Maximum sum rate of slotted Aloha networks with capture modeland SIC receivers C versus mean received SNR ρ. n = 20. . . . . . 55
4.7 (a) Optimal SINR threshold µ∗ and (b) optimal transmission prob-ability q∗,µ=µ
∗
0 of slotted Aloha networks with capture model andSIC receivers. n = 20. . . . . . . . . . . . . . . . . . . . . . . . . . 55
4.8 Network throughput of slotted Aloha networks with SIC receiversλout versus transmission probability q0. n = 20 and ρ = 20dB. (a)µ = 0.05. (b) µ = 1. . . . . . . . . . . . . . . . . . . . . . . . . . . 57
4.9 Sum rate of slotted Aloha networks with SIC receivers Rs versusSINR threshold µ. n = 20 and q0 = q∗0. (a) ρ = 0dB. (b) ρ = 20dB.(c) ρ = 40dB. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
4.10 Network performance of slotted Aloha with collision model, capturemodel and SIC receivers. n = 20. (a) Maximum network through-put λµ=µ
∗max and (b) maximum sum rate C versus mean received SNR
ρ. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
F.1 Sum rate of slotted Aloha networks with capture model over AWGNchannels versus SINR threshold. n = 50. K = 0 and q0 = q∗0. (a)ρ = 5dB. (b) ρ = 15dB. . . . . . . . . . . . . . . . . . . . . . . . . . 82
J.1dλoutdq0
versus transmission probability q0. n = 20 and ρ = 20dB. (a)
Ordered SIC. (b) Unordered SIC. . . . . . . . . . . . . . . . . . . . 92
K.1 (a) µh and µl versus mean received SNR ρ. (b) C1 and C2 versusmean received SNR ρ. n = 20. . . . . . . . . . . . . . . . . . . . . . 93
Abbreviations
1G first-generation
2G second-generation
3G third-generation
4G fourth-generation
AP Access Point
AWGN Additive White Gaussian Noise
BEB Binary Exponential Backoff
BS Base Station
CDMA Code Division Multiple Access
CSI Channel State Information
CSIT Channel State Information at the Transmitter side
CSMA Carrier Sense Multiple Access
FDMA Frequency Division Multiple Access
HOL Head-of-Line
i.i.d. Independent and identically distributed
LTE Long Term Evolution
M2M Machine-to-Machine
vii
viii Abbreviations
MAC Multiple Access Channel
MMSE Minimum Mean Square Error
MPR Multipacket Reception
OFDMA Orthogonal Frequency Division Multiple Access
PRACH Physical Random Access Channel
SNR Signal-to-Noise Ratio
SIC Successive Interference Cancellation
SIR Signal-to-Interference Ratio
SINR Signal-to-Interference-plus-Noise Ratio
TDMA Time Division Multiple Access
WLAN Wireless Local Area Network
Notations
hk Small-scale fading coefficient of node k
γk Large-scale fading coefficient of node k
gk Channel gain from node k to the receiver
Pk Transmission power of node k
P0 Mean received power
Nt The number of successfully decoded packets in time slot t
n Number of nodes
K Cutoff phase of HOL packets
ρ Mean received SNR
µ SINR threshold
µ∗ Optimal SINR threshold
πT Service rate of each node’s queue
R Information encoding rate of each packet
Rs Sum Rate
p Steady-state probability of successful transmission of HOL packets
pA Network steady-state point in saturated conditions
{qi}i=0,...,K Transmission probabilities of nodes
ix
x Notations
{q∗i }i=0,...,K Optimal transmission probabilities of nodes
rCi Conditional probability of one packet being successfully decoded given that it
has i concurrent packet transmissions for the capture model over fading
channels
rAWGNi Conditional probability of one packet being successfully decoded given that it
has i concurrent packet transmissions for the capture model over AWGN
channels
rNSi Conditional probability of one packet being successfully decoded given that it
has i concurrent packet transmissions for the unordered SIC over fading
channels
rOSi Conditional probability of one packet being successfully decoded given that it
has i concurrent packet transmissions for the ordered SIC over fading
channels
λ Aggregate input rate
λout Network throughput
λCmax Maximum network throughput with the capture model over fading channels
λAWGNmax Maximum network throughput with the capture model over AWGN channels
λNSmax Maximum network throughput with the unordered SIC over fading channels
λOSmax Maximum network throughput with the ordered SIC over fading channels
C Maximum sum rate
CAWGN Maximum sum rate with the capture model over AWGN channels
Ccollision Maximum sum rate with the collision model
Chapter 1
Introduction
Wireless communications have become an inseparable part of our daily life
since Heinrich Hertz demonstrated the existence of electromagnetic waves in 1888.
Driven by the soaring demand of applications, wireless communication systems
have experienced explosive growth over the past century. Diverse wireless systems
including wireless cellular networks, Wi-Fi networks based on the family of IEEE
802.11 standards, wireless sensor networks, and wireless ad-hoc networks have
been developed, offering a wide range of services to enrich peoples lives.
Due to the broadcast nature of wireless channels and spectrum limitation,
a fundamental problem in wireless communication systems is how to provide an
efficient means for multiple nodes to share a single channel. To address this issue,
various multiple access strategies have been developed for different types of wireless
networks, which can be broadly classified into two categories: centralized and
distributed.
The centralized scenario involves the use of a central controller that collects
all information and performs resource optimization for multiple users to satisfy
their performance requirements. For instance, a base station (BS) in cellular sys-
tems serves as the central controller and allocates resources to users located in its
cell. Typical examples of centralized multiple access technologies include Frequen-
cy Division Multiple Access (FDMA) in the first-generation (1G) system, Time
Division Multiple Access (TDMA) in the second-generation (2G) system, Code
Division Multiple Access (CDMA) in the third-generation (3G) system, and Or-
thogonal Frequency Division Multiple Access (OFDMA) in the fourth-generation
(4G) Long Term Evolution (LTE) system.
1
2 Chapter 1
Despite their high efficiency, centralized multiple access strategies entail high
implementation cost and system complexity. Therefore, distributed multiple ac-
cess strategies, which are also known as random access, are more appealing for
low-cost data networks than centralized multiple access strategies. Random-access
networks have no central controller to perform resource allocation among multiple
users. Instead, each user competes for the channel resource. Random access is
widely applied in many wireless networks, such as IEEE 802.11 networks, wire-
less ad-hoc networks, and machine-to-machine (M2M) networks [1–4], because of
minimum coordination and distributed control. The following section provides an
overview of the history and evolution of random access.
1.1 Random Access: Aloha and CSMA
The first random-access network, Aloha, was developed in 1968 to connect
multiple users with a central time-sharing computer in the University of Hawaii
under the leadership of Abramson [5]. The first commercial application of Aloha
was launched in 1976 by Comsat General in the Marisat maritime satellite com-
munications system [6]. Thereafter, Aloha has been widely adopted in cellular
systems for signaling and control purposes.
A graphic illustration of the slotted Aloha scheme is shown in Fig. 1.1. The
principle of slotted Aloha is that when a node has a packet to send, it transmits
the packet at the beginning of a time slot with a certain probability. When more
than one node transmit their packets simultaneously, a collision occurs and none of
them can be successfully decoded.1 Packets involved in collisions are retransmitted
after a random time delay according to the backoff strategy.
Node A
Node B
Node C
Collsion Success Collsion Success TimeIdle Idle
Figure 1.1: Graphic illustration of successful transmission and collision inslotted Aloha networks.
1This assumption on the receiver model is referred to as “classical collision model” and hasbeen widely adopted in early studies on random access.
Chapter 1 3
Note that the channel utilization of slotted Aloha is quit poor as each node
attempts to access the channel regardless of other nodes’ behavior. The maximum
channel utilization of slotted Aloha is shown to be e−1 [7], which indicates that
over 60% of the time is wasted when the network is either in collision or idle states.
If nodes can sense the channel first before transmission, then they may reschedule
the packet transmissions to avoid collisions when they sense that the channel busy
(i.e., the channel is dominated by other nodes’ transmissions). As a result, the
channel utilization can be significantly improved. Such an observation has led to
the development of Carrier Sense Multiple Access (CSMA) [8].
A graphic illustration of the CSMA scheme is shown in Fig. 1.2. With CSMA,
time is divided into multiple mini time slots with length a, where the value of a is
determined by the ratio of channel propagation delay and the packet transmission
time. It was shown in [8] that the maximum channel utilization approaches 1 as a
goes to zero, which is much higher than that of slotted Aloha. As a result, CSMA
has been widely adopted in commercial LANs, such as IEEE 802.11 WLANs.
Node A
Node B
Node C
CollsionSuccess Success Time
a
Idle Idle Success
Figure 1.2: Graphic illustration of successful transmission and collision inCSMA networks.
Despite the huge success in practical systems, it has been long observed that a
random-access network may suffer from significant performance degradation if the
parameters are not properly set. Yet due to the distributed nature, performance
analysis and optimization of random-access networks are much more difficult than
that in the centralized case, in which a central controller performs resource opti-
mization for the entire network. This thesis focuses on the performance optimiza-
tion of slotted Aloha, which is the simplest and one of the most representative
random-access schemes. A brief literature review on the performance analysis of
slotted Aloha networks is presented in the following section.
4 Chapter 1
1.2 Performance Analysis of Slotted Aloha
1.2.1 Network Throughput
The two basic features shared by most random-access schemes are summarized
as follows.
1. Independent Encoding and Uncoordinated Transmissions : Each node inde-
pendently encodes its information, and decides when to transmit. As the
subset of active nodes is random and time-varying, it is normally assumed
to be unknown at the transmitter side.
2. Time-Slotted and Packet-Based : Though it may not seem to be essential or
necessary for random access, a time-slotted and packet-based network has
been assumed in the literature since Abramson’s Aloha was proposed [7].
Due to the uncoordinated transmissions of nodes, not all the transmitted
packets can be successfully decoded in each time slot.
With the above features of random access, the number of successfully decoded
packets would be varying from time to time. As a result, most studies have focused
on the average number of successfully decoded packets per time slot, which is
referred to as the network throughput. Network throughput performance depends
on a series of key factors including the receiver model and protocol design. Early
studies on random access have focused on the classical collision model, which
assumes that collision occurs and none of the packets can be successfully decoded
when more than one node transmit their packets simultaneously. As at most one
packet can be successfully decoded at each time slot with the collision model, the
network throughput, which is also the fraction of time that an effective output
is produced in this case, cannot exceed 1. The maximum network throughput of
slotted Aloha is shown to be only e−1 with the collision model [7].
Despite extensive studies, how to maximize the network throughput has been
an open question for a long time. In Abramson’s landmark paper [7], by modeling
the aggregate traffic as a Poisson random variable with parameter G, the network
throughput of slotted Aloha with the collision model can be easily obtained as
Ge−G, which is maximized at e−1 when G = 1. To enable the network to operate
at the optimum point for maximum network throughput, nevertheless, it requires
Chapter 1 5
the connection between the mean traffic rate G and key system parameters such
as transmission probabilities of nodes, which turns out to be a challenging issue.
Various retransmission strategies have been developed to adjust the transmission
probability of each node according to the number of backlogged nodes to stabilize2
the network [9–12]. Yet most of them were based on the realtime feedback infor-
mation on the backlog size, which may not be available in a distributed network.
Decentralized retransmission control was further studied in [10, 13–15], where al-
gorithms were proposed to either estimate and feed back the backlog size [13, 15],
or update the transmission probability of each node recursively according to the
channel output [10, 14].
The difficulty originates from the modeling of random-access networks. As
demonstrated in [22], the modeling approaches in the literature can be roughly di-
vided into two categories: channel-centric [7–15] and node-centric [16–20, 23, 24].
Channel-centric approaches capture the essence of contention among nodes by
focusing on the state transition process of the aggregate traffic but ignores the
behavior of each nodes queue; thus, the effect of backoff parameters on the perfor-
mance of each single node is not sufficiently examined. In node-centric approaches,
modeling complexity becomes prohibitively high when the interactions among n-
odes queues are considered. To simplify the analysis, a key approximation is to
regard each node’s queue as an independent queuing system with identically dis-
tributed service time. This key approximation has been widely adopted and shown
to be accurate for the performance evaluation of large multi-queue systems [25].
Service time distribution is still crucially determined by the aggregate activities
of head-of-line (HOL) packets of all the nodes, which require proper modeling of
HOL packets behavior.
A unified analytical framework for two representative random-access proto-
cols, Aloha and CSMA, was recently established by L. Dai in [21, 22], where the
network steady-state points were characterized based on the fixed-point equations
of the limiting probability of successful transmission of HOL packets by assum-
ing the classical collision model. Both steady-state points were derived as explicit
functions of key system parameters including the aggregate input rate, the number
of nodes and the transmission probabilities of nodes, which enable the characteri-
zation of stable regions and performance optimization. In this thesis, the proposed
2Note that various definitions of stability have been developed in the literature. A widelyadopted one is that a network is stable if the network throughput is equal to the aggregate inputrate.
6 Chapter 1
analytical framework is further extended to incorporate some more advanced re-
ceiver structures, based on which explicit expressions of the maximum network
throughput and the corresponding optimal transmission probabilities of nodes are
derived.
1.2.2 Sum Rate
The network throughput evaluates how efficient the time is used for successful
transmission. However, it does not reflect how much information can be trans-
mitted in terms of bits per second per Hz. The data rate performance is usually
of more concern than the network throughput in real-life applications. Therefore,
many studies have focused on the sum rate analysis of random-access networks.
Specifically, from the information-theoretic perspective, random access can
be regarded as a multiple access channel (MAC) with a random number of ac-
tive transmitters. It is well known that the sum capacity of an n-user Additive-
White-Gaussian-Noise (AWGN) MAC is determined by the received SNRs, i.e.,
Csum = log2(1+∑n
i=1 SNRi). However, with random access, the number of active
transmitters is a random variable whose distribution is determined by the protocol
and parameter setting. Moreover, to achieve the sum capacity, joint decoding of all
transmitted codewords should be performed at the receiver side, which might be
unaffordable for random-access networks. Therefore, the sum rate performance of
random access is largely dependent on assumptions on access protocol and receiver
design.
A significant amount of effort has been exerted to explore the information-
theoretic limit of random-access networks. For instance, the concept of rate s-
plitting [26] was first introduced to slotted Aloha networks in [27], where a joint
coding scheme was developed for the two-node case. If each node independently
encodes its information, [28] showed that the sum rate performance of slotted Alo-
ha networks can be improved by adaptively adjusting the encoding rate according
to the number of nodes and the transmission probability of each node. [28] and
[27] are based on the assumption of joint decoding of multiple nodes packets at
the receiver side. In [29], each node was assumed to have its own channel state
information (CSI), and the classical collision model was adopted at the receiver
side. The scaling behavior of the sum rate of slotted Aloha as the number of nodes
Chapter 1 7
n goes to infinity was characterized, and shown to be identical to that of the sum
capacity of MAC.
1.3 Thesis Contributions and Outline
Most previous studies on the performance analysis of slotted Aloha focused on
the collision model, which assumes that when more than one node transmit their
packets simultaneously, a collision occurs and none of the packets can be suc-
cessfully decoded. Although it captures the essence of contention among multiple
nodes, the collision model could be overly pessimistic for wireless systems where
a large difference in the received power of nodes may exist as a result of distinct
channel conditions. In this case, a packet can be successfully decoded as long as its
received signal-to-interference-plus-noise ratio (SINR) is above a certain thresh-
old. Such a receiver is referred to as the capture model [38–51]. Efficiency can be
further improved but at the cost of increased receiver complexity if multiple pack-
ets can be jointly decoded by using multiuser detectors, such as Minimum Mean
Square Error (MMSE) and Successive Interference Cancelation (SIC) [27, 28, 52–
56]. Studies have been conducted on the performance analysis of slotted Aloha
with highly advanced receiver structures [27, 28, 38–56]. Although various ana-
lytical models were developed in these studies, many of them rely on numerical
methods to calculate the sum rate under specific settings. How to maximize the
sum rate by optimizing the system parameters remains largely unknown.
This thesis is devoted to the maximum sum rate characterization of slotted
Aloha under various receiver structures. The effects of backoff, power control,
multipacket reception and channel fading on the sum rate performance of slotted
Aloha networks are also discussed. The contributions of this thesis are summarized
as follows.
The analysis begins with the maximum sum rate characterization of slotted
Aloha with the capture model, which is presented in Chapter 3. By assuming that
the nodes are unaware of the instantaneous realization of channel gains and thus
have a uniform information encoding rate, the conditional probability of one pack-
et being successfully decoded given that it has i concurrent packet transmissions,
ri, is derived first. Based on it, the network steady-state point in saturated con-
ditions is further obtained as a function of the SINR threshold, which determines
8 Chapter 1
a fundamental tradeoff between the information encoding rate and the network
throughput. Explicit expressions of the maximum sum rate and the optimal set-
ting are obtained, which show that similar to the ergodic sum capacity of the
multiple access fading channel, the maximum sum rate of slotted Aloha also log-
arithmically increases with the mean received SNR, however, the high-SNR slope
is only e−1.
In Chapter 4, the analysis is further extended to incorporate the capacity-
achieving receiver structure, SIC. The reason why we consider SIC is two-fold.
First of all, as shown in Chapter 3, the high-SNR slope of the maximum sum rate
of slotted Aloha with the capture model is only e−1, which is far below that of the
ergodic sum capacity of multiple access fading channels. Nevertheless, whether the
gap is caused by the random access nature or suboptimality of the receiver (i.e.,
the capture model) is difficult to determine. By adopting the capacity-achieving
receiver structure, we can pinpoint the rate loss due to random access. Moreover,
SIC and the capture model both have the multipacket reception (MPR) capability.
The effect of MPR on the maximum sum rate performance of slotted Aloha can be
further evaluated by comparing the MPR receivers, including SIC and the capture
model, with the classical collision model.
Specifically, two representative SIC receivers, ordered SIC and unordered SIC
are considered. The maximum sum rates for both ordered SIC and unordered
SIC are characterized to address the first issue. Compared to the capture model,
the analysis shows that the rate gains are significant only with the ordered SIC
at moderate values of the mean received SNR. Meanwhile, the rate gap at the
high-SNR region sharply diminishes, and all of them have the same high-SNR
slope e−1, which is much lower than that of the ergodic sum capacity of fading
channels. This indicates that the rate loss is significant due to uncoordinated
random transmissions of nodes. For the second issue, by comparing both the SIC
receivers and the capture model to the classical collision model, it is demonstrated
that in contrast to the significant network throughput improvement brought by
MPR receiver structures at the low SNR region, the rate gain is marginal, and
becomes negligible at the high SNR region.
The effects of key factors, including backoff, power control and channel fad-
ing, on the sum rate performance of Aloha networks are also discussed, which shed
important light on the practical network design. For instance, by considering the
heterogeneous case, where nodes in the same group have identical mean received
Chapter 1 9
SNRs but the SNRs differ from group to group, it is shown that a large SNR
difference among nodes may be beneficial to the sum rate performance, though
serious unfairness is introduced. Moreover, the maximum sum rate with the cap-
ture model over the AWGN channel is characterized to demonstrate the effect of
channel fading. The comparison with that over fading channels shows that fading
always exerts negative effects if CSI is unavailable at the transmitter side.
The remainder of this thesis is organized as follows. Chapter 2 introduces
the system model and basic assumptions of this thesis. Chapter 3 presents the
maximum sum rate analysis of slotted Aloha networks with the capture model.
The analysis is further extended to SIC receivers in Chapter 4. Chapter 5 provides
the conclusions and suggestions for future work.
Chapter 2
System Model
This chapter presents the system model and basic assumptions which will be
used throughout the analysis in the following chapters. It is organized as follows.
Section 2.1 describes the assumptions on the channel model. The transmitter mod-
el and receiver model are introduced in Section 2.2 and Section 2.3, respectively.
The network throughput and sum rate are further defined in Section 2.4.
11
12 Chapter 2
HOL packet
Receiver
Node 1
Node 2
Node 3
Node n
Figure 2.1: Graphic illustration of an n-node slotted Aloha network.
Consider a slotted Aloha network where n nodes transmit to a single receiver,
as Fig. 2.1 illustrates.1 All the nodes are synchronized and can start a transmis-
sion only at the beginning of a time slot. We assume perfect and instant feedback
from the receiver and ignore the subtleties of the physical layer such as the switch-
ing time from receiving mode to transmitting mode and the delay required for
information exchange.
2.1 Channel Model
Let gk denote the channel gain from node k to the receiver, which can be
further written as gk = γk · hk, where hk is the small-scale fading coefficient of
node k which varies from time slot to time slot and is modeled as a complex
Gaussian random variable with zero mean and unit variance. The large-scale
fading coefficient γk characterizes the long-term channel effect such as path loss
and shadowing. Due to the slow-varying nature, the large-scale fading coefficients
are usually available at the transmitter side through channel measurement. Let us
first assume that power control is performed to overcome the effect of large-scale
fading.2 Specifically, denote the transmission power of node k as Pk. Then we
1Note that the MAC scenario considered in this thesis should be distinguished from the ad-hoc scenario which has been extensively studied in recent years [30–36]. In contrast to theMAC where multiple nodes transmit to a common receiver, multiple transmitter-receiver pairsexist in the ad-hoc case. Representative applications of the former one include cellular systemsand IEEE 802.11 networks, where in each cell/basic-service-set, multiple users transmit to thebase-station/access-point. The latter is usually considered in a wireless ad-hoc network, such aswireless sensor networks.
2In practical systems such as cellular systems, the base-station sends a pilot signal periodicallyfor all the users in its cell to measure their large-scale fading gains and adjust their transmissionpower accordingly to maintain constant mean received power. This process is usually referredto as open-loop power control.
Chapter 2 13
T 1 2 K0
01 q
0 tq p0 tq p
0 (1 )tq p 1(1 )tq p
11 q 21 q 1 K tq p
1 tq p 2 tq p
01 q
K tq p
0 (1 )tq p
...2 (1 )tq p 1(1 )K tq p
Figure 2.2: State transition diagram of an individual HOL packet in slottedAloha networks.
have
Pk · |γk|2 = P0. (2.1)
In this case, each node has the same mean received SNR ρ = P0/σ2. To demon-
strate the effect of power control, this assumption will be relaxed in Section 3.3,
where the analysis is extended to incorporate distinct mean received SNRs.
2.2 Transmitter Model
An n-node buffered slotted Aloha network is essentially an n-queue-single-
server system whose performance is determined by the aggregate activities of HOL
packets. The behavior of each HOL packet can be modeled as a discrete-time
Markov process shown in Fig. 2.2. Specifically, a fresh HOL packet is initially in
State T, and moves to State 0 if it is not transmitted. Define the phase of a HOL
packet as the number of unsuccessful transmissions it experiences. A phase-i HOL
packet stays in State i if it is not transmitted. Otherwise, it moves to State T if its
transmission is successful, or State min(K, i+1) if the transmission fails, where K
denotes the cutoff phase. Note that the cutoff phase K can be any non-negative
integer. When K = 0, States 0 and K in Fig. 2.2 would be merged into one state,
i.e., State 0.
In Fig. 2.2, pt denotes the probability of successful transmission of HOL
packets at time slot t. The steady-state probability distribution of the Markov
chain in Fig. 2.2 can be further obtained as
πT = 1∑K−1i=0
(1−p)iqi
+(1−p)KpqK
, (2.2)
14 Chapter 2
andπ0 = 1−pq0pq0
πT . K = 0
π0 = 1−q0q0πT , πi = (1−p)i
qiπT , i = 1, . . . , K − 1, πK = (1−p)K
pqKπT . K ≥ 1,
(2.3)
where p = limt→∞
pt. Note that πT is the service rate of each node’s queue as the
queue has a successful output if and only if the HOL packet is in State T.
Throughout the thesis, we assume that the transmitters are unaware of the
instantaneous realizations of the small-scale fading coefficients. As a result, each
node independently encodes its information at a constant rate R bit/s/Hz. Assume
that each codeword lasts for one time slot, i.e., no coding over successive packets.
Moreover, we consider the saturated conditions where each node always has packets
in its buffer.
2.3 Receiver Model
At the receiver side, let
µ = 2R − 1 (2.4)
denote the signal-to-interference-plus-noise ratio (SINR) threshold. A packet can
be successfully decoded if its received SINR is above the threshold µ.3 In Chapter
3 and 4, we will consider the following two receiver structures respectively:
1) Capture model: With the capture model, each packet is decoded inde-
pendently by treating others as background noise. A packet can be successfully
decoded as long as its received signal-to-interference-plus-noise ratio (SINR) is
above a certain threshold.
2) SIC: With the SIC receivers, once a packet is successfully decoded, it
is subtracted from the aggregate received signal before decoding other packets.
There are many variants of the SIC receiver. Specifically, two representative SIC
receivers are considered:
3More specifically, denote the received SINR of node k as ηk. If log2(1 + ηk) > R, then byrandom coding the error probability of node k’s packet is exponentially reduced to zero as theblock length goes to infinity. Here we assume that the block length is sufficiently large such thatnode k’s packet can be successfully decoded as long as ηk ≥ µ. Note that this is an ideal case. Inpractice, the threshold not only depends on the information encoding rate R, but also the errorprobability that is determined by the coding and decoding schemes.
Chapter 2 15
• Ordered SIC: The most widely used SIC assumes that packets are decoded
in descending order of the received power. In each iteration, the packet with
the highest received power is decoded, and is subtracted from the aggregate
received signal if it is successfully decoded. Otherwise, the decoding process
terminates4.
• Unordered SIC: As a comparison benchmark, we also consider the case that
packets are decoded in a random order. In each iteration, a packet is decod-
ed and is subtracted from the aggregate received signal if it is successfully
decoded. Otherwise, move to the next packet and repeat the process until
all the packets are decoded.
2.4 Network Throughput and Sum Rate
In this thesis, we focus on the long-term system behavior and define the sum
rate as the time-average of the received information rate:
Rs = limT→∞
1
T
T∑t=1
R ·Nt, (2.5)
where Nt denotes the number of successfully decoded packets in time slot t, which
is a time-varying variable. Define the network throughput as the average number
of successfully decoded packets per time slot, which can be written as
λout = limT→∞
1
T
T∑t=1
Nt. (2.6)
According to (2.5-2.6), we then have
Rs = R · λout. (2.7)
Note that the sum rate is closely dependent on the transmission probabilities
{qi}i=0,...,K and the SINR threshold µ. In this thesis, we aim at maximizing the
sum rate by optimally choosing the transmission probabilities {qi}i=0,...,K and the
SINR threshold µ: C = maxµ,{qi}Rs. By combining (2.4) and (2.7), the maximum
4If the packet with the highest received power cannot be successfully decoded, then otherpackets cannot be successfully decoded either.
16 Chapter 2
sum rate can be further written as
C = maxµ
λmax log2(1 + µ), (2.8)
where λmax = max{qi} λout denotes the maximum network throughput.
Both the information encoding rate R and the network throughput λout de-
pend on the SINR threshold µ. Intuitively, by reducing µ, more packets can be
successfully decoded at each time slot, yet the information encoding rate becomes
smaller. Therefore, the SINR threshold µ should be carefully chosen to maximize
the sum rate. Note that the network throughput λout is also crucially determined
by the protocol design and backoff parameters. In the following chapters, we will
consider the capture model and SIC separately, and characterize the maximum
sum rate of slotted Aloha networks in each case.
Chapter 3
Maximum Sum Rate with
Capture Model
This chapter characterizes the maximum sum rate of Aloha networks with
the capture model. This chapter is organized as follows. Section 3.1 presents
a literature review of previous related work. The steady-state point in saturat-
ed conditions is derived in Section 3.2, based on which the maximum network
throughput is further characterized. Maximum sum rate is characterized in Sec-
tion 3.3. The effects of backoff, power control and channel fading on the sum rate
performance of slotted Aloha networks are discussed in Section 3.4. Section 3.5
summarizes this chapter.
17
18 Chapter 3
3.1 Previous Work
It has been mentioned in Chapter 1 that early studies on random-access have
focused on the classical collision model, which assumed that a packet transmission
is successful only if there are no concurrent transmissions. Though an elegant and
useful simplification of the receiver, the classical collision model could be overly
pessimistic if there exists a large difference of received power. It was first point-
ed out by Roberts in [37] that even with multiple concurrent transmissions, the
strongest signal can be successfully detected as long as the signal-to-interference
ratio (SIR) is sufficiently high. This condition is referred to as the “capture ef-
fect”, which has been extensively studied in [38–51]. With the capture model,
each node’s packet is decoded independently by regarding the packets of others
as background noise. A packet can be successfully decoded as long as its received
signal-to-interference-plus-noise ratio (SINR) is above a certain threshold. In con-
trast to the classical collision model where at most one packet can be successfully
decoded, with the capture model, it is clear that multiple packets can be decoded
simultaneously if the SINR threshold is sufficiently low, which is referred to as
multipacket reception (MPR) capability [47, 51–63].
Many studies have been conducted on performance analysis of slotted Aloha
with the capture model. By assuming Poisson distributed aggregate traffic, for
instance, the network throughput was derived as a function of mean traffic rate
G and SIR in [38–40] under distinct assumptions on channel conditions. Similar
to that in the case of collision model, the maximum network throughput can be
obtained by optimizing G, yet how to properly tune the system parameters to
achieve the maximum network throughput remains unknown. The retransmission
control strategies developed in [9], [13] and [14] were further extended to the cap-
ture model in [43], [45] and [44], respectively. To evaluate the network throughput
performance for given transmission probabilities of nodes, various Markov chains
were also established in [41, 42, 48, 51] to model the state transition of each indi-
vidual user. The computational complexity, nevertheless, sharply increases when
sophisticated backoff strategies are further involved, which renders it extremely
difficult, if not impossible, to search for the optimal configuration to maximize the
network throughput.
In contrast to the network throughput, only few studies focused on the sum
Chapter 3 19
rate performance of slotted Aloha with the capture model [46, 49, 50]. Specifi-
cally, the effects of power allocation and modulation on the sum rate of slotted
Aloha in AWGN channels were analyzed in [46] and [49], respectively. Queueing
stability and channel fading were further considered in [50], where the sum rates
with various cross-layer approaches were derived. Most of them focused on eval-
uating the sum rate under specific settings, where how to maximize the sum rate
remains largely unexplored. As we will demonstrate in this chapter, the sum rate
optimization of slotted Aloha networks can be decomposed into two parts: 1) For
given information encoding rate R, or equivalently, SINR threshold µ, the network
throughput can be maximized by properly choosing backoff parameters (i.e., the
transmission probabilities of nodes). 2) As the information encoding rate and the
maximum network throughput are both functions of the SINR threshold µ, the
sum rate can be further optimized by tuning µ.
Specifically, in this chapter, by extending the unified analytical framework
[21, 22] from the classical collision model to the capture model, the maximum
network throughput is obtained as a function of the SINR threshold µ, based on
which the maximum sum rate is derived by further optimizing the SINR threshold
µ. The effects of backoff, power control and channel fading on the sum rate
performance of slotted Aloha networks are also discussed.
3.2 Maximum Network Throughput
As Fig. 2.1 illustrates, an n-node buffered slotted Aloha network is essen-
tially an n-queue-single-server system whose performance is determined by the
aggregate activities of HOL packets. In this section, we will first derive the net-
work steady-state point in saturated conditions as the single non-zero root of the
fixed-point equation of the steady-state probability of successful transmission of
HOL packets. Then the maximum network throughput will be obtained by opti-
mizing the transmission probabilities of nodes. Finally, simulation results will be
presented to verify the preceding analysis.
20 Chapter 3
3.2.1 Steady-state Point in Saturated Conditions
By regarding an n-node buffered slotted Aloha network as an n-queue-single-
server system, we can see that the network throughput λout is indeed the system
output rate, which is equal to the aggregate input rate λ if each node’s buffer has
a non-zero probability of being empty. As λ increases, the network will eventually
become saturated where each node is busy with a non-empty queue. In this case,
the network throughput is determined by the aggregate service rate, i.e.,
λout = nπT , (3.1)
which, as (2.2) shows, depends on the steady-state probability of successful trans-
mission of HOL packets p. In this section, we will characterize the network steady-
state point in saturated conditions based on the fixed-point equation of p.
Specifically, for HOL packet j, let Sj denote the set of nodes which have
concurrent transmissions. It can be successfully decoded at the receiver side if
and only if its received SINR is above the threshold µ, i.e.,Pj∑
k∈SjPk+σ2 ≥ µ, where
Pk = Pk|gk|2 = P0|hk|2 denotes the received power according to (2.1). Suppose
that |Sj| = i. The steady-state probability of successful transmission of HOL
packet j given that there are i concurrent transmissions, rji , can be then written
as
rji = Pr
{|hj |2∑
k∈Sj|hk|2+
1ρ
≥ µ
}, (3.2)
where ρ = P0/σ2 is the mean received SNR. With hk ∼ CN (0, 1), rji can be easily
obtained as [38, 51]
rji =exp(−µρ
)(µ+1)i
. (3.3)
The right-hand side of (3.3) is independent of j, indicating that all the HOL pack-
ets have the same conditional probability of successful transmission. Therefore,
we drop the superscript j, and write the steady-state probability of successful
transmission of HOL packets p as
p =n−1∑i=0
ri · Pr{i concurrent transmissions}. (3.4)
In saturated conditions, all the nodes have non-empty queues. According to
Chapter 3 21
the Markov chain shown in Fig. 2.2, the probability that the HOL packet is re-
questing transmission is given by πT q0+∑K
i=0 πiqi, which is equal to πT/p according
to (2.3). Therefore, the probability that there are i concurrent transmissions can
be obtained as
Pr{i concurrent transmissions} =
(n− 1
i
)(1− πT
p
)n−1−i (πTp
)i. (3.5)
By substituting (3.3) and (3.5) into (3.4), the steady-state probability of successful
transmission of HOL packets p can be obtained as
p = exp(−µρ
)·(
1− µµ+1· πTp
)n−1 for large n≈ exp
{−µρ− nµ
µ+1· πTp
}, (3.6)
where the approximation is obtained by applying (1 − x)n ≈ exp (−nx) for 0 <
x < 1.1 Finally, by substituting (2.2) into (3.6), we have
p ≈ exp
(−µρ− nµ
µ+1· 1∑K−1
i=0
p(1−p)iqi
+(1−p)KqK
). (3.7)
The following theorem states the existence and uniqueness of the root of the
fixed-point equation (3.7).
Theorem 3.1. The fixed-point equation (3.7) has one single non-zero root pA if
{qi}i=0,...,K is a monotonic non-increasing sequence.
Proof. See Appendix A.
As we can see from (3.7), the non-zero root pA is closely dependent on backoff
parameters {qi}i=0,...,K . Without loss of generality, let qi = q0 · Qi where q0 is the
initial transmission probability and Qi is an arbitrary monotonic non-increasing
function of i with Q0 = 1 and Qi ≤ Qi−1, i = 1, . . . , K. With the cutoff phase
K = 0, or the backoff function Qi = 1, i = 0, . . . , K, for instance, pA can be
explicitly written as
pA = exp(−µρ− nµ
µ+1q0
). (3.8)
1Note that with a small network size, i.e., n ≤ 5 for instance, the approximation error maybecome noticeable. It, nevertheless, rapidly declines as the number of nodes n increases.
22 Chapter 3
3.2.2 Maximum Network Throughput for Given µ and ρ
It has been shown in Section 3.2.1 that the network operates at the steady-
state point pA in saturated conditions. By combining (3.1) and (3.6), the network
throughput at pA can be written as
λout = (µ+ 1) ·(−pA ln pA
µ− pA
ρ
), (3.9)
where pA is an implicit function of the transmission probabilities qi, i = 0, . . . , K,
which is given in (3.7). It can be seen from (3.9) and (3.7) that the network
throughput is crucially determined by the backoff parameters {qi}. In this section,
we focus on the maximum network throughput λmax = max{qi}λout. The following
theorem presents the maximum network throughput λmax and the corresponding
optimal backoff parameters {q∗i }.
Theorem 3.2. For given SINR threshold µ ∈ (0,∞) and mean received SNR
ρ ∈ (0,∞), the maximum network throughput is given by
λmax =
µ+1µ
exp(−1− µ
ρ
)if µ ≥ 1
n−1
n exp(− nµµ+1− µ
ρ
)otherwise,
(3.10)
which is achieved at
q∗i =
q0Qi if µ ≥ 1n−1
1 otherwise,(3.11)
i = 0, . . . , K, where q0 is given by
q0 = µ+1nµ·
{K−1∑i=0
exp(−1−µ
ρ
)[1− exp
(−1−µ
ρ
)]iQi +
[1− exp
(−1−µ
ρ
)]KQK
}. (3.12)
Proof. See Appendix B.
Eq. (3.10) shows that for given SINR threshold µ, the maximum network
throughput λmax is a monotonic increasing function of the mean received SNR ρ.
As ρ→∞, we have
limρ→∞
λmax =
µ+1µe−1 if µ ≥ 1
n−1
n exp(− nµµ+1
)otherwise,
(3.13)
Chapter 3 23
10-4
10-3
10-2
10-1
100
101
102
10-4
10-3
10-2
10-1
100
101
102
1
1n
= 0 dB
= 20 dB
= 10 dB
max
(a)
m = 0.01
m = 1
m = 0.5
m = 5
(dB) r
max
λˆ
(b)
Figure 3.1: Maximum network throughput of slotted Aloha networks withcapture model λmax versus (a) SINR threshold µ and (b) mean received SNR
ρ. n = 50.
which approaches e−1 when µ� 1.
On the other hand, for given mean received SNR ρ, λmax monotonically de-
creases as the SINR threshold µ increases, as Fig. 3.1a illustrates. With a lower
µ, the receiver can decode more packets among multiple concurrent transmissions,
and thus better throughput performance can be achieved. It can be easily shown
that multipacket reception is possible when the SINR threshold µ is sufficiently
small. Specifically, for µ ≥ 1n−1 , λmax > 1 if and only if 1
n−1 ≤ µ < 1e−1 and
ρ > µ
lnµ+1µ−1
. Otherwise, λmax > 1 if and only if ρ > µ
lnn− nµµ+1
. As Fig. 3.1b
illustrates, with n = 50, if the SINR threshold µ = 0.01 < 1n−1 , λmax > 1 when
the mean received SNR ρ > −25.3dB. On the other hand, if µ = 0.5, we have1
n−1 < µ < 1e−1 ≈ 0.582. In this case, λmax > 1 when the mean received SNR
ρ > 7dB.
3.2.3 Simulation Results
In this section, simulation results are presented to verify the preceding anal-
ysis. In particular, we consider a saturated slotted Aloha network with Binary
Exponential Backoff (BEB) [66–69], i.e., qi = q0 · 12i
, i = 0, . . . , K. Section 3.2.1
24 Chapter 3
0.01 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.90
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
1q0
Ap
Simulation
Analysis
Simulation
Analysis
Simulation
Analysis
K = 0 K = 1
K = 3
(a)
0.01 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0.4
0.5
0.6
0.7
0.8
0.9
1
0.3
q0
Ap
Simulation
Analysis
Simulation
Analysis
Simulation
Analysis
n = 30
n = 50 n = 100
(b)
Figure 3.2: Steady-state point of slotted Aloha networks with capture modelpA versus initial transmission probability q0. (a) n = 50. µ = 1 and ρ = 10dB.
(b) K = 0. µ = 0.01 and ρ = 0dB.
has shown that it operates at the steady-state point pA, which is closely deter-
mined by the number of nodes n and the backoff parameters {qi}. The expression
of pA is given in (3.7) and verified by simulation results presented in Fig. 3.2.2
Fig. 3.3 illustrates the corresponding network throughput performance. The
network throughput λout has been derived as a function of pA in (3.9) in Section
3.2.2, which varies with the backoff parameters. As we can see from Fig. 3.3, the
network throughput performance is sensitive to the setting of the initial transmis-
sion probability q0. According to Theorem 3.2, when the SINR threshold µ ≥ 1n−1 ,
the maximum network throughput λmax is achieved when q0 is set to be q0. Oth-
erwise, λmax is achieved with qi=1, i = 0, . . . , K. The expressions of λmax and
the corresponding optimal backoff parameters q∗i are given in (3.10) and (3.11),
respectively, and verified by the simulation results presented in Fig. 3.3.
Note that in spite of the improvement on the maximum network throughput
by reducing the SINR threshold µ, the information encoding rate that can be
supported for reliable communications, i.e., R = log2(1 + µ), is quite low when µ
is small. It is clear that the SINR threshold µ determines a tradeoff between the
network throughput and the information encoding rate. In the next section, we
will further study how to maximize the sum rate by properly choosing the SINR
threshold µ.
2In simulations, the steady-state probability of successful transmission of HOL packets pA isobtained by calculating the ratio of the number of successful transmissions to the total numberof attempts of HOL packets over a long time period, i.e., 108 time slots.
Chapter 3 25
00.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
0.1
0.2
0.3
0.4
0.5
0.6
0.7
maxλ =0.67ˆ
0.040.07 0.15
denotes 0q
q0
out
λˆ
1
Simulation
Analysis
0K =
Simulation
Analysis
1K =
Simulation
Analysis
3K =
(a)
0.01 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10
5
10
15
20
25
=30
35
40
Simulation
Analysis
Simulation
Analysis
30
max
n=22
50
max
n
100
max
n=37
q0
ou
t
n = 30
n = 50 n = 100
Simulation
Analysis
(b)
Figure 3.3: Network throughput of slotted Aloha networks with capture modelλout versus initial transmission probability q0. (a) n = 50. µ = 1 and ρ = 10dB.
(b) K = 0. µ = 0.01 and ρ = 0dB.
3.3 Maximum Sum Rate
In this section, we will derive the maximum sum rate and the corresponding
optimal SINR threshold as functions of the mean received SNR ρ, and discuss
their characteristics at the high SNR and lower SNR regions, respectively.
Specifically, it has been demonstrated in Section 2.4 that the sum rate of
slotted Aloha networks is determined by the information encoding rate R and the
network throughput λout. Section 3.2.2 further shows that if backoff parameters
{qi} are properly selected, the network throughput is maximized at λmax, which
is a function of the SINR threshold µ. By combining (2.8) and Theorem 3.2,
the maximum sum rate can be further written as C = maxµ>0 f(µ), where the
objective function f(µ) is given by
f(µ) =
µ+1µ
exp(−1− µ
ρ
)log2(1 + µ) if µ ≥ 1
n−1
n exp(− nµµ+1− µ
ρ
)log2(1 + µ) otherwise.
(3.14)
The following theorem presents the maximum sum rate C and the optimal SINR
threshold µ∗.
26 Chapter 3
Theorem 3.3. For given mean received SNR ρ ∈ (0,∞), the maximum sum rate
is
C =
µ∗h+1
µ∗hexp
(−1− µ∗h
ρ
)log2(1 + µ∗h) if ρ ≥ ρ0
n exp(− nµ∗lµ∗l+1
− µ∗lρ
)log2(1 + µ∗l ) otherwise,
(3.15)
which is achieved at
µ∗ =
µ∗h if ρ ≥ ρ0
µ∗l otherwise,(3.16)
where µ∗h and µ∗l are the roots of the following equations:
(µ+ 1)µ+1ρ
+1µ = e, (3.17)
and
(µ+ 1)µ+1ρ
+nµ+1 = e, (3.18)
respectively, and
ρ0 =nn−1 ln
nn−1
1−(n−1) ln nn−1
. (3.19)
Proof. See Appendix C.
Note that ρ0 is a monotonic decreasing function of n ∈ [2,∞), and limn→∞ ρ0 =
2. When the number of nodes n is large, ρ0 is close to 3dB.
3.3.1 Optimal SINR Threshold µ∗
Theorem 3.3 shows that to achieve the maximum sum rate, the SINR thresh-
old µ should be carefully selected. Fig. 3.4a illustrates how the optimal SINR
threshold µ∗ varies with the mean received SNR ρ. At the low SNR region, i.e.,
ρ < ρ0, for instance, we can obtain from (3.16) and (3.18) that µ∗ρ<ρ0 = µ∗l ≈
e−W0
(− 1n
)− 1 for large n, where W0(z) is the principal branch of the Lambert W
function [64]. In this case, the effect of the mean received SNR ρ becomes negligi-
ble, and µ∗ρ<ρ0 reduces to a monotonic decreasing function of the number of nodes
n. With a large n, µ∗ρ<ρ0 � 1, implying that multiple packets can be successfully
decoded.
Chapter 3 27
3
(dB)r
*μ
denotes 0r
30n =
50n =
100n =
(a)
3
*
max
μ=μ
λ
30n =
50n =
100n =
(dB)r
denotes 0r
(b)
Figure 3.4: (a) Optimal SINR threshold µ∗ and (b) maximum network
throughput λµ=µ∗
max of slotted Aloha networks with capture model versus meanreceived SNR ρ.
At the high SNR region, we can obtain from (3.16-3.17) that µ∗ρ≥ρ0 = µ∗h ≈eW0(ρ) for large ρ. As we can see from Fig. 3.4a, with ρ � 1, the optimal SINR
threshold µ∗ρ≥ρ0 monotonically increases with the mean received SNR ρ.
By combining (3.16) with Theorem 3.2, we can also obtain the maximum
network throughput with µ = µ∗ as
λµ=µ∗
max =
µ∗h+1
µ∗hexp
(−1− µ∗h
ρ
)if ρ ≥ ρ0
n exp(− nµ∗lµ∗l+1
− µ∗lρ
)otherwise.
(3.20)
Fig. 3.4b illustrates how the maximum network throughput λµ=µ∗
max varies with the
mean received SNR ρ. As we can see from Fig. 3.4b, at the low SNR region, i.e.,
ρ < ρ0, the effect of ρ is negligible, and λµ=µ∗
max,ρ<ρ0 becomes a monotonic increasing
function of the number of nodes n. In this case, the optimal SINR threshold
µ∗ρ<ρ0 = µ∗l is decreased as n grows, and thus more packets can be successfully
decoded, though each at a smaller information encoding rate. For large n, we
have λµ=µ∗
max,ρ<ρ0 ≈ ne−1 according to (3.20).
At the high SNR region, Fig. 3.4a has shown that the optimal SINR threshold
µ∗ρ≥ρ0 = µ∗h is much larger than 1, with which at most one packet can be successfully
decoded at each time slot. Therefore, the maximum network throughput λµ=µ∗
max,ρ≥ρ0
quickly drops below 1, and eventually approaches e−1 as ρ→∞.
28 Chapter 3
5 10 15 20 25 300
0.5
1
1.5
2
2.5
3
High-SNR
approximation
(22)
(28)
(dB)
C(bit/s/Hz)
(a)
-20 -15 -10 -5 00
0.1
0.2
0.3
0.4
0.5
30n
100n
(22)
(29)
3
12loge e
(dB)
C(bit/s/Hz)
Large-n approximation
(b)
Figure 3.5: Maximum sum rate performance of slotted Aloha networks withcapture model. (a) Maximum sum rate C at the high SNR region. (b) Maximum
sum rate C at the low SNR region.
3.3.2 Maximum Sum Rate C at High SNR Region
Similar to Section 3.3.1, let us take a closer look at the maximum sum rate
C at different SNR regions.
With ρ ≥ ρ0, it has been shown in Section 3.3.1 that the optimal SINR
threshold µ∗ρ≥ρ0 = µ∗h ≈ eW0(ρ) for large ρ. The maximum sum rate in this case
can be then approximated by
Cρ≥ρ0 ≈(1 + e−W0(ρ)
)exp
(−1− eW0(ρ)
ρ
)log2(1 + eW0(ρ)), (3.21)
for ρ � 1. As Fig. 3.5a shows, the approximation (3.21) works well when the
mean received SNR ρ is large, i.e., ρ ≥ 15dB. Moreover, a logarithmic increase of
the maximum sum rate C can be observed at the high SNR region. The following
corollary presents the high-SNR slope of C.
Corollary 3.4. limρ→∞C
log2 ρ= e−1.
Proof. See Appendix D.
Recall that the high-SNR slope of the ergodic sum capacity of MAC is 1 when
single-antenna is employed at both the transmitters and the receiver. To achieve
the ergodic sum capacity, however, a joint decoding of all received signals is re-
quired and the codewords should span multiple fading states. With the capture
Chapter 3 29
model, in contrast, each node’s packet is decoded independently by treating oth-
ers’ as background noise at each time slot. When the mean received SNR is high,
at most one packet can be successfully decoded each time due to a large SINR
threshold µ∗ � 1. Corollary 3.4 shows that with the simplified receiver, the high-
SNR slope of the maximum sum rate of slotted Aloha networks is significantly
lower than that of the sum capacity.
3.3.3 Maximum Sum Rate C at Low SNR Region
For ρ < ρ0, it has been shown in Section 3.3.1 that the optimal SINR threshold
µ∗ρ<ρ0 = µ∗l ≈ e−W0
(− 1n
)− 1 for large n. The corresponding maximum sum rate
can be then approximated by
Cρ<ρ0 ≈ −nW0
(− 1n
)· exp
−n(1−eW0
(− 1n
))− e−W0
(−1n
)−1
ρ
log2 e, (3.22)
for n � 1. As we can see from Fig. 3.5b, the approximation (3.22) works well
when the number of nodes n is large. The following corollary further presents the
limiting maximum sum rate as n→∞ at the low SNR region.
Corollary 3.5. limn→∞Cρ<ρ0 = e−1log2 e.
Proof. See Appendix E.
Note that it has been shown in Section 3.3.1 that with ρ < ρ0, the maximum
network throughput λµ=µ∗
max ≈ ne−1, which grows with the number of nodes n
unboundedly. Although more packets can be successfully decoded, the information
carried by each packet decreases as n increases due to a diminishing information
encoding rate, i.e., R = log2(1 + µ∗ρ<ρ0) ≈1n
log2 e for large n. Therefore, as
the number of nodes n → ∞, the maximum sum rate reaches a limit that is
independent of the mean received SNR, as Corollary 3.5 indicates. It is in sharp
contrast to the ergodic sum capacity of MAC which linearly increases with n and
ρ at the low SNR region.
30 Chapter 3
0.02 2 4 6 8 10 12 14 16 18 200
0.2
0.4
0.6
0.8
1
2.9 9.2
0 dBC =0.53
10 dBC =0.73
=1.0215 dBC
Simulation
Analysis
0 dB
Simulation
Analysis
10 dB
Simulation
Analysis
15 dB
sR (bit/s/Hz)
denotes *
Figure 3.6: Sum rate of slotted Aloha networks with capture model Rs versusSINR threshold µ under different values of mean received SNR ρ. n = 50.
K = 0 and q0 = q∗0.
3.3.4 Simulation Results
In this section, simulation results are presented to verify the preceding anal-
ysis. Again we consider a saturated slotted Aloha network with the cutoff phase
K = 0. It has been demonstrated in Section 3.3 that as both the maximum net-
work throughput and the information encoding rate depend on the SINR threshold
µ, the sum rate can be maximized by optimally choosing µ. We can clearly observe
from Fig. 3.6 that the sum rate performance is sensitive to the SINR threshold
µ especially when the mean received SNR ρ is small. To achieve the maximum
sum rate, µ should be properly set according to ρ. The expressions of the optimal
SINR threshold µ∗ and the maximum sum rate C are given in Theorem 3.3, and
verified by the simulation results presented in Fig. 3.6.
3.4 Discussions
So far we have shown that to optimize the sum rate performance of slotted
Aloha networks, the SINR threshold µ and backoff parameters {qi} should be
properly set according to the mean received SNR ρ, and the maximum sum rate
logarithmically increases with ρ with the high-SNR slope of e−1. In this section,
Chapter 3 31
we will further discuss how the performance is affected by key factors such as
backoff, power control and channel fading.
3.4.1 Effect of Adaptive Backoff
Backoff is a key component of random-access networks. It has been shown
in Sections 3.2 and 3.3 that to achieve the maximum sum rate, backoff parame-
ters, i.e., the transmission probabilities {qi} of nodes, should be adaptively tuned
according to the number of nodes n and the mean received SNR ρ.3 In many stud-
ies, however, nodes are supposed to transmit their packets with a fixed probability
[39, 41, 42, 48, 51]. To see how the rate performance of slotted Aloha deterio-
rates without adaptive backoff, let us assume that each node transmits its packet
with a constant probability q at each time slot, i.e., qi = q, i = 0, . . . , K. In this
case, the network steady-state point in saturated conditions can be obtained from
(3.7) as pqi=qA = exp(−µρ− nqµ
µ+1
), and the corresponding network throughput is
λqi=qout = nq exp(−µρ− nqµ
µ+1
), according to (3.9). The sum rate can be then written
as Rqi=qs = nq exp
(−µρ− nqµ
µ+1
)· log2(1 + µ), which is an increasing function of the
mean received SNR ρ.
As ρ→∞, it can be easily obtained that Rqi=qs = limρ→∞R
qi=qs = nq exp
(− nqµµ+1
)·
log2(1 + µ), with the maximum
maxµ
Rqi=qs = nq exp
(−nq
(1−eW0
(− 1nq
)))· log2 e
−W0
(− 1nq
), (3.23)
which is achieved at
µ∗,qi=q = e−W0
(− 1nq
)− 1. (3.24)
Eq. (3.24) shows that the optimal SINR threshold µ∗,qi=q monotonically decreases
as the number of nodes n grows. For large n � 1, it can be obtained from
3Note that for practical random-access networks, the backoff parameters can be updatedthrough the feedback from the common receiver. In IEEE 802.11 networks, for instance, as eachnode associates with the access-point (AP) upon joining the network, the AP can count thenumber of nodes through the MAC header of the frame sent by each node, calculate the optimalbackoff parameters, and broadcast them in the beacon frame periodically. Each node can thenupdate its backoff parameters according to the received beacon frame. Such a feedback-basedupdate process can also be implemented in cellular systems where the base-station serves as thecommon receiver in each cell.
32 Chapter 3
(3.23-3.24) that µ∗,qi=q ≈ 1nq
, and
maxµ
Rqi=qs
n�1≈ e−1log2 e. (3.25)
Recall that it has been shown in Section 3.3.2 that the maximum sum rate increases
with the mean received SNR ρ unboundedly. Here (3.25) indicates that with a
constant transmission probability, the sum rate converges to a limit that is much
lower than 1 as ρ→∞. It corroborates that adaptive backoff is indispensable for
random-access networks.
It is interesting to note that when q = 1, all the nodes persistently transmit
their packets, and the slotted Aloha network reduces to a typical MAC. It is well
known that for an n-user AWGN MAC, if the capture model is adopted at the
receiver side and all the users have equal received power, the sum rate approaches
log2 e as n→∞ [65]. Here we can see from (3.25) that an additional factor of e−1
is introduced, which is mainly attributed to the effect of channel fading.4
3.4.2 Effect of Power Control
So far we have focused on a homogeneous slotted Aloha network where all
the nodes have the same mean received SNR ρ. In this section, the analysis will
be extended to the heterogeneous case, where nodes in the same group have an
identical mean received SNR but SNRs differ from group to group.
Specifically, assume that n nodes are divided into M groups. Group m has nm
nodes, and each node in Group m has the mean received SNR ρm, m = 1, . . . ,M .
For HOL packet j, let Sj denote the set of nodes that have concurrent transmis-
sions. It can be successfully decoded at the receiver if and only if its received
SINR is above the SINR threshold µ, i.e.,Pj∑
k∈SjPk+σ2 ≥ µ, where Pk denotes the
received power of node k’s packet. Suppose that Sj =⋃m=1,...,M Smj , where Smj
denotes the set of nodes which have concurrent transmissions in Group m, and
|Smj | = im, m = 1, . . . ,M . The steady-state probability of successful transmission
of HOL packet j given that there are {im}m=1,...,M concurrent transmissions, rj{im},
can be then written as rj{im} = Pr
{|hj |2∑M
m=1
∑k∈Sm
j|hk|2·
ρmρj
+1ρj
≥ µ
}. It can be easily
4Note that in this thesis, each codeword is assumed to last for one channel coherence timeperiod. Without coding over different fading states, the channel fluctuations cannot be averagedout.
Chapter 3 33
shown that with hk ∼ CN (0, 1), rj{im} is given by
rj{im} =exp
(− µρj
)∏Mm=1
(1+
ρmρjµ
)im . (3.26)
It can be seen from (3.26) that rj{im} is determined by the mean received SNR ρj
of HOL packet j. Suppose that HOL packet j belongs to Group l ∈ {1, · · · ,M}.As nodes in the same group have an identical mean received SNR, the superscript
j can be replaced by its group index l. The steady-state probability of successful
transmission of HOL packet j in Group l ∈ {1, · · · ,M} can then be written as
p(l) =
n1∑i1=0
· · ·nl−1∑il=0
· · ·nM∑iM=0
rl{im}
·M∏m=1
Pr{im concurrent transmissions in Group m}. (3.27)
For each node in Group m, the probability that it is busy with the HOL packet
requesting transmission in saturated conditions is given by π(m)T q0 +
∑Ki=0 π
(m)i qi,
which is equal to π(m)T /p(m) according to (2.3). Therefore, we have
Pr{im concurrent transmissions in Group m}
=
(nmim
) (1−π(m)
T /p(m))nm−im
·(π(m)T /p(m)
)imm6=l(
nl−1il
) (1−π(l)
T /p(l))nl−1−il
·(π(l)T /p
(l))il
m=l.(3.28)
By combining (3.26-3.28), the steady-state probability of successful transmis-
sion of HOL packet j in Group l ∈ {1, · · · ,M} can be obtained as
p(l)= exp(− µρl
)·(
1− µµ+1· π
(l)T
p(l)
)nl−1·
M∏m=1,m6=l
(1− µ
µ+ρl/ρm·π
(m)T
p(m)
)nmfor large n1,...,nM≈ exp
(− µρl−
M∑m=1
nmµµ+ρl/ρm
· π(m)T
p(m)
). (3.29)
Finally, by substituting (2.2) into (3.29), we have
p(l)= exp
− µρl−
M∑m=1
nmµµ+ρl/ρm
· 1∑K−1i=0
p(m)(1−p(m))i
qi+
(1−p(m))K
qK
, (3.30)
34 Chapter 3
l ∈ {1, · · · ,M}. We can see from (3.30) that in the heterogeneous case, HOL
packets in different groups have distinct steady-state probabilities of successful
transmission. With M groups, M non-zero roots {p(m)A }m=1,...,M can be obtained
by jointly solving M fixed-point equations given in (3.30). Note that nodes in the
same group have the same steady-state probability of successful transmission and
thus the same throughput performance. For each node in Group m, m = 1, . . . ,M ,
the node throughput can be obtained from (2.2) as
λ(m)out = π
(m)T = 1
∑K−1i=0
(1−p(m)
A
)iqi
+
(1−p(m)
A
)Kp(m)A qK
, (3.31)
and the network throughput is λout =∑M
m=1 nmλ(m)out .
To illustrate the above results, let us focus on the two-group case and as-
sume that the cutoff phase K = 0. The steady-state probabilities of successful
transmission of HOL packets in Group 1 and Group 2 can be obtained from (3.30)
as
p(1)A = exp
(− µρ1− n1µq0
µ+1− n2µq0
µ+ρ1/ρ2
),
p(2)A = exp
(− µρ2− n1µq0
µ+ρ2/ρ1− n2µq0
µ+1
). (3.32)
By combining (3.32) with (3.31), the node throughput can be obtained as
λ(1)out = q0 exp
(− µρ1− n1µq0
µ+1− n2µq0
µ+ρ1/ρ2
),
λ(2)out = q0 exp
(− µρ2− n1µq0
µ+ρ2/ρ1− n2µq0
µ+1
). (3.33)
Eq. (3.33) shows that the throughput performance is closely determined by the
mean received SNRs. If the two groups have equal mean received SNRs ρ1=ρ2=ρ,
for instance, we can see from (3.32) that all the HOL packets have the same steady-
state probability of successful transmission, i.e., p(1)A =p
(2)A . The node throughput
can be obtained from (3.33) as λ(1)out=λ
(2)out=q0 exp
(−µρ− (n1+n2)µq0
µ+1
). In this case,
each node has an equal probability of accessing the channel, thus achieving the
same throughput performance. As the difference between ρ1 and ρ2 grows, nev-
ertheless, the node throughput performance becomes increasingly polarized. We
can see from (3.32) that with ρ1�ρ2, p(1)A �p(2)A , which indicates that much more
packets from Group 1 can be successfully received than Group 2. The throughput
Chapter 3 35
1 100 200 300 400 500 600 700 800 900 10000.4
0.6
0.8
1
1.2
1.4
1.6
0 dB
10 dB
15dB
20 dB
1 2
C(b
it/s
/Hz)
Figure 3.7: Maximum sum rate of slotted Aloha networks with capture modelversus ρ1/ρ2 for a two-group slotted Aloha network. n1 = n2 = 25. K = 0.
performance of nodes in Group 1 is then much better than that in Group 2, i.e.,
λ(1)out�λ
(2)out according to (3.33), implying serious unfairness among nodes.
As the maximum network throughput λmax = maxq0 λout does not have an
explicit expression in general, we can only numerically calculate the maximum
sum rate C = maxµ λmax · log2(1 + µ). Fig. 3.7 illustrates how the maximum
sum rate C varies with the ratio of ρ1 and ρ2 by fixing the mean SNR of nodes
ρ =∑2m=1 nmρm∑2m=1 nm
to 0dB, 10dB, 15dB and 20dB. It is interesting to note from Fig.
3.7 that with a large SNR ratio ρ1/ρ2 � 1, the maximum sum rate is higher than
that with ρ1/ρ2 = 1, which suggests that despite serious unfairness, the sum rate
performance may be improved by introducing a large SNR difference among nodes.
Intuitively, the channel efficiency is maximized by allocating all the resources to
the strongest node(s). Here we can see that even without a central controller for
resource allocation, the fundamental tradeoff between efficiency and fairness still
holds true for random-access networks.
The tradeoff nevertheless becomes less significant when the network operates
at the low SNR region. It can be observed from Fig. 3.7 that with ρ = 0dB, the
maximum sum rate is insensitive to the SNR ratio. It indicates that power control
is desirable in this case, with which the fairness performance can be improved
without sacrificing the sum rate.
36 Chapter 3
3.4.3 Effect of Fading
Section 3.2 has shown that with the capture model, the network throughput
of slotted Aloha over fading channels can be much higher than e−1, the maximum
network throughput with the collision model in ideal channel conditions. In the
literature, it is sometimes mistaken as an improvement brought by fading channels:
the variation of the received power of nodes is enlarged due to fading, and thus the
chance that one of them can be captured, i.e., with much higher received power
than others, is increased. As we will demonstrate in this section, the gain comes
from the receiver rather than the channel fading. With the capture model, both
the maximum network throughput and the maximum sum rate of slotted Aloha
over AWGN channels are always higher than that over fading channels.
Let us first assume that the capture model is adopted. Specifically, by setting
the small-scale fading coefficients |hk| = 1, k = 1, . . . , n, the steady-state proba-
bility of successful transmission of a HOL packet given that there are i concurrent
transmissions over AWGN channels can be easily obtained from (3.2) as
rAWGNi = Pr
{1
i+ 1/ρ≥ µ
}=
1 i ≤ b 1µ− 1
ρc
0 i > b 1µ− 1
ρc.
(3.34)
By substituting (3.34) and (3.5) into (3.4), the steady-state probability of success-
ful transmission of HOL packets can be written as
p =
0 b 1
µ−1ρc<0
1 b 1µ−1ρc≥n−1
I1−πT
p
(n−1−
⌊1µ−1ρ
⌋,⌊1µ−1ρ
⌋+1)
otherwise,
(3.35)
where Ix(a, b) is the regularized incomplete beta function. The root of the fixed-
point equation (3.35) is closely determined by the backoff parameters {qi}i=0,...,K .
For the sake of illustration, let us focus on K = 0. In this case, the transmission
probability of each HOL packet is q0 regardless of how many collisions it has
Chapter 3 37
experienced. The root of (3.35) can be then written as
pA =
0 b 1
µ−1ρc<0
1 b 1µ−1ρc≥n−1
I1−q0
(n−1−
⌊1µ−1ρ
⌋,⌊1µ−1ρ
⌋+1)
otherwise.
(3.36)
and the corresponding network throughput can be obtained from (3.1) as
λout = npAq0
=
0 b 1
µ−1ρc<0
nq0 b 1µ−1ρc≥n−1
nq0I1−q0
(n−1−
⌊1µ−1ρ
⌋,⌊1µ−1ρ
⌋+1)
otherwise.
(3.37)
Appendix F shows that in this case, the maximum sum rate is given by
CAWGN =
e−1 log2(1 + ρ) ρ ≥ ρ0
n log2
(1 + 1
n−1+ 1ρ
)otherwise,
(3.38)
which is achieved when the SINR threshold is set to be
µ∗,AWGN =
ρ ρ ≥ ρ0
1n−1+ 1
ρ
otherwise,(3.39)
and the corresponding maximum network throughput is given by
λµ=µ∗,AWGN
max =
e−1 ρ ≥ ρ0
n otherwise,(3.40)
where ρ0 is the root of the following equation:
n log2
(1 +
1
n− 1 + 1ρ
)= e−1 log2(1 + ρ). (3.41)
For large number of nodes n, ρ0 ≈ ee − 1.
Fig. 3.8 illustrates how the maximum network throughput λµ=µ∗,AWGN
max and
the maximum sum rate CAWGN vary with the received SNR ρ. We can see from
(3.38) and Fig. 3.8 that at the low SNR region, i.e., ρ < ρ0, by setting the
38 Chapter 3
10-1
100
101
102
AWGN Channel
Fading Channel
(dB)r
*
max
μ=μ
λˆ
-20 -15 -10 -5 0 5 10 15 20 25 300r
(a)
10-1
100
101
C(b
it/s
/Hz)
(dB)r-20 -15 -10 -5 0 5 10 15 20 25 30
AWGN Channel
Fading Channel
0r
(b)
Figure 3.8: Network performance of slotted Aloha with the capture modelover AWGN and fading channels. n = 50. (a) Maximum network throughputversus mean received SNR. (b) Maximum sum rate versus mean received SNR.
SINR threshold µ = µ∗,AWGN = 1n−1+ 1
ρ
, we have⌊1µ− 1
ρ
⌋= n − 1, with which
all the packets can be successfully decoded according to (3.36). To maximize the
network throughput, all the nodes should transmit with probability 1, and thus
the maximum network throughput is equal to the number of nodes n. As n→∞,
the maximum sum rate C → log2 e according to (3.38), which is consistent to the
asymptotic sum rate of an n-user AWGN multiple access channel with the capture
model [65].
At the high SNR region, i.e., ρ ≥ ρ0, by setting the SINR threshold µ =
µ∗,AWGN = ρ, we have⌊1µ− 1
ρ
⌋= 0. According to (3.36), with
⌊1µ− 1
ρ
⌋= 0, a
packet can be successfully decoded if and only if there are no concurrent trans-
missions. In this case, the receiver reduces to the collision model. The maximum
network throughput is then e−1, and the maximum sum rate logarithmically in-
creases with the mean received SNR ρ with the high-SNR slope e−1.
Recall that it has been shown in Section 3.3 that in the fading case, the
maximum sum rate and the corresponding maximum network throughput at the
low SNR region are approximately given by e−1 log2 e and ne−1, respectively, for
large n, which are significantly lower than that over AWGN channels. At the
high SNR region, although the maximum sum rates in both cases have the same
high-SNR slope e−1, substantial gains are observed in the AWGN case. Therefore,
we can see from Fig. 3.8 that the gain comes from the receiver rather than the
Chapter 3 39
channel fading. It corroborates that the effect of channel fading is detrimental to
the network performance if CSI is not available at the transmitter side.
3.5 Summary
In this chapter, the maximum sum rate of Aloha with the capture model is
characterized. Specifically, explicit expressions of the maximum network through-
put and the corresponding optimal backoff parameters are obtained, based on
which the maximum sum rate is derived by optimizing the SINR threshold µ. The
analysis shows that with a low SNR, the maximum sum rate linearly increases with
the number of nodes n, and approaches e−1 log2 e as n → ∞. At the high SNR
region, a logarithmic growth of the maximum sum rate is observed as ρ increases,
with the high-SNR slope of e−1.
The analysis sheds important light on the practical network design. For in-
stance, it is demonstrated that to achieve the maximum sum rate, the transmission
probabilities of nodes should be adaptively tuned according to the network size
and the mean received SNR ρ. With a fixed transmission probability, the sum
rate may significantly deteriorate, and converges to a limit that is much lower
than 1 as ρ→∞. Moreover, the throughput performance of each node is found to
be closely dependent on its mean received SNR. Although a large SNR difference
among nodes may be beneficial to the sum rate performance, it introduces serious
unfairness. A uniform mean received SNR is shown to be crucial for achieving a
good balance between fairness and sum rate when the network operates at the low
SNR region.
Note that the capture model is essentially a single-user detector. It has been
shown in this chapter that the maximum sum rate of slotted Aloha with the capture
model is much smaller than the ergodic sum capacity of multiple access fading
channels. It is nevertheless difficult to identify whether the rate loss is caused by
the random access nature or suboptimality of the receiver. The following chapter
extends the analysis to incorporate the capacity-achieving receiver structure, SIC,
to pinpoint the rate loss caused by random access.
.
Chapter 4
Maximum Sum Rate with
Successive Interference
Cancellation (SIC)
In this chapter, the analysis will be further extended to incorporate SIC. This
chapter is organized as follows. Section 4.1 presents a literature review on previous
related work. Section 4.2 derives the network steady-state point in saturated
conditions, and Section 4.3 characterizes the maximum sum rate. The effect of
MPR on the sum rate performance of slotted Aloha and the rate loss due to
random access are discussed in Section 4.4. Section 4.5 provides a summary of
this chapter.
41
42 Chapter 4
4.1 Previous Work
As a representative type of multiuser detectors, it has been shown that suc-
cessive interference cancellation (SIC) can achieve the Shannon capacity region
boundaries for multiple access channels [26, 65, 70]. The key idea of SIC is that
once a packet is successfully decoded, it should be subtracted from the aggregate
received signal so that the next packet will experience less interference, which
greatly improves the performance of wireless networks [71–74].
By adopting SIC receivers in random-access networks, the network through-
put can be significantly enhanced compared with that in the classical collision
model thanks to the MPR capability [47, 51–63]. To study the effect of MPR, a
general model was proposed in [57], where the MPR capability was described by a
reception matrix whose entry in the k-th row and l-th column εkl is the conditional
probability that l packets are successfully decoded given that there are totally k
packet transmissions. Given the reception matrix, the stability regions and net-
work throughput performance of slotted Aloha networks were further studied in
[54, 57].
As the reception matrix depends on the receiver design, various MPR receiver
models have been considered in the literature. For instance, it was assumed in
[59–62] that the receiver can successfully decode up to M packets in each time
slot, that is, εkl = 1 if l ≤ M ≤ k and εkl = 0 otherwise. Here M represents
the MPR capability of the receiver, and its effects on the network throughput
performance were analyzed in [59–62]. For many receiver structures, however,
its MPR capability may not be simplified as a fixed number. With the widely-
adopted capture model [37, 38, 44, 47, 48, 51], for instance, if each node’s packet is
decoded independently by treating others’ as background noise and a packet can
be successfully decoded as long as its received SINR is above a certain threshold,
the reception probability εkl becomes a function of the SINR threshold [44]. In
general, the reception matrix {εkl} is determined by the specific receiver structure
and the minimum required SINR for successfully decoding a packet, which could be
difficult to obtain in some cases. It was shown in [56] that with a multi-stage SIC1,
the calculation of the reception probability εkl involves multiple nested integrals.
1Specifically, it was assumed in [56] that in each stage, the packets with power higher thanan SINR threshold are decoded, and subtracted from the total received signal. The processis repeated sequentially until no packets can be successfully decoded or the number of stagesreaches the maximum.
Chapter 4 43
To reduce the computational complexity, approximations were further proposed,
based on which a simple recursive algorithm was designed to estimate the network
throughput.
Note that despite a general description for MPR capability, the reception
matrix may not be necessary for throughput analysis of random-access networks.
Instead of deriving εkl, which could become intractable when sophisticated receiver
structures are adopted, the analysis can be established based on the conditional
probability of one packet being successfully decoded given that it has i concurrent
packet transmissions, ri. As we will show in this chapter, for an n-node saturated
slotted Aloha network with SIC, both the network throughput and the sum rate
can be obtained as explicit functions of {ri}i=0,...,n−1.
Specifically, in this chapter, two representative SIC receivers are considered,
i.e., ordered SIC and unordered SIC, which are described in Chapter 2.3. The max-
imum network throughput and the corresponding optimal transmission probability
of nodes in both cases are derived as functions of {ri}i=0,...,n−1, which depend on
SINR threshold µ and mean received SNR ρ. The maximum sum rate is further
characterized by optimizing the SINR threshold µ. To demonstrate the effect of
MPR on the sum rate performance, the maximum sum rates with SIC receivers
and the capture model are further compared to that with the classical collision
model.
It is worth mentioning that the SIC structure is also assumed in the line of
research on “coded random access” [75–79]. It is, however, used to recover the
coded successive packets of each node, and the classical collision model is assumed
for packets from different nodes. It should be distinguished from this thesis where
no coding/decoding is assumed over successive packets of each node, and the SIC
receivers are adopted for packets from multiple nodes.
4.2 Steady-State Point in Saturated Conditions
In this section, we will characterize the network steady-state point in satu-
rated conditions based on the fixed-point equation of p. We will start from the
derivation of the conditional probability of successful transmission of each HOL
packet given that it has i concurrent packet transmissions, ri.
44 Chapter 4
4.2.1 Conditional Probability of Successful Transmission
ri
For HOL packet j, denote rji as its steady-state probability of successful trans-
mission given that there are i concurrent packet transmissions. It is clear that rji
depends on the receiver design. In Chapter 3, it has been shown that with the
capture model, all the packets have the same conditional probability of successful
transmission of
rCi =exp
(−µρ
)(1 + µ)i
, (4.1)
where µ is the SINR threshold and ρ is the mean received SNR.
With SIC, in contrast, rji depends on the decoding order of packet j, and the
number of packets that have been successfully decoded before packet j. Specifi-
cally, it can be written as
rji =i+1∑m=1
Pr{packet j is decoded at the mth iteration}·m−1∑k=0
Pr{There are k
successfully decoded packets before packet j and packet j is
successfully decoded}. (4.2)
Note that all the packets have independent and identically distributed (i.i.d.) re-
ceived SNRs, and thus equal probability to be decoded at a given iteration. The
first item on the right-hand side of (4.2) is then given by
Pr{packet j is decoded at the mth iteration} =1
i+ 1. (4.3)
The second item on the right-hand side of (4.2) depends on whether the packets
are ordered. In the following, let us consider the ordered SIC and unordered SIC
cases separately.
Chapter 4 45
4.2.1.1 Ordered SIC
The second item on the right-hand side of (4.2) can be written as
Pr{There are k successfully decoded packets before packet j and packet j is
successfully decoded}
=Pr{There are k successfully decoded packets before packet j | packet j
is successfully decoded} · Pr{Packet j is successfully decoded}. (4.4)
With ordered SIC, packets are decoded in descending order of their received power.
If packet j is successfully decoded at the mth iteration, then all the m− 1 packets
before packet j must be successfully decoded. Therefore, we have
Pr{There are k successfully decoded packets before packet j | packet j is
successfully decoded}
=
1 if k = m− 1,
0 otherwise.(4.5)
Moreover, denote y(l)i as the probability that the packet at the lth iteration can be
successfully decoded given that there are i concurrent packet transmissions, i.e.,
totally i + 1 packets to be decoded. The probability that packet j is successfully
decoded is equal to the probability that all the packets before it, i.e., at iteration
1 to m, are successfully decoded, which can be written as
Pr{Packet j is successfully decoded} = Πml=1y
(l)i . (4.6)
46 Chapter 4
Appendix G further shows that
y(l)i =
(i+1)!
(i+1−l)!(l−1)!
d1/µe−1∑k=0
(i+1−lk
)(−1)k
l+kexp
− l+k
ρ(
1µ−k)( 1
µ−kl+ 1
µ
)i+1−l
l ≤ i+1− 1
µ
(i+1)!
(i+1−l)!(l−1)!
i+1−l∑k=0
(i+1−lk
)(−1)k
exp
(− k+lρ( 1µ−k)
)k+l
−i−l∑s=0
((i+1−l−k)s
(i+1)s+1·
Q
(1+s,
i+1
ρ( 1µ−i−1+l)
)+ exp
(− k+lρ( 1µ−k)
)( 1µ−k)s
( 1µ+l)s+1
·
(1−
Q
(1+s,− 1
ρ(i+1−l− 1µ)·(i+1−l−k)( 1
µ+l)
1µ−k
))) l > i+1− 1
µ,
(4.7)
for l = 1, ..., i+ 1, in which Q(s, t) = 1(s−1)!
∫∞te−xxs−1dx is the regularized upper
incomplete gamma function. With µ ≥ 1, (4.7) reduces to
y(l),µ≥1i =
(i+1
l
)exp
(− lµρ
)·(
1
lµ+ 1
)i+1−l
, (4.8)
for l = 1, ..., i + 1. By substituting (4.3-4.6) into (4.2), the steady-state probabil-
ity of successful transmission of HOL packet j given that there are i concurrent
transmissions for ordered SIC can be obtained as
rj,OSi =1
i+ 1
i+1∑m=1
Πml=1y
(l)i . (4.9)
The right-hand side of (4.9) is independent of j, indicating that all the HOL pack-
ets have the same conditional probability of successful transmission. Therefore,
we drop the superscript j in the following discussion.
Chapter 4 47
4.2.1.2 Unordered SIC
The second item on the right-hand side of (4.2) can be written as
Pr{There are k successfully decoded packets before packet j and packet j is
successfully decoded}
=Pr{Packet j is successfully decoded | there are k successfully decoded packets
before packet j} · Pr{There are k successfully decoded packets before packet j}.(4.10)
With unordered SIC, packets are decoded in a random order. Therefore, given
that there are k successfully decoded packets before packet j, the probability
that packet j can be successfully decoded is equal to rCi−k, i.e., the probability of
successful transmission given that there are i− k concurrent packet transmissions
for the capture model. According to (4.1), we have
Pr{Packet j is successfully decoded | there are k successfully decoded packets
before packet j }
=exp
(−µρ
)(1 + µ)i−k
= (1 + µ)k · rCi . (4.11)
It is difficult to obtain an explicit expression of Pr{There are k successfully de-
coded packets before packet j} due to the dependency of packets decoding, i.e.,
the decoding of each packet depends on how many packets at the preceding itera-
tions are successfully decoded. Let us consider the worst case that each packet is
subject to interference from i concurrent packet transmissions during its decoding
process. In this case, the probability that there are k successfully decoded pack-
ets before packet j at the mth iteration is given by(m−1k
) (1− rCi
)m−1−k (rCi)k
.
By combining (4.2-4.3) and (4.10-4.11), we can then obtain a lower-bound of the
steady-state probability of successful transmission of HOL packet j given that
there are i concurrent packet transmissions for unordered SIC as
rj,NSi ≥ rj,NS li = 1
i+1
i+1∑m=1
m−1∑k=0
(m−1
k
)(1+µ)k
(1− rCi
)m−1−k (rCi)k+1
=(1+µrCi )i+1−1
(i+1)µ.
(4.12)
Again, we drop the superscript j in the following discussion because the right-hand
side of (4.12) is independent of j.
48 Chapter 4
m
1nr-
Figure 4.1: Conditional probability of successful transmission rn−1 given thatthere are n − 1 concurrent packet transmissions versus SINR threshold µ in
slotted Aloha networks with SIC receivers. n = 20 and ρ = 20dB.
4.2.1.3 Comparison
We can see from (4.9) and (4.12) that both rOSi and rNS li depend on the SINR
threshold µ. Fig. 4.1 illustrates how they vary with µ. For the sake of comparison,
the result of the capture model obtained in Chapter 3 is also presented.
It can be clearly observed from Fig. 4.1 that both the ordered SIC and un-
ordered SIC have a much larger conditional probability of successful transmission
than the capture model when µ is small. It can be easily obtained from (4.1) that
with the capture model, rCn−1|µ= 1n
for large n≈ e−1. Appendix H further shows that for
large ρ and n,
rOSn−1|µ= 1n≈1, rNS l
n−1 |µ= 1n≈ee−1 − 1, (4.13)
both of which are significantly higher than e−1.
The gap, nevertheless, diminishes as the SINR threshold µ increases. As µ→∞, Appendix I shows that the conditional probability of successful transmission
in both the ordered SIC and unordered SIC cases converges to that of the capture
model.
limµ→∞
rOSirCi
= limµ→∞
rNS li
rCi= 1. (4.14)
Chapter 4 49
4.2.2 Steady-State Point in Saturated Conditions
The steady-state probability of successful transmission of HOL packets p can
be written as
p =n−1∑i=0
ri · Pr{i concurrent packet transmissions}. (4.15)
In saturated conditions, all the nodes have non-empty queues. According to the
Markov chain shown in Fig. 2.2, the probability that the HOL packet is requesting
transmission is given by πT q0+∑K
i=0 πiqi, which is equal to πT/p according to (2.2)-
(2.3). Therefore, the probability that there are i concurrent packet transmissions
can be obtained as
Pr{i concurrent packet transmissions}=(n−1
i
)(1−πT/p)n−1−i (πT/p)i . (4.16)
By substituting (4.16) into (4.15), we have
p =n−1∑i=0
ri ·(n−1
i
)(1−πT/p)n−1−i (πT/p)i . (4.17)
The service rate πT depends on the transmission probabilities of nodes {qi}i=0,...,K .
In this chapter, for the sake of simplicity, we assume that qi = q0 for i = 0, . . . , K.
In this case, the fixed-point equation (4.17) has a single non-zero root pA, which
is given by
pA =n−1∑i=0
(n− 1
i
)ri (1− q0)n−1−i · qi0. (4.18)
Specifically, with ordered SIC, pOSA can be obtained by combining (4.18) and (4.9).
With unordered SIC, pNSA can be approximated by combining (4.18) and the lower-
bound rNS li developed in (4.12).2 Fig. 4.2 illustrates how the steady-state point
pA varies with the SINR threshold µ for the ordered SIC and unordered SIC cases.
The result of the capture model is also presented for the sake of comparison.
Similar to the conditional probability of successful transmission illustrated in Fig.
4.1, we can see from Fig. 4.2 that the gains achieved by SIC receivers over the
capture model are significant only when µ is small.
2Unless otherwise specified, in the following analysis, rNSi is always approximated by its
lower-bound rNS li in the unordered SIC case.
50 Chapter 4
m
Ap
Figure 4.2: Steady-state point of slotted Aloha networks with SIC receiverspA versus SINR threshold µ. n = 20. q0 = 0.5 and ρ = 20dB.
4.2.3 Simulation Results
In this section, simulation results are presented to verify the preceding analy-
sis. In particular, we consider a saturated slotted Aloha network where each node
has the transmission probability q0. The simulation setting is the same as the
system model and thus we omit the details here.
In Section 4.2.1, the conditional probability of successful transmission ri for
ordered SIC has been derived in (4.9) and verified by simulation results presented
in Fig. 4.3. For the unordered SIC case, Fig. 4.3 shows that the lower-bound of
ri developed in (4.12) is quit close.
In Section 4.2.2, the steady-state point pA is further derived in (4.18) as
the non-zero root of the fixed-point equation of the steady-state probability of
successful transmission of HOL packets p, which is verified by simulation results
presented in Fig. 4.4. It can be clearly observed from Fig. 4.4 that the steady-
state point pA closely depends on the transmission probability of nodes q0. In
the next section, we will demonstrate how to optimally choose the transmission
probability of nodes q0 to maximize the network throughput and the sum rate.
Chapter 4 51
i
ir
(a)
i
ir
(b)
Figure 4.3: Conditional probability of successful transmission ri in slottedAloha networks with SIC receivers versus number of concurrent packet trans-
missions i. µ = 1. (a) ρ = 0dB. (b) ρ = 20dB.
0q
Ap
(a)
0q
Ap
(b)
Figure 4.4: Steady-state point of slotted Aloha networks with SIC receiverspA versus transmission probability q0. n = 20 and µ = 1. (a) ρ = 0dB. (b)
ρ = 20dB.
4.3 Maximum Sum Rate
In this section, let us first derive the maximum network throughput λmax by
optimizing the transmission probability of nodes q0.
52 Chapter 4
4.3.1 Maximum Network Throughput
In saturated conditions, the network throughput λout = nπT . By combining
(2.2) and (4.18), with qi = q0 for i = 0, . . . , K, the network throughput λout in
saturated conditions can be obtained as
λout = nq0pA = nn−1∑i=0
(n− 1
i
)ri (1− q0)n−1−i qi+1
0 , (4.19)
which closely depends on the transmission probability of nodes q0. For the maxi-
mum network throughput λmax = max0<q0≤1 λout, Appendix J shows that in both
the ordered SIC and unordered SIC cases, we have
λmax =
n∑n−1
i=0
(n−1i
)ri (1− q0)n−1−i qi+1
0 if µ ≥ µ0
nrn−1 otherwise,(4.20)
which is achieved at
q∗0 =
q0 if µ ≥ µ0
1 otherwise,(4.21)
where q0 is the root of
nn−1∑i=0
(n− 1
i
)ri (1− q0)n−2−i qi0(1 + i− nq0) = 0, (4.22)
and µ0 is the root ofrn−2rn−1
=n
n− 1. (4.23)
We can see from (4.20) that both λOSmax and λNSmax depend on the SINR thresh-
old µ. Fig. 4.5 illustrates how they vary with µ. Similar to the conditional
probability of successful transmission rn−1 and the steady-state point in saturated
conditions pA shown in Fig. 4.1 and Fig. 4.2, significant gains in maximum net-
work throughput can be achieved by SIC receivers over the capture model when
µ is small. Recall that with the capture model, we have λCmax|µ= 1n
for large n≈ ne−1.
It can be further obtained from (4.20) and (4.13) that
λOSmax|µ= 1n
for large n≈ n, λNSmax|µ= 1
n
for large n≈ n
(ee−1 − 1
), (4.24)
Chapter 4 53
m
max
l
Figure 4.5: Maximum network throughput of slotted Aloha networks withcapture model and SIC receivers λmax versus SINR threshold µ. n = 20 and
ρ = 20dB.
both of which are much higher than ne−1. With a large µ, nevertheless, the
maximum network throughput in both the ordered SIC and unordered SIC cases
converges to that of the capture model. As µ→∞, it can be obtained from (4.20)
and (4.14) that limµ→∞λOSmax
λCmax= limµ→∞
λNSmax
λCmax= 1.
It can be clearly seen from Fig. 4.5 that similar to the capture model, with
SIC receivers, the maximum network throughput λmax also increases as the SINR
threshold µ decreases. Such an improvement on maximum network throughput is,
nevertheless, achieved at the cost of a smaller information encoding rate that can
be supported for reliable communications, i.e., R = log2(1 + µ). In the next sub-
section, we will further study how to maximize the sum rate by properly choosing
the SINR threshold µ.
4.3.2 Maximum Sum Rate
By combining (2.8) and (4.20), the maximum sum rate can be further written
as C = max (C1, C2), where C1 = maxµ≥µ0 f1(µ) and C2 = maxµ<µ0 f2(µ), in which
f1(µ) = λµ≥µ0max log2(1+µ) = nn−1∑i=0
(n− 1
i
)ri (1− q0)n−1−i qi+1
0 · log2(1+µ) (4.25)
54 Chapter 4
for µ ∈ [µ0,∞) and
f2(µ) = λµ<µ0max log2(1 + µ) = nrn−1 · log2(1 + µ) (4.26)
for µ ∈ (0, µ0). Denote µh = arg max f1(µ) and µl = arg max f2(µ). Note that
µh and µl can be numerically obtained given the expressions of f1(µ) and f2(µ).
Appendix K shows that for both the ordered SIC and unordered SIC cases, the
maximum sum rate is given by
C =
f1(µh) if ρ ≥ ρ0
f2(µl) otherwise,(4.27)
which is achieved when the SINR threshold µ is set as
µ∗ =
µh if ρ ≥ ρ0
µl otherwise,(4.28)
where ρ0 is the root of f1(µh) = f2(µl).
The maximum sum rate C for both the ordered SIC and unordered SIC cases is
illustrated in Fig. 4.6. For the sake of comparison, the result of the capture model
is also presented. In contrast to Fig. 4.5 where the maximum network throughput
of the unordered SIC is shown to be significantly higher than that of the capture
model, we can see from Fig. 4.6 that only marginal gains in the maximum sum
rate can be achieved by the unordered SIC over the capture model. For the ordered
SIC case, substantial gains can still be observed at moderate values of the mean
received SNR ρ, i.e., −10dB< ρ < 35dB. The gap, nevertheless, diminishes at the
high SNR region, i.e., ρ ≥ ρOS0 , where the maximum sum rates of all the three
receivers have the high-SNR slope of e−1 as ρ→∞.
It is also interesting to note from Fig. 4.6 that ordering is crucial for the
rate performance of slotted Aloha networks with SIC receivers. Without proper
ordering of packets, the maximum sum rate could be drastically degraded and
becomes comparable to that of the capture model. To achieve the maximum sum
rate, the SINR threshold µ needs to be carefully tuned according to the mean
received SNR ρ based on (4.28). Fig. 4.7a illustrates the optimal SINR threshold
µ∗ for the ordered SIC, unordered SIC and the capture model cases. It can be
clearly seen from Fig. 4.7a that in all three cases, the SINR threshold µ, or
Chapter 4 55
=3.3=4.3 =36.1
C(bit/s/Hz)
(dB) r
0
Cr
0
NSr 0
OSr
Figure 4.6: Maximum sum rate of slotted Aloha networks with capture modeland SIC receivers C versus mean received SNR ρ. n = 20.
=3.3 =4.3 =36.1
*m
(dB) r
0
Cr
0
NSr 0
OSr
(a)
=3.3 =4.3 =36.1
(dB) r
**,
0q
mm
=
10.05
n=
0
Cr
0
NSr 0
OSr
(b)
Figure 4.7: (a) Optimal SINR threshold µ∗ and (b) optimal transmission prob-
ability q∗,µ=µ∗
0 of slotted Aloha networks with capture model and SIC receivers.n = 20.
56 Chapter 4
equivalently, the information encoding rate R, should be properly enlarged as the
mean received SNR ρ increases. Moreover, by combining (4.21) and (4.28), we
can further obtain the optimal transmission probability q∗,µ=µ∗
0 for maximizing
the sum rate as a function of the mean received SNR ρ. As Fig. 4.7b illustrates,
at the low SNR region, the optimal transmission probability is 1, indicating that
all the nodes should persistently transmit their packets. As the mean received
SNR ρ increases, nevertheless, the transmission probability of each node should
be reduced accordingly, and converges to 1n
as ρ→∞.
4.3.3 Simulation Results
In this section, simulation results are presented to verify the preceding anal-
ysis. Fig. 4.8 illustrates the network throughput performance. The network
throughput λout has been derived as a function of the transmission probability of
nodes q0 in (4.19). According to (4.20), if the SINR threshold µ ≥ µ0, the max-
imum network throughput λmax is achieved when q0 is set to be q0. Otherwise,
λmax is achieved when q0 = 1. With n = 20 and ρ = 20dB, we have µOS0 = 0.1205
and µNS0 = 0.0532 by substituting (4.9) and (4.12) into (4.23), respectively, for
the ordered SIC and unordered SIC cases. As we can see from Fig. 4.8a, with
µ = 0.05 < µ0, the network throughput in both cases monotonically increases
with q0, and is maximized at q0 = 1. With µ = 1 > µ0, on the other hand, the
network throughput is maximized at q0 = q0 which is given in (4.22), as Fig. 4.8b
illustrates. Simulation results presented in Fig. 4.8 verify that the analysis serves
as a good lower-bound in the unordered SIC case, and is accurate with ordered
SIC.
It is further shown in Section 4.3.2 that as both the information encoding rate
R and the maximum network throughput λmax depend on the SINR threshold µ,
the sum rate can be maximized by optimizing the SINR threshold µ. As Fig.
4.9 illustrates, the sum rate performance is sensitive to the setting of µ in both
ordered SIC and unordered SIC cases. To achieve the maximum sum rate, the
SINR threshold µ should be properly set according to the mean received SNR ρ.
The expressions of the maximum sum rate C and the corresponding optimal SINR
threshold µ∗ are given in (4.27)-(4.28), and verified by simulation results presented
in Fig. 4.9.
Chapter 4 57
0q
max
ˆNSl
max
ˆOSl
ˆ out
l
(a)
denotes 0q
0q
max
ˆNSl
max
ˆOSl
ˆ out
l
(b)
Figure 4.8: Network throughput of slotted Aloha networks with SIC receiversλout versus transmission probability q0. n = 20 and ρ = 20dB. (a) µ = 0.05.
(b) µ = 1.
4.4 Discussions
So far we have derived the maximum sum rates of slotted Aloha networks with
SIC receivers, and the corresponding optimal setting including the optimal SINR
threshold and the optimal transmission probability of nodes. In this section, we
will further demonstrate the effect of MPR on the sum rate performance of slotted
Aloha networks, and compare it to the ergodic sum capacity of multiple access
fading channels.
4.4.1 Effect of MPR
Note that both the SIC receivers and the capture model have the so-called
MPR capability: If the SINR threshold µ is small enough, then multiple packets
can be successfully decoded at each time slot. As a result, the network throughput
performance is expected to be substantially enhanced compared to the classical
collision model where at most one packet can be successfully decoded at each time
slot.
To further see the effect of MPR on the sum rate performance, let us consider
the classical collision model where a packet transmission is successful only if there
are no concurrent transmissions. Appendix L shows that in this case, the maximum
58 Chapter 4
*,NSm=0.05
*,OSm=0.10
=2.34OSC
=0.64NS
C
m
sR(bit/s/Hz)
(a)
*,NSm=24
*,OSm=0.12
=3
=1.46
OSC
NSC
m
sR(bit/s/Hz)
(b)
=
31.37 10´
*,NSm
*,OSm = =
m
sR
=3.43OSC
NSC
(bit/s/Hz)
(c)
Figure 4.9: Sum rate of slotted Aloha networks with SIC receivers Rs versusSINR threshold µ. n = 20 and q0 = q∗0. (a) ρ = 0dB. (b) ρ = 20dB. (c)
ρ = 40dB.
sum rate is given by
Ccollision = exp(−1− eW0(ρ)−1
ρ
)· log2(e
W0(ρ)), (4.29)
which is achieved when the SNR threshold is set to be µ∗,collision = eW0(ρ)− 1. The
corresponding maximum network throughput is given by
λµ=µ∗,collision
max = exp(−1− eW0(ρ)−1
ρ
). (4.30)
Fig. 4.10 illustrates the maximum sum rate and the corresponding maximum
network throughput with the collision model. For the sake of comparison, the
results of the capture model and SIC receivers are also presented. As we can see
Chapter 4 59
*
maxˆm
ml
=
(dB) r
(a)
(bit/s/Hz)
C
(dB) r
(b)
Figure 4.10: Network performance of slotted Aloha with collision model, cap-ture model and SIC receivers. n = 20. (a) Maximum network throughput λµ=µ
∗max
and (b) maximum sum rate C versus mean received SNR ρ.
from Fig. 4.10a, at the low SNR region, the maximum network throughput of the
collision model is significantly lower than that of SIC receivers and the capture
model, which can be much higher than 1 thanks to MPR. The gap, nevertheless,
diminishes as the mean received SNR ρ increases. All of them approach e−1 as
ρ→∞.
Similarly, Fig. 4.10b shows that the maximum sum rates of all four receivers
have the same high-SNR slope of e−1, and the rate difference becomes negligible
when the mean received SNR ρ is large. At the low SNR region, however, in
contrast to the significant throughput improvement, only marginal gains in the
maximum sum rate can be achieved by the SIC receivers and the capture model.
The rate difference between the collision model and the capture model is below
0.5 bit/Hz/s for the whole SNR region, which suggests that despite an overly pes-
simistic estimation on the network throughput, the collision model serves as a good
approximation for the capture model when analyzing the sum rate performance of
slotted Aloha networks.
Recall that it has been shown in Fig. 4.7a that at the low SNR region, the
optimal SINR thresholds with the SIC receivers and the capture model are small,
indicating that each packet has a small encoding rate. Therefore, although much
more packets can be successfully decoded compared to the collision model, the
gain in the maximum sum rate is limited. At the high SNR region, on the other
hand, both the SIC receivers and the capture model reduce to the collision model
60 Chapter 4
as the optimal SINR thresholds are much higher than 1. We can conclude from
Fig. 4.10 that although the network throughput performance at the low SNR
region can always be significantly improved thanks to MPR, the rate gain could
be marginal, and becomes negligible when the mean received SNR is large.
4.4.2 Rate Loss Due to Random Access
For an n-user multiple access fading channel, it has been well known that
the ergodic sum capacity without the channel state information at the transmitter
side (CSIT) is given by Csum = E [log2 (1 +∑n
k=1 |hk|2ρ)], where the expectation is
taken over all fading states of users. To achieve the ergodic sum capacity, an SIC
receiver should be adopted, and each node’s encoding rate is determined by its
decoding order, with codewords long enough to cover multiple fading states [65].
With slotted Aloha, in contrast, as each node transmits with a certain probability,
the subset of active nodes is time-varying, and supposed to be unknown at the
transmitter side. Therefore, rate allocation among nodes cannot be performed,
and a uniform information encoding rate is assumed for each packet with the
packet-based encoding, i.e., each packet contains one codeword lasting for one
time slot.
Fig. 4.10b illustrates the ergodic sum capacity in comparison to the maximum
sum rates of slotted Aloha under various receivers. It can be clearly seen from
Fig. 4.10b that even with the capacity-achieving receiver structure, the ordered
SIC, the maximum sum rate of slotted Aloha is still significantly lower than the
ergodic sum capacity. Specifically, different high-SNR slopes are observed, i.e., 1
for the ergodic sum capacity and e−1 for the maximum sum rates of slotted Aloha.
With the packet-based encoding/decoding and a uniform encoding rate of each
packet, even if the ordered SIC is adopted, the sum rate performance of slotted
Aloha is still bottlenecked by the first decoded packet, which can be improved by
increasing the mean received SNR ρ at the high SNR region only when there are
no concurrent packet transmissions. It has been shown in Fig. 4.7b that as ρ
increases, the optimal transmission probability of each node converges to 1/n. As
a result, the high-SNR slope of the maximum sum rates of slotted Aloha, which
is determined by the probability that there is a single packet transmission in each
time slot at the high SNR region, is only e−1. Here we can see that the huge
Chapter 4 61
rate loss of slotted Aloha at the high SNR region has its root in the contention of
packets caused by uncoordinated random transmissions of nodes.
4.5 Summary
In this chapter, the maximum sum rates of slotted Aloha with two repre-
sentative SIC receivers, ordered SIC and unordered SIC, are characterized. The
analysis shows that maximum sum rate with ordered SIC is significantly higher
than that with unordered SIC within a wide range of ρ. This result suggests that
ordering is crucial for the rate performance of slotted Aloha with SIC receivers.
Without proper ordering of packets, the maximum sum rate could be drastically
degraded and become comparable with that of the capture model. The comparison
with the classical collision model further reveals that although substantial gains
in network throughput can be achieved by SIC receivers and the capture model at
the low SNR region thanks to MPR, the rate difference could be limited, and they
all reduce to the collision model with the high-SNR slope of e−1 when the mean
received SNR ρ is large. Without proper rate allocation, even with the capacity-
achieving receiver structure, the maximum sum rate of slotted Aloha could remain
far below the ergodic sum capacity of fading channels, indicating that the rate loss
is significant due to uncoordinated random transmissions of nodes.
Chapter 5
Conclusion and Future Work
5.1 Conclusion
As a basic type of multiple access, random access has been widely applied to
a plethora of networks including cellular networks, IEEE 802.11 networks, wireless
ad-hoc networks and machine-to-machine (M2M) networks. In sharp contrast to
its simplicity in concept and success in applications, little was known about its
fundamental performance limits. Aiming to explore the information-theoretic limit
of slotted Aloha networks, this thesis presents a comprehensive analysis on the sum
rate of slotted Aloha by extending the unified analytical framework [21, 22] from
the classical collision model to the capture model and SIC receivers.
Specifically, the first part of this thesis focuses on maximum sum rate char-
acterization with the capture model. The maximum sum rate and corresponding
maximum network throughput are derived by assuming that the received SNRs
of each node’s packets are exponentially distributed with the same mean received
SNR ρ and saturated conditions. The optimal setting, including the optimal trans-
mission probability and the optimal SINR threshold, are also obtained. The anal-
ysis shows that at the low SNR region, the maximum sum rate linearly increases
with the number of nodes n, and approaches e−1 log2 e as n → ∞. At the high
SNR region, the maximum sum rate logarithmically increases with the mean re-
ceived SNR ρ, and the high-SNR slope is e−1, which is much lower than that of
the ergodic sum capacity of multiple access fading channels.
As the capture model is essentially a single-user detector, it is difficult to
determine whether the gap between the ergodic sum capacity of multiple access
63
64 Chapter 5
and the maximum sum rate of slotted Aloha with the capture model is caused by
the random-access nature or suboptimality of the receiver. Therefore, in the second
part of this thesis, the analysis is further extended to incorporate the capacity-
achieving receiver structure, SIC. Specifically, two representative SIC receivers,
ordered SIC and unordered SIC, are considered. The maximum sum rate in both
cases is obtained by optimizing the SINR threshold µ. The comparison to the
capture model shows that ordering is crucial for the rate performance of slotted
Aloha with SIC receivers: Without proper ordering of packets, the maximum
sum rate could be drastically degraded and becomes comparable to that with
the capture model. To further demonstrate the effect of MPR on the sum rate
performance, the maximum sum rates with SIC receivers and the capture model
are further compared to that with the classical collision model. It is found that
in contrast to the significant throughput improvement at the low SNR region,
only marginal gains in the maximum sum rate can be achieved by SIC receivers
and the capture model. The rate difference becomes negligible when the mean
received SNR ρ is large, and the maximum sum rates with all the receivers have
the same high-SNR slope of e−1, which is remarkably lower than that of the ergodic
sum capacity of multiple access fading channels. Evidently, the huge rate loss is
mainly due to the random access nature (i.e., uncoordinated random transmissions
of nodes).
To conclude, the main contributions of this thesis are summarized as follows:
1. The unified analytical framework in [21, 22] is extended from the classical
collision model to the capture model, where the network steady-state point in
saturated conditions is derived as a function of the SINR threshold. Explicit
expressions of the maximum network throughput, maximum sum rate and
optimal setting are obtained. The analysis shows that the maximum sum
rate of slotted Aloha with the capture model is much lower than the ergodic
sum capacity of multiple access fading channels.
2. The analysis is further extended to incorporate the capacity-achieving re-
ceiver structure, SIC, to determine whether the rate loss originates from the
random access nature or suboptimality of the receiver (i.e., the capture mod-
el). The maximum sum rate and corresponding optimal parameter setting,
which includes the transmission probability and SINR threshold for both
ordered and unordered SIC, are obtained. The comparison to the ergodic
Chapter 5 65
sum capacity of multiple access fading channels indicates that the rate loss
is due to the uncoordinated random transmissions of nodes, which make rate
allocation among nodes unfeasible.
3. The effects of adaptive backoff, power control, channel fading and multi-
packet reception on the sum rate performance of slotted Aloha networks are
further discussed. The main findings are summarized as follows:
• To achieve the maximum sum rate, the transmission probabilities of
nodes should be properly selected according to the network size and
the mean received SNR. With a fixed transmission probability, the sum
rate performance significantly deteriorates.
• Although a large SNR difference among nodes may potentially improve
the sum rate performance, serious unfairness is introduced. At the low
SNR region, power control is essential in order to achieve a good balance
between fairness and sum rate performance.
• With the capture model, both the maximum network throughput and
the maximum sum rate of slotted Aloha over AWGN channels are al-
ways higher than that over fading channels, which corroborates that
the effect of channel fading is detrimental to the network performance
if CSI is not available at the transmitter side.
• By comparing the maximum sum rates with SIC receivers and the cap-
ture model to that with the classical collision model, it is found that
substantial gains in network throughput can be achieved by MPR re-
ceivers at the low SNR region. However, the rate gain from MPR is
only marginal.
5.2 Future Work
The work in this thesis only provides a starting point. In the future, the
analysis should be extended to consider more practical assumptions. For instance,
this thesis focuses on the saturated conditions where the network throughput is
pushed to the limit, yet the mean queueing delay is infinite and the network could
be unstable. It is of great practical significance to further consider the unsaturated
scenario and study the maximum sum rate of slotted Aloha under certain system
constraints such as stability or delay requirements.
66 Chapter 5
Moreover, a key assumption throughout this thesis is that the nodes are un-
aware of the instantaneous realizations of small-scale fading, and they encode their
packets independently at the same rate. If CSI is available at the transmitter side,
each node can then adjust the information coding rate and the transmission prob-
ability according to its instantaneous channel conditions. How to characterize the
maximum sum rate with CSI at the transmitter side is another interesting and
challenging issue that deserves much attention in the future study.
From the practical perspective, the analysis can be further applied to real-
world networks, such as M2M networks. M2M communication has been regarded
as an emerging technology that can facilitate new business opportunities for both
consumers and enterprises. In M2M networks, a massive number of heterogeneous
devices such as smart meters, monitoring devices and intelligent transportation
devices operate in an automated manner, and generate small and infrequent pay-
loads. Therefore, random access is much more suitable than centralized multiple
access for M2M networks [82, 83]. Increasing attention has been provided to the
modeling and performance analysis of the current multiple access protocol for
M2M networks, namely, the Physical Random Access Channel (PRACH) of LTE
and LTE-A [3, 4, 84–91], which is a variant of slotted Aloha. Many studies have
pointed out that significant challenges for M2M networks include how to properly
tune the system parameters to optimize the network throughput performance un-
der certain delay requirement for time-restricted applications [3, 88–90] or how to
minimize the energy consumption of devices [4, 91, 92]. It would be of paramount
practical importance to further extend the proposed analytical framework in this
thesis to address the above challenging issues in M2M networks in the future.
Appendix A
Proof of Theorem 3.1
Proof. The right-hand side of (3.7) can be written as h(p) = exp(−µρ− nµ
µ+1· 1g(p)
),
where g(p) =∑K−1
i=0p(1−p)i
qi+ (1−p)K
qK. Define qi = 1/qi, for 0 ≤ i ≤ K − 1, and
qi = 1/qK for i ≥ K. g(p) can be then written as g(p) =∑∞
i=0 p(1−p)iqi = EX [qX ],
where X is a geometric random variable with parameter p.
Suppose that 0 < p1 < p2 ≤ 1. Let X1 and X2 denote geometric random
variables with parameters p1 and p2, respectively. Then we have X1 ≥st X2 [80].1
As {qi} is a monotonic non-increasing sequence, we have qX1 ≥st qX2 . We can
then conclude that g(p1) ≥ g(p2). Therefore, g(p) is a monotonic non-increasing
function with respect to p, which indicates that h(p) is a monotonic non-increasing
function. Moreover, as limp→0 h(p) = exp(−µρ− nµ
µ+1· qK
)> 0 and limp→1 h(p) =
exp(−µρ− nµ
µ+1· q0)< 1, we can then conclude that (3.7) has a single non-zero
root if {qi}i=0,...,K is a monotonic non-increasing sequence.
1X1 ≥st X2 denotes that a random variable X1 is larger than a random variable X2 in theusual stochastic order, i.e., Pr(X1 > x) ≥ Pr(X2 > x) for all x ∈ (−∞,∞).
67
Appendix B
Proof of Theorem 3.2
Proof. It is shown in (3.9) that the network throughput can be obtained as an
explicit function of pA. The following lemma first presents λpmax = maxpA∈(0,1]λout
and the corresponding optimal steady-state point p∗A.
Lemma B.1. For given SINR threshold µ ∈ (0,∞) and mean received SNR ρ ∈(0,∞), λpmax is given by
λpmax = µ+1µ
exp(−1− µ
ρ
), (B.1)
which is achieved at
p∗A = exp(−1− µ
ρ
). (B.2)
Proof. According to (3.9), the second-order derivative of λout with respect to pA
is given by −µ+1µpA
< 0, for pA ∈ (0,∞). Therefore, we can conclude that λout is a
strictly concave function of pA ∈ (0,∞) with one global maximum at p∗A, where
p∗A is the root of dλoutdpA
= 0, i.e., (µ + 1) ·(− ln pA−1
µ− 1
ρ
)= 0, which is given by
(B.2). Eq. (B.1) can be obtained by substituting (B.2) into (3.9).
We can see from Lemma B.1 and (3.7) that to achieve λpmax, the backoff
parameters {qi}i=0,...,K should be carefully selected such that pA = p∗A. For given
backoff function Qi, the optimal initial transmission probability for achieving λpmax
can be easily obtained by combining (3.7) and (B.2) as (3.12).
69
70 Appendix B
Note that the initial transmission probability q0 should not exceed 1. Lemma
B.2 shows that λpmax is achievable for qi ∈ (0, 1], i = 0, . . . , K, if and only if the
SINR threshold µ ≥ 1n−1 .
Lemma B.2. λpmax is achievable if and only if µ ≥ 1n−1 .
Proof. Define Qi = 1/Qi, for 0 ≤ i ≤ K − 1, and Qi = 1/QK for i ≥ K. Let
Y denote a geometric random variable with parameter exp(−1− µ
ρ
). Eq. (3.12)
can be then written as q0 = µ+1nµ· EY [QY ]. As Qi ≤ 1 for i = 0, 1, . . . , K, we have
EY [QY ] ≥ 1.
1) if : if µ ≥ 1n−1 , with Qi = 1 for i = 0, 1, . . . , K, we have EY [QY ] = 1 and
q0 = µ+1nµ≤ 1. In this case, λpmax can be achieved by setting q0 = q0.
2) only if : if µ < 1n−1 , we have q0 ≥ µ+1
nµ> 1, which indicates that λpmax is not
achievable.
For µ < 1n−1 , λpmax is not achievable for qi ∈ (0, 1], i = 0, . . . , K. The following
lemma shows that in this case, the maximum network throughput λmax is always
smaller than λpmax, which is achieved by setting qi = 1, i = 0, . . . , K.
Lemma B.3. For given SINR threshold µ < 1n−1 , the maximum network through-
put λmax is given by
λµ<
1n−1
max = n exp(− nµµ+1− µ
ρ
), (B.3)
which is achieved at q∗i = 1, i = 0, . . . , K.
Proof. According to (3.7), the initial transmission probability q0 can be written as
q0 = µ+1nµ
(− ln pA − µ
ρ
)· z(pA). (B.4)
where z (pA) =∑K−1
i=0pA(1−pA)iQi + (1−pA)K
QK. Similar to g(p) in Appendix A, it can
be proved that z (pA) is a monotonic non-increasing function of pA ∈ (0, 1]. Note
that − ln pA is also a monotonic non-increasing function of pA ∈ (0, 1]. Therefore,
we can conclude from (B.4) that pA is a monotonic non-increasing function of q0.
With 0 < q0 ≤ 1, we can obtain from (3.7) that
pA ≥ exp
(−µρ− nµ
µ+1· 1∑K−1
i=0
pA(1−pA)iQi +
(1−pA)KQK
), (B.5)
Appendix B 71
where “=” holds when q0 = 1. Note that the backoff functionQi ≤ 1, i = 0, . . . , K.
We can further obtain from (B.5) that
pA ≥ exp(−µρ− nµ
µ+1
), (B.6)
where “=” holds when q0 = 1 and Qi = 1, i = 0, . . . , K.
When µ < 1n−1 , we can see from (B.6) that pA > exp
(−µρ− 1)
= p∗A. Ac-
cording to the proof of Lemma B.1, the network throughput λout is a monotonic
decreasing function of pA when pA > p∗A. Therefore, in this case, λout is max-
imized when pA is minimized, i.e., pA = exp(−µρ− nµ
µ+1
)according to (B.6),
which is achieved at q∗i = 1. Eq. (B.3) can be then obtained by substituting
pA = exp(−µρ− nµ
µ+1
)into (3.9).
Finally, Eqs. (3.10) and (3.11) can be obtained by combining Lemma B.1,
Lemma B.2 and Lemma B.3.
Appendix C
Proof of Theorem 3.3
Proof. According to (3.14), we can rewrite the maximum sum rate as C = max (C1, C2),
where
C1 = maxµ≥ 1
n−1
µ+1µ
exp(−1− µ
ρ
)· log2(1 + µ), (C.1)
and
C2 = max0<µ≤ 1
n−1
n exp(− nµµ+1− µ
ρ
)· log2(1 + µ). (C.2)
Let us first focus on C1.
1) Denote the objective function of (C.1) as f1(µ) and let us first prove the fol-
lowing lemma.
Lemma C.1. f1(µ) is a monotonic decreasing function of µ ∈[
1n−1 ,∞
)if ρ < ρ0.
Otherwise, it has one global maximum at µ∗h, where µ∗h is the root of (3.17).
Proof. f1(µ) is a continuously differentiable function of µ ∈[
1n−1 ,∞
). The first-
order derivative of f1(µ) can be written as
f ′1(µ) = exp(−1− µ
ρ
)log2 e ·G1(µ), (C.3)
where
G1(µ) = 1µ− 1
µ2ln(1 + µ)− 1
ρ· 1+µ
µln(1 + µ). (C.4)
It can be easily obtained from (C.4) that
limµ→ 1
n−1
G1(µ) = (n− 1)− (n− 1)2 ln nn−1 −
nρ
ln nn−1 , (C.5)
73
74 Appendix C
and
limµ→∞
G1(µ) = −∞. (C.6)
Moreover, the first-order derivative of G1(µ) can be obtained from (C.4) as
G′1(µ) = − 1µ2
(2+µ1+µ− 2
µln(1 + µ)
)− 1
ρ
(1µ− ln(1+µ)
µ2
)< 0, (C.7)
for µ ∈[
1n−1 ,∞
), which indicates that G1(µ) is a monotonic decreasing function
of µ ∈[
1n−1 ,∞
).
i) If ρ ≥ ρ0, we can obtain from (C.5-C.6) that limµ→ 1
n−1G1(µ) ≥ 0 and limµ→∞G1(µ) <
0. As G1(µ) is a monotonic decreasing function of µ ∈[
1n−1 ,∞
), there must ex-
ist µ∗h ∈[
1n−1 ,∞
), such that G1(µ) > 0 for µ ∈
[1
n−1 , µ∗h
)and G1(µ) < 0 for
µ ∈ (µ∗h,∞), where µ∗h is the root of G1(µ) = 0, which is given in (3.17). We
can then obtain from (C.3) that f ′1(µ) > 0 for µ ∈[
1n−1 , µ
∗h
)and f ′1(µ) < 0 for
µ ∈ (µ∗h,∞), which indicates that f1(µ) has one global maximum at µ∗h.
ii) If ρ < ρ0, we can obtain from (C.5) that limµ→ 1
n−1G1(µ) < 0. As G1(µ)
is a monotonic decreasing function of µ ∈[
1n−1 ,∞
), we have G1(µ) < 0 for
µ ∈[
1n−1 ,∞
). According to (C.3), we can conclude that in this case f1(µ) is a
monotonic decreasing function as f ′1(µ) < 0 for µ ∈[
1n−1 ,∞
).
According to Lemma C.1, we can conclude that the optimal SINR threshold
for C1 is
µ∗1 =
µ∗h if ρ ≥ ρ0
1n−1 otherwise.
(C.8)
2) For C2, denote the objective function of (C.2) as f2(µ) and let us first prove
the following lemma.
Lemma C.2. f2(µ) is a monotonic non-decreasing function of µ ∈(0, 1
n−1
]if
ρ ≥ ρ0. Otherwise, it has one global maximum at µ∗l , where µ∗l is the root of
(3.18).
Proof. f2(µ) is a continuously differentiable function of µ ∈(0, 1
n−1
]. The first-
order derivative of f2(µ) can be written as
f ′2(µ) = n(1+µ)2
exp(− nµµ+1− µ
ρ
)log2 e ·G2(µ), (C.9)
Appendix C 75
where
G2(µ) = (1 + µ)−(
(1+µ)2
ρ+ n)
ln(1 + µ). (C.10)
It can be easily obtained from (C.10) that
limµ→0
G2(µ) = 1, (C.11)
and
limµ→ 1
n−1
G2(µ) = nn−1 − n ln n
n−1 −1ρ·(
nn−1
)2ln n
n−1 . (C.12)
Moreover, the first-order derivative of G2(µ) can be obtained from (C.10) as
G′2(µ) = 1− n1+µ− 1+µ
ρ(1 + 2 ln(1 + µ)) < 0, (C.13)
for µ ∈(0, 1
n−1
], which indicates that G2(µ) is a monotonic decreasing function of
µ ∈(0, 1
n−1
].
i) If ρ ≥ ρ0, we can obtain from (C.12) that limµ→ 1
n−1G2(µ) ≥ 0. As G2(µ) is a
monotonic decreasing function of µ ∈ (0, 1n−1 ], we have G2(µ) ≥ 0 for µ ∈
(0, 1
n−1
].
According to (C.9), we can conclude that in this case f2(µ) is a monotonic non-
decreasing function as f ′2(µ) ≥ 0 for µ ∈(0, 1
n−1
].
ii) If ρ < ρ0, we can obtain from (C.11-C.12) that limµ→0
G2(µ) > 0 and limµ→ 1
n−1
G2(µ) <
0. As G2(µ) is a monotonic decreasing function of µ ∈(0, 1
n−1
], there must exist
µ∗l ∈(0, 1
n−1
], such that G2(µ) > 0 for µ ∈ (0, µ∗l ) and G2(µ) < 0 for µ ∈
(µ∗l ,
1n−1
],
where µ∗l is the root of G2(µ) = 0, which is given in (3.18). We can then obtain
from (C.9) that f ′2(µ) > 0 for µ ∈ (0, µ∗l ) and f ′2(µ) < 0 for µ ∈(µ∗l ,
1n−1
], which
indicates that f2(µ) has one global maximum at µ∗l .
According to Lemma C.2, we can conclude that the optimal SINR threshold
for C2 is
µ∗2 =
1n−1 if ρ ≥ ρ0
µ∗l otherwise.(C.14)
3) By combining (C.8) and (C.14), we can see that if ρ ≥ ρ0, C1 = f1(µ∗h) and
C2 = f2(
1n−1
). As f2
(1
n−1
)= f1
(1
n−1
)and f1
(1
n−1
)≤ C1, we have C1 ≥ C2.
Therefore, we can conclude that in this case the maximum sum rate C = C1 and
the optimal SINR threshold µ∗ = µ∗h.
76 Appendix C
On the other hand, if ρ < ρ0, C1 = f1(
1n−1
)and C2 = f2(µ
∗l ). As f1
(1
n−1
)=
f2(
1n−1
)and f2
(1
n−1
)≤ f2(µ
∗l ), we have C2 ≥ C1. Therefore, we can conclude
that in this case the maximum sum rate C = C2 and the optimal SINR threshold
µ∗ = µ∗l .
Appendix D
Proof of Corollary 3.4
Proof. When ρ ≥ ρ0, the optimal SINR threshold µ∗ = µ∗h according to (3.16).
We can easily obtain from (3.17) that limρ→∞ µ∗h =∞, and
limρ→∞
µ∗hρ
lnµ∗h = limρ→∞
(1µ∗h
+1+µ∗hρ
)ln(1 + µ∗h) = 1. (D.1)
According to (D.1), we further have
limρ→∞
µ∗hρ
= limρ→∞
1lnµ∗h
= 0. (D.2)
Moreover, by applying L’Hopital’s rule on the left-hand side of (D.1), we can
obtain that
limρ→∞
dµ∗hdρ
(1 + lnµ∗h) = 1. (D.3)
Finally, by combining (3.15) with (D.1-D.3), we have
limρ→∞C
log2 ρ= e−1 · limρ→∞
log2 µ∗h
log2 ρ= e−1 · limρ→∞
ρµ∗h(1+lnµ∗h)
= e−1.
77
Appendix E
Proof of Corollary 3.5
Proof. When ρ < ρ0, the optimal SINR threshold µ∗ = µ∗l according to (3.16). We
can easily obtain from (3.18) that limn→∞ µ∗l = 0, and
limn→∞
n log2(1 + µ∗l ) = limn→∞
log2 e
1µ∗l+1
+µ∗l+1
nρ
= log2 e, (E.1)
limn→∞
nµ∗lµ∗l+1
= limn→∞
µ∗lln(1+µ∗l )
− µ∗l (µ∗l+1)
ρ= 1. (E.2)
Finally, by combining (3.15) with (E.1-E.2), we have
limn→∞Cρ<ρ0 = limn→∞ n exp(− nµ∗lµ∗l+1
− µ∗lρ
)· log2(1 + µ∗l ) = e−1 log2 e.
79
Appendix F
Derivation of (3.38-3.40)
We can see from (3.37) that the network throughput performance crucially
depends on⌊1µ− 1
ρ
⌋. Let us specifically consider the following cases:
1.⌊1µ− 1
ρ
⌋< 0: it can be easily obtained from (3.35) that pA = 0. In this case,
no packet can pass through due to an excessively high SINR threshold. Both
the network throughput and the sum rate are 0.
2.⌊1µ− 1
ρ
⌋≥ n− 1: it can be easily obtained from (3.35) that pA = 1. In this
case, all the packets can be successfully decoded, and the network through-
put is λout = nq0 according to (3.37). To maximize the network through-
put, all the nodes should transmit with probability q0 = 1, and the maxi-
mum network throughput is λµ≤ 1
n−1+ 1ρ
max = n. The corresponding sum rate is
n log2(1+µ), which is a monotonic increasing function of the SINR threshold
µ, and is maximized when µ = 1n−1+ 1
ρ
.
3. 0 ≤⌊1µ− 1
ρ
⌋< n − 1: in this case, the network throughput given by (3.37)
has one global maximum at q∗0, where q∗0 is the root of the following equation:∫ 1−q0
0
tn−2−b1µ− 1ρc(1− t)b
1µ− 1ρcdt = (1− q0)n−2−b
1µ− 1ρcqb
1µ− 1ρc+1
0 . (F.1)
The corresponding sum rate is then given by
Rs = nq∗0I1−q∗0
(n− 1−
⌊1
µ− 1
ρ
⌋,
⌊1
µ− 1
ρ
⌋+ 1
)log2(1 + µ). (F.2)
81
82 Appendix F
10-4
10-3
10-2
10-1
100
101
0
0.5
1
1.5
denotes
denotes
3.2
5 dBAWGN
C =1.45
0.02
sR (bit/s/Hz)
1
1
1n
(a)
10-4
10-3
10-2
10-1
100
101
102
0
0.5
1
1.5
2
15 dBAWGN
C =1.85
sR (bit/s/Hz)
0.02 31.6
denotes
denotes
1
1
1n
(b)
Figure F.1: Sum rate of slotted Aloha networks with capture model overAWGN channels versus SINR threshold. n = 50. K = 0 and q0 = q∗0. (a)
ρ = 5dB. (b) ρ = 15dB.
Note that when⌊1µ− 1
ρ
⌋= 0, i.e., µ = ρ, the optimal transmission probabili-
ty can be obtained from (F.1) as q∗0 = 1n. The maximum network throughput
with µ = ρ can be then obtained from (3.37) as λµ=ρmax = (1− 1n)n−1 ≈ e−1 for
large number of nodes n, and the corresponding sum rate is e−1 log2(1 + ρ).
Fig. F.1 illustrates how the sum rate varies with the SINR threshold µ. It can
be clearly observed from Fig. F.1 that there are two local maximum points at
µ = 1n−1+ 1
ρ
and µ = ρ, respectively. Which one is the global maximum point is
determined by the mean received SNR ρ. Let ρ0 denote the root of (3.41). If
ρ < ρ0, we have n log2
(1 + 1
n−1+ 1ρ
)> e−1 log2(1 + ρ). In this case, the maximum
sum rate C = n log2
(1 + 1
n−1+ 1ρ
), achieved when the SINR threshold µ = 1
n−1+ 1ρ
.
Otherwise, µ = ρ is the global maximum point, and the maximum sum rate is
given by C = e−1 log2(1 + ρ).
Appendix G
Derivation of (4.7)-(4.8)
The conditional probability that the packet at the lth iteration can be suc-
cessfully decoded given that there are i concurrent packet transmissions, y(l)i , can
be written as
y(l)i = Pr
{|hl:i+1|2∑i+1
s=l+1 |hs:i+1|2 + 1/ρ≥ µ
}, (G.1)
where |h1:i+1|2 ≥ |h2:i+1|2 ≥ · · · ≥ |hi+1:i+1|2 denotes {|hk|2}k=1,2,...,i+1 in descend-
ing order1. (G.1) can be further written as
y(l)i = Pr
{Z ≤ X
µ− 1
ρ
}=
∫ ∞0
∫ xµ− 1ρ
0
pX,Z(x, z)dzdx, (G.2)
where X = |hl:i+1|2 and Z =∑i+1
s=l+1 |hs:i+1|2, and the joint PDF pX,Z(x, z) is given
by [81]
pX,Z(x, z) =
S · exp (−lx− z)
∑i+1−lk=0
(i+1−lk
)(−1)k·
(z − kx)i−lU(z − kx) for x > 0, 0 < z < (i+ 1− l)x
0 otherwise,
(G.3)
with S = (i+1)!(i+1−l)!(l−1)!(i−l)! , and
U(a) =
1 a ≥ 0
0 otherwise.(G.4)
1Note that the order of received power of packets is solely determined by their small-scalefading gains because the effect of large-scale fading is removed according to (2.1).
83
84 Appendix G
For l = 1, ..., i+ 1, we consider the following two cases:
1) If l ≤ i + 1 − 1µ, we have (i + 1 − l)x > x
µ− 1
ρ. By combining (G.3) and
(G.2), y(l)i can be written as
y(l)i = S ·
∫ ∞µρ
∫ xµ− 1ρ
0
i+1−l∑k=0
(i+ 1− l
k
)(−1)k exp (−lx− z) (z − kx)i−lU(z − kx)dzdx
=(i+1)!
(i+1−l)!(l−1)!
d1/µe−1∑k=0
(i+1−lk
)(−1)k
l+kexp
− l+k
ρ(
1µ−k)( 1
µ−kl+ 1
µ
)i+1−l
.
(G.5)
2) If l > i + 1 − 1µ, we have (i + 1 − l)x > x
µ− 1
ρwhen x < 1
ρ(1µ−i−1+l
) and
(i+ 1− l)x < xµ− 1
ρwhen x > 1
ρ(1µ−i−1+l
) . By combining (G.3) and (G.2), y(l)i can
be written as
y(l)i = S·
∫ 1ρ(1/µ−i−1+l)
µρ
∫ xµ− 1ρ
0
i+1−l∑k=0
(i+1−lk
)(−1)k exp (−lx−z) (z−kx)i−lU(z−kx)dzdx
+ S·∫ ∞
1ρ(1/µ−i−1+l)
∫ (i+1−l)x
0
i+1−l∑k=0
(i+1−lk
)(−1)k exp (−lx−z) (z−kx)i−lU(z−kx)dzdx
=(i+1)!
(i+1−l)!(l−1)!
i+1−l∑k=0
(i+1−lk
)(−1)k
exp
(− k+lρ( 1µ−k)
)k+l
−i−l∑s=0
((i+1−l−k)s
(i+1)s+1·
Q
(1+s,
i+1
ρ( 1µ−i−1+l)
)+ exp
(− k+l
ρ( 1µ−k)
)( 1µ−k)s
( 1µ+l)s+1
·
(1−
Q
(1+s,− 1
ρ(i+1−l− 1µ)·(i+1−l−k)( 1
µ+l)
1µ−k
))) (G.6)
where Q(s, t) = 1(s−1)!
∫∞te−xxs−1dx is the regularized upper incomplete gamma
function. By combining (G.5) and (G.6), we can obtain (4.7).
With µ ≥ 1, for l = 1, ..., i, we have l ≤ i+1− 1µ. By substituting d1/µe−1 = 0
into (G.5), we have
y(l)i =
(i+1
l
)exp
(− lµ
ρ
)·(
1lµ+1
)i+1−l. (G.7)
Appendix G 85
For l = i+ 1, we have l > i+ 1− 1µ. By substituting i+ 1− l = 0 into (G.6), we
have
y(i+1)i = exp
(− (i+1)µ
ρ
). (G.8)
(4.8) can be obtained by combining (G.7) and (G.8).
Appendix H
Derivation of (4.13)
1) For ordered SIC, according to (4.9), we have
rOSn−1 = 1n
n∑m=1
Πml=1y
(l)n−1. (H.1)
With µ = 1n, i + 1 − 1
µ≤ 0 for i = 0, ..., n − 1. Therefore, according to (4.7), we
have
y(l)n−1|µ= 1
n= n!
(n−l)!(l−1)!
n−l∑k=0
(n−lk
)(−1)k
(exp
(− k+lρ(n−k)
)k+l
−n−l−1∑s=0
((n−l−k)sns+1 Q
(1+s, n
ρl
)+ exp
(− k+lρ(n−k)
)(n−k)s(n+l)s+1 ·
(1−Q
(1+s, 1
ρl· (n−l−k)(n+l)
n−k
)))), (H.2)
for l = 1, ..., n. Note that for large ρ, exp(− k+lρ(n−k)
)≈ 1, and the item
n−l−1∑s=0
((n−l−k)sns+1 Q
(1+s, n
ρl
)+ exp
(− k+lρ(n−k)
)(n−k)s(n+l)s+1 ·
(1−Q
(1+s, 1
ρl· (n−l−k)(n+l)
n−k
)))
can be approximated to∑n−l−1
s=0(n−l−k)sns+1 =
(1−
(n−l−kn
)n−l)/(k + l) because
Q(a, b) approaches 1 as b approaches 0. Therefore, y(l)n−1|µ= 1
nfor l = 1, ..., n can be
approximated to
y(l)n−1|µ= 1
n
for large ρ≈ n!
(n−l)!(l−1)!
n−l∑k=0
(n−lk
)(−1)kk+l
(n−l−kn
)n−l= 1. (H.3)
By combining (H.1) and (H.3), we can obtain that rOSn−1|µ= 1n≈ 1 for large ρ.
87
88 Appendix H
2) For unordered SIC, by combining (4.1) and (4.12), we have
rNS ln−1 |µ= 1
n=
(1 +
exp(− 1nρ
)n(1+
1n
)n−1
)n
− 1. (H.4)
For large n and ρ, we have exp(− 1nρ
)≈ 1 and
(1 + 1
n
)n−1 ≈ e. (H.4) can then be
approximated to rNS ln−1 |µ= 1
n
for large n, ρ≈
(1 + e−1
n
)n− 1
for large n≈ ee
−1 − 1.
Appendix I
Derivation of (4.14)
1) For ordered SIC, rOSi can be written as
rOSi = 1i+1
(y(1)i + y
(1)i y
(2)i + y
(1)i y
(2)i y
(3)i + · · ·+ y
(1)i · · · y
(i+1)i
), (I.1)
according to (4.9). For µ > 1, we havey(1)i
i+1= rCi by combining (4.8) and (4.1).
Therefore, we have
limµ→∞
rOSirCi
= limµ→∞
1 + y(2)i + y
(2)i y
(3)i + · · ·+ y
(2)i · · · y
(i+1)i . (I.2)
As µ → ∞, it can be easily obtained from (4.8) that limµ→∞ y(l)i = 0. We can
then obtain from (I.2) that limµ→∞rOSirCi
= 1.
2) For unordered SIC, we have
limµ→∞
rNS li
rCi= lim
µ→∞(1+µrCi )i+1−1
(i+1)µrCi, (I.3)
according to (4.12). Further note from (4.1) that
limµ→∞
µrCi = limµ→∞
ρ
exp(µρ
)(1+µ)i+1(1+µ+iρ)
= 0. (I.4)
We then have limµ→∞rNS li
rCi= 1 by combining (I.4) and (I.3).
89
Appendix J
Derivation of (4.20)-(4.23)
According to (4.19), the first-order derivative of λout with respect to q0 can
be written as
dλoutdq0
= nn−1∑i=0
(n− 1
i
)ri (1− q0)n−2−i qi0(1 + i− nq0). (J.1)
It can be easily obtained from (J.1) that
dλoutdq0|q0=0= (n− 1)r0 > 0, (J.2)
anddλoutdq0|q0=1= nrn−1 − (n− 1)rn−2. (J.3)
Let µ0 denote the root of nrn−1 − (n − 1)rn−2 = 0. By substituting (4.9) and
(4.12) into (J.1), we can obtain dλoutdq0
in the ordered SIC and unordered SIC cases,
respectively, and Fig. J.1 illustrates how dλoutdq0
varies with the transmission prob-
ability q0. We can see from Fig. J.1 that with µ < µ0,dλoutdq0
> 0 for q0 ∈ (0, 1).
Therefore, the maximum network throughput λmax is achieved at q∗0 = 1, and
we have λmax = nrn−1 according to (4.19). On the other hand, with µ ≥ µ0,
we have dλoutdq0|q0=1≤ 0. It can be clearly seen from Fig. J.1 that dλout
dq0> 0 for
q0 ∈ (0, q0) and dλoutdq0≤ 0 for q0 ∈ [q0, 1], where q0 is the root of dλout
dq0= 0. In this
case, the maximum network throughput λmax is achieved at q∗0 = q0, and we have
λmax = n∑n−1
i=0
(n−1i
)ri (1− q0)n−1−i qi+1
0 according to (4.19).
91
92 Appendix J
0
OSm m<
0
OSm m>
0qdenotes
m = 0.05
m = 0.1
m = 0.5
m = 1
0
ˆ out
d dql
0q
(a)
m = 0.05
m = 0.1
m = 0.5
m = 1
0
ˆ out
d dql
0q
0
NSm m<
0
NSm m>
0qdenotes
(b)
Figure J.1:dλoutdq0
versus transmission probability q0. n = 20 and ρ = 20dB.
(a) Ordered SIC. (b) Unordered SIC.
Appendix K
Derivation of (4.27)-(4.28)
By substituting (4.9) and (4.12) into (4.25) and (4.26), respectively, we can
numerically calculate µh and µl for both the ordered SIC and unordered SIC cases
as shown in Fig. K.1a, and further obtain C1 = f1(µh) and C2 = f2(µl) as shown
in Fig. K.1b. We can see from Fig. K.1b that in both cases, f1(µh) and f2(µl)
are monotonically increasing functions with respect to the mean received SNR ρ,
and there exists a unique point ρ0 such that when ρ < ρ0, f1(µh) < f2(µl) and
f1(µh) > f2(µl) when ρ > ρ0, where ρ0 is the root of f1(µh) = f2(µl). We can
then conclude that the maximum sum rate C = C1 = f1(µh) if ρ ≥ ρ0, which is
achieved when µ∗ = µh. Otherwise, the maximum sum rate C = C2 = f2(µl),
which is achieved when µ∗ = µl.
(dB) r0NS
r 0OS
r
lm
hm
and
hl
mm
(a)
(dB) r0NS
r 0OS
r
2 2( )lC f m=
1 1( )hC f m=
12
and
CC
(b)
Figure K.1: (a) µh and µl versus mean received SNR ρ. (b) C1 and C2 versusmean received SNR ρ. n = 20.
93
Appendix L
Derivation of (4.29)-(4.30)
Based on the collision model, a packet transmission is successful if and only if
there are no concurrent transmissions and its received SNR is above the threshold
µ. The steady-state probability of successful transmission of HOL packets, p, can
be then written as
p = Pr{no concurrent packet transmissions}·Pr{received SNR is above the threshold µ}.(L.1)
According to (3.5), the probability that there are no concurrent transmissions is
given by
Pr{no concurrent packet transmissions} =(
1− πTp
)n−1. (L.2)
Since the received SNR is exponentially distributed with mean ρ, the probability
that the received SNR is above the threshold µ is given by
Pr{received SNR is above the threshold µ} = exp(−µρ
). (L.3)
By substituting (L.2) and (L.3) into (L.1), we have
p =(
1− πTp
)n−1· exp
(−µρ
)for large n≈ exp
(−µρ− nπT
p
), (L.4)
which can be further written as
p = exp
(−µρ− n∑K−1
i=0
p(1−p)iqi
+(1−p)KqK
), (L.5)
95
96 Appendix L
according to (2.2). With qi = q0, i = 0, ..., K, the fixed-point equation (L.5) has
a single non-zero root pcA, which is given by pcA = exp(−µρ− nq0
). The network
throughput in saturated conditions can be then obtained as λcollisionout = nq0pcA =
nq0 exp(−µρ− nq0
), which is maximized at
λcollisionmax = exp(−1− µ
ρ
), (L.6)
when the transmission probability q0 is set to be q∗,collision0 = 1n. By combining
(2.8) and (L.6), the maximum sum rate can be written as
Ccollision = maxµ>0
exp(−1− µ
ρ
)· log2(1 + µ). (L.7)
Let f(µ) denote the objective function of (L.7). It can be easily shown that
f ′(µ) ≥ 0 for µ ∈ (0, µ∗,collision] and f ′(µ) < 0 for µ ∈ (µ∗,collision,∞), indicating
that f(µ) has one global maximum at µ∗,collision, where µ∗,collision is the root of
(µ+ 1)(µ+1)/ρ = e, which is given by
µ∗,collision = eW0(ρ) − 1. (L.8)
(4.29) and (4.30) can be then obtained by substituting (L.8) into (L.7) and (L.6),
respectively.
Bibliography
[1] C. S. R. Murthy and B. S. Manoj, Ad Hoc Wireless Networks: Architectures
and Protocols, Prentice Hall, 2004.
[2] J. F. Kurose and K. W. Ross, Computer Networking: A Top-Down Approach
(5th Edition), Addison Wesley, 2009.
[3] F. Ghavimi and H.H Chen, “M2M Communications in 3GPP LTE/LTE-A
Networks: Architectures, Service Requirements, Challenges, and Applications,”
IEEE Commun. Surveys Tuts., vol. 17, no. 2, pp. 525–549, 2015.
[4] A. Laya, L. Alonso and J. Alonso-Zarate, “Is the Random Access Channel of
LTE and LTE-A Suitable for M2M Communications? A Survey of Alternatives,”
IEEE Commun. Surveys Tuts., vol. 16, no. 1, pp. 4–16, 2014.
[5] N. Abramson, “The Aloha system: another alternative for computer commu-
nications,” in Proc. Fall joint Compet. Conf., vol. 44, pp. 281–285, Nov. 1970.
[6] D. W. Lipke, D. W. Swearingen, J. F. Parker, E. E. Steinbrecher, T. O. Calvit,
H. Dodel, “MARISAT - a Maritime Satellite Communications System,” COM-
SAT Technical Review, vol. 7, pp. 351–391, Nov. 1977.
[7] N. Abramson, “Packet switching with satellites,” in Proc. Nat. joint Comput.
Conf., vol. 42, pp. 695–702, 1973.
[8] L. Kleinrock and F. Tobagi, “Packet switching in radio channels: Part I–carrier
sense multiple-access modes and their throughput-delay characteristics,” IEEE
Trans. Commun., vol. 23, no. 12, pp. 1400–1416, 1975.
[9] A. Carleial and M. Hellman, “Bistable behavior of Aloha-type systems,” IEEE
Trans. Commun., vol. 23, no. 4, pp. 401–410, Apr. 1975.
97
98 Bibliography
[10] S. Lam and L. Kleinrock, “Packet switching in a multiaccess broadcast chan-
nel: Dynamic control procedures,” IEEE Trans. Commun., vol. 23, no. 9, pp.
891–904, Sept. 1975.
[11] M. J. Feguson, “On the control, stability, and waiting time in a slotted ALO-
HA random-access system,” IEEE Trans. Commun., vol. 23, no. 11, pp. 1306–
1311, Nov. 1975.
[12] G. Fayolle, E. Gelenbe, and J. Labetoulle, “Stability and optimal control of
the packet switching broadcast channel,” J. Assoc. Comput. Machinery, vol. 24,
no. 3, pp. 375–386, Jul. 1977.
[13] V. A. Mikhailov, Method of Random Multiple Access, Candidate Engineering
Thesis, Moscow Institute of Phy. and Tech., 1979.
[14] B. Hajek and T. Van Loon, “Decentralized dynamic control of a multiaccess
broadcast channel,” IEEE Trans. Automat. Contr., vol. 27, no. 3, pp. 559–569,
Jun. 1982.
[15] R. Rivest, “Network control by Bayesian broadcast,” IEEE Trans. Inf. The-
ory, vol. 33, no. 3, pp. 323–328, May 1987.
[16] B. Tsybakov and W. Mikhailov, “Ergodicity of slotted Aloha system,” Probl.
Inf. Transmission, vol. 15, no. 4, pp. 73–87, 1979.
[17] R. Rao and A. Ephremides, “On the stability of interacting queues in a
multiple-access system,” IEEE Trans. Inf. Theory, vol. 34, no. 5, pp. 918–930,
Sep. 1988.
[18] V. Anantharam, “The stability region of the finite-user slotted Aloha proto-
col,” IEEE Trans. Inf. Theory, vol. 37, no. 3, pp. 535–540, May 1991.
[19] W. Luo and A. Ephremides, “Stability of n interacting queues in random
access systems,” IEEE Trans. Inf. Theory, vol. 45, no. 5, pp. 1579–1587, July
1999.
[20] W. Szpankowski, “Stability conditions for some multiqueue distributed sys-
tems: buffered random access systems,” Adv. Appl. Prob., vol. 26, no. 2, pp.
498–515, June 1994.
[21] L. Dai, “Stability and delay analysis of buffered Aloha networks,” IEEE Tran-
s. Wireless Commun., vol. 11, no. 8, pp. 2707–2719, Aug. 2012.
Bibliography 99
[22] L. Dai, “Toward a coherent theory of CSMA and Aloha,” IEEE Trans. Wire-
less Commun., vol. 12, no. 7, pp. 3428–3444, Jul. 2013.
[23] T. Wan and A. Sheikh, “Performance and stability analysis of buffered slotted
Aloha protocols using tagged user approach,” IEEE Trans. Veh. Technol., vol.
49, no. 2, pp. 582–593, Mar. 2000.
[24] G. Bianchi, “Performance analysis of the IEEE 802.11 distributed coordina-
tion function,” IEEE J. Sel. Areas Commun., vol. 18, no. 3, pp. 535–547, Mar.
2000.
[25] J. Y. Hui, Switching and Traffic Theory for Integrated Broadband Networks,
Kluwer Academic Publishers, 1990.
[26] B. Rimoldi and R. Urbanke, “A rate-splitting approach to the Gaussian
multiple-access channel,” IEEE Trans. Inf. Theory, vol. 42, no. 2, pp. 364–375,
Mar. 1996.
[27] M. Medard, J. Huang, A. Goldsmith, S. Meyn, and T. Coleman, “Capacity
of time-slotted Aloha packetized multiple-access systems over the AWGN chan-
nel,” IEEE Trans. Wireless Commun., vol. 3, no. 2, pp. 486–499, 2004.
[28] P. Minero, M. Franceschetti, and D. Tse, “Random access: An information-
theoretic perspective,” IEEE Trans. Inf. Theory, vol. 58, no. 2, pp. 909–930,
2012.
[29] X. Qin and R. Berry, “Distributed approaches for exploiting multiuser diver-
sity in wireless networks,” IEEE Trans. Inf. Theory, vol. 52, no. 2, pp. 392–413,
Feb. 2006.
[30] F. Baccelli, B. Blaszczyszyn, and P. Mhlethaler, “An Aloha protocol for mul-
tihop mobile wireless networks,” IEEE Trans. Inf. Theory, vol. 52, no. 2, pp.
421–436, Feb. 2006.
[31] R. Ganti and M. Haenggi, “Spatial and temporal correlation of the interfer-
ence in ALOHA ad hoc networks,” IEEE Commun. Lett., vol. 13, no. 9, pp.
631–633, Sep. 2009.
[32] S. Weber, J. G. Andrews, and N. Jindal, “An overview of the transmission
capacity of wireless networks,” IEEE Trans. Commun., vol. 58, no. 12, pp.
3593–3604, Dec. 2010.
100 Bibliography
[33] P. H. J. Nardelli, M. Kaynia, P. Cardieri, and M. Latva-aho, “Optimal trans-
mission capacity of ad hoc networks with packet retransmissions,” IEEE Trans.
Wireless Commun., vol. 11, no. 8, pp. 2760–2766, Aug. 2012.
[34] F. Baccelli and C. K. Singh, “Adaptive spatial Aloha, fairness and stochas-
tic geometry,” in Proc. of Workshop on Spatial Stochastic Models for Wireless
Netw., pp. 7–14, May 2013.
[35] Y. Zhong, W. Zhang, and M. Haenggi, “Managing interference correlation
through random medium access,” IEEE Trans. Wireless Commun., vol. 13, no.
2, pp. 928–941, Feb. 2014.
[36] Y. George, I. Bergel and E. Zehavi, “Upper bound on the ergodic rate density
of ALOHA wireless ad-hoc networks,” IEEE Trans. Wireless Commun., vol. 14,
no. 6, pp. 3004–3014, Jun. 2015.
[37] L. G. Roberts, “Aloha packet system with and without slots and capture,”
ARPANET Satellite System Note 8, NIC Document No. 11290, Stanford Re-
search Institute, 1972.
[38] J. Arnbak and W. van Blitterswijk, “Capacity of slotted Aloha in Rayleigh-
fading channels,” IEEE J. Sel. Areas Commun., vol. 5, pp. 261–269, Feb. 1987.
[39] D. Goodman and A. Saleh, “The near/far effect in local Aloha radio commu-
nications,” IEEE Trans. Veh. Technol., vol. 36, no. 1, pp. 19–27, Feb. 1987.
[40] A. Sheikh, Y. D. Yao, and X. Wu, “The Aloha systems in shadowed mobile
radio channels with slow or fast fading,” IEEE Trans. Veh. Technol., vol. 39,
no. 4, pp. 289–298, Nov. 1990.
[41] C. Namislo, “Analysis of mobile radio slotted Aloha networks,” IEEE J. Sel.
Areas Commun., vol. 2, no. 4, pp. 583–588, Jul. 1984.
[42] I. Habbab, M. Kavehrad, and C. E. Sundberg, “Aloha with capture over slow
and fast fading radio channels with coding and diversity,” IEEE J. Sel. Areas
Commun., vol. 7, no. 1, pp. 79–88, Jan. 1989.
[43] C. V. der Plas and J. P. M. G. Linnartz, “Stability of mobile slotted Aloha
network with Rayleigh fading, shadowing, and near-far effect,” IEEE Trans.
Veh. Technol., vol. 39, no. 4, pp. 359–366, Nov. 1990.
Bibliography 101
[44] M. Zorzi and R. Rao, “Capture and retransmission control in mobile radio,”
IEEE J. Sel. Areas Commun., vol. 12, no. 8, pp. 1289–1298, Oct. 1994.
[45] M. Peh, S. Hanly, and P. Whiting, “Random-access over fading channels,” in
Proc. IEEE Globecom, San Francisco, CA, pp. 888–892, Dec. 2003.
[46] G. del Angel and T. Fine, “Information capacity and power control for slotted
Aloha random-access systems,” IEEE Trans. Inf. Theory, vol. 51, no. 12, pp.
4074–4090, 2005.
[47] J. Luo and A. Ephremides, “Power levels and packet lengths in random multi-
ple access with multiple-packet reception capability,” IEEE Trans. Inf. Theory,
vol. 52, pp. 414–420, Feb. 2006.
[48] S. Rasool and A. U. H. Sheikh, “An approximate analysis of buffered S-Aloha
in fading channels using tagged user analysis,” IEEE Trans. Wireless Commun.,
vol. 6, no. 4, pp. 1320–1326, 2007.
[49] J. Wieselthier, G. D. Nguyen, and A. Ephremides, “Throughput (bits/sec/hz)
of capture-based random-access systems with SINR channel models,” in Proc.
IEEE Int. Symp. Inf. Theory (ISIT), 2007, pp. 2401–2405.
[50] V. Naware and L. Tong, “Cross layer design for multiaccess communication
over Rayleigh fading channels,” IEEE Trans. Wireless Commun., vol. 7, no. 3,
pp. 1095–1103, 2008.
[51] A. Dua, “Random access with multi-packet reception,” IEEE Trans. Wireless
Commun., vol. 7, no. 6, pp. 2280–2288, Jun. 2008.
[52] T. Shinomiya and H. Suzuki, “Slotted Aloha mobile packet communication
systems with multiuser detection in a base station,” IEEE Trans. Veh. Technol.,
vol. 49, no. 3, pp. 948–955, 2000.
[53] S. Adireddy and L. Tong, “Exploiting decentralized channel state information
for random access,” IEEE Trans. Inf. Theory, vol. 51, no. 2, pp. 537–561, Feb.
2005.
[54] V. Naware, G. Mergen, and L. Tong, “Stability and delay of finite-user slotted
Aloha with multipacket reception,” IEEE Trans. Inf. Theory, vol. 51, no. 7, pp.
2636–2656, Jul. 2005.
102 Bibliography
[55] R. Yim, N. Mehta, A. Molisch, and J. Zhang, “Dual power multiple access
with multipacket reception using local CSI,” IEEE Trans. Wireless Commun.,
vol. 8, no. 8, pp. 4078–4088, 2009.
[56] A. Zanella and M. Zorzi, “Theoretical analysis of the capture probability in
wireless systems with multiple packet reception capabilities,” IEEE Trans. Com-
mun., vol. 60, no. 4, pp. 1058–1071, 2012.
[57] S. Ghez, S. Verdu, and S. Schwartz, “Stability properties of slotted Aloha with
multipacket reception capability,” IEEE Trans. Autom. Control, vol. 33, no. 7,
pp. 640–649, 1988.
[58] S. Ghez, S. Verdu, and S. Schwartz, “Optimal decentralized control in the
random access multipacket channel,” IEEE Trans. Autom. Control, vol. 34, no.
11, pp. 1153–1163, 1989.
[59] Y. J. Zhang, P. X. Zheng, and S. C. Liew, “How Does Multiple–Packet Re-
ception Capability Scale the Performance of Wireless Local Area Networks?”
IEEE Trans. Mobile Comput., vol. 8, no. 7, pp. 923–935, Jul. 2009.
[60] Y. J. Zhang, S. C. Liew, and D. R. Chen, “Sustainable Throughput of Wireless
LANs with Multipacket Reception Capability under Bounded Delay-Moment
Requirements,” IEEE Trans. Mobile Comput., vol. 9, no. 9, pp. 1226–1241,
Sept. 2010.
[61] D. S. Chan, T. Berger, and L. Tong, “Carrier sense multiple access commu-
nications on multipacket reception channels: Theory and applications to IEEE
802.11 wireless networks,” IEEE Trans. Commun., vol. 61, no. 1, pp. 266–278,
Jan. 2013.
[62] Y. H. Bae, B. D. Choi, and S. Alfa, “Achieving Maximum Throughput in
Random Access Protocols with Multipacket Reception” IEEE Trans. Mobile
Comput., vol. 13, no. 3, pp. 497–511, Mar. 2014.
[63] C. Xu, L. Ping, P. Wang, S. Chan, and X. Lin, “Decentralized power control
for random access with successive interference cancellation,” IEEE J. Select.
Areas Commun., vol. 31, no. 8, pp. 1–10, Nov. 2013.
[64] R. M. Corless, G. H. Gonnet, D. E. G. Hare, D. J. Jeffrey, and D. E. Knuth,
“On the Lambert W function,” Adv. Comput. Math., vol. 5, pp. 329–359, 1996.
Bibliography 103
[65] D. Tse and P. Viswanath, Fundamentals of Wireless Communication, Cam-
bridge University Press, 2005.
[66] R. Metcalfe and D. Boggs, “Ethernet: distributed packet switching for local
computer networks,” Commun. ACM, vol. 19, no. 7, pp. 395–404, July 1976.
[67] J. Goodman, A. G. Greenberg, N. Madras, and P. March, “Stability of binary
exponential backoff,” J. ACM, vol. 35, no. 3, pp. 579–602, July 1988.
[68] L.A. Goldberg, H. Al-Ammal and P.D. MacKenzie, “An improved stability
bound for binary exponential backoff,” Theory Comput. Syst., pp. 229–244, Jan.
2001.
[69] Byung-Jae Kwak, Nah-Oak Song, and L. E. Miller, “Performance Analysis
of Exponential Backoff,” IEEE/ACM Trans. Networking, vol. 13, no. 2, pp.
343–355, Apr. 2005.
[70] T. Cover, “Broadcast channels,” IEEE Trans. Inf. Theory, vol. IT-18, no. 1,
pp. 2–14, Jan. 1972.
[71] A. J. Viterbi, “Very Low Rate Convolutional Codes for Maximum Theoretical
Performance of Spreadspectrum Multiple-Access Channels,” IEEE J. Select.
Areas Commun., vol. 8, pp. 641–649, May 1990.
[72] P. Patel and J. Holtzman, “Analysis of a Simple Successive Interference Can-
cellation Scheme in a DS/CDMA System,” IEEE J. Select. Areas Commun.,
vol. 12, no. 5, pp. 796–807, June 1994.
[73] J. G. Andrews, “Interference cancellation for cellular systems: A contempo-
rary overview,” IEEE Wireless Commun., vol. 12, no. 2, pp. 19–29, Apr. 2005.
[74] M. Wildemeersch, T. Q. S. Quek, M. Kountouris, A. Rabbachin, and C.
Slump, “Successive interference cancellation in heterogeneous networks,” IEEE
Trans. Commun., vol. 62, no. 12, pp. 4440–4453, Dec. 2014.
[75] E. Casini, R. De Gaudenzi, and O. del Rio Herrero, “Contention resolution
diversity slotted ALOHA (CRDSA): an enhanced random access scheme for
satellite access packet networks,” IEEE Trans. Wireless Commun., vol. 6, no.
4, pp. 1408–1419, Apr. 2007.
104 Bibliography
[76] G. Liva, “Graph-based analysis and optimization of contention resolution
diversity slotted ALOHA,” IEEE Trans. Commun., vol. 59, no. 2, pp. 477–487,
Feb. 2011.
[77] C. Stefanovic and P. Popovski, “ALOHA random access that operates as a
rateless code,” IEEE Trans. Commun., vol. 61, no. 11, pp. 4653–4662, Nov.
2013.
[78] J. Goseling, M. Gastpar, and J. H. Weber, “Random access with physical-
layer network coding,” IEEE Trans. Inf. Theory, vol. 61, no. 7, pp. 3670–3681,
Jul. 2015.
[79] E. Paolini, C. Stefanovic, G. Liva and P. Popovski, “Coded random access:
applying codes on graphs to design random access protocols,” IEEE Commun.
Mag., vol. 53, no. 6, pp. 1408–1419, Jun. 2015.
[80] M. Shaked and J. G. Shanthikumar, Stochastic Orders, Springer Series in
Statistics, 2007.
[81] H. C. Yang and M. S. Alouini, Order Statistics in Wireless Communications:
Diversity, Adaptation, and Scheduling in MIMO and OFDM Systems, Cam-
bridge University Press, 2011.
[82] J. Kim, J. Lee, J. Kim and J. Yun, “M2M Service Platforms: Survey, Issues,
and Enabling Technologies,” IEEE Commun. Surveys Tuts., vol. 16, no. 1, pp.
61–76, 2014.
[83] A. Lo, Y. W. Law and M. Jacobsson, “A cellular-centric service architecture
for machine-to-machine (M2M) communications,” IEEE Wireless Commun.,
vol. 20, no. 5, pp. 143–151, Oct. 2013.
[84] 3GPP TR 23.888 V1.0.0, “System Improvement for Machine-Type Commu-
nications,” Sept. 2010.
[85] 3GPP TS 22.368 V11.0.0, “Service Requirements for Machine-Type Commu-
nications,” Dec. 2010.
[86] 3GPP TR 37.868 V11.0.0,“Study on RAN Improvements for Machine-Type
Communications,” Sep. 2011.
Bibliography 105
[87] S. Y. Lien, K. C. Chen, and Y. Lin, “Toward ubiquitous massive accesses
in 3GPP machine-to-machine communications,” IEEE Commun. Mag., vol. 49,
no. 4, pp. 66–74, April 2011.
[88] M. Y. Cheng, G. Y. Lin, H. Y. Wei, and A. C. Hsu, “Overload control
for Machine-Type-Communications in LTE-Advanced system,” IEEE Commun.
Mag., vol. 50, no. 6, pp. 38–45, June 2012.
[89] C. Y. Oh, D. Hwang and T. J. Lee, “Joint Access Control and Resource
Allocation for Concurrent and Massive Access of M2M Devices,” IEEE Wireless
Commun., vol. 14, no. 8, pp. 4182–4192, Aug. 2015.
[90] M. Koseoglu, “Lower Bounds on the LTE-A Average Random Access Delay
Under Massive M2M Arrivals,” IEEE Trans. Commun., vol. 64, no. 5, pp. 2104–
2115, May 2016.
[91] A. Aijaz, M. Tshangini, M. R. Nakhai, X. Chu and A. H. Aghvami, “Energy-
Efficient Uplink Resource Allocation in LTE Networks With M2M/H2H Co-
Existence Under Statistical QoS Guarantees,” IEEE Trans. Commun., vol. 62,
no. 7, pp. 2353–2365, July 2014.
[92] H. S. Dhillon, H. C. Huang, H. Viswanathan and R. A. Valenzuela, “Power-
Efficient System Design for Cellular-Based Machine-to-Machine Communica-
tions,” IEEE Wireless Commun., vol. 12, no. 11, pp. 5740–5753, Nov. 2013.