| Summary: | 博士 === 臺灣大學 === 資訊管理學研究所 === 95 === It is now possible to access data services anywhere, any time, via wireless networks
ranging from PWANs (Personal Wireless Area Networks) to office or home area networks,
from mesh networks and WiMax to satellite networks. As for the future of network
technologies, it is essential that research be directed toward improving person-to-person,
person-to-machine, and machine-to-machine communications. Thus, in this dissertation, we
focus on wireless networks, as well as the challenges and research avenues presented by
network planning and performance.
Primarily focusing on network architecture and network layers, the research scope of this
dissertation covers various network architectures, such as Wi-Fi hotspots, mesh networks,
and ad hoc networks (including sensor networks); and considers various network layers: the
application layer, the network layer, media access control (MAC), and the physical layer.
Previous related research is discussed in Chapter 1.
Both network planning and performance optimization issues are addressed. The following
is a brief summary of the presentation of these issues to be addressed in depth in the body of
this dissertation:
• Wi-Fi hotspots [Chapter 2]
In such hotspots, the transmission bit rate for a mobile device (MD) is dependent on its
distance from the nearest base-station. A problem arises when fast and slow MDs share
Abstract vii
a network in that, despite the higher capability of a fast MD, the throughput of that fast
MD is the same as that of a slow MD. Therefore, we address this problem and propose
an algorithm to achieve channel access time fairness. Our research includes
comparative studies of three adaptive MAC parameters: (i) the packet size, (ii) the
initial contention window size, and (iii) multiple back-to-back packets. On the basis of
that research, we have determined that adjusting the size of the initial contention
window provides the most significant optimization of the maximum system throughput.
It has been established that determining a global optimal solution is impossible in a
reasonable time; therefore, a modified binary search algorithm is implemented to solve
the problem. Experiment results show that the system throughput is 5.92 Mbps, which
is a 21% improvement over the original MAC protocol.
• Mesh networks [Chapter 3-4]
In mesh networks, the main issues are the performance and fairness of the system or
individual devices due to spatial bias. The issues addressed include: (i) top
load-balanced routing; (ii) end-to-end delay fairness; and (iii) backhaul assignment
problems, which have proven to be NP-complete. In this dissertation, these problems
are formulated as mixed-integer nonlinear programming problems. Lagrangean
Relaxation (LR) is used to solve the primal and Lagrangean dual problems, and to
obtain upper and lower bounds. Gaps between research issues (i) and (ii) are shown to
be less than 5%. Although a larger gap exists between issues (i) and (iii), i.e., 40%, the
improvement ratio is still 10% over other modified methods.
• Ad hoc networks [Chapter 5]
For ad hoc networks, the main concern addressed in this dissertation is the transmission
of multicast messages via broadcasting. The advantage of this method is that it obtains
the so-called “wireless broadcast advantage”. The same message is sent only once, but
it is received by many devices. Based on routing paths, we propose an
optimization-based integer- and nonlinear-programming model. The radius of each
node is calculated intelligently according to the structure of the broadcast tree, thus
minimizing the total power consumption required to broadcast each multicast message
to all receivers. This problem has also proven to be NP-complete. We adopt LR
methods to solve the problem, and determine the gaps to be within 10%.
viii Wen - Performance Optimization Algorithms for Wireless Networks
This static network research problem is extended to include mobility issues in mobile
networks. The message is broken down into smaller sub-sections. For a mobile node,
given the direction and speed, the duration of the current broadcast tree is found. New
broadcast trees are constructed to provide coverage to multicast group nodes until the
complete message is sent. Like the previous static case, this is also an NP-complete
problem. We solve it by LR, which obtains a gap of less than 30%. Our experiment
results show that the proposed algorithms outperform the MSPT, Prim MST, BIP, and
GIBT heuristics by at least 5%.
• Sensor networks [Chapter 6,7]
Sensors are typically scattered throughout an area of interest. As they may be located in
remote areas, recharging the sensors’ batteries is often infeasible. The network lifetime
of a wireless sensor network, which is interrupted when depleted batteries prevent the
transmission of environmental information, is dependent on battery capacity and
energy consumption efficiency, and has become a crucial issue in sensor network
research. Therefore, to prolong network lifetime starting from the physical layer and
extending all the way up to the application layer, we focus on: (i) multi-rate routing; (ii)
dynamic adjustment of the nodal transmission radius; (iii) duty cycle scheduling; (iv)
collision avoidance; (v) routing; and (vi) data aggregation. All combinations of these
six issues are considered within multi-sink and cluster-based architectures. These are
serial problems, formulated as mixed-integer nonlinear programming problems that
have proven to be NP-complete. Thus, the LR approach is used to find solutions to the
serial problems. Meanwhile, algorithms, including an O-MAC protocol and a serial
DAR (data aggregation routing) algorithm, are proposed to optimize energy
consumption. The feasible solution is derived from information provided by the
Lagrangean multipliers, and compared with the performance of other heuristics, such as
GIT, CNS, or SPT, which are modified to satisfy constraints on the research problem.
Our experiment results show that the proposed heuristic outperforms the others
approaches by 7%-43%.
Conclusions and extensions of the work in this dissertation are presented in the final
portion of the dissertation [i.e., Chapter 8], including additional issues that could be
addressed in future research, such as scheduling, admission control, and end-to-end delay in
Abstract ix
IEEE 802.16 broadband wireless area (BWA) networks. Accordingly, these issues are listed
as follows:
• Mesh networks + Wi-Fi hotspot networks
The signal may overshoot, even when the multi-channel is used. As the interference is
considered, the transmission error reduces the link capacity C(u,v), so that the traffic
flow is limited. In addition, if the interference issue is considered, it increases the
number of retransmissions which means increasing the node-to-node delay. Thus, the
interference issue is extended as one of our future work.
• Ad hoc and sensor networks
The proposed maximization of mobile network lifetime may be extended to include
balancing the power consumption of all nodes within a multiple session construction.
• IEEE 802.16 BWA networks
Potential future research in this area includes: (i) optimization of the relative
parameters and placing controls on scheduling and admission to minimize delay or
maximize performance under quality of service considerations; and (ii) minimization of
end-to-end delay with controls on scheduling in the IEEE 802.16j.
|
|---|
