Type of Document Dissertation Author Ma, Xiaoguang Author's Email Address email@example.com URN etd-07272010-163845 Title Spectrum Management in Wireless Networks Degree Doctor of Philosophy Department Electrical and Computer Engineering, Department of Advisory Committee
Advisor Name Title Ming Yu Committee Chair Bing W. Kwan Committee Member Bruce A. Harvey Committee Member Zhenhai Duan University Representative Keywords
- Resource Management
- Wireless Resource Management
- Cognitive Radio
- Sensor Networks
- Wireless Networks
- Power Management
- Radio Channel Allocation
Date of Defense 2010-06-15 Availability unrestricted AbstractThe limited spectrum provided by the IEEE 802.11 standard is not efficiently utilized in the existing wireless networks. The inefficiency comes from three issues in spectrum management. First, the utilization of the available non-overlapping channels is not evenly distributed, that is, closely deployed users tend to congregate in the same or interfering channels. This issue incurs an excessive amount of co-channel interference (CCI), causing collisions, and thus decreases network throughput. Second, the dynamic radio channel allocation (RCA) problem is non-deterministic polynomial-time hard (NP-hard). The employed heuristic optimization methods can not efficiently find a global optimum, including simple minimization or maximization processes, or certain slow learning processes. Third, the default transmission power of a user reserves unnecessarily large deference areas, in which the collision avoidance (CA) mechanisms prohibit simultaneous transmissions in a given channel. Consequently, the spatial channel reuse is significantly reduced. For the first issue, many RCA algorithms have been proposed. The objective is to minimize CCI among co-channel users while increasing network throughput. Most RCA algorithms use heuristic optimization methods, which have restricted performance limited by one or more of the following aspects.
1)Their evaluation variables may not properly reflect the CCI levels in a network, e.g., the number of co-channel users, the local energy levels, etc..
2)The dynamic RCA problem is non-deterministic polynomial-time hard (NP-hard). The employed heuristic optimization methods can not efficiently find a global optimum, e.g., simple minimization or maximization processes, or certain slow learning processes.
3)The information gathering and processing approaches in these RCA algorithms require prohibitive overheads, such as a common control channel or a central controller.
4)Some unrealistic premises are used, e.g., all users in the same channel can hear each other.
5)Most RCA algorithms are designed for some specific networks. For example, an algorithm designed for organized-or-information sharing (OIS) networks does not work properly in non-organized-nor-information-Sharing (NOIS) networks.
For the second issue, it is worth pointing out that the complexity of the existing distributed RCA algorithms has not been studied. For the third issue, various power control algorithms, including courtesy algorithms and opportunistic algorithms, have been introduced to restrain transmission power and thus to minimize deference areas, which in turn to maximize the spatial channel reuse. The courtesy algorithms assign a node with a specific power level according to the link length, which is the distance between the transmitter and the receiver, and the noise and interference power level. These algorithms can be further classified into linear power assignment algorithms and non-linear power assignment algorithms. The linear power assignment algorithms are so aggressive that they may introduce extra hidden terminals, which cause additional unregulated collisions. However, the non-linear algorithms are too conservative to maximize the power control benefits. The opportunistic power control algorithms allow conditional violations of the CA mechanisms, i.e., a deferring node can initiate a transmission with a deliberately calculated transmission power so that the ongoing transmission will not be affected. However, the power calculation is based on the constants that are only valid in certain wireless scenarios.
Related to this issue, a more difficult problem is how to improve network throughput when the demanded data rate within a certain area exceeds the limit of throughput density, which is defined as the upper limit of the total throughput constrained by the modulation techniques and CA mechanisms in the area. Note that no existing algorithm, neither RCA nor the power control, is able to solve this problem. In this work, we focus our study on the above issues in the spectrum management of wireless networks. Our contributions can be summarized as follows. Firstly, to solve the first issue, we propose an annealing Gibbs sampling (AGS) based distributive RCA (ADRCA) algorithm. The ADRCA algorithm has the following advantages:
1)It uses average effective channel utilization (AECU) to evaluate the channel condition. AECU has a simple relationship with CCI and can accurately reflect the channel congestion conditions.
2)It employs the AGS optimization method, which divides a global optimization problem into a set of distributed local optimization problems. Each of those problems can be solved by simulating a Markov chain. The stationary distribution of the Markov chains is a globally optimized solution.
3)It includes three different cases, namely AGS1, AGS2 and AGS3, which adapt to various types of wireless networks with different optimization objectives.
AGS1 is designed to search for a global optimal channel assignment in OIS networks; AGS2 is proposed to work in NOIS networks and pursue maximum individual performance. Added with a prerequisite for RCA procedures, AGS3 focuses on cost-effectiveness, reduces channel reallocation attempts, and enhances system stability without significantly downgrading its optimization performance. To further study the cost-effectiveness of ADRCA, an upper limit of the computational scale (CS) is found for AGS3 based on an innovative neighboring relationship model in a practical network scenario. Secondly, to solve the second issue, we propose a hybrid approach to study the computational scale (CS), which is defined as the number of channel reallocations until a network reaches a convergent state. First, we propose a simple relationship model to describe the interference relation between an AP and its neighboring APs. Second, for one of the simplest cases in the relationship model, we find an analytical solution for the CS and validate it by simulations. Third, for more general cases, we combine the cases with a similar CS means by using one-way analysis of variance (ANOVA) and find the upper bound of the CS with extensive simulations. The simulation results demonstrate that the hybrid approach is simple and accurate as compared to traditional intuitive comparison methods. Based on the aforementioned hybrid approach, an upper limit of the CS is found for AGS3 in a practical network scenario. Thirdly, to solve the third issue and also raise the limit of throughput density, we propose the channel allocation with power-control (CAP) strategy which integrates the ADRCA algorithm and the digitized adaptive power control (DAPC) algorithm, to achieve a synergetic benefit between power control and RCA, which is not considered by the existing RCA algorithms. The synergy comes from the following two aspects:
• By reducing the transmission power of each node, DAPC can lower CCI levels, allow more simultaneous transmissions within a certain area, increase spatial reuse, and raise the limit of the throughput density. It also reduces the number of nodes competing for a given channel, and thus significantly decreases the CS of ADRCA.
• By striving to assign interfering neighbors to non-overlapping channels, ADRCA minimizes the number of hidden terminals introduced by the power control processes. The integration causes two potential problems. First, since most RCA algorithms are heuristic, after a system converges, any change in transmission power may trigger unnecessary channel reallocation processes, which then would lead to extra computational costs. Second, channel reallocations can also invalidate current transmission power assignments.
The above two problems significantly impair system stability. The CAP strategy overcomes the following two problems as follows: to mitigate the impact of the first problem, a node must estimate the conditions of a new channel and use adaptive transmission power accordingly; for the second problem, the node must calculate the transmission power using linear power control algorithms and round it up to the next larger level in a given set of predetermined power levels. There are several statistical methods applied in our study, including the Markov chain Monte Carlo (MCMC) method, distribution model fitting, paired t-test, and the ANOVA test. They are more accurate and efficient than traditional intuitive comparison methods, making this study an important cornerstone for further research. In this work, we have conducted extensive simulations to demonstrate the effectiveness of the proposed methods. Our simulation results show that AGS1 can achieve a global optimum in most OIS network scenarios. With a 95% confidence level, it achieves 99.75% of the global maximum throughput. AGS2 performs on par with AGS1 in NOIS networks. AGS3 reduces the CS by as much as 98% compared to AGS1 and AGS2. The simulation results also demonstrate that compared with the standard MAC protocol, CAP increases the overall throughput by up to 9.5 times and shortens the end-to-end delay by up to 80% for UDP traffic.
Filename Size Approximate Download Time (Hours:Minutes:Seconds)
28.8 Modem 56K Modem ISDN (64 Kb) ISDN (128 Kb) Higher-speed Access Ma_X_dissertation_2010.pdf 5.96 Mb 00:27:35 00:14:11 00:12:24 00:06:12 00:00:31