Title page for ETD etd-04072005-115036


Type of Document Master's Thesis
Author Piron, Phat
Author's Email Address ppiron1@lsu.edu
URN etd-04072005-115036
Title Transform Techniques and Non-Stationarity with an Emphasis on Network Applications
Degree Master of Science in Electrical Engineering (M.S.E.E.)
Department Electrical & Computer Engineering
Advisory Committee
Advisor Name Title
Suresh Rai Committee Chair
Morteza Naraghi-Pour Committee Member
Subhash C Kak Committee Member
Keywords
  • wavelets
  • Hurst parameter
  • self-similarity
  • long-range dependence
  • non-stationarity
  • traffic
Date of Defense 2005-04-05
Availability unrestricted
Abstract
The recent years brought a phenomenal development of Internet. It is, therefore, important to find some ways to improve the performances. The first step in this direction is the characterization and modeling of the network traffic. It has been tested that the network traffic behaves like a self-similar process, while packets interarrivals time possess the long-range dependence property. In particular, we model them by using fractional Brownian motion and fractional Gaussian noise, respectively. Note that, the former is just the cumulative sum of the latter. By using these concepts, the traffic characterization reduces to the estimation of one value: the Hurst parameter. Numerous methods exist to evaluate this parameter. Nevertheless, a few studies take account of the inherent non-stationarity present in real data. For short samples, the stationarity hypothesis might hold. But for larger samples, this is hardly the case. As an example, for network traffic, the day cycle shows non-stationarity. By not considering the non-stationarity, an inaccurate or even inappropriate estimation may result. Our objective in this thesis is to test the robustness of several techniques such as aggregated variance method, rescaled range method, and wavelets method, in presence of a set of non-stationarity trends. We study the estimators on a known signal generated using Hosking, Davies and Harte method, or wavelets-based synthesis. We add various deterministic non-stationarity trends to the original signal. We considered polynomial, power-law, sinusoidal, and level-shift trends. Results help analyze the behavior of the estimators. All the simulations are carried out using Matlab. We show that, depending on the trend, the estimators react differently. We have also used real data to verify the effectiveness of estimators. Results confirm the observations that we have made with lab data. In particular, we show that the wavelets method provides several flaws. Especially, its results must be carefully analyzed when the data is non-stationary.
Files
  Filename       Size       Approximate Download Time (Hours:Minutes:Seconds) 
 
 28.8 Modem   56K Modem   ISDN (64 Kb)   ISDN (128 Kb)   Higher-speed Access 
  Piron_thesis.pdf 1.81 Mb 00:08:22 00:04:18 00:03:46 00:01:53 00:00:09

Browse All Available ETDs by ( Author | Department )

If you have more questions or technical problems, please Contact LSU-ETD Support.