Concentrated signal extraction using consecutive mean excision algorithms

Abstract

<div lang="en" class="abs"><p class="abs_header">Abstract</p><p>Spread spectrum communication systems may be affected by other types of signals called outliers. These coexisting signals are typically narrow (or <em>concentrated</em>) in the considered domain. This thesis considers two areas of outlier detection, namely the concentrated interference suppression (IS) and concentrated signal detection. The focus is on concentrated signal extraction using blind, iterative and low-complex consecutive mean excision (CME) -based algorithms that can be applied to both IS and detection. </p><p>A summary of results obtained from studying the performance of the existing IS methods, namely the CME, the forward CME (FCME) and the transform selective IS algorithms (TSISA), is presented. Accurate threshold parameter values for the FCME algorithm are defined. These accurate values are able to control the false alarm rate. The signal detection capability of the CME algorithms is studied and analyzed. It is noticed that the CME algorithms are able to detect signals, but they are not able to estimate signal parameters such as the bandwidth. The presented generic shape-based analysis leads to the limits of detection in which the concentrated signals can be detected. These limits enable checking fast whether the signal is detectable or not without time consuming computer simulations. The performance of the TSISA method is evaluated. Simulation results demonstrate that the TSISA method is able to suppress several types of concentrated interfering signals with a reasonable computational complexity. </p><p>Finally, new CME-based methods are proposed and evaluated. The proposed methods are the extended TSISA method for IS and the localization algorithm based on double-thresholding (LAD), LAD with normalized thresholds (LAD NT), LAD with adjacent cluster combining (LAD ACC) and two-dimensional (2-D) LAD methods for detection. The simulations indicate that the extended TSISA method has a good performance against several types of concentrated interfering signals. The narrowband signal detection capability of the LAD methods is studied. Numerical results show that the proposed LAD methods are able to detect and localize signals in their domain, and they are able to estimate the number of narrowband signals and their parameters, including, for example, bandwidths and signal-to-noise ratio (SNR) values. The simulations show that the LAD methods outperform the CME algorithms, and ACC and 2-D LAD methods outperform the original LAD method. The LAD methods are also proposed to be used for spectrum sensing purposes in cognitive radios. </p></div>