Tesis doctoral de Jorge Moragues Escrivá
This thesis is dedicated to the development of new energy detectors employed in the detection of unknown signals in the presence of non-gaussian and non-independent noise samples. To this end, an extensive study has been conducted on different energy detection structures, and novel techniques have been proposed which are capable of dealing with these problematic situations. the energy detector is proposed as an optimum solution to detect uncorrelated gaussian signals, or as a generalized likelihood ratio test to detect entirely unknown signals. In both cases, the background noise must be uncorrelated gaussian. However, energy detectors degrade when the noise does not fulfill these characteristics. Therefore, two extensions are proposed. The first is the extended energy detector, which deals with the problem of non-gaussian noise; and the second is the preprocessed extended energy detector, used when the noise also possesses non-independent samples. A generalization of the matched subspace filter is likewise proposed based on a modification of the rao test. In order to evaluate the expected improvement of these extensions with respect to the classical energy detector, a signal-to-noise ratio enhancement factor is defined and employed to illustrate the improvement achieved in detection. furthermore, we demonstrate how the uncertainty introduced by the unknown signal duration can decrease the performance of the energy detector. In order to improve this behavior, a multiple energy detector, based on successive subdivisions of the original observation interval, is presented. This novel detection technique leads to a layered structure of energy detectors whose observation vectors are matched to different intervals of signal duration. The corresponding probabilities of false alarm and detection are derived for a particular subdivision strategy, and the required procedures for their general application to other possible cases are indicated. The experiments reveal the advantages derived from utilizing this novel structure, making it a worthwhile alternative to the single detector when a significant mismatch is present between the original observation length and the actual duration of the signal. the important simulation results yielded by the new energy detectors offer promising opportunities for real-world applications, such as surveillance systems based on sound analysis. These systems present a suitable scope for verifying the robustness of the theoretical detectors presented in this thesis. Thus, several acoustic sources and a variety of real and simulated noise scenarios were tested and two novel approaches were presented. The first combines the information provided by an adaptive energy detector with the standard localization method. The localization rates are considerably improved with this original technique, mainly when the sound source is in presence of a background noise. Finally, a unique set of features are extracted from the multiple energy detector structure, evaluated, and compared with other common features used for the recognition of acoustic sounds. The results obtained with the new features considerably improve the classification accuracy, especially in low signal-to-noise ratios.
Datos académicos de la tesis doctoral «New energy detector extensions with application in sound based surveillance systems«
- Título de la tesis: New energy detector extensions with application in sound based surveillance systems
- Autor: Jorge Moragues Escrivá
- Universidad: Politécnica de Valencia
- Fecha de lectura de la tesis: 27/07/2011
Dirección y tribunal
- Director de la tesis
- Luís Vergara Domínguez
- Tribunal
- Presidente del tribunal: ramón Miralles ricós
- rudolf leonhard Rabenstein (vocal)
- Jesús Grajal de la fuente (vocal)
- Manuel Rosa zurera (vocal)