VGTU talpykla >
Doktorantūros skyrius / Department for Doctoral Studies >
Technologijos mokslų daktaro disertacijos ir jų santraukos >
Please use this identifier to cite or link to this item:
|Title: ||Development of self-organizing methods for radio spectrum sensing|
|Authors: ||Stašionis, Liudas|
|Issue Date: ||2016|
|Publisher: ||VGTU leidykla "Technika"|
|Citation: ||Stašionis, L. 2016: Development of self-organizing methods for radio spectrum sensing: Doctoral Dissertation. Vilnius: Technika, 180 p.|
|Abstract: ||A problem of wide-band radio spectrum analysis in real time was solved and presented in the dissertation. The goal of the work was to develop a spectrum sensing method for primary user emission detection in radio spectrum by investigating new signal feature extraction and intelligent decision making techniques. A solution of this problem is important for application in cognitive radio systems, where radio spectrum is analyzed in real time.
In thesis there are reviewed currently suggested spectrum analysis methods, which are used for cognitive radio. The main purpose of these methods is to optimize spectrum description feature estimation in real-time systems and to select suitable classification threshold. For signal spectrum description analyzed methods used signal energy estimation, analyzed energy statistical difference in time and frequency. In addition, the review has shown that the wavelet transform can be used for signal pre-processing in spectrum sensors. For classification threshold selection in literature most common methods are based on statistical noise estimate and energy statistical change analysis. However, there are no suggested efficient methods, which let classification threshold to change adaptively, when RF environment changes.
It were suggested signal features estimation modifications, which let to increase the efficiency of algorithm implementation in embedded system, by decreasing the amount of required calculations and preserving the accuracy of spectrum analysis algorithms.
For primary signal processing it is suggested to use wavelet transform based features extraction, which are used for spectrum sensors and lets to increase accuracy of noisy signal detection. All primary user signal emissions were detected with lower than 1% false alarm ratio. In dissertation, there are suggested artificial neural network based methods, which let adaptively select classification threshold for the spectrum sensors. During experimental tests, there was achieved full signals emissions detection with false alarm ratio lower than 1%.
It was suggested self organizing map structure modification, which increases network self-training speed up to 32 times. This self-training speed is achieved due to additional inner weights, which are added in to self organizing map structure. In self-training stage network structure changes especially fast and when topology, which is suited for given task, is reached, in further self-training iterations it can be disordered. In order to avoid this over-training, self-training process monitoring algorithms must be used. There were suggested original methods for self-training process control, which let to avoid network over-training and decrease self-training iteration quantity.|
|Appears in Collections:||Technologijos mokslų daktaro disertacijos ir jų santraukos|
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.