• Citer

Spectrum Sensing for Cognitive Radios: Improving Robustness to Impulsive Noise

par Renard, Julien
Président du jury Welch, Will
Promoteur Horlin, François ;Lampe, Lutz
Publication Non publié, 2016-06-07

Thèse de doctorat

• Accès en ligne
• Détails
• Contenu
• Statistiques

Résumé :

Many different types of promising spectrum sensing algorithms for Cognitive Radio (CR) have already been developed. However, many of these algorithms lack robustness with respect to signal statistical parameters uncertainties, such as the noise variance or the shape of its distribution (often assumed to be simply Gaussian). In conjunction with the low Signal-to-Noise Ratio (SNR) requirements, this lack of robustness can often render interesting sensing algorithms impractical for real-life applications. In this thesis, we primarily focus on the impact of heavy-tail noise distributions on different CR detectors and the use of signal limiters (mostly the spatial sign function) to improve their robustness to such noise distributions. Introducing a non-linear transformation of the received signal prior to its processing by the detector fundamentally changes the signal distribution which in turn modifies the distribution of the detector statistic. In order to parametrize the detector and study its performance, it is then necessary to know the shape of the modified distribution.Three types of detectors are investigated: a generic second-order cyclic-feature detectors, a Scaled-Largest Eigenvalue (SLE) detector studied in the context of stationary time-series and a new Sequential Likelihood Ratio Test (SLRT) detector. The analysis conducted for each detector revolves around the influence of its parameters, the distribution of the detector statistic and several comparisons with similar detectors for various detection scenarios. Our results indicate that at the cost of a moderate performance loss in a Gaussian noise environment, all the detectors fitted with a signal limiter become robust to impulsive noise and noise parameters uncertainties. We provide analytical approximations for the detectors statistical distribution that allow us to use the detectors in such configurations as well as to study their performance for different signal limiters and noise distributions.