Sensoriamento de espectro e classificação de sinais em rádio cognitivo por decomposição em subespaços e redes neurais RBF

Abstract

The possibility of spectrum shortage and saturation, combined with the increasing demands for higher transmission rates are driving factors for research within cognitive radio networks. Spectrum sensing is one of the major challenges for the commercial development of cognitive radio systems, since the verification of a primary user presence is a complex task that requires high reliability. The proposal of this work is to develop a signal classifier capable of verifying the primary user presence on a particular channel of the radio spectrum. The proposed classifier performs subspace decomposition of the signal covariance matrix, in order to obtain characteristics that may indicate the presence of a primary user. The subspace decomposition enables the design of filter banks to which new signals are submitted. RBF neural networks are used to analyze the filtered signal characteristics and to decide about the presence of a particular type of primary user. Based on IEEE 802.22 regulations, the classification process is performed at the cognitive radio base station, which is responsible for controlling all users and channels in its coverage area. The results indicate that the computational cost of subspace decomposition, which is cyclically performed in similar methods, can be reduced through the proposed approach without jeopardizing the detection quality.