Análise Comparativa de Técnicas de Sensoriamento Espectral baseadas em Redes Neurais, Matriz de Covariância e Densidade Espectral de Potência

Abstract

The CFCPSC (circular folding cooperative power spectral density split cancellation) is a periodogram-based algorithm designed to detect primary signals via cognitive ra dios, in centralized cooperative spectral sensing with data fusion. Its main advantages are: low implementation complexity, robustness in scenarios under non-uniform noise, in addition to being effective in scenarios under frequency-selective channels and/or with correlated shadowing. The weighted CFCPSC algorithm, or WCFCPSC, is the new weighted version of CFCPSC with greater statistical detection power. In this work, firstly, the analysis of the performance of the WCFCPSC technique under the afore mentioned scenarios is carried out in comparison with GLRT (generalized likelihood ratio test) and its predecessor CFCPSC. Subsequently, the analyses are expanded to other cooperative and centralized spectral sensing techniques with data fusion, na mely: GID (Gini index detector), PRIDe (Pietra-Ricci index detector) and an artificial neural network model called DenseNet. Simulations were conducted for a variety of scenarios, considering frequency-selective channels, spatially correlated shadowing, uniform and non-uniform noise, and also dynamic noise. The results showed the ro bustness and performance of the WCFCPSC and DenseNet detectors in the face of the proposed situations and variations in the sensed channel. Keyords: Spectrum sensing, cognitive radio, convolutional neural networks, WCFCPSC, dynamic noise.