Banca de QUALIFICAÇÃO: ARTHUR DIEGO DE LIRA LIMA

Uma banca de QUALIFICAÇÃO de DOUTORADO foi cadastrada pelo programa.
DISCENTE : ARTHUR DIEGO DE LIRA LIMA
DATA : 30/06/2017
HORA: 14:30
LOCAL: Sala de Videoconferência do PoP-RN/UFRN
TÍTULO:

Nonlinear Data Transformation Applied to Cyclostationary Analysis of Signals in Environments with Non-Gaussian Noise

PALAVRAS-CHAVES:

Cicloestacionarity, alpha-stable noise, data transformation, spectrum sensing.

PÁGINAS: 60
GRANDE ÁREA: Engenharias
ÁREA: Engenharia Elétrica
RESUMO:

Spectral scarcity is one of the main factors limiting the offer of
high data rates to wireless users. Techniques such as opportunistic
access to the spectrum and cognitive radio are among the potential
solutions to increase efficiency in the use of the radio spectrum.
However, these alternatives depend on robust algorithms for spectral
sensing. One of the most used methods for spectral sensing is based on
the detection of cyclostationary features, because it is capable of
high detection rates in high noisy environments, but with the
disadvantage of significant computational cost. However, the
second-order cycloestationary analysis, although resistant to additive
white Gaussian noise (AWGN), is very sensitive to signal contamination
by impulsive noise. Since impulsive noise presents unbounded variance,
typically its analysis is based on the substitution of second-order
statistics for fractional lower-order statistics. For example, instead
of sensing based on the detection of second-order cycloestationary
features, it is common to use the fractional lower-order
cyclestationarity in impulsive noise environments. In this work, we
propose the adoption of a non-linear data transformation to modulated
signals corrupted by symmetric non-Gaussian alpha-stable noise. The
transformed signal maintains its original cyclostationary and then
presents a Gaussian distribution, and can thus be examined with
second-order cyclostationary tools. In addition, the algorithm for the
extraction of cyclostationary features was parallelized to a 60-core
Xeon Phi architecture. Sensing results from alpha-stable noise
contaminated signals show a probability of detection similar to the
Gaussian case, reaching a rate of 100% for BPSK signals from the
geometric signal-to-noise ratio of -13 dB, and for QPSK signals from
-6 dB. Parallelization results indicate a parallel efficiency greater
than 90% when using up to 25 processing cores; And 70% when the all
the 59 available cores are used.

MEMBROS DA BANCA:
Presidente - 1543191 - LUIZ FELIPE DE QUEIROZ SILVEIRA
Interno - 1673543 - SAMUEL XAVIER DE SOUZA
Externo à Instituição - EDMAR CANDEIA GURJÃO - UFCG