Task description and Expected results
The cognitive radio technique is recognized as the most efficient method to improve the spectrum utilization, by using the
available spectrum effectively amongst the secondary users in an opportunistic manner. Different issues related to the dynamic
spectrum management like spectrum sensing, spectrum decision, spectrum sharing and spectrum scheduling need to be
addressed.
Spectrum Sensing is used to determine the state of the spectrum. A cognitive radio detects an unused spectrum or spectrum
hole. The identification of free frequencies might be viewed as a pattern recognition problem therefore machine learning
techniques are suitable approaches. Detection and classification of very low signal-to-noise ratio signals is one of the most
challenging tasks in C R systems. Various approaches rely on the extraction of cyclostationary features of the signals to perform
detection of the holes on spectrum. The detection task is a simple classification problem: feature extraction is applied to a time
series of the incoming radio channel and each channel is classified as “occupied” if a signal is present and “free” otherwise. The
second step includes classifiers to identify the signals types.
In this task we will study the application of linear dimensionality reduction techniques such as principal component analysis and
nonlinear ones such as the kernel principal component analysis combined with support vector machine. With these techniques we
expect to increase the success rate detection of the presence of primary users when compared with more traditional methods
such as energy detection.
We expect to combine these techniques with the cochlear radio of the task 6 in order to build a system able to perform spectrum
analysis over a large bandwidth.
Mem bers of the research team in this task: (BI) Bolseiro de Investigação (Mestre) 4; Ana Maria Perfeito Tome; José Manuel Neto Vieira; Nuno Miguel Gonçalves Borges de
C arvalho; Pedro Miguel Duarte C ruz; Teófilo José Marques Monteiro;
