Resumo:
Nowadays the blind signal processing is one of the areas of greater highlight in the signal processing. The signal processing techniques do not make use of any training sequence nor any information on the mixture of the system to which the signals are subjected to; being the blind separation one of the main areas of the blind processing.
The blind separation or the blind signal (source) separation problem consists of retrieval a set of unknown signals or sources by the observations done by sensors of mixture of this signals. It’s no shadow of doubt, a problem of great interest in the signal processing area, once to be solved it is necessary that a set of hypotheses a bit restrictive be carried out. Being that, the blind source separation techniques run across countless applications: data set processing; multi-users communications; voice and image recognition; biomedical signal processing. By means of separation techniques, one can, therefore, retrieve one or all the sources just basing on the information on observations or measurement done by the set of sensors. The blind adjective was incorporated to characterize the lack of information inherent in the separation process. To fulfill this lack of information some properties of the sources nature, mixture, and noise are taken into account for separation process. This way, the algorithms of blind separation try and restore at the exit of separation system one property known by the sources.
One of the main tools used to solve the blind separation problem has been the Independent Component Analysis. It is important to mention that the blind separation and the Independent Component Analysis terms are often mixed up or used like synonym, although they refer to a similar or equal pattern and are solved with similar or equal algorithms, under the restriction on the original sources be statistically independent. However, mainly in real problems, the independent component analysis and blind separation goals are a bit different: the blind separation goal is to estimate the original signals even if they are not completely independent; whereas, the independent component analysis goal is to determine one transformation that ensures the estimate signals are as independent as possible. One may yet observe that the independent component analysis methods use, in most cases, statistics of superior order, while the blind separation methods are apt to use just statistics of second order.
Based on the above-mentioned considerations, this dissertation presents one state-of-art review of the main techniques that deal with the separation problem; by means of comparison of three algorithms: AMUSE, JADE and FLEXICA that were compared through the application of them into test signals, telecommunication signals and a real world biomedical signal.