System Architectur antenna e

Approaches based on regression and classification have been developed by members of the ELEDIA Research Center exploiting the generalization capabilities of learning by example methodologies like neural networks (NN) and support vector machine (SVM) for the online DOA estimation of single and multiple signals. Multi-resolution strategies have been also suitably defined in order to further increase the performance of the proposed approaches, thus leading to

1. enhanced angular resolution of the detection process in the region of incidence of the impinging signals;
2. real-time evaluation of single and multiple DOAs through the definition of probabilities that a signal impinges on the antenna array. Both linear and planar geometries have been investigated;
3. high robustness in case of noisy data thanks to the generalization capabilities of the adopted methods.

Research activities concerning the DOA estimation have been focused in the mathematical formulation of the training and testing procedures of LBE algorithms. The antenna geometries have been modeled and the corresponding received signals have been defined in order to efficiently evaluate the signal covariance matrix. Moreover, feature selection strategies have been introduced in order to extract as much information as possible from the available data.