Research On Array Signal Processing Techniques Based On Independent Component Analysis

Array signal processing is an important branch of signal processing. It is widelyapplied to radar, communication, direction-finding, sonar, medicinal diagnosis andtherapy, astronomy and so on. In recent years, with the fast development of IndependentComponent Analysis （ICA） techniques, ICA-based array signal processing have beena research hotspot. This dissertation focuses on the array signal processing techniquesbased on ICA. The main contributions of the dissertation are listed as follows:（1） An ICA-based source number estimation method with uniform linear arrays（ULA） has been proposed. First, the proposed method defines the Vandermondesimilarity measurement based on the special structure of the steering matrix with ULA.Secondly, it estimates the mixing matrix by ICA and then constructs the Vandermondesimilarity measurement function versus the number of source signals. Finally, thenumber of source signals is estimated by checking the peak of the Vandermondesimilarity measurement function. Experiments have proved our method has a strongeradaptability in the cases that the noise is non-uniform or the number of source signals isequal to the sensors.（2） A joint estimation approach for the waveforms and DOAs of source signalsbased on ICA is presented. The waveforms of source signals and the mixing matrix arefirst estimated by ICA. Then, the DOAs of source signals are estimated by subspacemethod; At last, the DOAs are matched with the waveforms of source signals.Simulations have proved that our approach can estimate the waveforms as well as theDOAs of source signals.（3） An ICA-based method for the number estimation of coherent signals isproposed while many groups of coherent signals coexist. In addition, An ICA-basedDOA estimation method has been proposed in the present of coherent signals. Themixing matrix is firstly estimated by ICA. Then, the mixing matrix is reconstructed byeliminating the contributions of uncorrelated signals. Afterwards, several parameterequations are established upon the new mixing matrix. Finally, all DOAs of coherentsignals are estimated by solving these equations. We have demonstrated boththeoretically and empirically that our method is able to separate more signals than thenumber of sensors.