Blind source separation and bearing estimation using Fourier- and wavelet-based spectrally condensed data and artificial neural networks for indoor environments
Ebrahim A. Gharavol, Ooi Ban Leong, Koen Mouthaan · 2008
A new method for blind separation and bearing estimation of wavefronts in a smart antenna scheme, which is based on the usage of artificial neural networks (ANN) is presented here. Because of “the curse of dimensionality,” especially in the cases having many antenna elements, in uniform linear, circular or planar arrays, it is important to find a method which makes it feasible to use the ANNs. The proposed method, do not walk along the road of well-known method of correlation-coefficient training. In contrast this method uses the truncated version of their spectral representations. The Fast Fourier Transform (FFT) and Discrete Wavelet Transform (DWT) are employed to provide the spectral representations. The simulation scenario is set up to demonstrate that the results is applicable to realistic cases such as urban, non-line of sight, and indoor environments. For the sake of this purpose, coherent signals are employed in simulations. In this case, most conventional methods are not applicable, because they are built on some statistical assumptions which implies that the received signals by array must be independent.