{"title":"A Frequency Grouping Approach for Blind Deconvolution of Fairly Motionless Sources","authors":"E. S. Gower, T. Tsalaile, E. Rakgati, M. O. J. Hawksford","country":null,"institution":"","volume":56,"journal":"International Journal of Electrical and Computer Engineering","pagesStart":1015,"pagesEnd":1021,"ISSN":"1307-6892","URL":"https:\/\/publications.waset.org\/pdf\/4363","abstract":"A frequency grouping approach for multi-channel\r\ninstantaneous blind source separation (I-BSS) of convolutive\r\nmixtures is proposed for a lower net residual inter-symbol\r\ninterference (ISI) and inter-channel interference (ICI) than the\r\nconventional short-time Fourier transform (STFT) approach. Starting\r\nin the time domain, STFTs are taken with overlapping windows to\r\nconvert the convolutive mixing problem into frequency domain\r\ninstantaneous mixing. Mixture samples at the same frequency but\r\nfrom different STFT windows are grouped together forming unique\r\nfrequency groups.\r\nThe individual frequency group vectors are input to the I-BSS\r\nalgorithm of choice, from which the output samples are dispersed\r\nback to their respective STFT windows. After applying the inverse\r\nSTFT, the resulting time domain signals are used to construct the\r\ncomplete source estimates via the weighted overlap-add method\r\n(WOLA). The proposed algorithm is tested for source deconvolution\r\ngiven two mixtures, and simulated along with the STFT approach to\r\nillustrate its superiority for fairly motionless sources.","references":null,"publisher":"World Academy of Science, Engineering and Technology","index":"Open Science Index 56, 2011"}