
For FullText PDF, please login, if you are a member of IEICE,
or go to Pay Per View on menu list, if you are a nonmember of IEICE.

MLICABased Separation Algorithm for Complex Sinusoidal Signals with PDF Parameter Optimization
Tetsuhiro OKANO Shouhei KIDERA Tetsuo KIRIMOTO
Publication
IEICE TRANSACTIONS on Communications
Vol.E95B
No.11
pp.35563562 Publication Date: 2012/11/01 Online ISSN: 17451345
DOI: 10.1587/transcom.E95.B.3556 Print ISSN: 09168516 Type of Manuscript: PAPER Category: Sensing Keyword: radar signal processing, maximum likelihood independent component analysis (MLICA), complex sinusoidal signals, PDF parameter optimization,
Full Text: PDF>>
Summary:
Blind source separation (BSS) techniques are required for various signal decomposing issues. Independent component analysis (ICA), assuming only a statistical independence among stochastic source signals, is one of the most useful BSS tools because it does not need a priori information on each source. However, there are many requirements for decomposing multiple deterministic signals such as complex sinusoidal signals with different frequencies. These requirements may include pulse compression or clutter rejection. It has been theoretically shown that an ICA algorithm based on maximizing nonGaussianity successfully decomposes such deterministic signals. However, this ICA algorithm does not maintain a sufficient separation performance when the frequency difference of the sinusoidal waves becomes less than a nominal frequency resolution. To solve this problem, this paper proposes a superresolution algorithm for complex sinusoidal signals by extending the maximum likelihood ICA, where the probability density function (PDF) of a complex sinusoidal signal is exploited as a priori knowledge, in which the PDF of the signal amplitude is approximated as a Gaussian distribution with an extremely small standard deviation. Furthermore, we introduce an optimization process for this standard deviation to avoid divergence in updating the reconstruction matrix. Numerical simulations verify that our proposed algorithm remarkably enhances the separation performance compared to the conventional one, and accomplishes a superresolution separation even in noisy situations.

