ICA Algoritm with Likelihood Criterion to Separate Mixtures of Complex Sinusoidal Signals

In this paper we consider the blind source separation (BSS) of complex sinusoidal signals with different frequencies. We introduce a novel ICA (Independent Component Analysis) algorithm for the BSS. ICA requires no prior information of the source signals because it employs only the statistical independence of them. We have already confirmed that ICA was successfully applied to a deterministic signal like a complex sinusoidal signal. However, the frequency resolution of the former algorithm is not sufficient for the actual radar applications, such as pulse compression or clutter refection. To enhance frequency resolution, we propose a novel ICA algorithm in specifying the complex sinusoidal signal separation using the likelihood criterion. The conventional maximum likelihood based ICA algorithm typically selects the PDF (probability density function) from some promising candidates. Thus, we introduce the PDF of a complex sinusoidal signal. The results in numerical simulations verify that the proposed method successfully separates the multiple sinusoidal signals with frequencies close to less than DFT (Discrete Fourier Transform) resolution.