ICA techniques for more sources than sensors

In this paper we derive algorithms to identify the mixing matrix in the context of an independent component analysis with more sources than sensors. First, by exploiting the fact that for complex-valued observations, depending on the type of complex symmetry, 2 different fourth-order cumulants are available, we develop a technique that can cope with N(N+1)/2 sources for only N sensors. Secondly, the technique presented in Cardoso et al. (1994), based on a single cumulant, is modified to take both cumulants into account as well.