The learning rule of the unmixing matrix W is [36] (1)∆W∝[I−Ktanh(u)uT−uuT]W,W(m+1)=W(m)+μ∆W(m),ki=1: super-Gaussian,ki=−1: sub-Gaussian, where k i are elements of the N-dimensional diagonal matrix K, m is the iteration number, and μ is the step size. The switching parameter k i can be derived from the variation of the kurtosis sign. k i can be obtained as (2)ki=sign(E(ui4)−3(E(ui2))2(E(ui2))2)=sign(E(ui4)(E(ui2))2−3).
