Prewhitening of mixing matrices: impact on the contraction of their probability mass priors

The authors consider a linear mixture of n sources, x(t) = As(t) + w(t), where x(t) denotes the observations, A ∈ R^n×n is the mixing matrix, s(t) contains the n source signals, and w(t) models observation noise. The sources are assumed to be wide-sense stationary, zero-mean, R_s = E{s(t)s(t)^T} = I_n, and A is invertible.