Blind separation of source signals usually relies either on the nongaussian
ity of the signals or on their linear autocorrelations. A third approach wa
s introduced by Matsuoka et al., who showed that source separation can be p
erformed by using the nonstationarity of the signals, in particular the non
stationarity of their variances. In this paper, we show how to interpret th
e nonstationarity due to a smoothly changing variance in terms of higher or
der cross-cumulants. This is based on considering the time-correlation of t
he squares (energies) of the signals and leads to a simple optimization cri
terion. Using this criterion, we construct a fixed-point algorithm that is
computationally very efficient.