The aim of this paper is to present a study of polynomial functional-link n
eural units that learn through an information-theoretic-based criterion. Fi
rst the structure of the neuron is presented and the unsupervised learning
theory is explained and discussed, with particular attention being paid to
its probability density function and cumulative distribution function appro
ximation capability. Then a neural network formed by such neurons (the poly
nomial functional-link artificial neural network, or PFANN) is shown to be
able to separate out linearly mixed eterokurtic source signals, i.e. signal
s endowed with either positive or negative kurtoses. In order to compare th
e performance of the proposed blind separation technique with those exhibit
ed by existing methods, the mixture of densities (MOD) approach of Xu et al
, which is closely related to PFANN, is briefly recalled; then comparative
numerical simulations performed on both synthetic and real-world signals an
d a complexity evaluation are illustrated. These results show that the PFAN
N approach gives similar performance with a noticeable reduction in computa
tional effort.