The use of non-linear amplifiers near saturation in satellite channels caus
es severe distortion of the transmitted signal. These distortions make the
matter of power efficiency prevail over spectral efficiency in satellite co
mmunication systems. Nowadays, the use of satellite segments for applicatio
ns requiring ever increasing data rates puts this principle in question. Th
e present article proposes to fight non-linear distortion caused by satelli
te amplifiers by resorting to simple neural network equalisation devices in
the receiving earth terminals. Several neural network equalisers are intro
duced, and applied to a satellite Universal Mobile Telecommunication System
(S-UMTS) channel model. They are shown not only to outperform conventional
equalisation techniques like the linear transversal equaliser, but also to
have better performance than the non-linear Volterra equaliser. The perfor
mance improvement is all the more sensitive as the modulation scheme used i
s severely distorted. The case of 16-QAM transmission over a mobile satelli
te link is studied in details.