Neural networks in space communications

Neural networks (NN) are now widely used and known to solve recognition and classification problems. The paper aims at showing how their learning capability can be applied to adaptive signal processing problems, either in the replacement of conventional adaptive methods or to cope with the most difficult cases raised by mobile space communication: non linear and/or non stationary problems. A brief overview of NNs is given first, highlighting their key features: approximation capability, learning characteristic, etc. We then focus on 3 major applications to space communications: system and channel identification, equalization, and predistortion. The conclusion reached is that NN has no counterpart able to yield general solutions to these classes of problems with a single tool.