Title :
Predistortion of non-linear satellite channels using neural networks: Architecture, algorithm and implementation
Author :
Langlet, F. ; Abdulkader, H. ; Roviras, D.
Author_Institution :
ENSEEIHT-TeSA, Toulouse, France
Abstract :
This paper presents the adaptive linearisation of a non-linear digital satellite communication down link. That link is made up a 16-QAM modulator, followed by a non-linear High Power Amplifier, on board the satellite. When using the amplifier with low input back-off for a maximum power efficiency, two kinds of distortions occur on the input signal: amplitude (AM/AM conversion) and phase (AM/PM conversion). The satellite payload is regenerative. So, we use a predistortion on board to linearize the amplifier. We present the predistortion architecture realized with Multi-Layer Perceptron (MLP) Neural Networks (NN). Two algorithms associated to that predistorter are shown and compared: the ordinary and the natural gradient. The major problem to implement that predistorter is to get enough bandwidth (100 Mbits/s data rate). A mixed analog/digital implementation is one solution to solve it. We analyze the implementation imperfections effects in comparison with the theoretical algorithm.
Keywords :
gradient methods; neural nets; power amplifiers; quadrature amplitude modulation; satellite communication; satellite links; telecommunication computing; wireless channels; AM/AM conversion amplitude; MLP; NN; QAM modulator; adaptive linearisation; input signal; mixed analog/digital implementation; multilayer perceptron; natural gradient; neural networks; nonlinear digital satellite communication down link; nonlinear high power amplifier; nonlinear satellite channel predistortion; phase AM/PM conversion; power efficiency; satellite payload; Abstracts; Algorithm design and analysis; Artificial neural networks; Data models; Manifolds; Payloads; Predistortion;
Conference_Titel :
Signal Processing Conference, 2002 11th European
Conference_Location :
Toulouse