Skip to Main content Skip to Navigation

Linéarisation à base de réseaux de neurones pour amplificateurs de puissance

Blaise Mulliez 1
1 LAAS-OSE - Équipe Optoélectronique pour les Systèmes Embarqués
LAAS - Laboratoire d'analyse et d'architecture des systèmes
Abstract : The spectacular growth of space telecommunications during the last two decades requires an always higher data transmission speed and a flawless service quality. Nevertheless, in order to optimize the link budget and the spectral efficiency, the embedded High Power Amplifiers (HPA) are used close to their saturation point, which leads to strongly non-linear emitted signals. To circumvent this issue, a linearizer is often implemented before the amplifier. However, the linearization devices used today are not able adapt to different amplifiers or to HPA characteristics drift under the influence of aging and temperature variations : they are not adaptive. The objective of the work presented in this dissertation is the design of an innovating architecture capable of linearizing several HPA transfer characteristics. Analog Neural Networks (ANN) provide attractive performances for non-linear functions modelling and are reconfigurable. They are therefore a relevant choice to respond to this specific issue. First, a patented, innovating, generic, fast and accurate technique to determine the predistortion functions is detailed and used with the characteristics of three HPA provided by the French Space Agency (CNES). Then, the modelling of these predistortion functions with neural networks and behavioral static and dynamic simulations of these networks validate the concept of adaptive analog predistortion based on neural networks. Eventually, an analog predistortion ASIC, designed in a CMOS 0.35μm technology, including a neural network and an innovative configurable phase-shifting circuit, is described. The integrated circuit is able to generate the different predistortion functions and will be later embedded in a test-bench to demonstrate its ability to adaptively linearize several High Power Amplifiers.
Complete list of metadata
Contributor : Arlette Evrard <>
Submitted on : Tuesday, December 15, 2015 - 11:15:33 AM
Last modification on : Thursday, June 10, 2021 - 3:05:41 AM
Long-term archiving on: : Wednesday, March 16, 2016 - 10:12:27 AM


  • HAL Id : tel-01241354, version 1


Blaise Mulliez. Linéarisation à base de réseaux de neurones pour amplificateurs de puissance. Micro et nanotechnologies/Microélectronique. INP Toulouse, 2015. Français. ⟨tel-01241354⟩



Record views


Files downloads