PhD Thesis

These few pages are about my PhD thesis, which is about telecomunications, signal processing and artificial life. The name is "Reducing the non-linear distorsions of a multicarrier modulation using neural networks".

The multicarrier modulations are particularly adapted to frequency selective channels, and especially the multipath channels. Indeed in an urban environment the radio waves are reflected by the walls, causing unwanted echos during a ragio transmission. Multicarrier modulations, (and especially OFDM: Orthogonal Frequency Division Multiplexing) are a solution to this problem and manage to transmit correctly information on these difficult channels. OFDM is currently used in DAB/DVB (Digital Audio Broadcasting, Digital Video Broadcasting, the future replacements for FM radio and television), HiperLAN/2 et IEEE 802.11a (radio computer networks) and ADSL.

But an OFDM signal is sensitive to non-linear distortions, and the most important one is the high power amplifier. This non-linearity introduces errors in the transmitted signal. In this thesis we want to correct the received signal using a neural network.

The neural network I most used is the RPN (Ridge Polynomial Network). This network makes products in an intermediate layers, and this permits good approximations of higher order functions. I'm also interested in SVM (Support Vector Machine).

You can go and see my publications, and download the memoire (in French).