PhD student position: Toward smart transceivers: Machine/Deep learning for the Physical Layer With the emergence of 5G and its high-level performance targets, Waveform (WF) design has received considerable attention from the research community in recent years. To find an alternative to the classical OFDM, several multicarrier techniques addressing different 5G technical challenges, have been proposed like FBMC, UFMC and GFDM… Despite the adoption of OFDM for 5G Physical Layer, some of these post-OFDM WFs remain competitive and keep a chance in the race to beyond 5G. FBMC is one of the most powerful candidates offering numerous advantages compared to OFDM (better spectral properties, flexible access …). However due to the intrinsic interference of FBMC transceivers, the Multiple-Input and Multiple-Output (MIMO) compatibility is still an open issue. Indeed, some important techniques like maximum likelihood MIMO detection and Alamouti space-time coding are harmfully impacted by the FBMC inherent interference. In this thesis, we propose an innovative approach to overcome such challenges. This approach is based on Machine learning which is has become the main tool to tackle all sort of difficult problems in pretty much every Science Field. In particular, with ever-faster computers, deep neural networks have shown very high success in solving difficult problems. In fact, there is recently a growing consensus that Machine Learning methods can lead to a new vision of the communications system design. Indeed, such an approach can provide significant improvements in complex communications scenarios that are difficult to describe with tractable mathematical models. The purpose of this thesis is explore Machine Learning state-of-the-art in order to propose and design innovative solutions for the issues discussed above. We propose to use Deep Neural networks, that are a specific type of Machine Learning methods inspired from the human brain, and in which multiple layers specific operators (linear operations, convolutions, softmax, etc.) are used to build complex learning models. In this family of algorithms, autoencoders are unsupervised deep networks could be used for tasks such as encoding, decoding and denoising information, as well as for clustering tasks, making them good candidates for our problem. The thesis will be carried out at ISEP-École d'ingénieurs du numérique in Paris, France. The position is fully funded for 3 years by ISEP. Candidate profile -

Master’s degree in machine learning or signal processing or applied mathematics. Basic knowledge of machine learning algorithms Solid programming skills (Python, Matlab, …) Good proficiency in English (both written and spoken).

Application: To apply send an email to Michel Terré ([email protected]) and Yahia Medjahdi ([email protected]) with your curriculum vitæ (including contact details, and two references), full university transcripts and a motivation letter.