Joachim Rosseel (Storm)

Greetings everyone,

Joachim joined our team recently, but we haven’t had the chance to talk about his previous and current work. Well, we change this now! Well, not right now, but on June 7!

See you there!

Speaker

Joachim Rosseel

When

Friday, June 7, 2024 at 10:30 am.

Where

LaBRI’s amphitheater or online (Zoom link:https://u-bordeaux-fr.zoom.us/j/84690333219?pwd=apR8nXiSYBr5H9peZb2o32Se2bbGbK.1)

Title

Error correcting codes decoding for the IoT context and for 5G communication chain

Abstract

In this team seminar, I will first present my thesis, entitled "Error correcting codes decoding assisted by deep learning for IoT". My thesis was focused on improving the decoding of short LDPC codes with a Belief propagation algorithm (BP) as part of the Internet of Things (IoT) context, thanks in particular to machine learning tools such as neural networks. To do so, we developed new training methods specializing a BP modeled by a Recurrent Neural Network (BP-RNN) on decoding error events sharing similar structural properties. These specialization approaches were subsequently associated with decoding architectures composed of several specialized BP-RNNs, where each BP-RNN is trained to decode a specific kind of error events (decoding diversity). Secondly, we were interested in the post-processing of the BP (or the BP-RNN) with an Ordered Statistics Decoding (OSD) in order to close the gap to the maximum likelihood (ML) decoding performance. To improve the post-processing performance, we also implemented a decoding diversity strategy with multiple OSD post-processing. We showed that this strategy effectively takes advantage of the diversity, thus providing an effective way to close the gap with ML decoding. I will then describe my post-doc at INRIA and my advancement so far. The purpose of this post-doc is to develop software modules necessary for the implementation of 5G standard digital communication chains. The modules will be implemented in AFF3CT, a toolbox dedicated to Forward Error Correction. The performance of several communication chains will thus be evaluated in terms of error correcting performance. An energy consumption study shall also be explored. A showcase of AFF3CT’s capabilities will thus be constituted. As for now, I am currently getting to know AFF3CT with a first example of a 5G chain implemented with python and C++.