Error correction, very high throughput, DNA storage, AI contributions... A team of researchers from the Mathematical & Electrical Engineering (MEE) department at IMT Atlantique is working on the latest advances in coding, which plays a key role in all digital communication systems. In September, the institution will also organise the International Symposium on Topics in Coding, ISTC 2023, the main scientific conference on these subjects.
The continuous expansion of digital technology is generating dizzying quantities of data in the form of bits (sequences of 0's and 1's), which can be subject to errors due to a variety of causes: a signal that is too weak, parasitic noise, a breakdown, a defect in the medium, etc. All of which can affect the understanding of the message. Hence the need to design coding, storage and decoding systems capable of compensating for these errors, or at least of significantly reducing their number.
This is the subject on which the CODES team of the Mathematical & Electrical Engineering (MEE) department at IMT Atlantique is working. Composed of five associate professors, a dozen PhD students and three post-docs, it operates in conjunction with the Lab-STICC laboratory (1), which brings together half a dozen scientific establishments in Brittany. Its work consists of designing and developing "error correction" and source coding solutions for all digital communication and storage systems.
"Error correction is a function of the physical layer, which interacts with the other functions of the system," explains Raphaël Le Bidan, one of the MEE's associate professors. "The principle is to introduce certain redundancies into the message - as few as possible - to avoid or correct a maximum number of errors, while keeping the code simple to use". Despite this constraint, error-correcting coding, which was first used for space communications, offers a number of advantages: "it makes the message more 'robust', and therefore increases the range of the transmitter, and reduces battery power".
Catherine Douillard, Elsa Dupraz, Charbel Abdel Nour and Raphaël Le Bidan.
"Coding calls on multiple disciplines," says Catherine Douillard, professor and deputy head of the MEE department, who has been working in this field for nearly 30 years. "It is at the crossroads of information theory, algebra and discrete mathematics, and also requires skills in electronics."
As close as possible to the theoretical limits
The main challenge is to develop codes and decoding algorithms capable of approaching as closely as possible the theoretical limits in correction/compression stated by the American mathematician Claude Shannon in the late 1940s. Theoretically, it is known that very long codes, comprising tens or even hundreds of thousands of bits, can achieve such a level of performance. However, they pose formidable implementation problems when it comes to complying with practical constraints in terms of system or hardware specifications (limited energy consumption, guarantee of calculation integrity).
In the early 1990s, researchers at Télécom Bretagne (which became IMT Atlantique in 2017), including Claude Berrou and Alain Glavieux, made a decisive breakthrough with the development of a new class of corrective codes, the "turbo codes". These have been used in particular in mobile telephony standards: 3G, 4G and part of 5G.
"Today, we are faced with new challenges," explains Catherine Douillard. "Communication systems have become more complex, and the requirements in terms of data reliability have become more stringent: for example, for optical communications, we can hardly tolerate more than one error for every 10 to 15 bits transmitted. The aim is to achieve ever higher data rates while limiting the power consumption of the circuits. In some systems, we are aiming for terabit/second." One solution is to use parallel architectures.
The rise of the Internet of Things, for example, requires the use of very short packets. The behaviour and construction of short codes are not yet well understood, but the use of artificial intelligence appears to be a possible lever. Raphaël Le Bidan is leading a project funded by the ANR (2), 'AI4Code', which brings together half a dozen laboratories in France to develop coding and decoding devices intended in particular for the Internet of Things and all these new forms of communication systems based on the transmission of very small packets of data.
The promise of DNA storage
Another avenue of research is the storage of data on synthetic DNA strands. "DNA has many advantages: it is a very compact medium - a van would be enough to contain all the digital information available in the world," stresses Elsa Dupraz, Associate professor at the MEE. "What's more, it can be used at room temperature, and it is durable". Initial demonstrators have been developed - notably by Microsoft. However, DNA sequencing introduces many errors (random modification, insertion and deletion of symbols), different from those in the telecoms world. "We are mainly working in two directions: statistical modelling of these errors linked to the DNA medium, and the development of corrective codes," explains Elsa Dupraz. This work is the subject of an exploratory PEPR (Priority Programme and Equipment for Research) called MoleculArXiv, which brings together several teams in France. Led by the CNRS, it has a budget of 20 million euros over 7 years.
"Beyond the scientific issues themselves, there are also questions of sovereignty," adds Charbel Abdel Nour, also a professor in the MEE department and co-chair of the ISTC 2023 programme committee. It is indeed important to have cutting-edge teams in France and Europe working on these key technologies, which play a major role in innovation and economic development - particularly through standards. The Institut Mines-Télécom thus responded to a call for projects on digital sovereignty (PIA4), managed by the ANR. In association with SATT (3) Ouest Valorisation, the institute wishes to 'strengthen project support' and develop intellectual property by producing patent 'clusters'. Over the past few years, the team has already filed some twenty patents - one of which has even become an international standard for 2nd generation DTT.
An international symposium on coding in Brest
IMT Atlantique has been selected to organise the ISTC (International Symposium on Topics in Coding) conference in Brest, from 4 to 8 September, which welcomes the best coding specialists every other year. Between 150 and 200 researchers from all over the world are expected to attend. The topics to be discussed include quantum coding and the contribution of AI techniques - for example, in the acquisition of new high-performance coding models. This is a form of international recognition for the MEE department team and for IMT Atlantique, which initiated the first edition in 1997.
- (1) Laboratory of Information, Communication and Knowledge Sciences and Technologies (CNRS UMR 6285). The Lab-STICC is an international reference centre for communicating systems and has a staff of over 500, including 220 researchers andassociate professors from several institutions (CNRS, ENIB, ENSTA Bretagne, IMT Atlantique, UBO and UBS).
- (2) National Research Agency.
- (3) Société d'Accélération du Transfert de Technologies.
by Pierre-Hervé VAILLANT