(H/F) EID-5GSmartFact-CNRS-2: Transfer learning for optimizing intelligent radio environments (IREs)

Updated: 3 months ago
Job Type: FullTime
Deadline: 27 May 2021

The Laboratory of Signals and Systems (CNRS UMR 8506), Paris-Saclay University, Paris, France is seeking to appoint two Early Stage Researchers (ESRs) to join the Marie Sklodowska-Curie (MSCA) Industrial Doctorate Training Network on 'Future Wireless Connected and Automated Industry Enabled by 5G' (5GSmartFact).

About 5GSmartFact
The transition into the 4th industrial revolution promises to integrate Internet of things (IoT) and cyber-physical systems into the industrial domain and to boost the productivity of industrial verticals thanks to a radical automation of all the phases of production. Communications are key to enable Industry 4.0 (i4.0) but they are subject to the stringent requirements of automated applications in terms of availability, reliability, low latency, integrity, scalability, safety and positioning accuracy. A wirelessly connected factory enables novel mobile robots, easy reconfiguration of assembly lines and migration of embedded control functions to the virtually infinite computational/cache resources and flexibility of edge clouds. From a managerial perspective, integrated billing, and tracking capabilities of 5G facilitate novel models such as that can drive a business disruption. As a result, the i4.0 ecosystem is an opportunity for the wireless community and has become one of the key targets of 5G. From a technical side, the development of wireless i4.0 entails a paradigm shift from reactive and centralized networks towards massive, ultra-reliable and proactive networks that may operate in wide remote scenarios, with thousands of devices, uncertainty, high dynamics, rare events, unpredictable interference and harsh conditions. Merging 5G networks and i4.0 comes with its own difficulties, because these two domains have been disjointed so far. Here is the key opening identified by 5GSmartFact: The need of a surge of skilled researchers and engineers in the upcoming years to work at the crossroads of factory automation and 5G evolutions. Having this in mind, the objective of the research programme is to train young researchers to be able to analyse, design, develop and assess the deployment of 5G networks that target the i4.0 requirements and exploit them to integrate current robot applications which might lead to a complete redesign of robot architectures and hence to a leap forward in the automation industry.

The Role
CNRS-1 and CNRS-2 will be hosted by the Laboratory of Signals and Systems (CNRS UMR 8506), which is located in CentraleSupelec, Paris-Saclay campus, Paris, France. Both ESRs will be enrolled on the Ph.D. programme of Paris-Saclay University, Paris, France, and will write their thesis on a topic related to “Intelligent radio environments for Industry 4.0” (CNRS-1) and “Transfer learning for optimizing intelligent radio environments” (CNRS-2), under the supervision of Dr. Marco Di Renzo for the entire duration of their Ph.D. programme (three years). In according to the requirements of MSCA industrial training networks, the ESRs CNRS-1 and CNRS-2 will spend half of their time (eighteen months) in the premises of NEC Europe in Germany under the co-supervision of Dr. Vincenzo Sciancalepore (subject to legal feasibility).

Position: CNRS-2
Title: Transfer learning for optimizing intelligent radio environments.
Scientific context: Wired communications are the status quo in industry because of their high level of reliability and stable latency. The downside of using cables is the expensive, bulky, and inflexible deployment: an ensemble of robots cannot be freely manoeuvred in a warehouse. Wireless solutions provide much higher flexibility but are prone to channel fading, shadowing, interference, and disturbances from industrial machines. However, contemporary wireless networks are designed based on the postulate that only the transmitters and the receivers can be optimized for improving the network performance. The propagation environment that lies in between them (physical objects like walls, buildings, furniture, ceilings, floors, etc.) are out of control of the communication engineers. This approach to design and optimize wireless networks has three fundamental limitations: (i) the ultimate performance of wireless networks may not have been reached yet - By optimizing the transmitter, the receiver, and the environment, the performance of wireless networks may be further improved; (ii) in the industry Internet of Things, some devices are unlikely to be equipped with multiple antennas - Having the opportunity of customizing and controlling the environment may open new opportunities for network optimization; (iii) The radio waves are used inefficiently, since when reflected or refracted by an object, for example, the energy is scattered towards unwanted directions, thus reducing the efficiency of utilization of the emitted power - Equipping wireless networks with the functionalities of customizing the radio environment (i.e. controlling the propagation of radio waves and programming environmental objects) besides the capability of optimizing the end-points of multi-antenna distributed communication links, and optimizing the resulting wireless communications with the aid of machine learning computational techniques, constitute a paradigm-shift vision affecting the physical layer and the medium access control. The propagation environment within a given scenario has a unique structure that could be controlled using reconfigurable metasurfaces (RMS), complementing multi-antenna technologies at the transmitter and receiver. RMSs are an emerging technology that can alter the wavefront of a radio wave that impinges upon them by controlling an external stimulus and shape complex propagation environments, such as industrial environments. However, their fundamental performance limits and optimized design and operation are not known yet. In this context, the use of model-based methods and machine learning based methods (data based) constitute a promising solution for enhancing the performance in practical propagation environments.

Objectives: 1) To conduct research on an emerging approach for optimizing networks jointly combining mismatched model-based and small-size data driven (machine learning based) methods using tools of transfer learning and deep unfolding. 2) To jointly combine stochastic geometry and optimization methods in order to develop optimal resource allocation and scheduling algorithms bringing full potential to IRE-based deployments for industry Internet of Things. 3) To leverage the theory of point processes and Benders decomposition to solve general mixed-integer non-linear optimization scheduling and resource allocation problems with performance guarantee.
Expected results: 1) A new set of innovative approach to system optimization based on jointly combining models and data. 2) Full characterization of performance limits and guidelines for their optimal design and operation in realistic industry Internet of Things. 3) Publications in flagship conferences and leading journals. 4) Integration in the NEC system simulator and patent filing.

Planned secondment(s): NEC (Germany) for 18 months. (subject to legal feasibility)

Ph.D. enrolment: Paris-Saclay University (France).


View or Apply

Similar Positions