Periodic Reporting for period 4 - PERFUME (Smart Device Communication: A paradigm for high PERformance FUture Mobile nEtworking)
Période du rapport: 2020-04-01 au 2020-09-30
PERFUME has targeted the various axis of research below:
WP 1: Fundamental limits for future coordinated wireless networks
We aimed at development of information theoretic methods for decentralized device coordinated communications.
This area gives a fundamental layer to the project, is mathematically rooted and serves multiple key purposes: It lets us understand fundamental limits of communication performance, and expose new enabling analytic tools that play an important role in building systems that near those limits.
WP2: Smart device communication methods for the future Mobile Internet
2.1 Coordinated communication methods for 5G and beyond
This second area concerns the development of decentralized coordination mechanisms within the context of 5G mobile standards, i.e. using the 5G system assumptions as design constraints. This objective allowed for short-to-mid term exploitation of the novel system concepts pushed in the project.
2.2: Communications for dynamic and moving networks
PERFUME has pushed the development of algorithms and methods for the case where devices are both mobile and acting with some level of autonomy. A crucial case for this study is the case where radio devices are equipped on aerial devices, such as UAV (unmanned aerial devices).
In this context, PERFUME is envisioning the use of flying robots (autonomous drones) that are fully integrated with wireless network equipment in a number of scenarios. Flying relays or flying base stations can enable better and faster response in disaster recovery scenarios (after earthquakes or floods). They can also be used for intelligent IoT data collection, in monitoring smart cities and smart agriculture.
WP3: Experimental and prototyping development
As part of this research program on autonomous decision-making wireless flying devices, PERFUME has been developing prototypes for "autonomous aerial cellular relay robots". These use machine learning algorithms to self-optimize their parameters for improving overall radio network performance. The communication layers built on algorithms leveraging the world's leading opensource implementation of cellular LTE (OpenAirInterface), developed at EURECOM's Communications Systems Department.
WP 1: Fundamental limits for future coordinated wireless networks
We aimed at development of information theoretic methods for decentralized device coordinated communications.
This area gives a fundamental layer to the project, is mathematically rooted and serves multiple key purposes: It lets us understand fundamental limits of communication performance, and expose new enabling analytic tools that play an important role in building systems that near those limits.
WP2: Smart device communication methods for the future Mobile Internet
2.1 Coordinated communication methods for 5G and beyond
This second area concerns the development of decentralized coordination mechanisms within the context of 5G mobile standards, i.e. using the 5G system assumptions as design constraints. This objective allowed for short-to-mid term exploitation of the novel system concepts pushed in the project.
2.2: Communications for dynamic and moving networks
PERFUME has pushed the development of algorithms and methods for the case where devices are both mobile and acting with some level of autonomy. A crucial case for this study is the case where radio devices are equipped on aerial devices, such as UAV (unmanned aerial devices).
In this context, PERFUME is envisioning the use of flying robots (autonomous drones) that are fully integrated with wireless network equipment in a number of scenarios. Flying relays or flying base stations can enable better and faster response in disaster recovery scenarios (after earthquakes or floods). They can also be used for intelligent IoT data collection, in monitoring smart cities and smart agriculture.
WP3: Experimental and prototyping development
As part of this research program on autonomous decision-making wireless flying devices, PERFUME has been developing prototypes for "autonomous aerial cellular relay robots". These use machine learning algorithms to self-optimize their parameters for improving overall radio network performance. The communication layers built on algorithms leveraging the world's leading opensource implementation of cellular LTE (OpenAirInterface), developed at EURECOM's Communications Systems Department.
Results by Key Areas:
WP 1: Fundamental limits for future coordinated wireless networks
Results in this area rely on the exploitation of fundamental enabling tools, which are namely information theory as well as machine learning. Note that machine learning represents a quickly growing trend within wireless communication network design. While unforeseen at the time of submission of the project, The PI and his team, especially Paul de Kerret have played a pioneering role in this area, enhancing PERFUME visibility along the way.
Details are now given on the results, first on information theoretic results, then on machine learning results.
Based on the use of information theoretic tools, we have developed several algorithms for Robust Precoding for Transmitter Cooperation with imperfect CSIT Sharing.These algorithms allow to build robustness in the problem of decentralized decision making when different transmitters are endowed with different estimated from the radio environment.
In parallel to the algorithmic work, we have also developed a characterization of the fundamental limits for wireless agents with decentralized decision under information uncertainties. These results have been published in the information theoretic literature.
Application of Machine Learning to Team Decision problems
The PERFUME team was an early contributor and pioneer to the field of machine learning for decentralized decision making under uncertainties. We conducted fundamental studies in this area which raised the visibility of our project in the scientific community. In particular we introduced the new concept of Team Deep Neural Networks.
WP2: Smart device communication methods for the future Mobile Internet
2.1 Coordinated communication methods for 5G and beyond
A) Location-Aided Beam Alignment in mmWave communications
Among the enabling technologies for 5G wireless networks, millimeter wave (mmWave) communication offers the chance to deal with the bandwidth shortage affecting wireless carriers. Beam alignement is known to be a bottleneck in this area when mmwave devices are equipped with many antennas.
In PERFUME we have introduced several robust beam alignment methods in order to exhibit resilience with respect to the noise present in the position information obtained at the BS and the UE, in a single-user scenario.
The effectiveness of the robust beam alignment procedure, compared with classical designs, has been verified on simulation settings with varying location information accuracies.
2.2: Communications for dynamic and moving networks
Theory and algorithm of autonomous placement for flying radio devices
PERFUME conducted analysis and algorithm design in the area of autonomous placement and path planing for a flying radio device. The assumption of Lineof-sight (LoS) channels or the use of simple statistical blocking models (i.e. modeling the LoS probability) has proved an excellent way to derive early insights into the problem and to allow for closed-form average performance analysis. Unfortunately, the simplified or probabilistic nature of such approaches limit our ability to guarantee performance in an actual on-field UAV deployment. For example, a statistically optimized placement algorithm might suggest a UAV location which one eventually discovers to be severely affected by local blockage in practice (e.g. unforeseen presence of a tall building) forcing the robot to some suboptimal path recomputing. PERFUME has made substantial progress to go around this drawback. One of PERFUME’s achievement has been to identify and highlight the role played by the exploitation of suitable maps for UAV placement in the DaaR and DaaB settings. By maps, we here refer to a geographically indexed data set which can be used to better predict the actual channel conditions for any specific pair of UAV and ground node locations. Several types of inter-related maps can be considered, including throughput maps, radio (link strength) maps, and terrain maps. A suite of algorithms were designed by PERFUME team members that were capable of enhancing UAV trajectory design on the basis of such maps.
WP3 Experimentations
PERFUME started an experimentation activity allowing us to test our ideas, with emphasis on drone-aided communications. We obtained a working prototype with a flying relay onboard a drone is capable of steering itself to an optimized network location to help ground users.
WP 1: Fundamental limits for future coordinated wireless networks
Results in this area rely on the exploitation of fundamental enabling tools, which are namely information theory as well as machine learning. Note that machine learning represents a quickly growing trend within wireless communication network design. While unforeseen at the time of submission of the project, The PI and his team, especially Paul de Kerret have played a pioneering role in this area, enhancing PERFUME visibility along the way.
Details are now given on the results, first on information theoretic results, then on machine learning results.
Based on the use of information theoretic tools, we have developed several algorithms for Robust Precoding for Transmitter Cooperation with imperfect CSIT Sharing.These algorithms allow to build robustness in the problem of decentralized decision making when different transmitters are endowed with different estimated from the radio environment.
In parallel to the algorithmic work, we have also developed a characterization of the fundamental limits for wireless agents with decentralized decision under information uncertainties. These results have been published in the information theoretic literature.
Application of Machine Learning to Team Decision problems
The PERFUME team was an early contributor and pioneer to the field of machine learning for decentralized decision making under uncertainties. We conducted fundamental studies in this area which raised the visibility of our project in the scientific community. In particular we introduced the new concept of Team Deep Neural Networks.
WP2: Smart device communication methods for the future Mobile Internet
2.1 Coordinated communication methods for 5G and beyond
A) Location-Aided Beam Alignment in mmWave communications
Among the enabling technologies for 5G wireless networks, millimeter wave (mmWave) communication offers the chance to deal with the bandwidth shortage affecting wireless carriers. Beam alignement is known to be a bottleneck in this area when mmwave devices are equipped with many antennas.
In PERFUME we have introduced several robust beam alignment methods in order to exhibit resilience with respect to the noise present in the position information obtained at the BS and the UE, in a single-user scenario.
The effectiveness of the robust beam alignment procedure, compared with classical designs, has been verified on simulation settings with varying location information accuracies.
2.2: Communications for dynamic and moving networks
Theory and algorithm of autonomous placement for flying radio devices
PERFUME conducted analysis and algorithm design in the area of autonomous placement and path planing for a flying radio device. The assumption of Lineof-sight (LoS) channels or the use of simple statistical blocking models (i.e. modeling the LoS probability) has proved an excellent way to derive early insights into the problem and to allow for closed-form average performance analysis. Unfortunately, the simplified or probabilistic nature of such approaches limit our ability to guarantee performance in an actual on-field UAV deployment. For example, a statistically optimized placement algorithm might suggest a UAV location which one eventually discovers to be severely affected by local blockage in practice (e.g. unforeseen presence of a tall building) forcing the robot to some suboptimal path recomputing. PERFUME has made substantial progress to go around this drawback. One of PERFUME’s achievement has been to identify and highlight the role played by the exploitation of suitable maps for UAV placement in the DaaR and DaaB settings. By maps, we here refer to a geographically indexed data set which can be used to better predict the actual channel conditions for any specific pair of UAV and ground node locations. Several types of inter-related maps can be considered, including throughput maps, radio (link strength) maps, and terrain maps. A suite of algorithms were designed by PERFUME team members that were capable of enhancing UAV trajectory design on the basis of such maps.
WP3 Experimentations
PERFUME started an experimentation activity allowing us to test our ideas, with emphasis on drone-aided communications. We obtained a working prototype with a flying relay onboard a drone is capable of steering itself to an optimized network location to help ground users.
PERFUME has pushed beyond the state of the art subtantially in several domaines. First of all, we have derived some fundamental limits decentralized wireless coordination from information theory, statistics and machine learning. Secondly we have demonstrated the application of such tools in the derivation of implementable methods for future wireless networks. Thirdly, a subset of the tools are to be tried with live experiments at the cross-road between communication and robotics as this has proved so far a success both to gain visibility and to attract talented researchers in the team.
Overall, the research advances have been presented in 21 journal publications (A ranked) and 46 conference publications (also top ranked for the most part).
Overall, the research advances have been presented in 21 journal publications (A ranked) and 46 conference publications (also top ranked for the most part).