Periodic Reporting for period 2 - WINDMILL (Integrating wireless communication engineering and machine learning)
Reporting period: 2021-01-01 to 2023-06-30
As wireless communication networks evolve towards 5G and beyond, we are entering an era of massive connectivity, massive data, and extreme service demands. The unprecedented level of of connectivity enabled by 5G will be one of the cornerstones digital transformation, impacting all segments of society - education, health, security, transportation, industrial production, etc. It will also revolutionize the communications between humans and machines, merging physical and digital worlds.
However, it is challenging to handle such complex networks and the involved data volumes successfully. A promising approach to this issue is to develop new network management and optimization tools based on machine learning. This presents a major shift in the design and operation of wireless networks. At the same time, the approach demands a new type of expertise: a combination of engineering, mathematics and computer science disciplines. The ITN project WindMill addresses thus this issue by providing relevant interdisciplinary training. In the course of the project, 15 Early Stage Researchers (ESRs) were trained in integrating wireless communications and machine learning. The trainings were provided by a consortium of leading international research institutes and companies with experts in wireless communications and machine learning.
However, it is challenging to handle such complex networks and the involved data volumes successfully. A promising approach to this issue is to develop new network management and optimization tools based on machine learning. This presents a major shift in the design and operation of wireless networks. At the same time, the approach demands a new type of expertise: a combination of engineering, mathematics and computer science disciplines. The ITN project WindMill addresses thus this issue by providing relevant interdisciplinary training. In the course of the project, 15 Early Stage Researchers (ESRs) were trained in integrating wireless communications and machine learning. The trainings were provided by a consortium of leading international research institutes and companies with experts in wireless communications and machine learning.
The work carried out in the project can be divided in the following general topic:
- Applying machine learning methods to optimize the performance of the new generation of wireless cellular systems: A major part of WindMill project addressed novel generations of cellular systems, like 5G and beyond 5G networks. These systems are incredibly complex, such that even their analytical modelling is often impossible. Machine learning tools are well suited for such scenarios and in the project they were successfully applied to a number of related research challenges, e.g. channel charting (creating the radio maps of mobile network users), use of intelligent surfaces (advanced antenna arrays) for localization purposes, and cell-free massive MIMO systems.
- Optimization of the use of resources in wireless access networks: With the ongoing transition to wireless everywhere and anywhere, wireless access is becoming the critical segment that requires careful optimization of the use of time, spectrum, and energy resources. WindMill project investigated application of deep reinforcement learning for provision of an efficient and fair access to massive number of users in the wireless access (characteristic for IoT networks), use of neural networks to improve spectral efficiency, and use of learning-based methods for resource allocation and scheduling in WiFi networks.
Tailoring and optimizing machine learning algorithms and frameworks for the application in wireless communications: Problems in wireless communications come with their own challenges and limitations that require adjustment and adaptation of machine learning algorithms. In WindMill project, it was shown that robust Bayesian learning can deal with the lack of training data and model inconsistencies, such that it can be effectively used for a number of important applications, like localization and spectrum sensing.
Every week the Windmill project posted on the KSP either a personal blog addressing topics for general viewers or a academic topic for those with a specialised knowledge of the field. In total the views seen through Windmill's YouTube channel goet 9697 views.
The list of publications totals 35 papers. They can be found at https://windmill-itn.eu/publications/
The ESRs presented papers and topics at a number of conferences and workshops in 2022, Helsinki and Serbia, and 2023, Spain. Plans for an earlier introduction to academic presentations were made but changed due to repeated Corona shutdowns.
- Applying machine learning methods to optimize the performance of the new generation of wireless cellular systems: A major part of WindMill project addressed novel generations of cellular systems, like 5G and beyond 5G networks. These systems are incredibly complex, such that even their analytical modelling is often impossible. Machine learning tools are well suited for such scenarios and in the project they were successfully applied to a number of related research challenges, e.g. channel charting (creating the radio maps of mobile network users), use of intelligent surfaces (advanced antenna arrays) for localization purposes, and cell-free massive MIMO systems.
- Optimization of the use of resources in wireless access networks: With the ongoing transition to wireless everywhere and anywhere, wireless access is becoming the critical segment that requires careful optimization of the use of time, spectrum, and energy resources. WindMill project investigated application of deep reinforcement learning for provision of an efficient and fair access to massive number of users in the wireless access (characteristic for IoT networks), use of neural networks to improve spectral efficiency, and use of learning-based methods for resource allocation and scheduling in WiFi networks.
Tailoring and optimizing machine learning algorithms and frameworks for the application in wireless communications: Problems in wireless communications come with their own challenges and limitations that require adjustment and adaptation of machine learning algorithms. In WindMill project, it was shown that robust Bayesian learning can deal with the lack of training data and model inconsistencies, such that it can be effectively used for a number of important applications, like localization and spectrum sensing.
Every week the Windmill project posted on the KSP either a personal blog addressing topics for general viewers or a academic topic for those with a specialised knowledge of the field. In total the views seen through Windmill's YouTube channel goet 9697 views.
The list of publications totals 35 papers. They can be found at https://windmill-itn.eu/publications/
The ESRs presented papers and topics at a number of conferences and workshops in 2022, Helsinki and Serbia, and 2023, Spain. Plans for an earlier introduction to academic presentations were made but changed due to repeated Corona shutdowns.
The project advanced state-of-the art in the following areas:
- channel charting, localization and sensing techniques,
- advancing robust Bayesian learning for wireless systems,
- beamforming for massive MIMO setups,
- network optimization in industrial control setups,
- optimization of resource allocation in wireless networks, etc.
The project also established pioneering results on:
- autonomous learning of communication protocols by network devices using reinforcement learning,
- use of neural networks to recognize human gestures using measurements obtained by millimeter wave radars - this line of research may enable novel applications in industrial, vehicular, and smart home scenarios.
The most important outcome is that the project achieved training a cohort of young and talented researchers which are now ready to develop and design new generations of wireless systems. Although the project has just finished, several ESRs have already made excellent career moves. For example, two accepted postdoc offers (one from King’s College London and the other in Swiss Data Science Center), two are currently working in Nokia and Nokia Bell Labs and one has advanced to a permanent position in Bosch.
- channel charting, localization and sensing techniques,
- advancing robust Bayesian learning for wireless systems,
- beamforming for massive MIMO setups,
- network optimization in industrial control setups,
- optimization of resource allocation in wireless networks, etc.
The project also established pioneering results on:
- autonomous learning of communication protocols by network devices using reinforcement learning,
- use of neural networks to recognize human gestures using measurements obtained by millimeter wave radars - this line of research may enable novel applications in industrial, vehicular, and smart home scenarios.
The most important outcome is that the project achieved training a cohort of young and talented researchers which are now ready to develop and design new generations of wireless systems. Although the project has just finished, several ESRs have already made excellent career moves. For example, two accepted postdoc offers (one from King’s College London and the other in Swiss Data Science Center), two are currently working in Nokia and Nokia Bell Labs and one has advanced to a permanent position in Bosch.