• Brisch, Fabian fabian.brisch@hs-osnabrueck.de
• Khalid Ali khalid.ali@th-deg.de
• Özgür Kaynak oezguer.kaynak@th-deg.de

• Mohsen Meamarian mohsen.meamarine@kau.se  Data plane programming
• Hamza Chahed Hamza.Chahed@kau.se 
• Phil Aupke Phil.Aupke@kau.se Smart Grid Forecasting
• Amal Nammouchi Amal.Nammouchi@kau.se Smart Grid Control
• Rubens Figueiredo  rubens.figueiredo@kau.se Industrial PhD candidate with B-ISDN GmbH

• Florian Aigner

• Ali Kotb ali_kotb97@hotmail.com
• Anton Johansson gurkor211@gmail.com

• Lecture Internet Technologies (4 SWS) Bachelor AI
• Lecture Internet Technologies (4 SWS) Bachelor Bioinformatics
• Lecture Datacenter Network, Programming (FWP) (4 SWS) Masterlevel

• Mobile and wireless Networks (4 SWS) 
• Netze Operating Systems and Networks (4 SWS) Bachelor Artificial Intelligent Networks
• Research Methodology & Statistics (4 SWS)

• Lecture Datacenter Network, Programming (FWP) Master Level
• Lecture Internet Technologies (4 SWS) Bachelor Bioinformatics

• As a close partnership between Karlstad University and Deutsche Telekom
•  Demanding academic challenges with high relevance for the industry
• Well-equipped laboratories and test environments available
• Possible visits or implementation in Germany

choose from 2 main focus areas and let's define topics together

5G ATSSS multipath communication

Multipath communication based on 5G ATSSS enables the customer to use Wi-Fi and cellular access simultaneously via a cell phone. Compared to traditional single access usage, increased reliability and throughput rates are provided. These multipath concepts offer numerous opportunities for new traffic engineering approaches and rely on multipath network protocols such as MPTCP. Although this offers clear advantages at concept level, there are still open questions when it comes to implementation.

• How can consistent multipath performance be ensured across current and expected traffic mixes?
• What are the necessary metrics for multipath traffic distribution and reunification strategies?

Skills you should bring along:

• Understanding of multipath concepts and network protocols (MPTCP, MP-DCCP, MP-QUIC)
• Linux kernel hacking / C programming language
• Network traffic analysis

Literature:

• 5G ATSSS [1, 2],
• Non-TCP multipath support [3],
• MPTCP [4],
• MP-DCCP [5, 6],
• MP-QUIC [7]

Open 5GS testbed for 5G feature exploration

5G ATSSS multipath communication 5G Core Wi-FiCellular features introduced with the first 5G 3GPP release 15 and expected with the upcoming release 16 offer mobile network operators new opportunities that have not yet been fully explored. Technologies such as network slicing and multipath communication (5G ATSSS) are mentioned here only as examples.

In order to gain practical experience, such new functions are to be examined in an open 5G system test bed under various conditions, e.g. based on free5GC or OpenAirInterface. Such a test bed still needs to be implemented and missing functions need to be implemented. Analysis and evaluation are an essential part of gaining insight into the respective function and estimating its value. In addition, new ideas can be evaluated that can later be incorporated into standards.

Skills you should bring along:

• Understanding the 5G system + 3GPP documents
• 5G control plane (service-based architecture, protocols, policies)
• Linux-based implementation and operation

Literature:

• free5GC [1],
• OpenAirInterface [2],
• 5G System [3],
• 5G Policy Framework [4] 2 

Forecasting electricity generation from renewable energy sources such as solar modules is important for smart grid management. Together with a good forecast quality, a precise characterization of the forecast error is required. However, such a forecasting algorithm should be integrated into an energy management system to be useful for controlling where the energy is used (e.g. whether the battery is charged or the electricity is sold to other members or the smart grid, etc.).

In this work, you will familiarize yourself with the IoT platform ThingsBoard and integrate a solar power generation prediction algorithm into this platform using Apache Kafka as an intermediate publish/subscribe framework. In detail, the work includes:

• Adaptation of an existing prediction algorithm and/or development of a new one using e.g. time series clustering and random forest or LSTM for prediction
• Integration of the ThingsBoard platform with a distributed/publish-subscribe platform, Apache Kafka
• Implementation of a Kafka consumer that reads the weather features and stream readings and invokes the ML algorithm with the Faust stream processing framework
• Implementation of a Kafka producer (using the Faust stream framework) that produces the forecast, analyzes the forecast error and feeds forecast and error metrics back into the ThingsBoard IoT platform
• Implementation of a visualization of the forecast error in the ThingsBoard IoT platform
• If time permits, development of a new forecasting method that combines a physical model of solar power generation with machine learning based time series forecasting

Excellent programming skills in Python are required

Time Critical Networks AB (TCN) develops TCN TimeAnalysis™ for digital modeling and simulation of primarily electrical architectures and networks in the automotive industry that integrate Time-Sensitive Networking (TSN), the new IEEE standards for making traditional Ethernet networks that involve time-critical data streams more deterministic and robust.

Target groups of the tool can be found in the automotive industry, for example, who use the software to design future communication systems for autonomous driving inside and outside the vehicle (V2V/V2X).

Background

TSN networks are Ethernet networks that provide QoS guarantees for highly critical traffic in terms of latency, jitter and bandwidth. To achieve this, the configuration of the TSN elements (switches and endpoints) must be optimized and properly tested.

Suggested work

The aim of this project is to automate the configuration of the TSN network elements using TCN's simulation tool. This includes providing scheduling algorithms to derive the configuration, testing the derived configuration in the simulator, checking the results and transferring the configuration to the network elements.

Technical Details

Specifically, we have developed a CNC unit (an SDN-based controller for TSN networks) that can configure the TSN-enabled switch in our testbed using the Netconf/Yang protocol. The configuration is derived by an external network optimizer outside the CNC using a network optimization interface. We also have access to TCN's simulator engine, which we use to simulate the network under the optimized configuration to get an idea of the expected performance. The task of the thesis is to integrate the three parts (CNC, TCN's simulator and network optimizer) by defining and implementing the interface between them as well as designing the workflow for automating the optimization - simulation - deployment. The task also includes the deployment of the optimizer and its interface to the CNC in the best possible way to perform the above workflow.

Note

The development of the optimization algorithm is not part of the project as it is provided to you.

External Partner: Time Critical Networks AB

The integration of renewable energy (RE) sources into electricity grids and advances in information and communication technology, such as cloud and edge computing, have significant potential for the development of smart energy grids in which a customer has production, consumption and storage capabilities, leading to the concept of the prosumer. The variability of RE production gives the prosumer an active role in the operational management of the smart grid and contributes to CO2 reduction for flexible and sustainable energy systems. Therefore, coordination for energy production and management between distributed prosumers is essential to realize the potential of large-scale renewable energy trading.

In this thesis, you will deepen your understanding of AI-powered intelligent networks and the further requirements for edge/cloud integration beyond the initial intelligent network architecture. This will enable you to provide research and innovation such as the development of comprehensive AI and ML algorithms that facilitate the optimization of distributed energy management plans enabled by cloud/edge integration, the development of extensions and updates to cloud/edge of virtual power plant platforms to integrate with ML algorithms. The results will lead to more flexibility and robustness of distributed RE production and trading, which is central to the evolution of the smart grid concept towards AI-powered future energy systems. In terms of robustness, it is important to develop methods that enable better prediction of RE supply and demand as well as quantification of RE uncertainties. In this context, the partner Glava has solar panels and wind farms with associated weather stations where data from the last 5 years is available. Furthermore, it is possible to connect to relevant data in real time. The main objective of this thesis is to develop new methods based on machine learning to better quantify the variability and uncertainty of RE production and/or supply. Alternatively, the student may focus on integrating machine learning methods for predicting RE supply and demand into the Thingsboard IoT platform for the SmartGrid use cases.

External Partner: Glava Energy Center

With the evolution of mobile networks towards 5G, we are able to offer services that require low latency, high data rates and massive connectivity. In this context, user/device location awareness is an essential function, especially for Internet of Things (IoT) applications, and is therefore considered an integral part of the design of 5G and beyond systems.

Global Navigation Satellite Systems (GNSS) may not be suitable in some cases due to the additional power consumption and cost required to support GNSS chips. Therefore, alternatives that enable reliable positioning via mobile cellular technologies, such as 4G Long Term Evolution (LTE), 5G New Radio (NR) and Narrowband IoT (NB-IoT), as highlighted by current investigations in the research community, are required.

This Thesis will focus on the development and testing of positioning techniques in multi-technology scenarios, utilizing the application of improved machine and deep learning techniques to empirical measurements [1][2].

Literature: 

[1] K. Kousias, et al., “A Large-Scale Dataset of 4G, NB-IoT, and 5G Non-Standalone Network
Measurements,” 2022. [Online]. Available: https://dx.doi.org/10.21227/7a8s-nt68. (DATASET)
[2] L. De Nardis, et al., “Positioning by Fingerprinting with Multiple Cells in NB-IoT Networks,” in
Proceedings of International Conference on Localization and GNSS. IEEE, 2022, pp. 01– 07

With the evolution of mobile networks towards 5G, we are able to offer services that require low latency, high data rates and massive connectivity. In this context, it is crucial for traffic management and network optimization to be able to reliably predict the performance experienced by end users (e.g. in terms of latency, throughput and connection stability).

Recent studies on commercial networks show that the performance of 5G can vary significantly over time and from one location to another due to the complex interactions between system protocols and network conditions. Therefore, further research is needed to provide solutions to reliably predict such performance.

This thesis will focus on the development and testing of techniques for reliable network performance prediction, utilizing the application of improved machine and deep learning techniques to empirical measurements [1][3].

Literature: 

[1] K. Kousias, et al., “A Large-Scale Dataset of 4G, NB-IoT, and 5G Non-Standalone Network
Measurements,” 2022. [Online]. Available: https://dx.doi.org/10.21227/7a8s-nt68. (DATASET)

[3] K. Kousias, et al., “Coverage and Performance Analysis of 5G Non-Standalone Deployments,” in
Proceedings of the ACM MobiCom Workshop on Wireless Network Testbeds, Experimental
Evaluation and Characterization, 2022, pp. 61–68.

The technical details of (provider-specific) RAN implementations are complex. Although general functions and interfaces are defined by 3GPP, the details remain proprietary. The RAN is typically configured for optimized coverage, handovers, quality, etc. via technical guidelines specific to each RAN provider (and specific to other criteria: Area, urban situation, frequency layer, mobility...).

To automate the generation of policies, an ML model must be trained to understand the details of the RAN provider (data models, parameters, dependencies, etc.). Based on the target (intention: desired goal) as text input, an “Intent Interpreter” uses the trained model and suggests possible policy approaches (or requests clarifications). Through iterative and interactive “ChatGPT-like” refinements, valid declarative policy code is generated (a kind of “improved program co-pilot”).

Policies can be validated by humans and must generate valid RAN gNB/cell configurations (artifacts). It remains to be clarified whether also included in the scope of work: simulation of RAN configuration changes and closed-loop based policy optimizations. Not in the scope: Policy conflict resolution (a separate topic in itself...?); Complex policies beyond RAN; Non-technical policies; Historical data; Measurements; Network topology.

Ail:

Translation of the technical intention (of a specific adaptation) into declarative guidelines (human text input -> comprehensible transparent code/expression).

External Partner: Deutsche Telekom

The configuration of Time-Sensitive Networks (TSN) is complicated due to the many available parameters and options. An important aspect is the proper configuration of network cards and network stacks of the end devices in order to coordinate real-time packet processing within the network and the end devices. In this direction, there have been initial efforts by Intel and the open source community to develop the “detd” configuration framework that properly configures multiple components of the Linux network stack, including optional hardware offloading capabilities for various Ethernet cards. In this thesis, you will extend the configuration toolchain by adding support for AF_XDP streams, which requires diving deeper into the 802.1Qcc standard, the Linux kernel network stack and especially XDP as well as Taprio in hardware mode. As a final step, add support for Docker and Kubernetes (optional) to enable TSN capabilities for cloud-native data paths. The project concludes with a performance comparison.

Requirements:

• Excellent programming skills in Python and C
• Willingness to work with the Linux kernel, Docker containers, cloud-native Dataplane (optional) and the latest TSN standards

External Partner: Intel Munich Labs

The operation of smart energy grids traditionally involves a variety of complex decision problems. In recent decades, smart grids have benefited significantly from advances in computation that have enabled the transition from manual processing to automation. Nevertheless, various stakeholders in the smart grid still need to make significant efforts to explain and interpret the optimization results. In this work, recent advances in Large Language Models (LLMs) are integrated into smart grid optimization and human sensing and confidence building.

Required knowledge:

• Python programming
• Knowledge in the solution of optimization problems (e.g. Gurobi)

Time-Sensitive Networking (TSN) is a group of Ethernet standard extensions that enable the provision of real-time guarantees for Ethernet frames. In modern Industry 4.0 environments, end devices are connected to sensors and actuators via TSN to enable software-based real-time control loops. However, it is unclear which configurations of Linux endpoints provide which real-time guarantees. In this master thesis, the impact of different configuration decisions of applications, the Linux operating system and the network stack on the real-time performance of Ethernet traffic will be investigated. An existing TSN testbed is used and scripts and test applications are implemented to evaluate the influence of different parameters and hardware configurations on latency and jitter.

External Partner: Intel Munich

• We are a member of the one6G Association! More info: https://one6g.org

• 11 - 14.03.24 ICIN 2024 Congress in Paris - Professor Kassler takes part
  Our research team is growing! We are pleased to announce that Khalid Ali and Özgur Ozan are now part of the research team.

• Our two Swedish doctoral students Phil Aupke and Amal Nammouchi are doing research with us at TH DEG for a month

• Professor Kassler takes part in the RNDM international Workshop on Resilient Networks Design and Modeling