Thesis projects cover a broad range of subjects. Students work closely with our researchers to gain hands-on experience in current research.
Open Topics in the Area of Software Defined Radio, Cellular Networks, High-Altitude Platforms, Meshed Networks.
Adaptive Resource Management in O-RAN
Supervisor: Binh V. Duong
This thesis focuses on designing and evaluating a machine-learning framework for intelligent network state estimation and adaptive resource control in Open RAN (O-RAN). The student will collect multimodal communication data, develop ML models to infer real-time network conditions, and integrate these models into the Near-RT RIC for closed-loop control. The work includes end-to-end validation on a testbed and adherence to O-RAN specifications via the E2 interface. The goal is to improve the accuracy, interpretability, and responsiveness of RAN control within the O-RAN ecosystem.
A Study on DECT NR+ as Access Network for 3GPP 5G/6G
Supervisor: Stefan Senk
DECT NR+ (New Radio Plus) is the first and only non-cellular 5G technology approved by the ITU, designed to meet the demanding requirements of industrial and enterprise communications. It brings 5G-grade performance — including ultra-reliable low-latency communication (URLLC) and massive machine-type communication (mMTC) — to private, license-exempt networks.
Built on proven cellular principles, NR+ should deliver low latency (down to 1 ms), high reliability, and flexible network topologies such as point-to-point, star, and mesh. This makes it suitable for diverse applications including industrial automation, smart metering, building management, and professional audio.
The technology emphasizes affordability, scalability, and ease of deployment, featuring self-organizing and self-healing mesh networks, automatic joining, and minimal maintenance. Operating in a dedicated spectrum below 6 GHz, it avoids interference common in ISM bands.
Technically, NR+ employs OFDM modulation, HARQ and Turbo coding for reliability, and supports IPv6 for secure, direct IP connectivity without complex gateways. With adaptive transmission power and coexistence with classic DECT systems, DECT NR+ offers a future-proof, global standard for robust wireless communication in professional and industrial environments.
Energy Consumption Optimization in Wireless Sensor Network
Supervisor: Juan Cabrera, Isikcan Yilmaz
The amount of sensor data gathered for various automated operations, such as building temperature control, automated factory operations, automated lighting control, etc. is increasing exponentially. In many instances this data is gathered by a network of sensor nodes that are battery powered, which is why optimization efforts in terms of energy consumption are crucial.
Reception and transmission of Radio Frequency signals are among the most energy consuming tasks done by a wireless sensor node. One way of optimizing for energy consumption could be the lowering of sensor sampling rates, however the application at hand may or may not allow for this. An enhancement to this could be the interpolation of the missing samples, however this can mean extra compute time, and the interpolated values may have errors. The main goal here is to evaluate, with real hardware, the tradeoff between the accuracy and the energy consumption of the sensor nodes in the network.
Software Defined Radio (SDR) for Computing Network Nodes
Supervisor: Juan Cabrera
Radio communication systems used to be implemented mostly with hardware solutions. It was difficult to make quick changes since they involved hardware replacement, e.g., filters, and modulators, and demodulators. This changed in the later years with Software Defined Radios SDR. This technology allows developers to build up complete radio communication stacks with off-the-shelf hardware. At the Comnets Chair, we have access to a testbed of multiple SDR devices that allows us to test a plethora of communication schemes for communication networks. We are interested in deploying and evaluating the performance of multiple communication schemes for wireless network nodes that perform computation, storage, and transport of information. These types of nodes tend to be at the edge of the network, and using them for computation and storage can drastically reduce latency and increase the reliability of communication.
Motivation links:
https://youtu.be/DEeOFE_DreU
https://youtu.be/1bgC3AjCnA4
https://youtu.be/xQVm-YTKR9s
At a glance
Type: Student thesis; Diploma/Master thesis (with task extention)
Starting Time: immediate
Requirements: Python, basic knowledge of
digital communication (digital modulation,
CDMA, OFDM)
Study of cloud radio access network and edge computing in High-Altitude Platforms (HAP) and nanosatellites
Supervisor: Riccardo Bassoli
High-Altitude Platforms (HAP) and nanosatellites represent the new way to provide connectivity and computing in remote/tactical areas, where no infrastructure is available. This can also become a useful solution in urban areas in case of natural disasters (e.g. earthquakes). However, HAP-based or satellite-based cloud radio access network open various fundamental challenges in edge computing and network virtualisation. The thesis’ work will be devoted to study, analyse and test (via simulation) specific characteristics of these systems. The details of the thesis’ topic and the level of the targets will be adapted according to the student’s preferences, motivation and talent.
Motivation links:
https://ieeexplore.ieee.org/document/9046846
https://ieeexplore.ieee.org/document/9172316
https://ieeexplore.ieee.org/document/9316545
At a glance
Type: Diploma/Master Thesis
Starting Time: immediate
Requirements: background on
telecommunications, programming
and ComNets2
Study and Development of eSIM Infrastructure for IoT Devices in 5G
Supervisor: Thomas Höschele
While GMSA has successfully specified and rolled out the SGP.22 standard, which enables remote provisioning of eSIM for smartphones; the final version of the SGP.32 standard is still in development. The SGP.32 standardizes the remote provisioning of IoT devices, which is very important to support cellular use cases for industrial scenarios like warehouses and manufacturing. As these IoT devices will have limit input capabilities the provisioning process faces challenges. The thesis aims to integrate, qualify and extend open source tools for an architecture to provisioning eSIM on 5G-Devices and develop a way to provision eSIM in the IoT world. The details of the thesis’ topic and the level of the targets will be adapted according to the student’s preferences, motivation and talent.
Motivation links:
https://www.gi-de.com/de/spotlight/digital-security/neue-spezifikation-zur-remote-sim-bereitstellung
https://iot-analytics.com/role-of-esim-for-iot-better-security-simplified-roaming-easier-provisioning/
At a glance
Type: Student/Diploma/Master Thesis
Starting Time: immediate
Requirements: background on
telecommunications, programming
and ComNets2
Simplified IMS in Private 5G Networks
Supervisor: Thomas Höschele
Voice communication is a major feature of 5G networks. Within the 5G-Core the IP Multimedia Subsystem (IMS) is responsible for voice calls and voice transfer (Voice over NR). A typical IMS is a heavily large function, even comparable to the 5G core itself. Particularly for private 5G networks which require a lean core installation this would be too much. A solution is a simplified IMS, only consisting of the minimal necessary functionalities, procedures and services to enable voice calls a connection to a SIP client. The thesis aims to develop, implement and test a simplified IMS in a private 5G network enabling native voice calls. The details of the thesis’ topic and the level of the targets will be adapted according to the student’s preferences, motivation and talent.
Motivation links:
https://en.wikipedia.org/wiki/IP_Multimedia_Subsystem
https://www.researchgate.net/publication/285648601_Cloudifying_the_3GPP_IP_Multimedia_Subsystem_for_4G_and_Beyond_A_Survey
https://portal.3gpp.org/desktopmodules/Specifications/SpecificationDetails.aspx?specificationId=821
At a glance
Type: Student/Diploma/Master Thesis
Starting Time: immediate
Requirements: background on
telecommunications, programming
and ComNets2
Get in touch with Tung Doan. He will help you find a supervisor who will work with you on a tailored topic.
Time-sensitive Networking (TSN) describes a set of IEEE standards to ensure deterministic transmissions over Ethernet networks. It is especially of interest for industrial applications, such as robot control or process automation where both (ultra) low latencies and reliability must be guaranteed. Recently, the integration of 5G and TSN started. In order to leverage 5G communication for industrial applications, TSN is targeted as an enabling solution. In ongoing research projects (at the chair, such as TICCTEC or stic5G), the integration of 5G as a virtual TSN bridge is under investigation. Part of the tasks is to extend the 5G core network with network functions (translator functions), to ensure correct QoS management within the 5G system, to research direct device-to-device (D2D) communication, and further associated fields of interest, such as implementing new scheduling algorithms or multi-path communication for enhancing the reliability.
A Study on DECT NR+ as Access Network for 3GPP 5G/6G
Supervisor: Stefan Senk
DECT NR+ (New Radio Plus) is the first and only non-cellular 5G technology approved by the ITU, designed to meet the demanding requirements of industrial and enterprise communications. It brings 5G-grade performance — including ultra-reliable low-latency communication (URLLC) and massive machine-type communication (mMTC) — to private, license-exempt networks.
Built on proven cellular principles, NR+ should deliver low latency (down to 1 ms), high reliability, and flexible network topologies such as point-to-point, star, and mesh. This makes it suitable for diverse applications including industrial automation, smart metering, building management, and professional audio.
The technology emphasizes affordability, scalability, and ease of deployment, featuring self-organizing and self-healing mesh networks, automatic joining, and minimal maintenance. Operating in a dedicated spectrum below 6 GHz, it avoids interference common in ISM bands.
Technically, NR+ employs OFDM modulation, HARQ and Turbo coding for reliability, and supports IPv6 for secure, direct IP connectivity without complex gateways. With adaptive transmission power and coexistence with classic DECT systems, DECT NR+ offers a future-proof, global standard for robust wireless communication in professional and industrial environments.
Challenges and Opportunities in Using the Age of Information for the Tactile Internet
Supervisor: Hosein Kangvara Nazari, Tobias Scheinert
The thesis aims to explore how Linux-based TSN features can be used to improve the temporal relevance of transmitted data, with a focus on delivering fresh information. Age of Information (AoI) is considered an alternative performance metric to latency, aiming to assess whether it better captures the quality of experience in real-time systems. Furthermore, the thesis will evaluate whether existing TSN tools and implementations in Linux are sufficient for the demands of Tactile Internet use cases or if additional mechanisms and optimizations are required.
Evaluation of DDS-TSN with Götting KATE AGV
Supervisor: Stefan Senk
Automated Guided Vehicles (AGVs) are used in various ways in the industry such as production, manufacturing, and other domains. Typically, an AGV is controlled via a fleet management system that localizes an AGV with markings using sensors from an AGV. Another way is to remotely control an AGV from a terminal through direct user input. Depending on the implementation, the requirements for the communication channel between control unit and AGV can be more or less demanding. Commonly, closed loop control systems require almost deterministic communication behavior. Time-sensitive Networking (TSN) is a set of IEEE standards that aims to make wired Ethernet communication deterministic.
More and more robots rely on the Robot Operating System (ROS). “ROS is a set of software libraries and tools that helps to build robot applications.” [ros.org] In the latest version, ROS2, the Data Distribution Service (DDS) is used as a middleware communication protocol for exchanging messages in a publish and subscribe transport architecture.
Enhancing Reliability in Private 5G Communication via TSN FRER Vector Recovery
Supervisor: Hosein Kangvara Nazari, How-Hung Liu
With the convergence of TSN and 5G networks, new challenges arise from wireless delay, jitter, and packet reordering. This work aims to study, implement, and evaluate the VRA within the INET simulation framework and validate its performance using data from a 5G-TSN testbed. The goal is to assess VRA’s reliability under hybrid wired–wireless conditions and explore optimizations to improve its robustness and efficiency.
Investigation of Intel Time-coordinated Computing for TSN
Supervisor: Hosein Kangvara Nazari, How-Hung Liu
Time-sensitive applications are characterized by stringent timing requirements, where even minor deviations such as jitter or latency exceeding a specified threshold can degrade communication quality. Such applications are increasingly emerging in domains such as the Tactile Internet and industrial automation. To address these challenges, Time-Sensitive Networking (TSN) provides deterministic communication through preplanned and time-triggered packet transmissions.
Investigation of 5G E2 Interface for 5G-TSN using OAI Box and FlexRIC
Supervisor: Tobias Scheinert, Ricardo Pousa
Industrial and real-time applications demand communication systems with guaranteed latency, reliability, and time synchronization. While 5G introduced URLLC features to address such requirements, achieving the deterministic behavior of Time-Sensitive Networking (TSN) over 5G remains a challenge.
A key factor lies in the Radio Access Network (RAN), which strongly impacts latency, jitter, and reliability in wireless transmission. To meet TSN requirements, the RAN must be dynamically configurable and responsive to changing traffic conditions.
The O-RAN architecture provides a promising solution by enabling intelligent RAN control through the E2 interface. This allows near-real-time adaptation of radio resources via eXtended applications (xApps), supporting flexible scheduling and precise control essential for TSN traffic.
Investigation of Time-sensitive Networking for 5G Fronthaul using OAI Box
Supervisor: How-Hung Liu, Tobias Scheinert
OpenAirInterface (OAI) is an open-source software platform that implements 5G cellular network functions, enabling end-to-end experimentation from radio access to core network. The research question investigates how Time-Sensitive Networking (IEEE 802.1CM) can be integrated into an OAI-based 5G fronthaul to achieve deterministic ultra-low-latency RU–DU communication, and how the OAI Box and OAI software can be practically configured or extended to realize a TSN-enabled fronthaul.
Investigation of Selected 5G-TSN Translator Features using OAI Box
Supervisor: Stefan Senk, Hosein Kangvara Nazari, Tobias Scheinert
This thesis addresses the challenge of integrating 5G/6G with Time-Sensitive Networking by investigating extensions of the User Plane Function (UPF) with deterministic features and the incorporation of TSN mechanisms such as traffic scheduling and precise synchronization on both the network and device side. Building on the OAIBOX platform based on OpenAirInterface, the work aims to design and experimentally validate an integration framework in which 5G and TSN operate seamlessly, enabling communication with guaranteed latency, reliability, and timing accuracy. The outcome of this research contributes to advancing industrial communication systems capable of meeting the stringent requirements of future connected production and automation.
Comparative Analysis of Deterministic Communication Protocols
Supervisor: Stefan Senk, Hosein Kangvara Nazari, How-Hung Liu, Tobias Scheinert
This task revolves around addressing the growing demands of emerging industrial applications, which increasingly require deterministic communication protocols to ensure guaranteed performance levels. Industries often encounter situations where different deterministic communication techniques are needed, each suited for specific scenarios. The primary objective of this work is to conduct a detailed comparative analysis of various deterministic communication protocols, with a focus on answering critical questions related to their characteristics and capabilities.
Analyzing TSN’s Frame Replication and Elimination Protocol for Enhanced Reliability
Supervisor: Stefan Senk, Hosein Kangvara Nazari, How-Hung Liu, Tobias Scheinert
Time-Sensitive Networking (TSN) is a set of IEEE standards to achieve deterministic communication over Ethernet networks. This is especially relevant for industrial domains, such as medical, banking, avionics, or automotive. The communication is characterized by strict requirements on delay, packet delay variations, and packet loss. In order to achieve certain guarantees, the TSN standards provide different algorithms, metrics, and tools.
One of the key TSN protocols to ensure reliability is Frame Replication and Elimination. The primary objective of this undertaking is to delve into the inner workings of this protocol and assess its impact on enhancing the reliability of communication.
Quality of Service Mapping for 5G-TSN Integration
Supervisor: Stefan Senk, Hosein Kangvara Nazari, How-Hung Liu, Tobias Scheinert
Time-Sensitive Networking (TSN) is a set of IEEE standards to achieve deterministic communication over Ethernet networks. This is especially relevant for industrial domains, such as medical, banking, avionics, or automotive. The communication is characterized by strict requirements on delay, packet delay variations, and packet loss. In order to achieve certain guarantees, the TSN standards provide different algorithms, metrics, and tools.
On the other hand, the 5G System (5GS) is itself very complex. With enhancements for Ultra-Reliable Low-Latency Communication (URLLC) in Release 15 (R15), the 5GS paved the path for latency-aware communication. The 5GS already provides possibilities to treat packets similarly to in TSN.
The integration of 5G and Time-Sensitive Networking (TSN) is a prominent and evolving area of research, facilitating deterministic communication and extending it over the air. One critical aspect of this integration is the translation of TSN traffic priorities into 5G’s Quality of Service Indicator (5QI) framework. This task aims to create a QoS mapping table for seamless translation of QoS between TSN and 5G, focusing on industrial automation.
The Evaluation of Determinism in Different TSN Configuration Models
Supervisor: Stefan Senk, Hosein Kangvara Nazari, How-Hung Liu, Tobias Scheinert
TSN, a set of standards designed to ensure deterministic communication over Ethernet cables and wired connections, is particularly valuable for applications with strict deadlines and the need for guaranteed performance. The objective of this task is to conduct a comprehensive investigation into the impact of configuration models on the determinism of Time-Sensitive Networking (TSN). TSN offers various configuration models to achieve determinism over wired links. This task comprises two main steps:
• Comparative Analysis of TSN Configuration Models: The first step involves an in-depth exploration of the distinct TSN configuration models specified by IEEE standards. The student will highlight and examine the significant differences between these models.
• Evaluation of Determinism Levels: In the second step, the student will assess the degree of determinism provided by each TSN configuration model. This assessment will be conducted using the OMNeT++ simulator, allowing for a quantitative analysis of determinism levels.
Implementation of Asynchronous Traffic Shaper in Linux Kernel Space using High-Performance Data Processing
Supervisors: Stefan Senk, Hosein Kangvara Nazari, How-Hung Liu
The IEEE standard 802.1Qcr provides an Asynchronous Traffic Shaper (ATS). In contrast to the Time-aware Shaper (TAS) defined by IEEE 802.1Qbv, it does not rely on tight time synchronization. Unfortunately, to best of our knowledge, there is no hardware implementation. Therefore, the thesis should investigate the performance of the ATS under real world conditions.
Furthermore, the implementation should leverage eBPF in conjunction with Express Data Path (XDP) to build on top of the existing Linux network features. eBPF and XDP offer great performance and possibilities for packet processing on x86 hardware. As a result, the thesis should provide a comparison with a Commercial-of-the-Shelf (COTS) TAS-capable switch.
There is usually the possibility, that project work topics can be extended for bachelor/master/diploma thesis. If you don’t find a topic for yourself, we are also flexible and open to prepare a new topic together with you! The following ComNets chair members are working on Time-Sensitive Networking: Stefan Senk, Hosein Kangvara Nazari, How-Hung Liu and Tobias Scheinert. Come and talk to us about what you would like to research!
Bridging the Gap: 5G-TSN Integration for Industrial Robotic Communication Proceedings Article
In: European Wireless (EW), Rome, Italy, 2023.
Incremental Joint Scheduling and Routing for 5G-TSN Integration Proceedings Article
In: European Wireless (EW), Rome, Italy, 2023.
Flexible Measurement Testbed for Evaluating Time-Sensitive Networking in Industrial Automation Applications Proceedings Article
In: 2022 IEEE 8th International Conference on Network Softwarization (NetSoft) (NetSoft 2022), Milan, Italy, 2022.
TSN-FlexTest: Flexible TSN Measurement Testbed (Extended Version) Journal Article
In: 2022, (arXiv pre-print).
Healing Hands: The Tactile Internet in Future Tele-Healthcare Journal Article
In: Sensors, vol. 22, no. 4, 2022, ISSN: 1424-8220.
Open Topics in the Area of Software Defined Networks (SDN) & Network Functions Virtualization (NFV).
Improving Performance of Service Mesh for Cloud Native Applications
Supervisor: Tung Doan
Unlike traditional monolithic applications, cloud-native applications are the
collection of small and independent services, which are so-called microservices. As cloud-native applications have gained tremendous interest in recent years, many cloud vendors such as Google Cloud and Amazon Web Service already provided cloud platforms for cloud-native applications. Service meshes have been considered as a de facto communication subtrate for cloud-native applications. Specifcially, each service in a cloud-native application communicate to each other via a software proxy, called sidecar. A sidecar intercepts cloud traffic reaching a service and thus provides various control functions such as security and traffic management. However, as each sidecar is co-located with each service, this design introduces overhead (e.g., increasing latency and lower throughput) for cloud-native applications, especially for applications that include a significant number of services. This work aims to improve the performance of service mesh for cloud-native applications.
At a glance
Type: Student/Diploma/Master Thesis
Starting Time: immediate
Requirements: Basic knowledge of computer networks, basic knowledge of Linux, programming languages: C, Python, Go (preferably)
Investigating State Transfer For Programmable Network Devices
Supervisor: Tung Doan
Applications are classified into stateless and stateful. Stateless applications do not require acknowledging application states (i.e., historical processing values) to handle users‘ requests. Meanwhile, stateful applications rely on application states for proper processing. More importantly, application states are used to provide the fault tolerance and scalability of applications that require state transfer between applications. For instance, this demo shows how we used application states.
Existing studies have been proposed to utilize the states of applications deployed on general-purpose servers. Due to the emerge of latency-sensitive use cases such as autonomous driving and robots, there is a possibly to deploy applications directly on programmable network devices such as Tofino switches, thus allowing applications to process users‘ requests at line rate and consequently reducing latency. While ensuring line-rate processing for applications, programmable network devices have to maintain application states, thus making fault tolerance and scalability challenging. This work aims to investigate a solution for the state transfer of applications deployed on programmable devices.
At a glance
Type: Student/Diploma/Master Thesis
Starting Time: immediate
Requirements: Basic knowledge of computer networks, basic knowledge of Linux, programming languages: C and Python
Improving The Performance of Ractive Programable Network Devices
Supervisor: Tung Doan
Traditionally, applications are deployed on general-purpose servers. Recently,
modern network devices such as Tofino switches are able to deploy applications with line-rate processing. However, applications on programable network devices have the abilities to rapidly react to the unexpected network behaviors such as link failure or traffic overload, thus so-called reactive programmable network devices. This requirement leads to the involvement of network control with low latency.
Recent studies such as Mantis have been proposed to employ switch’s CPU (Central Processing Unit) to provide control programs that reactive to unexpected network behaviors. To ensure the line-rate processing of applications, the goal of this work is to futher reduce the reactive time of control programs.
At a glance
Type: Student/Diploma/Master Thesis
Starting Time: immediate
Requirements: Basic knowledge of computer networks, basic knowledge of Linux, programming languages: C and Python
Leveraging Network Programmability to Improve The Performance Of Network Coding
Supervisor: Tung Doan
Network coding has demonstrated its great potential to improve the performance in various scenarios such as unreliable communication networks and distributed storage systems. Due to the potential of network coding, recent years witness tremedous variations (e.g., Fulcrum coding). NCKernel, network coding support in Linux kernel, has been proposed to prove the practicality of network coding.
However, NCKernel is mainly used for handling network coding functions. The network support has not been considered yet in NCKernel. For instance, forwarding function is needed to allow recoder to receive the packets from encoder and then forward them to decoder. To tackle this issue, NCkernel relies on NCnet that is a seperate software entity dedicated to provide the network support for NCKernel. The use of NCnet in NCKernel introduces latency overhead. Therefore, the goal of this work is to improve the performance of NCKernel. Particularly, the forwarding function (i.e., similar to a layer-2 switch) will be directly implemented in NCKernel. The forwarding function should be transparent to sender and receiver, i.e., the sender only needs to know the IP address of the receiver and the network coding functions will be implemented in the network.
At a glance
Type: Student/Diploma/Master Thesis
Starting Time: immediate
Requirements: Basic knowledge of computer networks, basic knowledge of Linux, programming languages: C and Python
Enhancing Traffic Classification with Programmable Data Plane Switches
Supervisor: Mingyu Ma, Yushan Yang
Machine learning has become an essential tool for solving networking challenges, such as traffic classification, anomaly detection, and network configuration. By leveraging machine learning algorithms, networks can dynamically manage data traffic and optimize overall performance. Recently, in-network machine learning solutions have emerged, driven by the evolution of network devices that are both high- performance and programmable. Technologies like Switch-ASICs, network interface cards (NICs), and FPGA-based network devices now utilize P4, a domain-specific language that enables the definition and customization of network protocols and processing functions directly within the data plane. This level of programmability opensnew possibilities for offloading traffic classification tasks to network hardware. In this project, students will identify an appropriate machine learning algorithm suited for network communication scenarios and deploy it on a Switch-ASIC (Tofino Switch) testbed. The goal is to enable real-time detection and classification of network traffic conditions directly within the switch hardware.
At a glance
Type: Student/Diploma/Master Thesis
Starting Time: immediate
Requirements: Familiarity with Linux, programming skills in C and Python
DU Lossless Migration
Supervisor: Ricardo Pousa
With the advent of 5G, a continuous effort was seen to virtualize the functions supporting mobile communications. With an architecture based on NFV (Network Function Virtualization), it becomes possible to separate and freely allocate resources according to one’s network’s necessities. Coupled with the ever-increasing throughput requirements for new applications, edge computing presents itself as an attractive solution. In line with these trends, RAN (Radio Access Networks) are being rethought and partitioned into CU (Centralized Units) and DU (Distributed Units). This change can lead to lower bandwidth requirements, and give rise to in-network processing opportunities. The objective of this task is to implement an (Intra or inter-computer) migration procedure, that makes use of the containerized nature of next-generation networks while mitigating loss of functionality.
At a glance
Type: Student/Diploma/Master Thesis
Starting Time: immediate
Requirements: Basic knowledge of Computer Networks, Linux, Docker, programming languages: C and Python, markup languages: YAML
Get in touch with Tung Doan. He will help you find a supervisor who will work with you on a tailored topic.
Secure Data Transmission in Device-to-Device (D2D) Communication via HMAC
Supervisor: Hosein K. Nazari, Mehmet Akif Kurt
Network coding has emerged as a promising technique for optimizing network communication, particularly over lossy links. It is especially effective in wireless networks, where intermediate nodes can perform recoding to enhance reliability and throughput. A widely adopted approach is Random Linear Network Coding (RLNC). In RLNC, packets are typically equipped with CRC checksums to detect errors caused by intentional pollution attacks or unintentional channel noise. However, once a packet is corrupted, new CRC checksums cannot be calculated for the recoded packets since access to the original packet is lost. A potential solution is to replace CRC checksums with Homomorphic Message Authentication Code (HMAC) tags, enabling the combination of polluted packets without the need for checksum re-calculation, thus allowing these packets to participate in the recoding process. This study aims to evaluate the performance of HMAC tags in detecting both intentional and unintentional packet modifications and to analyze the trade-offs involved in using tags instead of CRC checksums.
Early HMAC Verification in the Linux Networking Stack Using
XDP and AF_XDP
Supervisor: Hosein K. Nazari, Mehmet Akif Kurt
Network coding holds great promise for resilient communication, but it also raises security concerns. HMAC has been recognized as a suitable technique for ensuring integrity in such systems, though its computational overhead can hinder low-latency applications.
The main focus of this work is the implementation of HMAC verification at the earliest possible stage in the Linux networking stack. This includes exploring eBPF/XDP for in-kernel verification and AF_XDP for user-space packet processing. The objective is to investigate whether performing HMAC verification closer to the data path reduces the delay compared to conventional implementations (e.g., AF_PACKET or AF_INET).
Enhancing Security Mechanisms for Low-Latency
Communication in Network Coding
Supervisor: Hosein K. Nazari, Mehmet Akif Kurt
Network coding holds great promise for resilient communication, but security remains a critical concern. While Homomorphic MAC (HMAC) is recognized as an effective technique for securing random linear network coding, its inherent complexity and associated overhead can degrade system performance, rendering it unsuitable for low-latency communication applications.
This task focuses on conducting an extensive literature study on HMAC, comparing different methods based on computational complexity, overhead, and security aspects. The primary objective is to propose strategies and approaches that reduce the complexity of HMAC procedures while taking into account the necessary trade-offs between complexity reduction and security measures. The ultimate goal is to enhance communication latency in network coding systems.
Hybrid Packet Recovery and HMAC Verification in Network
Coding Systems
Supervisor: Hosein K. Nazari, Mehmet Akif Kurt
Packets in network coding can become corrupted either through deliberate adversarial pollution or accidental bit flips caused by the communication channel. Such corrupted packets jeopardize the decoding process and may significantly reduce the reliability of the overall system.
The main focus of this work is to investigate strategies that combine packet recovery mechanisms with HMAC-based verification. The objective is to detect corrupted packets early and, where possible, recover them without retransmission. This includes evaluating approaches for mixing recovery strategies with lightweight cryptographic checks, and comparing them against conventional mechanisms such as ARQ (Automatic Repeat reQuest). The task includes measuring the speed and effectiveness of detection and recovery, assessing the impact on latency and throughput, and identifying the trade-offs between additional overhead and reliability gains.
Get in touch with Tung Doan. He will help you find a supervisor who will work with you on a tailored topic.
Adaptive Resource Management in O-RAN
Supervisor: Binh V. Duong
This thesis focuses on designing and evaluating a machine-learning framework for intelligent network state estimation and adaptive resource control in Open RAN (O-RAN). The student will collect multimodal communication data, develop ML models to infer real-time network conditions, and integrate these models into the Near-RT RIC for closed-loop control. The work includes end-to-end validation on a testbed and adherence to O-RAN specifications via the E2 interface. The goal is to improve the accuracy, interpretability, and responsiveness of RAN control within the O-RAN ecosystem.
Post–Training and Reinforcement Learning for Large Language
Models
Supervisor: Fatima Rani, Juan Cabrera
Large Language Models (LLMs) such as GPT and Llama have transformed natural language understanding and generation. However, pretrained LLMs often display issues like hallucinations, inconsistent reasoning, and unsafe responses, making them unsuitable for direct deployment in production environments. Post–training including fine–tuning and reinforcement learning with human feedback (RLHF) has emerged as a crucial step to align model behavior with human values and specific real–world tasks. This thesis aims to systematically study and apply these post–training techniques to enhance reasoning, reliability, and safety in LLMs, bridging the gap between research prototypes and production–ready AI systems.
Motivation links:
https://arxiv.org/abs/2507.21931
https://developers.redhat.com/articles/2025/11/04/post-training-methods-language-models
https://huggingface.co/blog/royswastik/reinforcement-learning-for-llms
At a glance
Type: Diploma/Master Thesis
Starting Time: immediate
Requirements: Profound knowledge in Python Programming, ML frameworks (PyTorch TensorFlow), Basic knowledge of LLM architectures and fine-tuning methods.
Energy–Efficient Deployment of Large Language Models (LLMs): A Systematic Benchmarking Study for Green AI
Supervisor: Fatima Rani, Pit Hofmann
Large Language Models (LLMs) have become central to modern AI systems, driving applications in natural language understanding, reasoning, and generation. However, their immense computational and energy demands raise sustainability concerns. The field of Green AI seeks to optimize the trade–off between performance and environmental cost by valuating efficiency across hardware and algorithmic dimensions.
This thesis investigates the energy efficiency, latency, and accuracy trade–offs of deploying LLMs across heterogeneous hardware platforms (e.g., GPUs, NPUs, edge accelerators). By systematically benchmarking transformer–based architectures under varied inference configurations, the project contributes empirical evidence toward sustainable deployment strategies for LLMs and provides
insights into the environmental impact of large–scale language models.
Motivation links:
https://www.youtube.com/watch?v=aKBALONUAyY
https://www.youtube.com/watch?v=sTz2tXG1T0c
https://www.youtube.com/watch?v=l0BdmevNhuc
https://www.youtube.com/watch?v=7xTGNNLPyMI
https://pytorch.org/blog/optimize-llms/
At a glance
Type: Diploma/Master Thesis
Starting Time: immediate
Requirements: Profound knowledge in Python Programming, ML frameworks (PyTorch TensorFlow), Basic knowledge of LLM training/inference pipelines & optimization techniques
Neural Network–Based Signal Detection in Noisy Channels: Performance and Robustness Evaluation
Supervisor: Pit Hofmann
In this project, a (feedforward) NN will be designed, trained, and evaluated for
signal detection in a simulated communication system. The focus is on
understanding the robustness and reliability of such NN-based detectors under
realistic constraints, including limited training data, channel noise variation, and
neuron-level faults.
Energy Profiling & Carbon-Aware AI Model Comparison
Supervisor: Sifat Rezwan
The growing deployment of large-scale machine learning models has led to rapidly increasing energy consumption and environmental impact. Green AI research focuses on optimizing models and systems not only for accuracy but also for efficiency and sustainability. This project investigates the energy usage and carbon footprint of different neural network architectures and precision settings during inference on GPUs. Students will perform systematic energy profiling using available measurement tools and convert these measurements into carbon-aware metrics (e.g., CO2 per inference). The results will enable fair comparisons between models under a sustainability perspective and foster awareness of energy-efficient AI design.
Benchmarking Neuromorphic and Conventional AI for
Energy–Efficient Inference
Supervisor: Sifat Rezwan, Juan Cabrera
Neuromorphic computing, particularly through spiking neural networks (SNNs), promises significant energy savings compared to conventional deep learning. However, systematic benchmarks of SNNs versus traditional neural networks on real hardware accelerators are scarce. With access to heterogeneous hardware (Axelera, Hailo, Jetson Nano, Asus, GPUs), this project investigates whether neuromorphic models provide measurable benefits in accuracy, energy, latency, and robustness, contributing to the evidence base for Green AI.
At a glance
Type: Master Thesis
Starting Time: immediate
Requirements: Python (must), ML frameworks (PyTorch/TensorFlow), Familiarity with computer vision architectures, Familiarity with spiking neural network libraries (optional), Basic knowledge of Linux and hardware benchmarking.
Energy–Accuracy Trade–offs in Edge AI Accelerators for
Real–Time Computer Vision
Supervisor: Sifat Rezwan, Juan Cabrera
Deploying AI workloads at the edge demands a balance between model accuracy, inference latency, and energy consumption. Green AI emphasizes minimizing the environmental and energy footprint of AI while retaining reliable performance. With diverse hardware platforms available (Axelera AI accelerator, Hailo AI accelerator, Nvidia Jetson Nano, Asus AI accelerator, and Nvidia GPUs), this project explores how computer vision models behave under varying constraints, providing a systematic study of the energy–accuracy trade–offs across hardware.
At a glance
Type: Master Thesis
Starting Time: immediate
Requirements: Python (must), ML frameworks (PyTorch/TensorFlow), Familiarity with computer vision architectures, Familiarity with spiking neural network libraries (optional), Basic knowledge of Linux and hardware benchmarking.
Motion Capturing and Data Analysis in Physiotherapy
Supervisor: Giang Nguyen
The vision of the spin-off project Veiio in Dresden is to develop an intelligent suit with fully integrated sensors and actuators that uses motion capture to provide real-time vibrotactile feedback to the wearer.
We offer a diploma thesis project in Motion Capture (MoCap) Data Analysis based on Inertial measurement unit (IMU) sensors. In this project, you should develop a prototypical system that recognizes selected movement exercises and that evaluates the movement execution in real-time. Included tasks are also: acquiring the data with provided hardware, testing, and comparing different approaches for motion classification and evaluation, providing a simple interface for examining the algorithms.
You should bring:
We provide:
That sounds exciting?
Then please contact us: info@veiio.de or Giang Nguyen
At a glance
Type: Student/Diploma/Master Thesis
Starting Time: immediate
Requirements: Profound knowledge in Python programming, experience with approaches from machine learning (e.g., scikit-learn, PyTorch)
Get in touch with Tung Doan. He will help you find a supervisor who will work with you on a tailored topic.
Simulating quantum-mechanical properties of quantum communication networks is not an easy task and every existing simulator is based on some initial assumptions to allow classical modelling of quantum behaviours. Each quantum simulator has pros and cons to be analysed in order to use it correctly, to model a specific aspect (or a set of aspects) of future quantum communication networks. This is fundamental to be able to correctly interpret the results obtained after simulations. The details of the thesis’ topic and the level of the targets will be adapted according to the student’s preferences, motivation and talent.
Predictive Resilience Curves and Rerouting for Threshold-Safe Quantum Networking
Supervisor: Vignesh Raman, Abdelkrim Menina
Develop theory and methods to predict dips in a quantum network’s resilience curve — that is, performance as a function of stress or fault level — and trigger reactive or proactive rerouting so that performance remains above a specified robustness threshold.
Students will formalize resilience-curve dynamics under noise and failure processes, design sequential dip detectors based on derivative or curvature signals and change-point tests with false-alarm guarantees, build short-horizon predictors with error bounds, and connect predictions to routing and control policies with provable safety-style guarantees.
Resilience Metrics for Quantum Network Performance: A Comparative Study via Resilience Curves
Supervisor: Vignesh Raman
Students will investigate, formalize, and compare resilience metrics for quantum networks by constructing and analyzing resilience curves—performance vs. stress/fault level—for a range of operating conditions and failure models. Performance may include end-to-end distributed fidelity, task completion latency for entanglement distribution, throughput/secret-key rate, availability, and fairness across flows.
Stress can include link/node failures, stochastic loss, depolarizing/dephasing noise, queue contention, hardware drift, and adversarial disruptions. The goal is a principled, apples-to-apples comparison of metrics that illuminates trade-offs (e.g., fidelity vs. latency) and reveals which metrics are most decision-useful for design and operations.
Quantum Synchronisation Technologies for Communication Networks: Theoretical Exploration and TCEP-Based Simulation
Supervisor: Swaraj Shekhar Nande
Synchronisation is fundamental to modern communication networks, supporting accurate timestamping, low-latency transmission, and coordinated operations across distributed nodes. In classical systems, synchronisation relies on protocols such as the Network Time Protocol (NTP), the Precision Time Protocol (PTP), and White Rabbit, which achieve nanosecond-level precision under favourable conditions. However, these methods remain constrained by propagation delay asymmetry, network jitter, and hardware timing noise, limiting their scalability and robustness for future 6G and quantum communication infrastructures.
Co-opetitive Game Theory for Resilience in Quantum Networks
Supervisor: Vignesh Raman
Study quantum network resilience when multiple administrative domains (or services) both compete for scarce quantum resources (repeaters, memory time, purification time) and cooperate to maintain minimum service levels (fidelity/availability) during faults.
Develop a co-opetitive model that blends non-cooperative equilibria (pricing/routing/priority) with cooperative coalitions (mutual aid, shared standby capacity). Prove existence, efficiency, and stability guarantees; characterize when cooperation is self-enforcing; and design cost/reward splits that sustain resilient operation.
Design and Implementation of a Distributed Quantum Key Management System for Three-Node QKD Systems
Supervisor: Riccardo Bassoli
Quantum Key Distribution (QKD) is a core technology in the field of quantum communication, enabling theoretically unconditionally secure key distribution. However, with the rapid development of QKD technologies, quantum key management has become a critical research topic, especially in multi-node networks where key management faces challenges distinct from those in classical communication systems. This project aims to design and implement a distributed quantum key management system for three-node QKD systems, providing technical support for the future construction of quantum communication networks
At a glance
Type: Diploma/Master Thesis
Starting Time: immediate
Requirements: background on classical computer science and programming, ComNets2 and ComNets3
Design and Implementation of a Three_Node QKD Communication Framework
Supervisor: Riccardo Bassoli
Quantum Key Distribution (QKD) is a core technology in the field of quantum communication, enabling theoretically unconditionally secure key distribution. With the rapid development of quantum communication technologies, the design and implementation of QKD systems have become a research hotspot. This project aims to design and implement a three-node QKD communication framework and develop a fully functional frontend UI for monitoring network status, deploying tasks, and managing quantum keys.
At a glance
Type: Diploma/Master Thesis
Starting Time: immediate
Requirements: background on classical computer science and programming, ComNets2 and ComNets3
Applications of FPGA in Quantum Key Distribution
Supervisor: Riccardo Bassoli
Quantum Key Distribution (QKD) is a secure communication technology based on the principles of quantum mechanics, enabling theoretically unconditionally secure key distribution. Field-Programmable Gate Arrays (FPGAs), with their high parallel processing capabilities and econfigurability, have broad application prospects in QKD systems. This project aims to explore specific application examples of FPGAs in QKD to enhance the performance and security of QKD systems.
At a glance
Type: Diploma/Master Thesis
Starting Time: immediate
Requirements: background on classical computer science and programming, ComNets2 and ComNets3
Get in touch with Tung Doan. He will help you find a supervisor who will work with you on a tailored topic.
The pathway to carbon neutrality is reshaping our traditional electricity consumption patterns. Historically reliant on centralized power utilities, we are witnessing a paradigm shift towards decentralized energy models. An increasing number of households and companies are shifting towards renewable energy solutions, notably solar panels, leading to the establishment of microgrids. These microgrids not only function autonomously but also possess the potential to feed excess energy back into the primary grid. As we advance into this new energy frontier, electric vehicles (EVs) are set to play a pivotal role. With the vehicle-to-grid (V2G) capability, EVs can serve as both consumers and providers of energy, enhancing grid flexibility but also adding further complexity. Leveraging cutting-edge communication technology such as 5G, we can seamlessly integrate multiple microgrids with EVs on the move. By tapping into the synergy among these entities and utilizing AI coupled with advanced optimization techniques, we pave the way for a more sustainable future.
Tasks
General tasks encompass three main areas: the development of testbeds, the creation of simulations, and the design of algorithms. Specifics can be tailored based on individual student preferences and interests.
At a glance
Type: Student/Diploma/Master Thesis, project work
Starting Time: immediate
Requirements: Experience in one of the following programming languages: Python, Java, or C++, knowledge of optimization algorithms and AI is a plus.
We’re seeking enthusiastic and skilled students to join our team. If you are interested in this, please contact Razan Habeeb or Shiwei Shen.
Open Topics in the Area of Human-Machine-Interfaces and Haptic Devices, Mobile Robotics, Navigation and Recognition, CeTIBAR Robot-aided Kitchen and Bar Service, Multi-Robot Coordination and Assembly.
Our dedicated team is continuously researching and developing our cutting edge technology to stay ahead of your needs. The robotic group stands for innovation, efficiency, and excellence in the areas of robotics.
Be part of this revolution and discover new ways of productivity with us!
Development of an Electromechanical End-Effector for Automated Sauce Dispensing
Supervisor: Marius Matzke
The core goal is to conceptualize, design, and test a novel robotic end-effector (a “sauce squeezer”). This device will be mounted onto a standard robot flange and will utilize an lectromechanical system to precisely dispense various sauces. The project involves the complete development cycle, from the initial concept and design to the creation of a functional prototype. The final system will be tested in a practical application within the Cetibar, demonstrating its capability for automated food handling.
Development of an Electromechanical Gripper System for Robotic Handling
Supervisor: Marius Matzke
The primary objective is to design, develop, and test a specialized end-effector (gripper) for a robotic arm. This involves creating a custom adapter that allows for the rapid and secure lifting of containers. The final outcome of this project will be a functional prototype of a fast electromechanical locking mechanism, enabling the safe and efficient manipulation of KLT containers by a robot. Students will be supported throughout the design, construction, and testing phases by our in-house experts and will have access to our 3D printing facilities for rapid prototyping.
Object Detection and Grip Pose Estimation for Robotic Manipulation
Supervisor: Roby Gehler
This project investigates how visual perception systems can be designed and optimized for robotic manipulation tasks in complex domestic settings.
Students will explore state-of-the-art methods for object detection (e.g., YOLO, DETR, Mask R-CNN) and grasp pose estimation (e.g., GraspNet, Dex-Net, keypoint-based 6D pose estimation).
The goal is to understand how perception pipelines can support safe, flexible, and human-aware manipulation within the CeTIBAR research environment.
Vision-Based Distance Estimation for Robotic Collision Avoidance
Supervisor: Marius Matzke
A fundamental challenge in robotics is enabling autonomous systems to perceive their environment for safe navigation. This project addresses the complex task of depth perception—calculating the distance to objects—using only the input from a single standard RGB camera (monocular vision). The primary objective is to develop a purely software-based solution that can generate a dense depth map, providing a distance value for every pixel in the camera’s image.
This capability is crucial for detecting potential obstacles, such as objects or people, and allowing a robot to automatically adjust its position to maintain a safe distance and prevent collisions. The core scientific challenge lies in extracting 3D information from a 2D image without the aid of stereo cameras, LiDAR, or other depth sensors. This project requires a thorough investigation of existing concepts and algorithms, from traditional computer vision techniques to modern deep learning-based approaches, to implement and test a robust system.
Then follow the link and check out the website of the team robotics.
Open Topics in the Area of Post-Shannon Communication and Guesswork
Functional Compression for Practical Networked Control Applications
Supervisor: Sifat Rezwan
Functional compression (FC) is a novel communication paradigm that aims to reduce communication overhead by transmitting only the necessary bits of information to achieve the desired goal. FC is a post-Shannon communication paradigm based on the idea of goal-oriented communication. The compression gains of this approach are potentially more significant than traditional compression techniques because the encoders transmit only the necessary information to compute the function and not to reconstruct the original sensor data. The system achieves the same goal with fewer bits, but the decoder cannot reconstruct the sensor data; it can only compute the goal function.
Common Randomness generation strategies in future communication systems
Supervisor: Prashanth K. H. Sheshagiri
Common randomness as a resource is gaining immense benefits in wireless communication. Some examples include improved security and trustworthiness. More novel applications include identification codes and integrated sensing and communications. However, there are several challenges in implementing common randomness generation in 5G/6G communication.
Guessing Noise to Decode Messages
Supervisor: Juan Cabrera
An ideal channel decoder would implement a maximum likelihood decoding technique to guess what message was transmitted. This is guessing which codeword was sent by maximizing the probability of receiving the obtained message. Because this is computationally complex, channel codes are designed backward. I.e., the design of a low-complexity decoder comes first followed by the encoder. This limits the type of codes that can be used because not all codes can be decoded in practical time. However, researchers from MIT and Maynooth University have proven that by guessing the noise in the transmission channel instead of the message, you can obtain results similar to a maximum likelihood decoder. The mathematical proof is complicated, yet the principle of operation is quite simple: If you receive a stream of bits that is not a valid codeword, you can flip one bit and ask if the new codeword is a valid one. If it is not, flip a different bit and repeat the process. If the probability of an error bit is low, then with a few flips and questions it is possible to decode. This opens the door to new codes since the decoding process is universal and potentially independent of the code used. We want to implement these novel techniques into our wireless system. To do that, we want to use Software Defined Radio to build the wireless channel and benchmark the novel decoder with state of the art codecs.
Motivation Links:
https://youtu.be/xQVm-YTKR9s
https://youtu.be/1bgC3AjCnA4
At a glance
Type: Student Thesis,
Diploma/Master Thesis (with task extention)
Starting Time: immediate
Requirements: Python, basic knowledge of
digital communication (digital modulation,
CDMA, OFDM)
Get in touch with Tung Doan. He will help you find a supervisor who will work with you on a tailored topic.
Drones and Simultaneous Localization and Mapping (SLAM)
Supervisor: Christian Vielhaus, Johannes Hofer
We are looking for a motivated student for a diploma thesis in the area of drones and Simultaneous Localization and Mapping.
The topic will involve the simulation of drones, the creation of maps, and path planning algorithms. You’ll get in touch with the following software/topics:
Are you interested? Then please go ahead, write us a message, come by for a coffee, and we talk about the details of the topic in person.
At a glance
Type: Diploma Thesis
Starting Time: immediate
Get in touch with Tung Doan. He will help you find a supervisor who will work with you on a tailored topic.
Microfluidic Molecular Communication (MMC) with AI-Assisted Encoding/Decoding
Supervisor: Pit Hofmann, Juan Cabrera
Microfluidic channels are compelling communication media for lab-on-chip and constrained environments, such as in-body communication, but they offer extremely low data rates and highly non-Gaussian noise due to diffusion, advection, and cross-talk between chemicals. This project explores AI-assisted, function/semantics-aware communication over our microfluidic testbed: Rather than fighting the channel to send raw bits, we co-design the source, channel, and task so that intelligent agents assist both encoder and decoder.
A concrete toy example system will be built end-to-end: an “audio-over-fluid” link where the transmitter listens to speech, converts it to text, communicates that text via chemicals in the microfluidic channel, and the receiver reconstructs intelligible audio using an AI text-to-speech decoder. The baseline will be extended with rate-distortion controls (e.g., summaries, punctuation-aware coding, unequal error protection, grammar-constrained symbols) and language-model-aided error correction.
At a glance
Type: Student Thesis,
Diploma/Master Thesis (with task extention)
Starting Time: immediate
Requirements: Python (must), basic command of Linux tooling, Familiarity with digital communications/signal processing (modulation, detection, synchronization) or a strong willingness to learn fast, Experience with at least one ML framework (PyTorch/TF) for ASR/TTS or language models
Molecular Communication for Leakage Detection Scenarios
Supervisor: Pit Hofmann, Pengjie Zhou
Inspired by our surrounding nature, molecular communication uses molecules and nano-particles as information carriers. Mainly intended for communication at the micro- and nano-scale level due to its biocompatibility and energy efficiency compared to conventional wireless systems, there is also a promising application for molecular communication in the macro-scale range with leakage detection. Leakage detection describes the process of identifying, locating, and assessing the presence of leaks in various system setups, such as pipelines, containers, or structures. The task involves investigating the potential of employing molecular communication principles for detecting leaks in a predefined system. The setup contains a mobile robot, e.g., a robot arm or the Boston Dynamics Spot robot dog, for exploring a room and a sensory part. The aim is to develop a robust and efficient system capable of detecting and localizing leaks in environments where traditional methods may be impractical or insufficient. The thesis will delve into the fundamentals of molecular communication, exploring how signaling molecules can be utilized to convey information about the presence and location of leaks. Furthermore, the research will focus on designing and implementing experimental setups to validate the effectiveness of the proposed approach. Through this study, the goal is to contribute to the advancement of leak detection technologies, particularly in scenarios where conventional methods face limitations.
At a glance
Type: Student/Diploma/Master Thesis
Starting Time: immediate
Requirements: basic knowledge of electrical engineering/communication science, Python
Get in touch with Tung Doan. He will help you find a supervisor who will work with you on a tailored topic.
