Seminar on Topics in Communications Networking

Lecturer (assistant)
TypeAdvanced seminar
Duration3 SWS
TermWintersemester 2021/22
Language of instructionEnglish
Position within curriculaSee TUMonline
DatesSee TUMonline


Admission information

See TUMonline
Note: Participation in the Kick-Off lecture (first lecture - see TUMonline) is mandatory to get a topic assigned. Prior registration in TUMonline is also mandatory. In case of more student registrations than topics available, topics assignment will follow a predefined selection process.


The goal is to teach skills in literature review, scientific writing and presentation. The seminar is intended to introduce students to current research literature to bridge the gap between coursework and thesis research. The student will gain understanding of a particular research area, learn how to apply the understanding, analyze the results, and evaluate their impact.


Topics (WS 21/22)

If you are interested to do your seminar on one of the topics offered by LKN, kindly contact the respective supervisor to show your interest (including the motivation to select this topic). If you reach an agreement with the respective supervisor before 20th October and the supervisor informs us, then the topic will be assigned to you. The remaining topics will be available for assignment based on a lottery system in the Kick-off meeting for the seminar on 22nd October, 2021. Only students with assigned topics will be accepted to our seminar.


List of Topics:

1) Soft Failure Detection and Localization Techniques in Elastic Optical Networks (Assigned, not Available!)
Supervisor: Saquib Amjad (
Description: Failure detection and management is crucial for reliable network performance and guarantee Quality of Transmission (QoT). Hard failures, such as fiber cuts, are easily detectable and have fixed mean time to repair and network downtime, unlike soft failures. Various channel impairments and optical component aging can lead to soft failures in the network. They can degrade signal quality and induce anomalies, leading to degradation of network over time, and possible service disruption. Recent research has used different monitoring parameters and machine learning techniques for soft failure detection, identification and localization in EON.

The goal of this seminar is to conduct a comparative study of two different detection and localization methods, and provide a clear analysis of the advantages and drawbacks of these solutions.

[1] Kayol S. Mayer, Jonathan A. Soares, Rossano P. Pinto, Christian E. Rothenberg, Dalton S. Arantes, and Darli A. A. Mello. "Soft failure localization using machine learning with SDN-based network-wide telemetry." In 2020 European Conference on Optical Communications (ECOC), pages 1–4, 2020
[2] Sima Barzegar, Marc Ruiz, Andrea Sgambelluri, Filippo Cugini, Antonio Napoli, and Luis Velasco. "Soft-failure detection, localization, identification, and severity prediction by estimating QoT model inputparameters." IEEE Transactions on Network and Service Management, 18(3):2627–2640, 2021


2) Intelligent Reflective Surfaces in LiFi-enabled 6G networks: Opportunities and Challenges (Assigned, not Available!)
Supervisor: Hansini Vijayaraghavan (
Description: Reconfigurable intelligent surfaces (RISs), also known as intelligent reflecting surfaces, are considered a promising technology in 6G networks due to their potential to improve the capacity and coverage of wireless networks by intelligently programming the wireless channel or environment.

Light-fidelity (LiFi), or visible light communications (VLC), is envisioned to be a part of 6G systems. The major obstacle to readily adopting LiFi is that it is subject to line-of-sight blockages and random receiver orientations. However, the integration of RISs in LiFi-enabled 6G networks is expected to combat adverse effects due to blockages and and random receiver orientation to achieve ubiquitous connectivity. This also open up opportunities to consider the channel as an additional degree of freedom instead of an impediment. This work involves carrying out a survey of existing articles on RIS assisted LiFi/RF networks and providing a vision of the role that RISs can play in LiFi-enabled 6G networks and challenges involved in their implementation in real-life applications.

[1] Marco Di Renzo, Alessio Zappone, Merouane Debbah, Mohamed-Slim Alouini, Chau Yuen, Julien De Rosny, and Sergei Tretyakov. "Smart radio environments empowered by reconfigurable intelligent surfaces: How it works, state of research, and the road ahead." IEEE Journal on Selected Areas in Communications, 38(11):2450–2525, 2020.
[2] Hanaa Abumarshoud, Lina Mohjazi, Octavia A Dobre, Marco Di Renzo, Muhammad Ali Imran, and Harald Haas. "LiFi through reconfigurable intelligent surfaces: A new frontier for 6G?" arXiv preprint arXiv: 2104.02390, 2021.


3) Federated Learning for Digital Twin Driven Industrial Internet of Things (Assigned, not Available!)
Supervisor: Alba Jano (
Description: In 6G mobile networks, resource management is supported by the digital representation of the communication network, known as Digital Twin (DT). Due to limited radio, computing, power, storage resources, resource management is necessary to fulfil the requirements from diverse use cases and support DTs, which perform communications, data analysis and AI-based computation. In the Industrial Internet of Things (IIoT) use case, a large amount of data needs to be communicated and processed from DT. Therefore, to avoid performance bottlenecks, caused by the large amount of data, failure in the operation of the DTs and address security and privacy issues, the distributed concept of DT was proposed, supported by federated learning. 

It is challenging to build distributed machine learning system over multiple nodes, as proposed by federated learning because synchronization and collaboration of AI algorithms are required. This work involves surveying existing articles on federated DT and providing a vision of the advantages and challenges presented by the federated DT network.

[1] Hamed Ahmadi, Avishek Nag, Zaheer Khan, Kamran Sayrafian, and Susanto Rahadrja. "Networked twins and twins of networks: an overview on the relationship between digital twins and 6G." arXiv preprint arXiv:2108.05781, 2021.
[2] Qiang Song, Shiyu Lei, Wen Sun, and Yan Zhang. "Adaptive federated learning for digital twin driven industrial internet of things." In 2021 IEEE Wireless Communications and Networking Conference(WCNC), pages 1–6. IEEE, 2021.


4) A Fault Management in SDN-enabled Elastic Optical Transport Networks (Assigned, not Available!)
Supervisor: Sai Kireet Patri (
Description: With the rapid deployment of 5G capabilities by major network operators across the world, there is a need to quickly upgrade back-haul capabilities, which are provided by re-configurable and elastic optical transport network.

Software Defined Networking (SDN), coupled with streaming telemetry has enabled network operators not only to extract information from their network, but also to find and localize soft or hard failures, which may appear either in the form of alarms in the network controller, or could be detected from performance monitoring data.

The aim of this seminar is to survey various experimental or simulation based studies on fault detection and localization for elastic optical transport networks while comparing them in terms of solution generality, scalability, and feasibility.

[1] Zhuotong Li, Yongli Zhao, Yajie Li, Sabidur Rahman, Feng Wang, Xiangjun Xin, and Jie Zhang. "Fault localization based on knowledge graph in software-defined optical networks." Journal of Lightwave Technology, 39(13):4236–4246, 2021.
[2] C. Delezoide, K. Christodoulopoulos, A. Kretsis, N. Argyris, G. Kanakis, A. Sgambelluri, N. Sambo, P. Giardina, G. Bernini, D. Roccato, A. Percelsi, R. Morro, H. Avramopoulos, E. Varvarigos, P. Castoldi, P. Layec, and S. Bigo. "Pre-emptive detection and localization of failures towards marginless operations of optical networks." In 2018 20th International Conference on Transparent Optical Networks (ICTON), pages 1–4, 2018.
[3] Francesco Musumeci. "Machine learning for failure management in optical networks." In Optical Fiber Communication Conference (OFC) 2021, page Th4J.1. Optical Society of America, 2021.


5) End-to-End Learning of Communication Systems without Known Channel (Assigned, not Available!)
Supervisor: Serkut Ayva ̧sık (
Description: The wireless system is separated to individual blocks and developed individually such as channel encoder/decoder, modulator, demodulator etc. throughout the generations. However, finding the optimum of such individual blocks does not guarantee the global optimum of a wireless communication systems. To overcome this problem a novel concept of End-to-End (E2E) learning of communication systems is proposed [1]. This novel concept is utilizing the deep learning techniques to jointly optimize the whole wireless communication system. Although quite appealing theoretically, the practicality of this idea is questionable due to the unavailability of the wireless channel knowledge at the transmitter during the training. To mitigate the problem, there are several proposed approaches such as using a generative adversarial net (GAN) to represent the channel effects at the transmitter as in [2].

In this work it is expected from student to comprehend the existing literature on the E2E learning of communication systems as well as GAN usage for unknown channels and provide a discussion about the advantages and challenges of the E2E learning concept in the wireless communication systems.

[1] Timothy O’Shea and Jakob Hoydis. "An introduction to deep learning for the physical layer." IEEE Transactions on Cognitive Communications and Networking, 3(4):563–575, 2017.
[2] Hao Ye, Le Liang, Geoffrey Y. Li, and B. Juang. "Deep learning-based end-to-end wireless communication systems with conditional GANs as unknown channels." IEEE Transactions on WirelessCommunications, 19:3133–3143, 2020.
[3] Hao Jiang and Linglong Dai. End-to-end learning of communication system without known channel. In ICC 2021 - IEEE International Conference on Communications, pages 1–5, 2021.


6) Load Balancing in Datacenter Networks (Assigned, not Available!)
Supervisor: Johannes Zerwas (
Description: Modern datacenter fabrics provide a rich path diversity between servers.  While this provides high throughput and connectivity in case of failures, traffic must be distributed over the available paths in order to operate the network efficiently. A common approach for distributing traffic among parallel paths is equal cost multi path routing. In the most basic version, it considers only header information when assigning flows to the paths.  However, depending on the flows' sizes, the actual load distributions can be imbalanced which, in turn, can lead to performance degradation. To mitigate this several other approaches for assigning flows to equal cost paths have been proposed, e.g., re-allocating flows more frequently in order to balance the traffic.

The goal of this topic is to provide an overview of the existing scheduling approaches for multi path routing and extract major assumptions as well as advantages and disadvantages.

[1] Jinbin Hu, Jiawei Huang, Wenjun Lyu, Weihe Li, Zhaoyi Li, Wenchao Jiang, Jianxin Wang, and Tian He. "Adjusting switching granularity of load balancing for heterogeneous datacenter traffic." IEEE/ACM Transactions on Networking, 2021.


7) Machine Learning in Data Planes (Assigned, not Available!)
Supervisor: Hasanin Harkous (
Description: P4 is a domain-specific language for programming the data plane of packet processors. It enables describing the behavior of the packet processing pipeline of switches and NICs in high-level syntax. There is a continuous increase in networking devices that support P4 programmability. Many innovative applications have been developed to run on P4 programmable devices for solving common networking issues. One of these applications is applying machine learning algorithms in the data plane for predicting relevant metrics.

In this work, the student will analyze different papers that employ machine learning techniques on the data plane. The objectives, contributions, and implementations of these papers will be summarized. Conclusions on the effectiveness of this approach will be conducted.

[1] Amedeo Sapio, Marco Canini, Chen-Yu Ho, Jacob Nelson, Panos Kalnis, Changhoon Kim, Arvind Krishnamurthy, Masoud Moshref, Dan Ports, and Peter Richtarik. "Scaling distributed machine learning with in-network aggregation." In 18th USENIX Symposium on Networked Systems Design and Implementation (NSDI 21), pages 785–808. USENIX Association, April 2021.
[2] Davide Sanvito, Giuseppe Siracusano, and Roberto Bifulco. "Can the network be the AI accelerator?" In Proceedings of the 2018 Morning Workshop on In-Network Computing, NetCompute ’18, page 20–25, New York, NY, USA, 2018. Association for Computing Machinery.