Theses in Progress

Enumerative sphere shaping [1] is an efficient method to perform energy-optimal distribution matching for short to medium block lengths. It is an alternative approximate shell mapping schemes based on CCDM [2].

The goal of this thesis is to extend enumerative sphere shaping to general target distributions using the ideas from [3] and compare the performance of this distribution matcher to CCDM-based approximate shell mapping.

1. https://doi.org/10.1109/TWC.2019.2951139
2. https://doi.org/10.1109/TIT.2015.2499181
3. https://doi.org/10.1109/LWC.2018.2890595

.

Federated learning is a machine learning paradigm where decentralized entities (clients) collaboratively learn using their private data. A central server acts as a coordinator of the learning process. Due to the sensitivity of the private data involved,  the data cannot be transferred. A salient problem of federated learning is the presence of malicious clients, which are clients that try to destroy the learning process. Malicious clients can do this by corrupting their data and/or by modifying their local model updates. The goal of this project is to understand how model poisoning attacks and defense strategies perform under different scenarios of federated learning using experiments. 

References: 

[1]- https://www.ndss-symposium.org/wp-content/uploads/ndss2021_6C-3_24498_paper.pdf

[2]- https://arxiv.org/pdf/1903.03936.pdf

[3]- https://arxiv.org/pdf/2304.00160.pdf

• Basic knowledge of machine learning
• Python programming skills, knowledge of PyTorch is an advantage 

DiVincenzo in 2000, provides a comprehensive framework for identifying the necessary requirements to achieve successful quantum computation. In this project, we search for ideal conditions necessary for buidling distributed quantum devices. 

We apply information-theoretic perspectives from rate-distortion theory to optimize wildfire notifications from earth observation satellites.

We investigate the security of decentralized learning with gradient compression via Zero-order methods.

.

.

We study the impact of different measures of fairness on the privacy guarantees for individual clients in federated learning.

We study the usage of channel codes for the setting of federated learning with over-the-air computation so that the sum of codewords over reals can be efficiently and reliably decoded at the federator to obtain the average of partial model updates.

The main aim of the thesis is to make a comparison between different multicarrier modulations, in particular OFDM and OTFS, inside a wireless system.

The considered wireless system consists in a Delay-Doppler channel, which is typical in vehicular communications. A Hybrid IRS is considered in order to be able to achieve Integrated Sensing and Communications.

In paper 

"Correcting a Single Error in Feedback Channels" the problem of correcting a single error with feedback was investigated. It was mainly devoted to binary channels, namely, binary symmetric and asymmetric channels. A general theorem, which allows constructing codes correcting one error with one feedback, was proved. For the symmetric channel with one error, it was proved that with two feedbacks one can transmit as many messages as with complete feedback.

In this project, it is proposed to generalize these results to the case of two errors. Namely, it is required to prove the theorem, which describes optimal codes, correcting 2 errors with one feedback. After that, some optimization problems should be solved to find the parameters of optimal codes.

This work shall deal with post-quantum secure schemes utilizing the Lee metric. The student's task is to get familiar with this metric and to design a signature scheme. The student shall show that known attacks in the Hamming metric are not applicable for the Lee metric.

Channel Coding

Due to the recent advances in quantum computers, the search for cryptosystems that survive quantum attacks is of great interest. Code-based cryptography is a promising candidate, since it is build on the NP-hard problem of decoding a random code [1].

Recently, different variants of the classical syndrome decoding problem (SDP) in the Hamming metric have been proposed [2,3].
The main reason for this is that it appears hard to build an efficient digital signature scheme around the classical SDP.
One such variant is the restricted syndrome decoding which was introduced in [2].

The goal of this the construction of codes for this error model, which has not been done before.
For this, a first approach is to follow the general idea given in [4].

Open questions are:

- discussion of the appropriate choice of the error set of a McEliece-type cryptosystem
- optimality bounds for codes in the restricted setting
- construction of short codes that are efficiently decodable and/or optimal
- construction of longer codes from short codes and the evaluation of their perfomance

References:

[1] Weger, V., Gassner, N., & Rosenthal, J. (2022). A Survey on Code-Based Cryptography. arXiv preprint arXiv:2201.07119.

[2] Baldi, M., Battaglioni, M., Chiaraluce, F., Horlemann-Trautmann, A. L., Persichetti, E., Santini, P., & Weger, V. (2020). A new path to code-based signatures via identification schemes with restricted errors. arXiv preprint arXiv:2008.06403.

[3] Baldi, M., Bitzer, S., Pavoni, A., Santini, P., Wachter-Zeh, A., & Weger, V. (2023). Generic Decoding of Restricted Errors. arXiv preprint arXiv:2303.08882.

[4] Rohweder, D., Freudenberger, J., & Shavgulidze, S. (2018). Low-density parity-check codes over finite Gaussian integer fields. In ISIT.

Security in Communications and Storage

Channel Coding

Compared with traditional frame-based cameras, an event-based camera has the advantages of low latency, high dynamic range, (almost) no motion blur, etc., and can respond fast to a brightness change at the image plane with thresholds determined by the previous state of brightness. It can generate event data of structural features without signal processing, such as edges and corners, which saves time, energy and computing effort.

However, it still lacks standard processes for analyzing, characterizing and evaluating event-based camera information. Moreover, data acquisition of this camera demands changes in brightness on a respective pixel. This can be introduced artificially by relative movement of the camera to the object. This might miss out part of the data due to linear translation along structural elements like edges or introduce some error due to error motion or vibration.

This thesis aims to ensure safety and quality when implementing event-based cameras in the field of industrial inspection, by dedicated experiments and related discussion of results. Event based camera imageswill be characterized and evaluated. New methods for event generation and signal processing will be proposed which will make use of the special characteristics of event-based cameras and their special characteristics arising from their working principle.

In theory, the check nodes operations of belief propagation rely on tanh() and arctanh() functions which require high computational power. Hence, in most of the practical applications, an approximation called “MinSum” is used. This method aims to exploit the structure of tanh() function where the absolute value of output sufficiently converges to 1 with increasing absolute value of input. Hence, in a series multiplication of absolute outputs, the most dominant effect comes from the minimum element and other contributions are considered negligible. However, neglection of other attenuation factors cause overcalculated outputs in “MinSum” algorithm which can accumulated by multiple iterations. This drawback can be compensated through adding attenuation or/and offset factors. These factors are mostly iteration specific and intuitively determined, which means one factor which is determined by educated guess is applied to all leaving edges. However, every edge in an unfolded Tanner graph has its own unique identity corresponding to the previous nodes and edges that the message is transmitted.

In addition to approach aiming to close the performance gap between main algorithm and “MinSum ” approximation, we can intend to improve the qualities of main algorithm. Even though belief propagation decoding in LDPC codes is considered as highly successful, it is still a “suboptimal” method compared to very expensive but accurate Maximum A posteriori Probability (MAP) estimation. It means there might be some room for improvement in performance. Additionally, belief propagation requires multiple iterations to converge and the required number of iterations can dramatically increase by decreasing signal-to-noise ratio (SNR). Additional correction weights imposed on iterated messages can be a candidate to improve performance in overall.

5G specification for channel coding is using protograph based LDPC codes. Every node duplicated from same base matrix node is keen to show similar properties, it may be possible to use same weights for these nodes by preserving the good decoding results. This detail can help us to using additional correction weights by minimum additional memory burden.

In this project the student should calculate the zero-error capacity for 

a multi-user model with feedback.

Information theory

Whenever an automobile manufacturer develops a new car model, the aim is to surpass its predecessors in terms of performance, efficiency, safety, and overall user experience. In order to do so, logged data has to be analyzed to find out where improvements can be made. Whenever a vehicle returns from a test drive, the collected data will be stored as an MDF (Measurement Data Format) file. These files are then read and analyzed with a software tool, but as of now, only a few files can be processed simultaneously due to hardware limitations. This research internship explores how data can be efficiently organized to maximize analyzability. Eventually, a program should be created which allows users to analyze all MDF files a vehicle has produced in parallel. 

In this internship, the student will explore the literature on privacy-preserving synchronization methods and summarizes the findings.

The Student is working on direction measurements for Radar communications. The scenario considered is a bi-static scenario in which a signal from a static vehicle is collected from a fixed antenna. From this radar communication, the student has to analyze the DoA estimation and try to find techniques in order to improve it.

In this internship, the student should learn and implement the decoding of Linearized Reed-Solomon in the scenario where the encoding is joint but distributed.

- Channel Coding

In this internship, the student will read and summarize the recent progress on codes for privacy and security in federated learning.

Depending on the student's progress, we can investigate new ideas for security in private federated learning that improve upon the state0of0the-art.

In this project, we investigate the interplay between redundancy and straggler tolerance in distributed learning.

The setting is that of a main node distributing computational tasks to available workers as part of a machine learning algorithm, e.g., training a neural network. Waiting for all workers to return their computations suffers from the presence of stragglers, i.e., slow or unresponsive nodes. Mitigating the effect of the stragglers can be done through the use of redundancy or by leveraging the properties of the convergence of the machine learning algorithm. 

The goal of this work is to compare when redundancy is helpful. In this case, we aim to analyze the convergence speed with and without redundancy. Then, we compare the convergence as a function of time of all the schemes.

Further reading: 

R. Bitar, M. Wootters and S. El Rouayheb, Stochastic Gradient Coding for Straggler Mitigation in Distributed Learning, IEEE Journal on Selected Areas in Information Theory (JSAIT), Vol. 1, No. 1, May 2020. arXiv:1905.05383

S. Kas Hanna, R. Bitar, P. Parag, V. Dasari and S. El Rouayheb, Adaptive Stochastic Gradient Descent for Fast and Communication-Efficient Distributed Learning, preprint, arXiv:2208.03134.

Knowledge in the following topics:

• Probability Theory
• Gradient descent and stochastic gradient descent
• Coding theory

Independence and motivation to work on a research topic

Knowledge of implementing neural networks is a plus

We are interested in computing upper bounds (on capacity) for bandlimited channels with a memoryless nonlineary at the transmitter/receiver.

One application for these bounds are short-reach fiber-optic communication systems with a single photodiode at the receiver. The photodiode is a memoryless nonlinearity, as it produces an output that is proportional to the squared magnitude of the input signal.

A simple upper bound for the above model is given in [Sec. III D, 2].

D. Plabst et al., "Achievable Rates for Short-Reach Fiber-Optic Channels With Direct Detection," in Journal of Lightwave Technology, vol. 40, no. 12, pp. 3602-3613, 15 June15, 2022, doi: 10.1109/JLT.2022.3149574.

Information Theory

Linear System Theory

Forschungspraxis

.

.

.

.

the students should translate functions from Matlab and C to julia. the functions involve calculation of infomation theoretic quantities to basic function of a discrete time transmission chain.

The students task is to learn julia and matlab to a extend that the translation from one to the other language is possible. We furthermore expect the student to learn how to use git and gitlab for managing larger projects.

basic programming knowledge

.