You may also contact one of our doctoral candidates directly, if you are interested in a Bachelor or Master thesis, a student job, an "Ingenieurspraxis" or a "Forschungspraxis". It often happens, that topics are being prepared which are not adverdised yet. It may also be possible to find a topic matching your specific interests.
Please include a curriculum vitae together with a list of attended courses when applying for a thesis.
If your "Ingenieurspraxis" is selected to be supervised by one of our professors, please hand in the documents to Doris Dorn (Room N2401).
|LDPC Decoder implementation on FPGA|
|Feasible decoding complexity is a major criteria for any decoding algorithm. LDPC Codes have linear decoding complexity, which makes them interesting for various applications. The aim of this bachelor thesis is to implement one specific LDPC Code on hardware starting from a software implementation.
Week 1: get familiar with LDPC Codes
Week 2: implement LDPC decoder in software
Week 3-8: implement LDPC decoder on FPGA
Week 8-9: write your thesis
Requirements: some experience with FPGAs
|Supervisor: Peihong Yuan, Patrick Schulte, Georg Böcherer|
|A Deterministic Approach to the Gaussian Multiple-Input Multiple-Output (MIMO) Broadcast Channel|
|Multi-user communication systems (such as wireless systems) are different than point to point channels in two key aspects; namely, the broadcast nature of transmissions (e.g., wireless transmissions can be picked up by any receiver in the vicinity of the transmitter) and the superposition (e.g., a wireless receiver picks up a linear combination of transmissions from all transmitters in its vicinity). Recently a deterministic approach to multi-user information theory has proved promising by achieving progress in some of the long-standing open problems by finding approximate solutions. The idea behind a deterministic approach, on a high level, is to focus on the interaction of signals rather than the background noise, through a simpler appropriate model; find exact solutions for the simpler model; and then translate the intuitions and techniques to new techniques and/or approximate solutions on noisy models.
The goal of thesis is to undertake a deterministic approach to Gaussian MIMO broadcast channels.
Prerequisite: Linear Algebra, Information Theory
|Supervisor: Shirin Saeedi|
|Role of Feedback in Broadcasting|
|This project intends to investigate the role of feedback in broadcast channels. More information available upon request.
Pre-requisite (at least one of the following):
Basic knowledge of Network Coding (see )
Basic knowledge of Algorithms
Basic knowledge of Information Theory
|Supervisor: Shirin Saeedi|
|Channel Code Design for Security Primitives|
We are looking for motivated students to work on (convolutional) code design for security applications. For realistic channel models, we want to test a set of codes to analyse their complexity, reliability, and secrecy performance. The thesis basically consists of code design for security primitives with given range of channel parameters, desired block lengths and BERs. Details will follow.
Interested students should directly send their CVs and transcripts.
|Supervisor: Onur Günlü|
|Probabilistic Signal Shaping for Short-Reach Optical Fiber Systems|
|Optical fiber systems form the backbone of the internet. Despite their large usable bandwidth, they are going to running out of capacity due to an increasing demand for services such as high-definition video streaming. In order to obtain the largest possible transmission rates for a fixed link, a non-uniformly spaced input can replace the common uniform input. Prerequisites for this thesis: digital communications (must!), basics in optical communication systems (important), information theory (equally important). If you are interested in this work, please contact me at email@example.com to set up a personal meeting.|
|Supervisor: Tobias Fehenberger|
|Efficient Algebraic Decoding of Subspace Codes for Network Coding|
|Subspace codes were proposed by Kötter and Kschischang in 2008 for error control in random linear network coding (RLNC). The main idea is to encode information by the choice of a subspace rather than a choice of a vector. A basis for the selected subspace is then transmitted over a channel. The decoder returns the subspace that is closest to the received subspace according to some metric.
I offer topics on efficient (list and unique) decoding algorithms for subspace and rank-metric codes that allow to correct errors beyond the unique decoding radius.
You should be interested in Channel Coding and should have good knowledge in (linear) Algebra.
|Supervisor: Hannes Bartz|
|Designing codes for secret key generation and extracting the secret bits in left over hash lemma|
|The source model of secret key generation deals with the idea of Alice and Bob generating a key in a distributed manner from correlated observations. This key must be kept secret from an evesdropper. In this internship/thesis the student will start by looking at a simpler model which corresponds to the left over hash lemma. The student will build on some preliminary work done by me to design codes for the extraction of left over hash in a simple setting. After this the student will extend the work to design codes for distributed secret key generation for the source model.|
|Supervisor: Ali Amjad|
|MIMO Multiport Communications|
|The topic of the thesis is to develop a simulator for MIMO communications including the transmit and receive RF frontends of a single user link. The simulator should be able to support different types of antenna elements and array configurations as well as matching circuits and amplifier models. In addition it should be able to estimate rates of communication over deterministic and stochastic channels.|
|Supervisor: Andrei Nedelcu|
|Rate-Distortion Theory for Sparse Sources|
|Rate-Distortion Theory for Sparse Sources
This project investigates information theory limits for the lossy compression of sparse sources.
Motivated by the recent mathematical theory Compressed Sensing that deals with the efficient acquisition and reconstruction of sparse signals, we try to determine the fundamental limits of digitally storing data created by such sources.
A sparse source emits a string of symbols of which - loosely speaking - only a few contain information. There are different ways to model these sources. An example is the Gaussian spike source which emits a Gaussian symbol with probability p and a zero with probability 1-p.
Possible directions of a project are:
- Investigating different source models
- (Numerically) determining the rate distortion function of a sparse source
- Computing finite length performance bounds for sparse sources
- Information Theory (must)
- Pleasure with mathematics (must)
- Multiuser information theory (beneficial)
|Supervisor: Lars Palzer|
|Migration of media content for the LNTwww|
|The LNTwww is based on a custom content management system. Many learning examples are provided in Flash. However, Flash is not supported on many platforms, e.g. iOS, and will likely not be supported in the future at all. The task is to evaluate possible alternatives for Flash multimedia content and to implement a preview of possible new alternatives to make the content accessible for as many students and platforms as possible. The student should also be able to have a good understanding of technologies used for web sites.|
|Supervisor: Markus Stinner|
Forschungspraxis or MSCE Internships
|Code design for Physical Layer Security|
|Wiretap channel represents the basic setup for physical layer security. It has been extensively studied in the last four decades and the fundamental limits of communication for this channel are known in a wide variety of scenarios. Nevertheless the only explicit code construction that can achieve wiretap secrecy capacity uses Polar codes. Designing codes for secrecy involve the combined design of codes for reliability and channel resolvability. In 2015 a new coding scheme for channel resolvability was introduced by Amjad and Kramer. The aim of this internship is to combine this channel resolvability code with existing channel codes in order to design wiretap code.|
|Supervisor: Ali Amjad|
|Implementing a Constant Composition Distribution Matcher on a Zynq FPGA|
|Constant composition distribution matching has recently received a lot of attention because it may be used to increase spectral efficiency using Probabilistic Amplitude Shaping proposed in . The combination of both inventions was awarded the 3rd prize in renominated Ball Labs Prize . In  an arithmetic coding scheme is proposed to index sequences of same weight or same composition, i.e., sequences of same length that consist of the same letters but permuted. The task of the Internship is to transform the encoder and decoder found at  from the programming language C into a hardware description language and evaluate its performance and complexity. A consecutive master's thesis could include improvement of the algorithm. We require the student to have some basic knowledge of VHDL and prior experience with the generic FPGA design flow and VHDL synthesis. Linux basics (using the command line, file system structure, ...) are also mandatory.  http://arxiv.org/abs/1503.05133  http://arxiv.org/abs/1502.02733  https://www.tum.de/en/about-tum/news/press-releases/short/article/32802/  http://www.lnt.ei.tum.de/en/projects/constant-composition-distribution-matching/|
|Supervisor: Patrick Schulte, Karl-Tasnad Kernetzky, Georg Böcherer|
|Code Design for Secret Key Generation/ Left over Hash Lemma|
|The source model of secret key generation deals with the idea of Alice and Bob generating a key in a distributed manner from correlated observations. This key must be kept secret from an evesdropper. In this internship/thesis the student will start by looking at a simpler model which corresponds to the left over hash lemma. The student will build on some preliminary work done by me to design codes for the extraction of left over hash in a simple setting. After this the student will (if time permits) extend the work to design codes for simple cases of distributed secret key generation for the source model.|
|Supervisor: Ali Amjad|
External Master Theses
|Fundamental limits of cooperation with imperfectly shared state information|
POSITION: EXTERNAL MASTER THESIS[PDF]
OPEN POSITIONS (FUNDED FROM ERC ADVANCED PROJECT "PERFUME")
Technical Context: The “PERFUME” project benefits from an “ERC Advanced” (European flagship) funding to investigate the so-called Smart Device Communications within the future Mobile Internet. Despite successes with past and current (5G) research, new paradigms leading to greater spectral efficiencies and intelligent network organizations will be in great demand to absorb the continuous growth in mobile data. Classical wireless design places the radio device under the tight control of the network. Pure network-centric designs, such as optical cloud- supported ones raise cost and security concerns and do not fit all deployment scenarios. More importantly these conventional approaches under-utilize growing local memory and computing capabilities at the device side (tablets, sensors, robotic objects such as wirelessly connected aerial drones etc.).
Our project envisions a radically new approach to designing the mobile Internet, which taps into radio devices’ new capabilities. Our approach recasts devices as distributed computational nodes solving together multi-agent problems, allowing to maximize the network performance by exploiting local measurement and information exchange capabilities. The projects sets novel objectives such as understanding new information theory limits for systems with decentralized information, the development of novel device communication methods, jointly with so-called team decisional statistical signal processing algorithms. The project will demonstrate the gains associated with exploiting the devices’ collective, network friendly, intelligence in pushing the frontiers of mobile Internet performance. Please visit http://www.eurecom.fr/cm/gesbert/erc for more details on the project.
Expertise and skills: Several master thesis are proposed for talented students with an excellent academic track . The project is at the frontiers of signal processing, communication theory, network information theory, game and optimization theory, artificial intelligence, as well as real-time implementation and experimentation, such that the subjects can be adapted to the desire and the skills of the students.
Practical Conditions: This position is located at EURECOM, in the French Riviera, near Antibes and Nice. The position being offered as part of the ERC project, the student will have a team of phd students and researchers providing interactions and a motivating work atmosphere. The starting date is flexible and can be discussed.
The student would be paid around 900 euros after taxes. Conditioned on the quality of the work realized, travels to international conferences could be funded. Ideally, this thesis would be the starting point of a doctorate at EURECOM.
Supervisor: David Gesbert and Paul de Kerret
|Supervisor: Georg Böcherer, Prof. David Gesbert (EURECOM), Paul de Kerret (EURECOM)|
Theses in Progress
|Alberto Diago Gallardo : Atmospheric Influence on Free Space Optical Communications|
|The work is based on Free Space Optical (FSO) communications and will include the following parts: • Case Study: Ground-to-Ground laser Links: o Link budget calculations; (Done) o Tracking Power budget calculations; o Scintillation analysis (optional: atmospheric simulations and fade statistics); (Ongoing) o Atmospheric Attenuation: Link availability calculations. (Ongoing) • Support the design, integration of testing of the “Frame-Steering Concept” as a demonstrator terminal for static links. (Ongoing) • Outdoor link demonstration and measurement campaign. • Analysis and evaluation of the measurement data.|
|Supervisor: Norbert Hanik, Luis Martin Navajas|
|Robert Markl: Development of an OFDM Physical Layer on a Zynq FPGA|
|The student's task is to develop an OFDM PHY layer on a Zynq FPGA located on a ZC706 evaluation board. The work will include include an FPGA design flow in System Generator for DSP (MATLAB/SIMULINK) and programming in VHDL.|
|Supervisor: Karl-Tasnad Kernetzky|
|Aya Ben Salha: n-Markov Language Generator|
|The aim of the thesis is to build a program that learns a language based on plain text and generates a markov chain that can be used for imitating language. Language can be german, english or symphonies.|
|Supervisor: Patrick Schulte, Bernhard Geiger|
|Rami Ezzine: Lossy Source Coding Performance of One-Bit Compressive Sensing|
|The subject of this Bachelor’s thesis is the performance analysis of recent one-bit com- pressive sensing (CS) algorithms in a lossy source coding framework.
Based on the seminal works of Donoho as well as Candes and Tao that deal with the efficient linear encoding of a sparse real-valued vector into a real-valued vector of smaller dimension, Boufounos and Baraniuk  proposed a one-bit CS algorithm where each linear measurement is quantized to +1 or -1. From these binary measurements, the original signal can then be approximated using various optimization techniques.
Since the one-bit CS procedure discards the amplitude information of the source signal, a popular source model is the uniform distribution on a sphere. However, applying a simple scalar quantizer to the amplitude of a signal, one might extend those algorithm to, for instance, Gaussian distributions. Another important source model is the binary source.
In lossy compression, the performance of a quantization code is usually measured with respect to its average distortion (the average quantization error) or excess distor- tion, that is, the probability of exceeding a given quantization error. Recently, there has been significant progress in finding the fundamental limits for the excess-distortion performance of fixed length vector quantizers, see  and the references therein as well as [5, 6]. These results can be seen as benchmarks for the performance of practical algorithms.
The aim of this thesis is thus to implement state of the art one-bit CS algorithms and investigate their lossy source coding performance for these source models.
The following tasks will serve as a guideline.
• Familiarize yourself with the concept of one-bit CS and perform a literature review of state of the art one-bit CS algorithms. Good starting points are [1, 2, 3]
• Implement selected algorithms in MATLAB
• Perform Monte-Carlo simulations to numerically evaluate the average quantization errors and excess-distortion behavior of the algorithms for suitable source models
• Compare the results to existing bounds for fixed-length lossy source coding
 P. T. Boufounos and R. G. Baraniuk, “1-bit compressive sensing,” 42nd Annual Conference on Information Sciences and Systems (CISS), Mar 2008, pp. 16–21.
 J. N. Laska, “Regime change: Sampling rate vs. bit-depth in compressive sensing,” Ph.D. dissertation, Rice University, 2011.
 V. Kostina and S. Verdu ́, “Fixed-length lossy compression in the finite blocklength regime,” IEEE Trans. Inf. Theory, vol. 58, no. 6, pp. 3309 – 3338, Jun 2012.
 L. Palzer and R. Timo, “A converse for lossy compression in the finite blocklength regime,” 2016 International Zurich Seminar on Communications, Mar 2016
 L. Palzer and R. Timo, “Fixed-length compression for letter-based fidelity measures in the finite blocklength regime,” preprint, Jan 2016
|Supervisor: Lars Palzer|
|Amine M'Charrak: Clustering Alorithms for Lossy Data Compression|
The subject of this Bachelor’s thesis is the application and performance evaluation of clustering algorithms to lossy data compression.
In lossy data compressoion, one is concerned with quantizing a random source symbol X drawn according to some distribution PX into another symbol Y , where Y can take on fewer values than X. The performance of a quantizer is usually measured with respect to its average distortion (the quantization error average over PX) or excess distortion, that is, the probability of exceeding a given quantization error.
With the recent interest in big data, there is large and growing interest in efficient clustering of large data sets. The aim of clustering is the meaningful grouping of a given data set. That is, given a set of data points K, we would like to map these points into a smaller number of clusters C while minimizing a certain cost on the whole data set. Usually, clustering algorithms do not incorporate any knowledge about the distribution of the source but only consider the individual points in K. Further, the number of clusters is typically much smaller than the number of data points which one may see as compression with relatively large quantization errors.
The purpose of this Bachelor thesis is to investigate the possible applicability of clus- tering algorithms to lossy data compression. As a first example, we consider the K- means algorithms for binary sequences. In some sense, K-means can be interpreted as an optimized version of random coding - a proof technique providing insights into the fundamental limits of data compression. Indeed, a straightforward implementation for the equiprobable binary source with Hamming distortions shows that the K-means algo- rithm can yield a close to optimal vector quantizer in certain scenarios. We would like to investigate this behavior further and particularize it do biased binary source. Departing from this simple approach, we will consider computationally efficient extensions of this algorithm to handle larger block lengths.
|Supervisor: Lars Palzer|
|Anes Belkacem: Reliability Analysis for Secret-key Binding for Physical Identifiers|
|Secret-key binding for Physical Unclonable Functions (PUFs) is a promising technique to provide security to Internet of Things (IoT) rather than storing a key in Non-volatile Memory (NVM). Key-binding schemes using PUFs require an error-correction step due to the noisy nature of PUF outputs. The PUF post-processing scheme that uses the Discrete Cosine Transform (DCT) to decorrelate the source is a suboptimal but efficient approach. The DCT-based scheme uses an algorithm to satisfy the secrecy and reliability constraints simultaneously by optimizing a reliability metric for each DCT coefficient so that a certain block error probability can be achieved for the secret-key by using a linear block code. This thesis aims to find a better reliability metric to give performance guarantees to extract a fixed number of bits. Real PUF outputs will be used to determine the source and channel models for the DCT-based scheme and to check the security and reliability performance of the scheme with the new metric.|
|Supervisor: Onur Günlü, Bernhard Geiger|
|Wafa Labidi: Resynchronization of Variable Length Coding|
|For variable length coding, local errors can lead to error propagation and synchronization problems. Consider the following exmple:
a -> 0
b -> 1 0
c -> 1 1 0
d -> 1 1 1
and the sequence
a a b d c a b -> 0, 0,10,111,110,0,10
. Suppose there was a transmission error of bit 4:
0, 0,111,111,10,0,10 -> a,a,d,d,b,a,b
There is error propagation, but interestingly, it is limited and resynchronization occurs. The topic of this thesis is to investigate this effect.
|Supervisor: Georg Böcherer|
|Houda Bannour: Design of a FEC Scheme for High Data Rate-Anti Jam speech waveform inteligibility optimization|
|The High Data Rate-Anti Jam waveform from Rohde&Schwarz allows audio signal transmission, supported by the MELPe vocoder with 2.4kbps. For multiple transmissions, the waveform uses the TDMA scheme. A TDMA slot contains MELPe frames and a header information with a total payload of 488bits. Under the Wideband mode, a TDMA slot is transmitted with a higher data rate, which means more bits per slot. These additional free bits shall be used in order to implement an FEC (Forward Error Correction) scheme to increase the robustness against channel noise and fading effects. Error Concealment (EC) techniques, such as bits interleaving, can be exploited on header and MELPe frames to reduce the effects of losses and burst channel errors. FEC combined with EC shall be designed in a way to find a balance between header and MELPe frames encoding schemes, in order to achieve an optimized bit error rate (BER), given by the case where all additional bits are used to protect the MELPe frames, without considering the slots’ losses due to the errors in header. The optimized BER ensures a minimum MOS (Mean Opinion Score) required for speech intelligibility.|
|Supervisor: Andrei Nedelcu, Dipl.-Ing. (FH) Stefan Kerger M.Sc|
|Clemens Bloechl: Aggregation of Hidden Markov Models - Theory and Applications|
|The topic of the thesis is to develop and analyze information-theoretic aggregation methods to reduce the state space and/or the observation space of hidden Markov models. Using the Kullback-Leibler divergence rate as cost function, we expect connections with the information bottleneck method, lumpability, and spectral aggregation techniques.
In the second part of the thesis, the developed methods shall be applied to practical examples, such as speech recognition systems. As a further example, the techniques shall be used to collapse output states of a discrete memoryless channel, without affecting the error probability of a Viterbi algorithm decoding a convolutional code.
|Supervisor: Bernhard Geiger|
|Nicolas Blum: Blind symbol rate and timing synchronization estimators for a Non-Cooperative Multitone Receiver|
|Supervisor: Yingkan Chen, Joerg Biedermann (Rohde & Schwarz)|
|Marcel Grec: Fountain Codes under inactivation decoding|
|Fountain codes are a class of erasure correcting codes which can potentially generate an unbounded
number of encoded symbols. Due to their flexibility they are a very efficient solution for the reliable
multicasting setting, where a transmitter wants to deliver the same object (data file, video streaming
etc.) to a number of receivers that have independent channels. In fact, fountain codes have been
adopted in a number of standards, among others in:
|Supervisor: Patrick Schulte, Francisco Lazaro-Blasco|
|Janis Surof: Intradyne Reception in Optical Satellite Links|
|The master thesis is set out to perform analytical and experimental feasibility study of an optical BPSK and QPSK optical link in GEO satellite system scenario. The task is to analyse and simulate a coherent phase-modulation optical link with respect to optical signal and noise power limitations, linear and nonlinear system e ects and atmospheric distortions. Then, series of equalising and correcting algorithms will be developed in MATLAB and applied to the transmitted signal in order to compensate for the simulated distorting e ects. The analytical results will be later experimentally demonstrated in laboratory conditions.|
|Supervisor: Tobias Fehenberger, Dr. Juraj Poliak (DLR)|
|Xianhe Yangzhang: Capacity Estimation using Nonlinear Frequency-Division Multiplexing for Optical SMF Channel|
|The capacity of the single-mode fiber (SMF) at high-SNR region is a critical topic for researchers in optical and information theory society. Due to the application of linear methods such as Nyquist-pulse shaping, OFDM, WDM, SDM, etc. in optical SMF transmission, the fiber Kerr-effect induced nonlinearity causes significant intra-/inter-channel interference which might lead to an underestimation of the capacity limit for SMF channel. Nonlinear frequency-division multiplexing (NFDM) based on the nonlinear Fourier Transform (NFT) is a promising approach to overcome the bottleneck of the data rate at high-SNR region. It treats fiber nonlinearity and chromatic dispersion as essential parts of the transmission such that the channels don't interfere with each other in nonlinear frequency domain. In this master thesis, an NFDM system at the defocusing case using forward and backward NFT will be developed and evaluated , aiming at giving a better lower bound of the capacity for optical SMF channel.|
|Supervisor: Yingkan Chen, Prof. Mansoor I. Yousefi|
|Sirui Zhang: Non-Asymptotic Complexity of Polar List Decoding|
|Supervisor: Peihong Yuan|
|Zeng Fan: Soft-Output Decoding of Polar Codes|
|Supervisor: Peihong Yuan|
|Julian Renner: Multidimensional Probabilistic Shaping for the Nonlinear Fiber Channel|
|Probabilistic shaping has shown to give large sensitivity gains for the additive white Gaussian noise (AWGN) channel and also for the nonlinear fiber channel. For a fiber-optic communication system, the optimal input distribution is unknown, and typically, heuristics are used to realize shaping. In this thesis, an approximate model of the fiber channel is used to numerically find and evaluate optimized input distributions, while exploiting the multi-dimensional nature of the considered channel.|
|Supervisor: Tobias Fehenberger|
|Delcho Donev: Channel Coding for Pulse Position Modulation|
|Supervisor: Georg Böcherer, Gianluigi Liva (DLR)|
|Manuela Meier: Syndrome Decoding of Interleaved Subspace Codes|
|Subspace codes are used for error control in random linear network coding (RLNC) and can be constructed from the Reed-Solomon like Gabidulin codes. A channel model for RLNC is the operator channel which incorporates insertions (of additional code symbols) as well as deletions (of code symbols). It has been shown, that interleaved or punctured subspace codes are very resilient against insertions if an interpolation-based decoder is used. The goal of this thesis is to analyze syndrome-based decoding schemes for interleaved subspace codes that are more resilient against insertions than deletions. The results should be verified by simulations.|
|Supervisor: Hannes Bartz, Vladimir Sidorenko|
|Berkay Köprü: Coding Schemes for Multiple Access Communications|
|Multiple access channel (MAC) can be seen as the model where two or more transmitters send information to a common receiver. For instance, multiple mobile phones communicating to a base station can be modeled as a MAC. The main challenge in MAC is that the receivers may share the transmission channel in an uncoordinated way, such that their transmissions can interfere. This problem is solved in many modern communication systems by allocating orthogonal resources (like time slots or frequency bands) to the transmitters such that the interference is avoided. However, although it is a practical solution, orthogonalization is not the optimal solution if considered from an information theoretic point of view. According to the theory, non-orthogonal transmission in MAC can be beneficial in many scenarios, however for these scenarios the coding schemes should be designed carefully to benefit from the non-orthogonal communication. Next generation cellular systems will support a multitude of services where multiple nodes are involved in the communications simultaneously, including vehicle-to-vehicle communications and massive machine communications. For these highly challenging communication scenarios, the design of the multiple access schemes will be of practical importance. In this thesis, practical coding schemes for MAC should be investigated, implemented and compared to the theory and to the existing solutions. Task Description: The tasks of the student can be summarized as follows: Studying the multiple access related scenarios for the next generation communication systems. Determining the correct channel models for the related scenarios. Basic theoretical evaluation of the scenarios according to the given models and comparison to the performance of the orthogonal schemes. Studying the already proposed solutions like IDMA and SCMA. Understanding the basic tools for channel code design, like EXIT analysis. Development and implementation of practical coding schemes for the selected multiple access scenarios. Simulations and comparisons to the state-of-the-art solutions and to the theory The actual tasks will be defined jointly with the candidate according to his background and the achievements. The focus can be put either on the theoretical analysis of multiple access schemes for different scenarios, or the more practical design and evaluation of coding schemes and receiver algorithms.|
|Supervisor: Patrick Schulte, Dr. Onurcan Iscan (HUAWEI), Dr. Ronald Boehnke (HUAWEI)|
|Mahmoud Marcam Helmy: Erstellung einer App für Smartphones der BITS mobile ERP Linie auf Basis von IOS Plattformen|
|Supervisor: Gerhard Kramer, BITS better it solutions GmbH, Wallenmahd 47, A-6850 Dornbirn|
|Ulrich Patrick Sandoval: Weiterentwicklung des Netzwerkanalysetools (NAT) mit Schwerpunkt auf Testfälle|
|Supervisor: Gerhard Kramer, Fa. Infineon Technologies, Neubiberg, Hr. Lippmann|
|Hasan Yagiz Özkan: Image Tracking Algorithm Development|
|Supervisor: Gerhard Kramer, Fa. Bases Teknoloji Arge Sanay, Hr. T. Cirak, Istanbul|
|Lukas Püttner: Evaluierung verschiedener Tools zur Extraktion und Auswertung von Metadaten aus IP-Verkehrsdaten|
|Supervisor: Norbert Hanik, ESG Elektroniksystem- und Logistik GmbH, 82256 Fürstenfeldbruck|
|Francesco Montanari: Dynamic Dead Pixel Correction|
|Supervisor: Norbert Hanik, Fa. ARRI, Arnold & Richter Cine Technik GmbH & Co KG, München, Fr. E. Kögel|
|Clarissa Hamann: Test|
|Supervisor: Gerhard Kramer, Dedan Kimathi University, Kenia|
Forschungspraxis (12 ECTS)
|Liu Xiaotong: Simulator in Matlab for MIMO Multiport Communications|
|The purpose is to build a Matlab simulator that can incorporate both circuit and channel models for MIMO communications in order to evaluate the achievable rates with different RF front-end designs in various propagation scenarios.|
|Supervisor: Andrei Nedelcu|
|Edward Wall: Finite-Precision Gaussian Mixture Models|
|In practical systems, Gaussian mixture models can only be presented with finite-precision. The first goal of this research internship is to survey the literature about how this problem is usually dealt with: Can we trade parameter precision for the number of mixture components? Can we restrict the covariance matrix to be diagonal? To be an identity matrix? What kind of cost functions are used to characterize these trade-offs? As a second goal, relative entropy shall be used as a cost function. For a simple multivariate Gaussian distribution and a given finite precision, the Gaussian distribution with quantized parameters minimizing relative entropy shall be computed.|
|Supervisor: Bernhard Geiger|
|Michael Wedrat: Ordered Statistics Decoding|
|Ordered Statistics Decoding (OSD) can perform close to Maximum Likelihood decoding with reduced complexity. The topic of this research internship is to implement an OSD decoder and to validate OSD results presented in literature.|
|Supervisor: Georg Böcherer|
|Afghahi Keikavoos: Development of adaptive radio systems for industrial applications|
|Wireless real-time communication plays an increasingly pivotal role in industrial automation. Next generation high performance wireless technologies have to meet highest demands regarding latency, efficiency and reliability. Further, they have to be smart and adapt to an ever changing radio environment. In this internship you will participate in the research and development of adaptive wireless communication systems for application in the novel field of Industry 4.0. Using a Software Defined Radio (SDR) system you will develop a spectrum sensing procedure, which enables higher efficiency and reliability of wireless communication technologies for industrial applications. This internship will consist of the following tasks: · Getting familiar with the SDR development using USRP (Universal Software Radio Peripheral) and GNU Radio · Development of a spectrum sensing algorithm using the USRP hardware platform · Software development for interferer detection and smart adaptive channel allocation · Creation of a wireless network prototype, which dynamically adapts to a changing radio environment in an industrial scenario|
|Supervisor: Ginni Khanna, Neda Petreska|
|Nathan Gibbs: Estimation of (Generalized) Mutual Information for Real-time Applications|
|Evaluating the post-FEC BER of a coded-modulation system using the uncoded (pre-FEC) BER is a commonly used method in fiber-optics. However, this approach can be inaccurate. Instead, mutual information (MI) is a natural figure of merit to consider as it is an achievable rate for a decoder that operates on symbol metrics. For the practical bit-interleaved coded modulation systems, the generalised mutual information (GMI) has been shown to accurately predict the post-FEC performance. In this Forschungspraxis, GMI estimation under typical constraints for real-time applications is investigated.|
|Supervisor: Tobias Fehenberger|
|Niklas Jünger: Energy-efficiency of massive MIMO and network MIMO in a local area scenario|
Multiple-input multiple-output (MIMO) methods increase spectral efficiency. These methods can also be used to decrease energy consumption. Using many more antennas at the base station then served users is called massive MIMO. With each added antenna at the base station spectral efficiency is increased. It is however not obvious if the energy efficiency is decreased with every added antenna.
The student should review how much energy is consumed and review existing precoding and power allocation strategies for energy efficient transmission. The most well suited strategies should be adopted to the local area scenario deployments. Simulations should be run in the 3GPP two stripe building.
|Supervisor: Stefan Dierks|
|Muhammad Firas Hammosh: Is Online PCA Information-Preserving?|
|In this research internship, and overview over existing online (i.e., iterative, recursive, etc.) algorithms for Prinicipal Components Analysis (PCA) should be given. We try to find our which (if any) of these algorithms is invertible in the sense that one can reconstruct the original data from only looking at the rotated data. For those algorithms for which this is not possible, the (relative) information loss should be computed.
This work thus builds the bridge between PCA given knowledge of the covariance matrix (given-statistics) and PCA given only the sample covariance matrix (given-data). While no information is lost in the former, the information loss in the latter was shown to be substantial. We believe that the information loss of online PCA lies somewhere in between.
|Supervisor: Bernhard Geiger|
|Muhammad Umer Anwaar: Coding Techniques for Natural Language Processing|
|In this internship the student will review current state-of-the-art techniques for Natural Language Processing (with a focus on Machine Translation). Specifically, the student will check which of these techniques employ Hidden Markov Models, and whether they have connections to decoding algorithms for channel codes. Finally, the student should present a recommendation if, and how, list decoding methods can be applied in machine translation. If you are interested to pursue this topic for your Master's thesis that is also possible.|
|Supervisor: Ali Amjad, Bernhard Geiger|