Semester Projects (Semesterarbeiten)

If you are interested in one of the following topics, please contact the person listed with the respective topic.

If you don't find a compelling topic in our list, we are always keen on hearing about your ideas in the area of our research interests (see the Research section of our website). You are most welcome to discuss your interests with us.

Also, we have a list of finished student projects on our website.

List of Projects

Implementation of an RF channel emulator (SA)

In the development process of radio transceivers, it is important to test and optimize the performance of the implemented hardware, coding schemes, and signal processing algorithms under different channel conditions. This can be done with real-world channels in over-the-air tests. However, this strategy has several disadvantages, namely interference, logistical problems, and legal restrictions on certain frequency bands. Furthermore, the results will depend on environmental conditions and are therefore not reproducible.

A channel emulator, i.e., a device that emulates a channel and applies it to a communication signal, is a tool built exactly to solve this problem. The parameters of the channel (e.g. Doppler spread, delay spread, noise, ...) can be controlled through the configuration of the emulator [1]. Most channel emulators also have multiple inputs/outputs to emulate MIMO channels. MIMO channel emulation devices are available on the market [2], but expensive and sometimes inflexible.

The aim of this project is to implement a channel emulator that works in the digital baseband domain and can be controlled from an external PC. An FPGA platform including an RF transceiver and analog/digital (ADC) and digital/analog converters (DAC) was already selected in a previous project and can be used as a starting point.

Type of project: 70% VHDL programming and simulation, 20% measurements, 10% theory
Prerequisites: Interest in RF hardware and wireless systems, Matlab, VHDL
Supervisor: Michael Lerjen
Professor: Helmut Bölcskei

[1] S. Yang and Z. Can, "Design and implementation of multiple antenna channel emulator for LTE system," Proc. 9th International Conference on Communications and Networking in China (CHINACOM), pp. 208–213, 2014 [Link to Document]

[2] Keysight Technologies, Inc., "Wireless device test sets & wireless solutions" [Link to Document]

Implementation of a compressed sensing radar architecture (SA/DA)

Radar systems are typically modeled as linear systems that induce weighted superpositions of delayed and Doppler-shifted versions of the probing signal. Identifying the delay-Doppler shifts yields the position and relative speed of the object of interest. A similar problem arises in the identification of wireless channels, where the (unknown) delay-Doppler shifts correspond to point scatterers in the propagation environment. For suitably chosen probing signals the problem of identifying the delay-Doppler shifts can be reduced to that of recovering a sparse vector from (highly) undersampled measurements [1, 2, 3], i.e., to a compressed sensing problem. Standard approaches to solving the corresponding problem include l1-minimization or greedy algorithms such as orthogonal matching pursuit. It was shown in [3] that recovery of the delay-Doppler shifts in the radar and the system identification problem can be formulated as a multiple measurement vector (MMV) problem, which can be solved efficiently using low-complexity subspace algorithms such as, e.g. MUSIC [4].

The goal of this project is a hardware implementation of a compressed sensing radar system. Specifically, we consider a system consisting of RF instruments to generate and capture the signals and a hardware channel emulator. The subspace algorithm will be implemented in MATLAB in a first step and, time permitting, on an FPGA in a second step. The overall system shall be tested in terms of performance and practical applicability.

Type of project: 50% implementation (RF measurements, Matlab programming, hardware design), 30% simulation, 20% theory
Prerequisites: Interest in RF hardware and wireless systems, Matlab, possibly VHDL, and linear algebra
Supervisor: Michael Lerjen, Céline Aubel
Professor: Helmut Bölcskei

[1] W. U. Bajwa, K. Gedalyahu, and Y. C. Eldar, "Identification of parametric underspread linear systems and super-resolution radar," IEEE Trans. Signal Process., vol. 59, no. 6, pp. 2548–2561, Jun. 2011. [Link to Document]

[2] M. Herman and T. Strohmer, "High-resolution radar via compressed sensing," IEEE Trans. Signal Process., vol. 57, no. 6, pp. 2275–2284, 2009. [Link to Document]

[3] R. Heckel and H. Bölcskei, "Identification of sparse linear operators," IEEE Trans. Inf. Theory, vol. 59, no. 12, pp. 7985–8000, 2013. [Link to Document]

[4] R. Schmidt, "Multiple emitter location and signal parameter estimation," IEEE Trans. Ant. Propag., vol. 34, no. 3, pp. 276–280, Mar. 1986. [Link to Document]

Deformation sensitivity bounds for deep convolutional neural networks (SA/DA)

Feature extractors based on so-called deep convolutional neural networks (DCNNs) have been applied with tremendous success in a wide range of practical signal classification tasks [1] such as, e.g., in handwritten digit classification. The features to be extracted in this case correspond to the edges of the digits, and we would want these features to be robust with respect to handwriting styles. This can be accomplished by demanding that the feature extractor have limited sensitivity to certain non-linear deformations.

Recently, [2], [3] established deformation sensitivity bounds for a wide class of DCNN-based feature extractors. These bounds apply to a variety of input signal classes such as band-limited functions, cartoon functions, and Lipschitz-functions. Many signals of practical interest (such as textures) exhibit, however, sharp oscillations and are therefore not captured by these results.

The goal of this project is to use the theory of approximately time- and band-limited functions [4] to develop general deformation sensitivity bounds.

Type of project: 80%-100% Theory, 0-20% Simulations, depending on the student's preference
Prerequisites: Analysis, Linear Algebra, Signals and Systems I
Supervisor: Thomas Wiatowski
Professor: Helmut Bölcskei

[1] I. Goodfellow, Y. Bengio, and A. Courville, “Deep Learning,” MIT Press , 2016. [Link to Document]

[2] T. Wiatowski and H. Bölcskei, “A mathematical theory of deep convolutional neural networks for feature extraction,” arXiv:1512.06293, 2015. [Link to Document]

[3] P. Grohs, T. Wiatowski, and H. Bölcskei “Deep convolutional neural networks on cartoon functions,” Proc. IEEE Int. Symp. on Inf. Theory (ISIT), pp. 1163–1167, 2016. [Link to Document]

[4] D. Slepian, "On bandwidth," Proc. of the IEEE, pp. 292–300, 1976. [Link to Document]

Control policy learning for a flight simulator (SA)

Reinforcement learning (RL) [1] refers to the problem of finding suitable actions to take in a given situation in order to maximize a reward. RL algorithms that recently led to breakthrough results in practical machine learning tasks (such as learning to play Atari games [2] or the strategy board game Go [3]) typically employ a deep convolutional neural network [4] for feature extraction.

This project aims at developing an RL-based approach for learning control policies for a flight simulator. The scope of the project may encompass value function design, training data acquisition, as well as a first implementation in Python.

Type of project: 90% Theory/Literature research, 10% Programming
Prerequisites: Python, machine learning basics is a plus
Supervisor: Michael Tschannen, Michael Lerjen
Professor: Helmut Bölcskei

[1] R. S. Sutton and B. A. Barto, "Reinforcement learning: An introduction," MIT press, 1998.

[2] V. Mnih et al., "Human-level control through deep reinforcement learning," Nature, vol. 518 (7540), pp. 529–533, 2015. [Link to Document]

[3] D. Silver et al., "Mastering the game of Go with deep neural networks and tree search," Nature, vol. 529 (7587), pp. 484–489, 2016. [Link to Document]

[4] Y. LeCun, Y. Bengio, and G. Hinton, "Deep learning," Nature, vol. 521 (7553), pp. 436–444, 2015. [Link to Document]

Structured convolutional neural networks (SA)

In recent years, deep convolutional neural networks (CNNs) [1,2] have proven tremendously successful in many practical classification tasks. However, conventional CNNs often result in models with a large number of parameters [3] and are often highly sensitive to adversarial noise [4] (see image left, image credit: [4]).

The goal of this project is to explore how structure can be built into CNNs in order to reduce model size and/or improve robustness to adversarial noise.

Type of project: 20%-50% Theory, 50%-80% Programming
Prerequisites: Programming (preferably Python), linear algebra, machine learning basics is a plus
Supervisor: Michael Tschannen
Professor: Helmut Bölcskei

[1] Y. LeCun, Y. Bengio, and G. Hinton, “Deep learning,” Nature, vol. 521, pp. 436–444, 2015. [Link to Document]

[2] I. Goodfellow, Y. Bengio, and A. Courville, "Deep Learning," MIT Press, 2016. [Link to Document]

[3] S. Han, J. Pool, J. Tran, and W. Dally, "Learning both weights and connections for efficient neural networks," Advances in Neural Information Processing Systems, pp. 1135–1143, 2015. [Link to Document]

[4] C. Szegedy, W. Zaremba, I. Sutskever, J. Bruna, D. Erhan, I. Goodfellow, and R. Fergus, "Intriguing properties of neural networks," International Conference on Learning Representations, 2014 [Link to Document]

Deep convolutional neural networks for scene labeling (SA/DA)

Deep convolutional neural networks (DCCNs) with pre-defined filters [1,2] extract characteristic features from signals by recursively applying the composition of the following three operations: convolution with a set of filters, a non-linearity, and a sub-sampling step. These networks, combined with a classifier (such as, e.g., a support vector machine) were successfully employed in a number of practical classification tasks [1,3,6]. In contrast to traditional DCNNs which learn the filters from training data [4], DCNNs with pre-defined filters rely on wavelets [1], curvelets [2], or shearlets [2]. While the use of pre-defined structured filters leads to less flexibility, it allows for faster implementations.

This project shall explore the application of DCNNs with pre-defined filters to the problem of scene labeling, where the aim is to assign a class label such as "street", "tree", or "building" to every pixel of an image. Applications of scene labeling include situational awareness systems [4], which often demand low-complexity scene labeling.

The goal of this project is to extend the work carried out in prior semester theses on the topic. In particular, the classification stage should be improved and the existing pipeline should be optimized for speed.

Type of project: 0%-20% Theory, 80%-100% Programming, depending on the student's preference
Prerequisites: Python, C programming, linear algebra
Supervisor: Michael Tschannen, Thomas Wiatowski, Lukas Cavigelli (IIS), Michael Lerjen
Professor: Helmut Bölcskei, Luca Benini (IIS)

[1] J. Bruna and S. Mallat, "Invariant scattering convolution networks," IEEE Trans. Pattern Anal. Mach. Intell., vol. 35, no. 8, pp. 1872-1886, 2013. [Link to Document]

[2] T. Wiatowski and H. Bölcskei, "A mathematical theory of deep convolutional neural networks for feature extraction," IEEE Transactions on Information Theory, (revised version, Aug. 2016), Dec. 2015, submitted. [Link to Document]

[3] J. Andén and S. Mallat, "Deep scattering spectrum," IEEE Trans. Sig. Process., vol. 62, no. 16, pp. 4114-4128, 2014. [Link to Document]

[4] Y. LeCun, L. Bottou, Y. Bengio, and P. Haffner, "Gradient-based learning applied to document recognition," Proc. of the IEEE, pp. 2278-2324, 1998. [Link to Document]

[5] L. Cavigelli, M. Magno, and L. Benini, "Accelerating real-time embedded scene labeling with convolutional networks," Proc. of ACM/EDAC/IEEE Design Automation Conference (DAC), pp. 1-6, 2015. [Link to Document]

[6] I. Goodfellow, Y. Bengio, and A. Courville, "Deep Learning," MIT Press, 2016. [Link to Document]