Currently the DSS group offers only one seminar. You can find some details below.
Topics and first dates have been updated for WS 19/20.
Selected Topics in Digital Signal Processing
|Lecturers:||Gerhard Schmidt and group|
|Language:||English or German|
|Target group:||Master students in electrical engineering and computer engineering|
|Prerequisites:||Fundamentals in digital signal processing|
If you want to sign up for this seminar, you need to register with the following information in the registration form
Please note that the registration period starts 01.10.2019 at 8:00 h and ends 25.10.2019 at 23:59 h. All applications before and after this registration period will not be taken into account.
Registration will be possible within the before mentioned time under the following subsite - Seminar Registration.
During the registration process you will also choose your seminar topic. Only one student per topic is permitted (first come - first serve).
Students write a scientific report on a topic closely related to the current research of the DSS group. Potential topics, therefore, deal with digital signal processing regarding:
Students will also present their findings in front of the other participants and the DSS group.
Topics offered in WS 19/20
Speech and Audio:
Analysis of the Intelligibility of Speech
A lot different diseases can affect the intelligibility of the speech of the impaired people. A typical example is Morbus Parkinson which often leads to a too quiet speech and unclear articulation of the patients. In literature, there are different approaches to classify those cases using recordings of different vowels to extract certain features. For this topic, those different features extracted from vowels shall be examined and the usage of these in classifiers like neural networks to estimate the intelligibility shall be further evaluated.
Frameworks for Speech Recognition
In some cases, using a custom speech recognizer instead of a readily available (commercial) solution is advantageous. However, such a custom recognizer needs to first be implemented and trained, which may be done within an open-source framework. For this seminar topic, open-source frameworks for the development of a speech recognizer are to be analyzed regarding their properties, handling and workflow. As a result, a well-reasoned recommendation for the selection of a framework should be given.
Microphone Array Signal Processing
Beamforming is usually used for speech signal acquisition in challenging environments. High background noise may deteriorate the quality of the microphone signals, and the desired signal may be reverberated due to room acoustics. The task of a beamformer is to selectively pick up signals impinging from a predefined direction, the so-called steering direction. In this seminar, the most recent adaptive beamformer structures should be considered by means of literature study. A comparison of different microphone array topologies used for beamforming should be considered and the spatial sampling of a sound field should also discussed. A discussion of integrating beamformer with other acoustic systems (like echo canceller) is also preferable.
Acoustic Artifacts Imposed by Speech Enhancement Systems
Communication systems such as ICC systems or communication units in firefighter breathing mask are implemented to enhance communication between two or more persons or parties. Often, such systems have to deal with several challenges at one time. Therefore, some unwanted effects may occur. The aim of this seminar is to review the mentioned systems regarding possible unwanted effects.
Neural Network-based Optimal Step-Size Estimation
Normalized least mean squares (NLMS) based adaptive filters are used in many applications such as for echo cancellation in hands-free telephony. Here, a fast convergence and a good accuracy is mandatory for the system's overall performance. The speed of convergence depends on the so-called step-size, where a large step-size means a fast convergence and a small step-size leads to a slow convergence but a better steady state performance. Whenever a fixed step-size is chosen, a tradeoff has to be made. However, there are a lot of adaptive step-size approaches, which are mostly based on an estimation of the undisturbed error signal. Neural networks are versatile, so they are used for many different applications. In this work, neural network-based approaches for the optimal step-size estimation should be investigated by means of a literature research.
Voice Tremor Investigation
A tremor can occur as a symptom of various diseases and can be studied to distinguish them. In addition to body parts such as the hands, arms or the head, whose tremors can be measured with an acceleration sensor, a voice tremor can also occur. This seminar investigates how different diseases can be classified using voice tremors. First, the features of different voice tremors should be discussed. Subsequently, classification algorithms used in the literature to differentiate between different diseases should be investigated.
Simulation of Motor Unit and Nerve Potentials and Typical Model Parameters
Electroneuography (ENG) is the current gold standard of nerve assessment to investigate different signals of the nerve in the clinical environment. The electrical signal of peripheral nerves is quite low. Peak amplitudes in the mV range can be observed by using an external electric stimulation of the nerve. The electrical signal produced by the muscles also lies in the mV range. Different simulations are already published with different model parameters to access the individual signals of nerve fibers. This helps to simulate different human nerve diseases. Available simulations and also typical simulation parameters should be investigated from a signal analysis point of view in this seminar.
Neural Networks for Noise Reduction
Neural networks can be used for different applications, for instance classification of different diseases. Besides these typical approaches, neural networks can also be used for the reduction of noise in measured signals. In this seminar, an overview about attempts using neural networks for noise reduction should be considered by means of literature study. Finally, a comparison of the different attempts should be given.
Non-linear Drift Removal to Estimate Gait Kinematics
Inertial measurement units (IMUs) are combinations of 3D accelerometers and 3D gyroscopes that nowadays are often used for unconstrained human movement analysis. The naïve method to obtain position data and/or joint angles from the sensors is to integrate measured accelerations and angular velocities. However, IMUs suffer from time-varying and temperature-dependent offset, therefore drift arises in the final results of (double) integration. A typical drift removal technique is to subtract a linear fit from the integrated data. In this seminar, you will research about different drift removal techniques with an emphasis on non-linear methods.
Achieving Orthogonality under Constraints Preset by given Hardware Limitations of SONAR Systems
The essential advantage of MIMO SONAR systems yields from the possibility to transmit nearly perfectly orthogonal sequences. Given the assumption that all transmitted orthogonal sequences can be recovered at the receive side, this leads to a more sophisticated processing and more flexibility e.g. due to “offline” changeability of “transmit” beamforming directions.
In this seminar, first orthogonality should be specified and based on this, possible (not necessarily perfect) techniques to achieve orthogonality of signals should be examined and discussed. All investigations should be done under the consideration of given hardware/software limitations. Conclusively, the impact of non-perfect orthogonality should be considered given previous findings.
Tracking of Multiple Targets using Joint Probabilistic Data Association
Target tracking in both SONAR and Radar applications make use of the Probabilistic Data Association (PDA) method. The method is suitable for tracking of one or more targets, in case they are well separated. However, if there are several target objects, events can occur in which a measurement is associated with several tracks. This is exactly the case when the measurement is in the gating range of two or more targets. In order to master these events, the Joint Probabalistic Data Assiciation (JPDA) method is often used. As part of this seminar topic, the last five scientific papers on this topic are to be found and summarized into one paper. For the beginning, some papers of renowned authors will be provided.
Comparison of Underwater Communication Modems
The underwater acoustic channel is one of the most challenging channels for communication. Because of damping and dispersion, high-frequency communication is not usable for longer distances underwater. Nevertheless, underwater modems are frequently used for monitoring purposes, controlling AUVs or speech communication, using comparably low data-rates. Nearly all commercially available modems use different modulation schemes and are capable of different data-rates. In this seminar the used modulation schemes, the data-rates and transmit power of the available modems should be examined and discussed. Conclusively, the robustness and the possible range of the modems should be addressed and a recommendation for a suitable modulation scheme for a short to mid-range speech communication should be given.
Multi-Channel Wiener Filter
In current systems for localization of underwater objects multiple receiving elements are used to achieve a certain directivity for detection. These receiving elements are arranged in an array and therefore exposed to coherent noise like waterflow and mechanical vibrations. While classic delay-and-sum-beamformers don't accommodate for these noise sources, a multi-channel Wiener filter will take these disturbances into account and will obtain a more robust solution. The goal of this seminar is to find and summarize literature about the multi-channel Wiener filter and it's application in beamforming purposes.