Sprungmarken

Servicenavigation

TU Dortmund Webseite TU Dortmund

Hauptnavigation

You are here:

Bereichsnavigation

Hauptinhalt

Research



The Information Processing Lab is engaged in designing and implementing algorithms of information technology and signal processing. The implementation of these methods in terms of certain computer-architectures and the integration as system-on-chip components is analyzed for different technical applications. While designing new methods, the interaction of algorithms and architectures becomes particular relevant but also the constraints of the concrete analyzed technical application (real time, noise, power consumption) will be incorporated.

Considered technical applications are:

  • Cooperative positioning in mobile Ad-Hoc networks
  • Positioning of RFID tags
  • OFDM systems with reduced cyclic prefix for LTE
  • Adaptive decoding of convolutional codes
  • Signal processing methods for electric power systems
  • Sparse matrix problems for network-on-hhip architectures
  • Convex optimization in communication problem
  • FPGA and ASIC Implementation of the derived methods


Biomedical Signal Processing and Classification in Body Sensor Networks

Signal processing methods have become increasingly important within the scope of biomedicine. Body-mounted sensor networks, henceforth referred to as „Body Sensor Networks“, consist of several sensor types and ensure a continuous measurement of the bodily functions. Among others, sensors for surface electromyography, electrocardiography as well as acceleration sensors or gyroscopes are used.

The data of the sensors in a Body Sensor Network are processed and combined so that the condition of a person can be evaluated with the help of classification methods. These results can be very important as far as injury prevention and assessment of the treatment process are concerned.

Induced by the further development of Body Sensor Networks, complicated measurement setups can be replaced in future. These Sensor Networks hold a lot of promise, because the collection of data can be done during everyday life independently from a test laboratory. Therefore the medical assessment of the general health condition can be done more precise and in the long term.

The goal of our studies is the development of tools for a skillful classification of biomedical output signals. In this context, the combination of the individual sensor signals in a Body Sensor Network plays an important role. In contrast to an isolated inspection of one signal, the efficiency of the results can be enhanced by involving more sensors.

Before regarding the interplay or classification of the signals, the data coming from the sensors must be processed first. However, the separation of useful information and artefacts is rather challenging. The disturbing signals are of different origins and have a serious influence on the sensor output signals. The varying sweat secretion during a measurement, which leads to a shift of the skin resistance, can be seen as an example of this. Therefore the aim is to eliminate these artefacts with signal processing techniques. In addition to this filtering procedure, the extraction of several features(e.g. standard deviation, mean value, median frequency, subspaces, etc.) is inevitable for the classification .

Finally the mapping of these extracted features to previously defined classes allows an assessment of the regarded system(human being).

Adaptive Modeling and Real-Time Identification of Transcontinental Energy Transmission Systems

Electrical transmission networks in Europe are operated closer to their allowable limits as a result of liberalization and fluctuating feed by renewable energies. To ensure the reliability and safety of the operation and to avoid large-scaled Blackouts, an interdisciplinary research group founded by the DFG is searching for profitable methods for the Evaluation of Network Status, Decision Logics on Countermeasures and Simulative Evaluation of Protection and Control Systems. Research focuses on the latest statistical and data-processing methods in context of transcontinental energy systems.

Static Clustering such as Spectral Clustering provides the opportunity to devide network models in topologically connected regions. Dynamic Clustering devide networks in similar sections based on Stability Indicators of synchronized measured values. For the stability analysis an ARMAX model is created for a node-specific eigenvalue analysis.The results can be used for the Decision Logic of Protective Functions considering Subnetworks and for creating Adaptive Models for the analysis of low-frequency oscillations.

Condition monitoring for Lithium accumlators

In the turn from fossile to regenerative energies, the electification of vehicles is currently one of the important tasks. In that context, especially lithium ion batteries have proven to be competitive due to its high specific energy. Due to specific characteristics of that cell chemistry, existing methods from other chemistries cannot be applied easily.

The objectives of the efforts are new methods for the condition monitoring for Lithium batteries for use in vehicles, either for online or offline measurements.

Sparse Matrix Problems for Network-on-Chip Architectures

Sparse Matrix-Vector-Multiplication (SMVM) appears in many scientific and engineering applications. Iterative methods (Jacobi method, conjugate gradient method, Lanczos method) are used to solve the underlying numerical problems, as e.g. solving sparse systems of linear equations or computing the eigenvalue decomposition of sparse matrices.

Since all these methods are based on SMVMs, various ways to speed up the SMVM on general purpose processors as well as parallel hardware structures were presented. However, all these approaches are strongly depending on a specific sparsity structure of the given matrix. Here, by using a Network-on-Chip (NoC) an approach for dealing with arbitrary sparsity structures is taken.

NoC architecture is the proposed concept for replacing the traditional bus-based on-chip interconnections by packet-based switch network architecture. Therefore, the packets (vector elements, matrix elements) can be freely distributed over the parallel hardware structure. Furthermore, there is a wide range of topologies of the NoC architectures and the used routing schemes. In this project the NoC architecture is used to deal with the highly irregular communication structure of parallel SMVM operations. The proposed SMVM-NoC realizes a chip–internal packet–based switch network as the main transmission network for the data transfers required for the SMVM computations. Meanwhile, the concept will also be realized in FPGA prototypes as well as an ASIC implementation using TSMC 45nm technology library.