Achtung:

Sie haben Javascript deaktiviert!
Sie haben versucht eine Funktion zu nutzen, die nur mit Javascript möglich ist. Um sämtliche Funktionalitäten unserer Internetseite zu nutzen, aktivieren Sie bitte Javascript in Ihrem Browser.

Info-Icon This content is partly available in English

Research topic Signal Processing

Overview

The evaluation of measurement data often requires the use of signal processing, like the Fourier analysis which is used for rating spectral components. Furthermore it is necessary to reduce and interpret data. Possible techniques are Principal Component Analyse (PCA) and the use of Artificial Neural Networks (ANN). Also adapted transformations can help to extract the wanted information out of a measured signal.

Principal Component Analysis (PCA)

Large amounts of data can be a problem in the course of evaluation. One possibility of data reduction is PCA. PCA extracts properties in a mathematical manner, by using the fact that a lot of measured data is statistically dependent. The data is then transformed, so that the few independent properties can be extracted. This results in a new grid showing the largest variances and enabling conclusions about certain properties of the original data.

Artificial Neural Networks (ANN)

Artificial Neural Networks (ANN) are learning methods, which can i.e. be used to estimate not directly reachable quantities. This method is mainly utilised when a system's behaviour is complex, incomplete or can't be described exactly. By using reference- and model-processes as well as rule-based learning, different weights for the input data can be adapted.

One kind of applications uses ANN as a classification method, by i.e. adding patterns to a class. Here, a discreet factor is existent. Other applications use ANN as an interpolator between temporary quantities and offer an analogue result, process-tomography being an example.

Transformations

Transformations can often be very helpful when examining measured data. They can strengthen the wanted signal properties or can be a necessity to make them visible. For example when the determination of a material properties results in sonagram, where the typical illustration on a frequency-time scale will, due to the very short time pulses, only offer a very rough view on the frequency. Here different transformations like the Shorttime Fast Fourier-Transformation (SFFT), the Wavelet Transformation (WT) or DXP-algorithms can be very helpful. All these methods will result in errors when evaluating systems with a limited band. For this reason the Measurement Engineering Group has enhanced the DXP-algorithm. The new Enhanced-DXP (E-DXP) considers the locally frequency dependent bandwidth with a complex time window. This way the accuracy can be increased in both time- and frequency view.

The University for the Information Society