|
|
|||
| |
|
||
| Adaptive signal processing systems refer to systems for processing of data, which based on some recursive algorithm, are able to
operate in an unknown or time-varying environment or possibly both.
Hence, the basic nature of
the problem is that some elements of it is unknown and must be
learned from the data, or some components of the system is changing in an
unknown manner, and therefore must be tracked. So far most of these problems
have been formulated as linear filtering problems.
Nevertheless, adaptive filters are commonly classified
as linear or non-linear. An adaptive filter is said to be linear if
the estimated quantity of interest is computed adaptively as a linear
combination of the observable applied data. Else, the adaptive filter is
characterized as non-linear.
In this project we are studying algorithms for non-linear adaptive signal
processing. Basically, we are studying a model consisting of three sections:
a bank of linear filters followed by an array of non-linear elements and a
linear combiner. The objective is to find effective, stable algorithms for
adjusting the parameters of the system.
Presently, we are studying two applications: adaptive speech enhancement and
non-linear time series prediction.
A neural network is a large system of interconnected non-linear processing elements called neurons. Its motivation comes from the way the human brain performs its operations. Neural networks have the following characteristics: they have non-linear processing elements, they make few or no assumptions about the environment, they have learning and generalization capabilities, and they have fault tolerance. We are interested in the design of neural networks for specific application, such as: adaptive signal processing, time series prediction, clustering and pattern recognition. For more information please contact
Torbjørn Eltoft.
|
|||
|
Department of Physics, Faculty of Science University of Tromsø, N-9037 Tromsø, tel. 77 64 51 50 Responsible editor: Trond Brattli Web responsible: Svein Jacobsen; Last update: September 11th, 2002 |
|||