Use this URL to cite or link to this record in EThOS: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.630445
Title: Real time model adaptation for non-linear and non-stationary systems
Author: Chen, Hao
Awarding Body: University of Reading
Current Institution: University of Reading
Date of Award: 2013
Availability of Full Text:
Access through EThOS:
Abstract:
This thesis studies the modelling for the non-linear and non-stationary systems. In a highly non-stationary environment, not only the model coefficients but also the model structure should be adapted with time. A number of novel on-line modeling approaches are proposed in this thesis. The proposed approaches are validated using several benchmark signal processing applications including time series prediction, noise cancellation and channel equalization. First, a novel tunable radial basis function network is proposed. in which the number of nodes (or the model size) of the network is fixed and a new structured node is used to replace the worst performing node whenever the current network does not fit the input data. Two schemes are proposed to optimize t.he structure of the new node: a powerful version based on the quantum particle swarm optimization algorithm and a fast version based on the "gradient search" approach. Secondly, a new online multiple modelling approach is proposed for nonstationary systems. The proposed multimodel approach is based on two level structures of linear sub-models. The advantage of the proposed method is that it is very fast, making it particularly suitable for real time applications. Finally a new adaptive channel equalizer is developed based on minimum biterror- rate. A key issue in the minimum bit-error-rate equalizer is how the probability density function of an associated signed decision variable can be estimated on-line. In the proposed equalizer, a novel online probability density function based on Gaussian mixture model is derived, which has significant better performance than existing approaches.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.630445  DOI: Not available
Share: