Use this URL to cite or link to this record in EThOS: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.664232
Title: Blind deconvolution : techniques and applications
Author: Zheng, Fu-Chun
Awarding Body: University of Edinburgh
Current Institution: University of Edinburgh
Date of Award: 1992
Availability of Full Text:
Access through EThOS:
Full text unavailable from EThOS. Please try the link below.
Access through Institution:
Abstract:
This thesis is primarily concerned with developing new parameter based blind deconvolution algorithms and studying their applications. The blind deconvolution problem for minimum phase (MP) systems is well understood, and in this case the well known predictive schemes can be employed. When systems are nonminimum phase (NMP), however, the predictive deconvolution methods can only generate the spectrally equivalent MP solution. This is because the predictive schemes are based only on autocorrelations, which are completely blind to the phase properties of systems. In order to solve the blind deconvolution problem of NMP systems, higher order cumulant (HOC) analysis is adopted in this thesis. The reason for this is that HOC carry the phase information of systems only to a linear phase shift, the parametric approach is adopted due to its advantages in terms of variance and resolution over nonparametric methods. Both MA and AR based models are studied in this work. A new robust blind deconvolution algorithm for MP systems: variance approximation and series decoupling (VASD), is presented first. It is shown that this algorithm possesses some advantages over the existing ones with the same purpose. Then, based on a MA system model, we proposed a HOC-based two-step relay algorithm, in which the close-form formula for MA parameters are combined with an optimal fitting scheme, and the thorny problem of multimodality is overcome to a very great degree. Thus, the optimal identification of the MA parameters of NMP systems can be obtained. In the study of the AR based model, six new families of HOC based linear equations with respect to the AR parameters are derived. Since the inverse filter coefficients are simply the solution of a set of linear equations, their uniqueness can normally be guaranteed. In comparison with the existing AR based methods, only diagonal slices of cumulants are used in our algorithms, in which simplicity and elegance are fully embodied. It has been shown that our algorithm can offer more accurate results than the existing ones.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.664232  DOI: Not available
Share: