Institutional Repository
Technical University of Crete
Technical University of Crete
EN
|
EL
| URI: | http://purl.tuc.gr/dl/dias/D060CBBF-3261-4192-AD8A-AF7AF873BEA7 | ||
| Year | 1998 | ||
| Type of Item | Conference Full Paper | ||
| License |
|
||
| Bibliographic Citation | A. P. Liavas, P. A. Regalia ,J-P. Delmas.(1998).Blind channel approximation: Effective channel length determination.Presented at Asilomar Conference on Signals, Systems, and Computers.[online].Available:http://www-public.tem-tsp.eu/~regalia/personal/articles/sp-dec-99.pdf | ||
| Appears in Collections |
A common assumption of blind channel identificationmethods is that the order of the true channel is known. Thisinformation is not available in practice, and we are obliged toestimate the channel order by applying a rank detection procedureto an “overmodeled” data covariance matrix. Informationtheoretic criteria have been widely suggested approaches for thistask. We check the quality of their estimates in the context oforder estimation of measured microwave radio channels andconfirm that they are very sensitive to variations in the SNRand the number of data samples. This fact has prohibited theirsuccessful application for channel order estimation and has createdsome confusion concerning the classification into underandover-modeled cases. Recently, it has been shown that blindchannel approximation methods should attempt to model only thesignificant part of the channel composed of the “large” impulseresponse terms because efforts toward modeling “small” leadingand/or trailing terms lead to effective overmodeling, which isgenerically ill-conditioned and, thus, should be avoided. Thiscan be achieved by applying blind identification methods withmodel order equal to the order of the significant part of the truechannel called the effective channel order. Toward developing anefficient approach for the detection of the effective channel order,we use numerical analysis arguments. The derived criterionprovides a “maximally stable” decomposition of the range spaceof an “overmodeled” data covariance matrix into signal andnoise subspaces. It is shown to be robust to variations in theSNR and the number of data samples. Furthermore, it providesuseful effective channel order estimates, leading to sufficientlygood blind approximation/equalization of measured real-worldmicrowave radio channels.
Copyright © DIAS 2013
