Chair: John Treichler, Applied Signal Technology (USA)
A.A. Rontogiannis, University of Athens
S. Theodoridis, Computer Technology Institute (GREECE)
This paper presents two new, closely related adaptive algorithms for LS system identification. The starting point for the derivation of the algorithms is the inverse Cholesky factor of the data correlation matrix, obtained via a QR decomposition (QRD). Both are of $O(p)$ computational complexity with $p$ being the order of the system. The first algorithm is a fixed order QRD scheme with enhanced parallelism. The second is a lattice type algorithm based on Givens rotations, with lower omplexity compared to previously derived ones.
Peter Strobach, Fachhochschule Furtwangen (GERMANY)
A class of adaptive filters based on sequential eigen decomposition
of the data covariance matrix is presented. These new algorithms
are completely rank-revealing and hence they can perfectly handle
the following two relevant data cases where conventional RLS methods
fail to provide satisfactory results: (1) Highly oversampled "smooth"
data with rank deficient or almost rank deficient covariance matrix.
(2) Noise-corrupted data where a signal must be separated effectively
from superimposed noise. This paper corrects the widely held belief
that eigen-based algorithms must be computationally more demanding
than conventional RLS techniques. A spatial RLS adaptive filter has
a principal complexity of O(N*N) operations per time step. Somewhat
ironically, though, the corresponding new eigen subspace adaptive
filter requires only O(N*r) operations per time stepwhere r
George-Othon Glentis, University of Twente (THE NETHERLANDS)
In this paper a highly modular normalized adaptive lattice algorithm
for multichannel Least Squares FIR filtering and multivariable system
identification, is presented. Multichannel filters with different
number of delay elements per input channel are allowed. The main features
of the proposed multichannel adaptive lattice least squares algorithm
is the use of scalar only operations, multiplications/divisions and
square roots, and the local communication which enables the development
of a fully pipelinable architecture.
J. Tuthill, Curtin University of Technology (AUSTRALIA)
In many adaptive signal processing applications, it is often desired
to impose linear and quadratic constraints on the adaptive filter
weights in order to meet certain performance criteria. This paper
presents a modification of the well known adaptive RLS or FLS algorithm
to achieve this. By way of illustration, the paper considers an adaptive
narrowband beamformer with first and second order spatial derivative
constraints. The performance of the algorithm is studied via computer
simulations.
Paulo S.R. Diniz, Federal University of Rio de Janeiro (BRAZIL)
A detailed analysis of the QR-RLS algorithm in finite and infinite
precision implementations is presented, emphasizing the case where
the input signal samples are correlated. The expressions for the mean
square values of all internal variables in steady state are first
derived. These expressions are key to determine the dynamic range
of the internal signals, and to derive the analytical expressions
for the mean square values of the deviations in the output variables
of the algorithm in finite wordlength implementations. Previous works
address this problem considering the input signal a white noise, a
situation not so often encountered in practice. The accuracy of all
analytical results are verified through a number of computer simulations.
M. Milosavljevic, University of Belgrade (YUGOSLAVIA)
In this paper a new method of estimation time-varying AR models using
weihted recursive least square algorithm with variable forgetting
factor is described. The variable forgetting factor is adapted to
a nonstationary signal by a generalized likelihood ratio algorithm
through so-called discrimination function which gives good measure
of nonstationarity. In this way we connect results from nonstationary
signal estimation and jump detection area and obtain algorithm which
exhibits good tracking performance together with parameter estimation
accuracy. The feasibility of the approach is demonstrated with both
simulation data and real speech signals.
Geoffrey A. Williamson, Illinois Institute of Technology
Adaptive IIR filters implemented in cascade-form are attractive due
to the ease with which their stability may be monitored. In this paper,
four cascade-form structures are compared for use in adaptive filtering
with respect to complexity of implementation, error surface geometry,
and adaptation speed. The four structures include a cascade of second
order pole/zero sections, a cascade of second order all-pole sections
followed by a tapped delay line and two new structures introduced
by this paper. The latter pair includes a tapped cascade, which is
a cascade of second order all-pole sections whose output is constructed
as a weighted combination of signals tapped from the cascade. The
second new structure is a modification of the tapped cascade that
yields orthogonal signals at the taps of the cascade. It is shown
that the tapped cascade provides the best overall performance in the
respects noted above.
Robert Shorten, Daimler-Benz Research (GERMANY)
The recursive least squares (RLS) algorithm with exponential forgetting
((lambda)RLS) is perhaps the best known and most widely used algorithm
for tracking the time varying parameters of a linear regression model.
The implicit assumption in using the (lambda)RLS algorithm is that
the information is uniformly distributed over the time horizon. Frequently
this assumption does not hold and seriousdifficulties can be encountered
when using many model structures. These include convergence of the
parameters to local system or noise characteristics and output bursting,
i.e. a large error when the operating point changes. In this paper
several simple alternatives to the standard (lambda)RLS algorithm
are proposed.The proposed algorithms extend the idea of a sliding
window by quantising the input space.These algorithms considerably
reduce the risk of forgetting useful information and eliminate the
possibility of output bursting by relating the adaptation capabilities
of the algorithm to the amount of input stimul
H.J.W. Belt, Eindhoven University of Technology (THE NETHERLANDS)
A second-order IIR filter is considered as the basic component of an
Adaptive Line enhancer (ALE). As a new feature, the bandwidth of the
proposed ALE is adapted simultaneously with the center frequency.
This leads to the possibility to combine convergence speed and accuracy.
The adaptation of the filter poles is controlled by a sign algorithm.
The stepsizes are chosen such that transients caused by the retuning
of the filter are ensured to remain much smaller in amplitude than
the response of the filter to the input signal. When the input signal
consists of a sinusoid corrupted by wide-band noise, an accurate frequency
parameter estimate can be obtained with an algorithm given in this
paper.
300dpi TIFF Images of pages:
Acrobat PDF file of whole paper:
TOP
A Highly Modular Normalized Adaptive Lattice Algorithm for Multichannel Least Squares Filtering
Authors:
Cornelis H. Slump, University of Twente (THE NETHERLANDS)
Volume 2, Page 1420
Abstract:
300dpi TIFF Images of pages:
Acrobat PDF file of whole paper:
TOP
Adaptive RLS Filters with Linear and Quadratic Constraints
Authors:
Y.H. Leung, Curtin University of Technology (AUSTRALIA)
I. Thng, Curtin University of Technology (AUSTRALIA)
Volume 2, Page 1424
Abstract:
300dpi TIFF Images of pages:
Acrobat PDF file of whole paper:
TOP
Analysis of the QR-RLS Algorithm for Colored- Input Signals
Authors:
Marcio G. Siqueira, Federal University of Rio de Janeiro (BRAZIL)
Volume 2, Page 1428
Abstract:
300dpi TIFF Images of pages:
Acrobat PDF file of whole paper:
TOP
Estimation of Nonstationary AR Model Using the Weighted Recursive Least Square Algorithm
Authors:
M. Dj. Veinovi, University of Belgrade (YUGOSLAVIA)
Branko Kovacevic, University of Belgrade (YUGOSLAVIA)
Volume 2, Page 1432
Abstract:
300dpi TIFF Images of pages:
Acrobat PDF file of whole paper:
TOP
Structural Issues in Cascade-form Adaptive IIR Filters
Authors:
James P. Ashley, Motorola Inc.
Majid Nayeri, Michigan State University (USA)
Volume 2, Page 1436
Abstract:
300dpi TIFF Images of pages:
Acrobat PDF file of whole paper:
TOP
Robust Parameter Tracking Through Regional Forgetting
Authors:
Andreas Schutte, Daimler-Benz Research (GERMANY)
A.D. Fagan, Daimler-Benz Research (GERMANY)
Volume 2, Page 1440
Abstract:
300dpi TIFF Images of pages:
Acrobat PDF file of whole paper:
TOP
Adaptive Line Enhancement Using a Second-Order IIR Filter
Authors:
A.C. den Brinker, Eindhoven University of Technology (THE NETHERLANDS)
F.P.A. Benders, Eindhoven University of Technology (THE NETHERLANDS)
Volume 2, Page 1444
Abstract:
300dpi TIFF Images of pages:
Acrobat PDF file of whole paper:
TOP