Organizer:
Witold
Kinsner
A. Signals and Computing
From the perspective of applied and industrial mathematics, a signal
is an electrical representation of various ubiquitous, quantifiable
physical variables such as temperature, pressure, flow, and light
intensity. Such signals can be either analog (continuous with very
high resolution), or discrete (sampled, but very high resolution),
or digital (sampled and quantized to a finite resolution), or boxcar
(step-wise analog) functions over time (e.g., speech), space (e.g.,
images) or both (e.g., volumetric Doppler radar). The digital signals
are of particular importance to computer-based signal processing
which deals with the modelling, analysis/synthesis, feature extraction,
and classification of such signals in order to gain insight into
the underlying physical process, or to perform specific control
tasks with the process.
Signal processing is used in nearly all fields of human endeavour
from signal detection in the presence of noise, to fault diagnosis,
advanced control, audio and image processing (restoration, enhancement,
segmentation, reconstruction, coding, compression), communications
engineering, intelligent sensor systems with reconfigurable architectures,
business, and humanistic intelligence (HI) [1] which utilizes the
natural capabilities of the human body and mind, as well as cognitive
informatics (CI) [2].
B. Autonomic Computing
In this theme, signal processing is also put in the context of the
emerging autonomic computing (AC) [3,4] systems which are evolving
earnestly because the cost-performance of hardware improvements
(speed and capacity) have lead to escalating complexity of software
(features and interfaces). However, this increased complexity requires
elaborate managing systems that are now six to ten times the cost
of the equipment itself. Autonomic computing is intended to simplify
this problem by making the systems self-configuring, self-optimizing,
self-organizing, self-healing, self-protecting, and self-telecommunicating,
thus leading to increased reliability, robustness, and dynamic flexibility.
This involves not only the traditional fault tolerant computing
(i.e., tolerating hardware and software faults), but also tolerating
various faults made by human operators and users, thus shifting
attention from the mean-time-between failures (MTBF) to the mean-time-to-recover
(MTTR) in order to make the systems more available. AC applies to
both desktop computing, portable computing, pervasive computing,
and embedded systems.
C. The Present and Future of Signal Processing for Autonomic
Computing
There are many disciplines involved in the design and implementation
of the necessary features of such self-aware AC systems. In addition
to the modern hardware/software/radio-frequency design techniques
and the statistical signal processing (SSP) [5,6], one must add
intelligent signal processing (ISP) [1] with pattern recognition,
as well as CI because of the required autonomy and interaction with
humans.
The classical SSP includes either a time-domain signal analysis,
or spectral analysis and estimation, using either parametric methods
or nonparametric methods. The traditional signal processing has
been concerned with mathematical models that are linear, stationary,
Gaussian, and local in order to simplify their analysis.
Since many real-world physical systems are time varying, complex
(high-dimensional), nonlinear, statistically nonstationary, non-Gaussian,
nonlocal, sometimes chaotic, and subjected to unwanted signals (noise),
the classical SSP must be augmented by ISP. ISP has been found to
be a more useful approach as it employs adaptation and learning
to extract the essential information from the acquired signals and
noise, without any assumed statistical models of the signals or
theirs sources. These signals no longer exhibit additive invariance
(short-range dependence), but multiplicative invariance (self-affinity
with long-range dependence). The ISP tools include supervised and
unsupervised learning through adaptive neural networks, wavelets
and their variations, fuzzy rule-based computation [7] and rough
sets, genetic algorithms, and blind signal estimation.
CI is concerned with (i) the extraction of characteristic features
from signals obtained from measurements and observations, and (ii)
the measurement and characterization of patterns (i.e., order and
correlation) in processes related to perception and cognition (i.e.,
interaction with humans).
Signals obtained from physical dynamical processes appear to be
very complex. Much attention has been given to deterministic and
stochastic linear-time-invariant (LTI) signals with a limited-bandwidth
power spectrum density and short-tail distributions, leading to
processing with finite moments. However, many physical signals are
fundamentally different from the LTI signals in that they are invariant
to scale rather than to translation [8]. Such signals have different
degrees of singularity as measured by their noninteger (fractal)
dimensions. Correlation in such signals persists from short to very
long ranges, with distributions having long tails (infinite moments).
In contrast to the well-established LTI system theory, the nonlinear
scale invariant (NSI) system theory and applications are still developing.
There is also another class of signals, the chaotic signals, originating
from nonlinear dynamical systems, such as the AC systems. Research
is being conducted to measure and characterize such systems.
This theme covers some elements of this large and diverse area.
[1] S. Haykin and B. Kosko, Intelligent Signal Processing. New York,
NY: Wiley-IEEE, 2001, 573 pp.
[2] Y. Wang, "On cognitive informatics," in Proc. 1st
IEEE Intern. Conf. Cognitive Informatics (Calgary, AB; 19-20 August
2002) pp. 34-42, 2002. {ISBN 0-7695-1724-2}
[3] A.G. Ganek and T.A. Corbi, "The dawning of the autonomic
computing era," IBM Systems J., vol. 42, no. 1, pp. 34-42,
2003.
(Available from http://www.research.ibm.com/journal/sj/421/ganek.pdf)
[4] IBM Autonomic Computing Manifesto.
(Available from http://www.research.ibm.com/autonomic/)
[5] A.V. Oppenheim, R.W. Schafer, and J.R. Buck, Discrete-Time Signal
Processing. Prentice Hall, 1999 (2nd ed.), 870 pp.
[6] J.G. Proakis and D.G. Manolakis, Digital Signal Processing:
Principles, Algorithms and Applications. Upper Saddle River, NJ:
Prentice-Hall, 1995 (3rd ed.), 1016 pp.
[7] W. Pedrycz and F. Gomide, An Introduction to Fuzzy Sets: Analysis
and Design. Cambridge, MA: MIT Press, 1998, 465 pp.
[8] G.W. Wornell, Signal Processing with Fractals: A Wavelet-Based
Approach. Upper Saddle River, NJ: Prentice-Hall, 1996, 177 pp.
Plenary
Speaker:
Simon Haykin (McMaster University)
Abstract
Invited Speakers:
Mark Alexiuk (Institute for Biodiagnostics, National Research Council
Canada) Abstract
Jeff Diamond (TRLabs Wpg) Abstract
Zahra Moussavi (University of Manitoba)
Abstract
Michael Potter (University of Manitoba) Abstract
Gabriel Thomas (University of Manitoba) Abstract
Yingxu Wang (University of Calgary) Abstract
|
|
