Lunchtime Strathclyde EEE Dept DSP Seminar (all welcome),
JA8.13, John Anderson Bldng (12:45-1:30pm)
The Role of Genomic Signal Processing in Systems Biology, Ilya Shmulevich
►Monday AFTERNOON tutorials (1:45pm to 4:45pm):
♦ Sparse Sampling of Structured Information (T. Blu, P.L. Dragotti, P. Marziliano, M. Vetterli)
♦ Biometric Authentication: Theory, Algorithms and Emerging Applications (A. Drygajlo)
♦ Robust Statistics (A. Zoubir)
►Monday AFTERNOON industry presentations (1:45 to 4:45pm):
♦ Advanced DSP design with SystemVue 2009 (Agilent EEsof)
Tutorial Abstracts
Tutorial 1: Image and Video Coding - From Principles to Systems
Bernd Girod, Stanford University
Image and video coding has become one of the most ubiquitous signal processing technologies, enabling everything from digital photography to graphically rich web pages, from video streaming to digital cinema, from digital TV broadcasting to HD optical disks. Based on a graduate-level course taught in the Stanford Electrical Engineering Department, this tutorial reviews the most important algorithms used in image and video coding, emphasizing the underlying rate-distortion principles. Topics discussed range from entropy coding, transforms, and wavelet decompositions to motion estimation and compensation. We will put into context various coding standards, such as JPEG-2000 and the family of MPEG standards. The tutorial should be of interest to industry practioners, academic researchers and graduates students working in the area, as well as newcomers seeking a comprehensive overview of the field.
(back to top)
Tutorial 2: Statistical Methods for Single and multi-Pitch Estimation
Mads G. Christensen, Aalborg University
Andreas Jakobsson, Lund University
Periodic signals can be decomposed into sets of sinusoids having frequencies that are integer multiples of a fundamental frequency. The problem of finding such fundamental frequencies from noisy observations is important in many speech and audio applications, where it is commonly referred to as pitch estimation. In this tutorial, an introduction to pitch estimation is given and a number of statistical methods for pitch estimation are presented and discussed. The basic signal models and associated estimation theoretical bounds are introduced, and the properties of speech and audio signals are discussed and illustrated. The presented methods include both single- and multi-pitch estimators based on statistical methods, filtering methods, and subspace methods.
(back to top)
Tutorial 3: Generalized Discrete Fourier Transform: Non-Linear Phase DFT for Improved Multicarrier Communications
Ali Akansu, New Jersey Institute of Technology
Recently, the Generalized DFT (GDFT) exploiting the entire phase space has been forwarded in the literature as an extension to DFT. GDFT with non-linear phase functions has reduced correlations compared to DFT. GDFT framework is a powerful mathematical tool to design optimal constant modulus sets adaptively tracking channel variations in order to minimize BER degradations due to ISI, ICI and PAPR charasteristics. We will show that GDFT based OFDM methods significantly outperform the widely used DFT based systems. We will also present design methods offering computationally efficient implementations of GDFT as a low cost modification to the celebrated FFT algorithms.
(back to top)
Tutorial 4: Sparse Sampling of Structured Information
Thierry Blu, Chinese University of Hong Kong
Pier Luigi Dragotti, Imperial College London
Pina Marziliano, Nanyang Technological University
Martin Vetterli, EPFL
The problem of reconstructing or estimating partially observed or sampled signals is an old and important one, and finds application in many areas of signal processing and communications. Traditional acquisition and reconstruction approaches are heavily influences by the classical Shannon sampling theory which gives an exact sampling and interpolation formula for bandlimited signals. Recently, the classical Shannon sampling framework has been extended to classes of non-bandlimited structured signals, which we call signals with Finite Rate of Innovation. In these new sampling schemes, the prior that the signal is sparse in a basis or in a parametric space is put to contribution and perfect reconstruction is possible based on a set of suitable measurements. This leads to new exact reconstruction formulas and fast algorithms that achieve such reconstructions. The main aim of this tutorial is to give an overview of these new exciting findings in sampling theory. The fundamental theoretical results will be reviewed and constructive algorithms will be presented, both for 1-D and 2-D signals. We also discuss the effect of noise on the sampling and reconstruction of structured signals. Finally a diverse set of applications of these new concepts will be presented to emphasize the importance and far reaching implications of these new theories.
(back to top)
Tutorial 5: Biometric Authentication: Theory, Algorithms and Emerging Applications
Andrzej Drygajlo, EPFL
This tutorial provides an ample coverage of theoretical and applied state-of-the-art research work as well as new trends and directions in the biometrics field. It offers attendees a thorough understanding of how core signal processing and pattern recognition building blocks of a biometric authentication system are developed, implemented and tested. While this tutorial covers a range of biometric traits including face, fingerprint, iris, hand palm/geometry, vein structures, dynamic signature and voice, its main emphasis is placed on the generic chain of processing and statistical/probabilistic methods used, starting from signal sensing and its quality estimation, passing through features extraction and their statistical modelling for template creation and ending at classifier/decision stage of single-classifier, multi-classifier and multimodal systems.
(back to top)
Tutorial 6: Robust Statistics for Signal Processing
Abdelhak Zoubir, Universität Darmstadt
The tutorial concerns robust statistics and their use in signal processing. Robust statistics continue to gain importance due to an increase of impulsive measurement environments and outliers in practical engineering systems. Classical estimation or detection theory does not apply in such situations and robust methods (in the statistical sense) are sought for. The tutorial aims at equipping the attendee with the most fundamental concepts of robust statistics and at showing their power to solving signal processing problems. First, we highlight the motivation for using robust statistics in real-life situations and how robust statistics can be expected to remedy problems in such practical systems. After a brief overview of concepts from classical estimation theory, including Maximum Likelihood (ML) estimation, we focus on robust estimation. We first introduce the qualitative and the quantitative definitions of robustness and treat in detail Huber’s robust M-estimator (ML-type estimator). We show how robust M-estimators for location and scale are constructed. Joint estimation for location and scale and the estimation of covariance and correlation matrices are also discussed. We then discuss semi-parametric adaptive estimation and give examples of its use. The theoretical treatment is followed by three application examples. First, we discuss a robust method for filtering in image processing. Then, we introduce various robust multi-user detectors for wireless communications, and finally, we give an example on robust direction of arrival estimation.
(back to top)
INDUSTRY LAB PRESENTATION #2:
Advanced DSP Design with SystemVue 2009
Frank Ditore, Agilent (EEsof), Atlanta, USA
In this session, attendees will get hand-on experience using SystemVue for advanced DSP communication design. A lab of 25 machines will be fully licenced with the software, and an experienced instructor lead follow-me session will be run, allowing attendees to experience the complete design environment. SystemVue is a focused EDA environment for electronic system-level (ESL) design that enables system architects and algorithm developers to innovate the physical layer (PHY) of next-generation wireless and aerospace/defense communications systems. SystemVue also provides unique value to RF, DSP, and FPGA/ASIC implementers who rely on signal processing to deliver the full value of their hardware platforms. SystemVue replaces general-purpose digital, analog, and math environments by offering a dedicated platform for ESL design and signal processing realization. SystemVue "speaks RF", cuts PHY development and verification time in half, and connects to your mainstream EDA flow.
INDUSTRY LAB PRESENTATION #1:
High Speed FPGA DSP Design and Implementation
In this session XUP (Xilinx University Programme) will present the abridged version of the DSP Primer. Attendees will be given a suitable set of notes and lab materials and a session will be run to allow attendees to experience the easy-to-use System Generator design flow, starting with the SysGen design, using the ISE tools, then finally downloading to an FPGA Virtex board. The design example will be a QAM based digital transceiver featuring downvconverters, NCOs and timing circuits. All university professors and academic staff attending and completing this session will be elligible at the end to apply for support under the Xilinx University Programme, where as appropriate XUP can make available sofware licences, extensive DSP teaching materials (including slides, labbooks and realtime examples), and in some cases provide support for acquiring FPGA hardware for teaching and research.
Strathclyde Lunctime DSP Seminar
The Role of Genomic Signal Processing in Systems Biology
Dr Ilya Shmulevich
It is now possible to interrogate biological systems with a broad spectrum of powerful measurement technologies that generate high dimensional data on genomic, transcriptomic, and proteomic levels. Such data may take the form of symbolic signals derived from nucleic acid or protein sequences, high-dimensional temporal measurements of transcriptional (mRNA or microRNA) and protein expression, genome-wide measurements of protein-DNA interactions or chemical modifications (e.g. DNA methylation, histone acetylation), population-wide measurements of single-cell intracellular signaling, and others. Frequently, there are class labels or response variables associated with such data sets, which may be determined by a phenotype (cancer grade or subtype, cellular growth rate) or clinical outcome (response to a drug, survival).
Complex dynamical biomolecular systems govern virtually all biological processes, on time scales ranging from development to physiology. A paramount problem, and a central goal of systems biology, is to understand the structural and dynamical properties of such systems and their role in cellular function and dysfunction. The integration of information from heterogeneous measurement data in conjunction with model inference approaches using a variety of modeling formalisms, including systems of ordinary differential equations, nonlinear discrete dynamical systems, and their probabilistic or stochastic counterparts, is making it possible to predict various aspects of system behavior under environmental or genetic perturbations. In particular, the predictive nature of such models sets the stage for optimal intervention strategies intended to control system behavior, particularly in the context of disease.
At the same time, accurate and early diagnostic markers are critical to the prevention and treatment of diseases, such as cancer. Genome-wide measurements of cancer tissues, combined with statistical pattern recognition and machine learning approaches, allow us to determine sets of informative genes or proteins whose measurements may be used for prognosis or diagnosis. I will discuss a number of approaches in Genomic Signal Processing for addressing these challenges and will give examples in the areas of cancer research and immunology.
Dr. Ilya Shmulevich joined the ISB faculty in April 2005. His work focuses on the computational, mathematical, and statistical aspects of systems biology, complex systems theory, and applications of signal processing and machine learning to genomics and proteomics Dr. Shmulevich received his Ph.D. degree in Electrical and Computer Engineering from Purdue University, West Lafayette, IN, USA, in 1997. In 1997-1998, he was a postdoctoral researcher at the Nijmegen Institute for Cognition and Information at the University of Nijmegen and National Research Institute for Mathematics and Computer Science at the University of Amsterdam in The Netherlands, where he studied computational models of music perception and recognition. In 1998-2000, he worked as a senior researcher at the Institute of Signal Processing in Tampere University of Technology, Tampere, Finland. Prior to joining the ISB, he was an Assistant Professor in the Department of Pathology at The University of Texas M. D. Anderson Cancer Center and also held an Adjunct Professor position in the Department of Statistics in Rice University.
