Trends in Signal Processing

Organizers: Jonathan MANTON and Liuqing YANG
Session Chair: Chong-Yung CHI

This special session hosts six leading international researchers, each invited to present an overview of an area of signal processing of his choosing. Each talk will be approximately 20 minutes long, followed by 10 minutes of questions from the audience.

Topic and Abstract
Chong-Yung Chi,
National Tsing Hua University
Topic: Convex Geometric Analysis for Non-negative Blind Source Separation
Abstract: Despite the tremendous advances made in imaging methodologies and equipments, often most of the real world observations are mixtures of the true sources. Blind Source Separation (BSS) is a signal processing methodology to extract the sources from the mixed observations, devoid of (or with very limited) prior knowledge about the sources and how those sources are mixed in the observations. Inherently, many of the real world sources and their observations are non-negative in nature (e.g., imaging applications such as biomedical imaging, hyperspectral imaging, micro-array data etc.), and thereby naturally leading to a specific class of BSS, namely non-negative BSS (nBSS). Till date, many useful nBSS algorithms, centered around independent component analysis (ICA) and non-negative matrix factorization (NMF), have been reported. Originated from philosophies that are different from successfully developed nICA and NMF algorithms, there is a branch of nBSS methods exploiting the intriguing convex geometry of the sources. In this talk, we intend to give a historical review of nBSS methods, and present some representative and state-of-the-art convex geometry based nBSS (CG-nBSS) algorithms in a unified perspective manner, together with novel experimental results for some practical applications. Finally, we conclude with the research trend of nBSS for solving hidden challenges and bottlenecks in related science and engineering applications.
Robert Cui,
Texas A&M University
Topic: Big Data Oriented Network Information Processing
Abstract: In this talk we will focus on large-scale signal processing and information delivery over networks. We will start with the highlights of some key challenges, and argue that one promising solver is distributed information processing. Afterwards, we will instantiate the problem with several concrete examples namely, the large-scale distributed estimation problem, distributed detection problem, and distributed storage problem. Then we will open the floor for discussions.
Tan Lee,
The Chinese University of Hong Kong
Topic: Robust Pitch Estimation for Speech and Music
Abstract: Pitch is an important attribute of auditory signals, including speech and music. In human speech communication, pitch carries abundant information, which could be linguistic, paralinguistic and non-linguistic in nature. For music signals, pitch is the most prominent and ubiquitous feature in almost all aspects. Pitch estimation refers to the process of automatically determining the fundamental frequency (F0) of an acoustic signal. It can be done by exploiting the time-domain waveform periodicity, and/or the frequency-domain harmonicity of the signal. These signal characteristics would be greatly distorted when the signal is contaminated by background noise or interfered by other sound sources. Effective approaches to robust pitch estimation include the use of multiple and complementary pitch representations, the incorporation of human auditory processing mechanisms, and the use of prior signal and noise models with advanced machine learning techniques. In this talk, we will discuss the major challenges of robust pitch estimation for speech and music signals in both single-source and multi-source scenarios. By comparing and evaluating the state-of-the-art algorithms, we attempt to identify the key directions for future research in this area. New applications of auditory pitch analysis will also be presented.
Ta-Sung Lee,
National Chiao Tung University

Topic: Signal Processing for Next Generation Mobile Broadband Communications

Abstract: Signal processing has been an active research area in mobile communications in the past two decades. In recent years, there has been an explosive growth of signal processing research addressing different aspects of next generation mobile broadband communications to meet new challenges such as super high spectrum efficiency, real-time access, and uniform user experiences in a cell. In this talk, we will briefly review the current status of the LTE-Advanced standard and identify several challenges as well as research opportunities for signal processing in the development of future mobile broadband technologies. In particular, challenges and potential solutions for heterogeneous networks, interference management and large scale multi-antenna systems will be addressed.

Xiang-Gen Xia,
University of Delaware and Chonbuk National University

Topic: Robust Remaindering and Signal Processing

Abstract: Robust remaindering problem is how to robustly determine a large integer from its erroneous remainders. This problem has many applications including frequency determination from multiple undersampled waveforms, such as phase unwrapping in SAR imaging of moving targets. When the remainders are error free, Chinese remainder theorem (CRT) provides a solution. However, it is well-known that CRT is not robust. This talk is about the latest developments on this topic.

Abdelhak Zoubir,

Technische Universität Darmstadt

Topic: Robust Statistics for Signal Processing  

Abstract:Statistical signal processing often relies on strong and precise assumptions, e.g. optimal estimators, detectors and filters are derived based on a particular parametric model or a probability distribution of the signal and/or noise or interference. Optimality, however, is only achieved when the underlying assumptions hold, and the performance of optimal procedures deteriorates significantly, even for minor departures from the assumed model. Measurement campaigns have revealed the presence of heavy tailed or impulsive interference, which can cause conventional signal processing techniques, especially the ones derived using the nominal Gaussian probability model, to be biased or to even break down. The occurrence of impulsive interference has been reported, for example, in outdoor mobile communication channels, due to switching transients in power lines, and in radar and sonar systems as a result of natural or man-made electromagnetic and acoustic interference. Moreover, impulsive interference occurs in biomedical sensor array measurements of the brain activity (MRI) in various regions of the human brain, where a complex tissue structure is known to exist. In geolocation position estimation and tracking, non-line-of-sight signal propagation, caused by obstacles such as buildings or trees, results in outliers in the measurements, to which conventional position estimation methods are very sensitive. Consequently, in these situations, there is a need for robust statistical signal processing methods. With the increase of complexity of engineering system design and the lack of predictability of natural as well as man-made interference, systems that are resistant to deviations from the assumed model are more important than ever before.