ASU SenSIP Exposition Session

Session: ASU SenSIP Exposition

Time: Monday, May 12, 14:30-15:45
Chair: Andreas Spanias, Arizona State University

ASU-1: Multi-resolution Pattern Learning using Wavelet and Curvelet Domain Statistical Models
Karthikeyan N. Ramamurthy, Jayaraman J. Thiagarajan and Andreas Spanias Arizona State University, Tempe AZ
Multi-resolution techniques such as the wavelet and curvelet transforms have the ability to provide optimal sparse representations for 1-D and 2-D signals respectively. Multi-resolution representations have been successfully used in template learning with the assumption that the coefficients are statistically independent. In this study, we model the inter-dependencies between the wavelet coefficients with the existing contextual Hidden Markov Models (HMMs) and use that statistical model in template learning. We extend the contextual HMM to curvelet setting also and explore using the curvelet-domain statistical model in the pattern learning application.

ASU-2: Agile Sensing and Signal Processing
Antonia Papandreou-Suppappola Arizona State University, Tempe, AZ, Darryl Morrell Arizona State University Polytechnic Campus, Mesa AZ, Douglas Cochran, Ying Li, Ioannis Kyriakides, Jun Zhang, Shwetha Edla, Bhavana B. Manjunath and Thomas Trueblood Arizona State University
We present our work on agile sensing and signal processing that has resulted in significant improvements in single or multiple target tracking, high sea clutter detection, environment characterization, and non-myopic sensor scheduling.

ASU-3: Improving Connectivity and Routing in Cooperative Sensor Networks
Qinghai Gao and Junshan Zhang Arizona State University, Tempe AZ
In general, probabilistic coverage/connectivity is an under-explored area. In this focused effort, we will first investigate node cooperation for improving coverage and connectivity in sparse sensor networks. Based on the interesting idea of using mobile Mules for collecting data from sensor node in sparse networks, we will also study optimum mobility strategies of mobile Mules for improving probabilistic coverage and intermittent connectivity.

ASU-4: Signal Processing and Communications Algorithms for MIMO Underwater Acoustic Communications
Tolga M. Duman, Yunus Emre, Jatinder Bajwa, Hyunjun Kim, Sanjay Mani and Vinod Kandasamy Arizona State University, Tempe AZ

We review some of the underwater acoustic (UWA) communications work we have been carrying out for the past several years. Specifically, we are interested in multi-input multi-output (MIMO) communications techniques for shallow water UWA communications. We have considered different alternatives for MIMO communications including spatial multiplexing, grouped space-time trellis codes, and space-time block codes. More recently, we have also incorporated orthogonal frequency division multiplexing (OFDM) into our system. For error control, we have employ turbo codes at the transmitter and iterative equalization/decoding schemes at the receiver. Through extensive simulations and experiments carried out at-sea, we have been able to demonstrate that impressive data rates with very low error probabilities can be achieved over a range of 2-3 km (for which the available bandwidth is very limited). To give some examples, were able to attain data rates in the order of 50 kbps over a bandwidth of 20 kHz or so, and in some cases spectral efficiencies in excess of 4 bits/sec/Hz.

ASU-5: Algorithm-Architecture Co-Design of Signal Processing Systems
Lanping Deng, Aarul Jain, Tao Liu, Veera Papirla, Qi Qi, Aaron Williams, Chi-Li Yu and Chaitali Chakrabarti Arizona State University, Tempe AZ
Traditionally, signal processing and communication algorithms have been designed with no thought given to the efficiency of their implementation on hardware platforms. This has resulted in inefficient implementations which could have been avoided had the algorithms been tailored to match the constraints imposed by the existing architecture. We illustrate the concept of algorithm-architecture co-design through several representative examples.

ASU-6: Distributed Detection in Ultra-Wideband Wireless Sensor Networks
Cihan Tepedelenlioglu and Kai Bai Arizona State University, Tempe AZ
We consider distributed detection in ultra-wideband (UWB) wireless sensor networks over frequency-selective channel with the amplify-and-forward scheme. In particular, we are interested in the following questions: (i) What is the tradeoff between the detection performance and the feedback overhead? (ii) How to achieve the optimal performance at least asymptotically under the practical limits? (iii) How is the asymptotical optimality affected by the system bandwidth and power? To reveal some answers for the above questions, we investigate three amplify-and-forward schemes with different requirements on CSI, and compare their performances using large deviation analysis. We first derive the performance of a log-likelihood ratio detector when no CSI is available at the sensors. Then, we show that if sensors know the sum of their own multipath gain, a coherent combining alike scheme can achieve asymptotically optimal performance under some conditions.

ASU-7: A Fast Loudness Estimation Algorithm
Harish Krishnamoorthi, Visar Berisha and Andreas Spanias Arizona State University, Tempe AZ
Audio processing applications such as rate determination, bandwidth extension, compression, and noise reduction make use of loudness metrics. Most loudness estimation algorithms are computationally expensive and often not suitable for real time applications. In this paper, we present a low-complexity loudness estimation algorithm applicable to both steady and time-varying sounds. The model computes an estimate of the excitation pattern by simultaneously pruning the frequency components and detector locations. Comparative results indicate that the proposed algorithm performs consistently well for different types of audio signals at a reduced complexity.

ASU-8: A Low-Complexity Sinusoidal Modeling of Audio Based on Loudness Patterns
Harish Krishnamoorthi Arizona State University, Tempe AZ
Transform based audio coding and irrelevancy reduction techniques usually form the core components of an audio coder. Additional coding gain can be obtained through a parametric model of audio, where the incoming audio is split into sinusoidal, transient and noise objects. In this paper, we propose a novel strategy for selection of sinusoids based on loudness patterns. We also propose a multi-resolution window analysis scheme for transient detection. Experimental results indicate that the proposed approach has lower computational complexity and provides comparable performance.

ASU-9: Hidden Markov Modeling and Sensor Fusion for Structural Health Monitoring
Narayan Kovvali, Wenfan Zhou, Debejyo Chakraborty, Donna Simon, Antonia Papandreou-Suppappola and Douglas Cochran Arizona State University, Tempe AZ
We describe an algorithm for the classification of structural damage based on hidden Markov models (HMMs). Time-frequency-scale damage features are first extracted from structural data using the matching pursuit decomposition (MPD). The features are then modeled using HMMs and classification is performed in a Bayesian framework. Both discrete and continuous observation density HMMs are utilized. A sensor data fusion procedure is considered for the integration of the discriminatory information gathered by the multiple (distributed) active sensors, so as to increase overall classifier performance. The fusion is implemented using the Bayesian decision fusion approach, where local classification decisions are first made at each sensor and these are then combined at a decision fusion center. Results are presented from an example application to the classification of delamination damage in a laminated composite.

ASU-10: Speech Recognition Using Rank-Order Coding
Alexandros Kyriakides, Costantinos Pitris, Julius Georgiou University of Cyprus, Bharatan Konnanath and Andreas Spanias Arizona State University, Tempe AZ
Rank-order coding (RAC) has been recognized as a viable alternative to rate-order coding for modeling a visual system. Recently, speech recognition using rank-order networks has been evaluated and found to outperform HMMs in certain isolated word recognition experiments, particularly in the presence of noisy background. RACs have additional advantages in terms of training. In this project, we develop a simple word recognition system using rank order neural networks and evaluate its performance and computational complexity.

ASU-11: Multi-University Development and Dissemination of Java Software for DSP and Signals and Systems Courses
Venkataraman Atti, Andreas Spanias, Susan Haag, Antonia Papandreou-Suppappola, Cihan Tepedelenlioglu, Junshan Zhang Arizona State University, Tempe AZ, G. Faye Boudreaux-Bartels University of Rhode Island, Kingston RI, Mike Stiber University of Washington-Bothell, Brothell WA, Takis Kasparis University of Central Florida, Orlando FL, Philippos Loizou University of Texas at Dallas, Dallas TX and Constantinos Pattichis University of Cyprus
This collaborative project effort involves six universities, namely, Arizona State University, the University of Washington-Bothell, the University of Texas at Dallas, the University of Rhode Island, the University of Central Florida, and the University of Cyprus. We describe educational technology innovations and Java software extensions that enable the on-line software Java-DSP to be used in DSP and signals and systems courses at six different universities.

ASU-12: Analysis of the MPEG-I Layer-3 Audio Coding Algorithm
Jayaraman J. Thiagarajan Arizona State University, Tempe AZ
Digital compression of audio data is highly important due to bandwidth and storage limitations. High-fidelity and small bit-rate are the most desirable properties of the audio coding algorithms. In this project, we analyze the different functionalities of the MPEG-I Layer-3 (MP3). Furthermore, performance evaluation of the algorithm at different bit rates is reported.

ASU-13: MPEG-7 for Audio Information Retrieval
Gordon Wichern Arizona State University, Tempe AZ
The MPEG-7 standard is a collection of contest-based indexing and description techniques, that make searching for multimedia possible. MPEG-7 builds on the representation of previous standards, MPEG-1, -2, and -4, but whereas these previous MPEG standards made content available, MPEG-7 helps the user search for content. The descriptors in MPEG-7 audio are feature vectors whose MPEG-7 compliant extraction algorithms are investigated in terms of computational complexity and retrieval performance.

ASU-14: Query-By-Example for Perceptually Coded Audio Signals
Jiachen Xue Arizona State University, Tempe AZ
Query-By-Example (QBE) technique in content-based audio analysis domain aims at retrieving the most similar audio signal given the query audio signal provided by the user. Most research conducted in QBE uses a model-based approach, and lossless encoded audio files. This paper extends the research to use audio signals with perceptual coding, and analyzes how the perceptual audio coding would affect the signal features. This paper also compares the performance between the model based retrieval strategy and retrieval using a compressed based similarity measure.

ASU-15: DRM and Watermarking in Audio Files
Mahesh K. Banavar Arizona State University, Tempe AZ
In this project, we cover the basics of Digital Audio Watermarking, and some of the more widely used DRM techniques, such as Apple Fairplay the Sony OpenMG. We will demonstrate a simple watermarking example and the effect of noise on watermarked audio files. Finally, we will also explore some of the controversy and problems surrounding the DRM.

ASU-16: Nonlinear Acoustic Echo Cancellation
Tushar Gupta Arizona State University, Tempe AZ
Standard approaches to cancel echo signals model the acoustic path using a linear transfer function. Therefore, they cannot capture the effect of nonlinearly distorted components. This leads to a reduced echo attenuation impairing the quality of speech communication system. One of the main sources of nonlinear distortion is an overdriven loudspeaker that often results in memory-less soft clipping. Another source of nonlinearity is distortion introduced by different loudspeaker parts: the acoustical, electromagnetic and mechanical parts which can be modeled by Volterra filters. In this project, we survey the different methods for nonlinear echo cancellation and we simulate nonlinear Volterra filters.

ASU-17: Sparse Representations for Audio Signals and Their Application in Audio Coding
Lakshminarayan Ravichandran Arizona State University, Tempe AZ
As a combination of time and frequency modeling, sparse representations provide a means for effective time-frequency modifications of signals. The matching pursuit decomposition technique is used to obtain a sparse representation of a signal using a set of functions that form a dictionary. We use orthogonal transforms based on Modified Discrete Cosine Transform (MDCT) and Modulated Complex Lapped Transform (MCLT) FOR this analysis. We will address sparse decompositions of audio signals and their use in audio coding.

ASU-18: Blu Ray Audio Standard
Sreejana Sharma Arizona State University, Tempe AZ
Blu Ray, also known as Blu-ray Disc (BD) is the name of a next-generation optical disc format. The format was developed to enable recording, rewriting and playback of high-definition video (HD), as well as storing large amounts of data. Blu Ray was developed as a successor to the very popular Digital Versatile Disk (DVD). For any blu Ray disc player to be compliant to the Blu Ray standard it must be able to decode and play DTS, Dolby Digital Plus, and LPCM coded bit streams. This project deals with a review of the Blu Ray standard with special emphasis on the mandatory audio coder algorithms.

ASU-19: Adaptive, Real-Time, Software-only Video Compression and Transmission
Wei-Jung Chien Arizona State University, Tempe AZ, Glen Abousleman General Dynamics, Scottsdale AZ and Lina Karam Arizona State University
A real-time, software-only video compression and transmission system, developed jointly by Arizona State University and General Dynamics C4 Systems, will be presented. The presented system has been commercialized as SelectFocusVIDEO (www.gdc4s.com/selectfocus) by General Dynamics. It can operate over ultra-low-bandwidth links, with a much lower bandwidth than the ones supported by current video compression standards, in addition to higher bandwidth links, including IP-based networks. The presented system allows on-the-fly adjustment of coding parameters and incorporates a sophisticated bandwidth throttling mechanism, which enables it to be tailored to a wide variety of applications. SelectFocusVIDEO will be demoed live and in real-time.