Luca Cazzanti's UW Web Site

Applied Physics Laboratory, University of Washington

CALL FOR PAPERS - SPECIAL SESSION
Learning from pairwise relationships: non-metric proximities, indefinite kernels, and graphs

I am an engineer in the Environmental & Information Systems Department. I am also an affiliate assistant professor with the Dept. of Electrical Engineering. I work on various projects developing algorithms in machine learning, signal processing, digital communications, and acoustics. A sample of my projects is below.

Similarity-based statistical learning - Basic research into new statistical learning architectures that can recognize patterns based on the pairwiase similarity of samples and effectively fuse many disparate data sources. This research seeks answers to two main questions: How does one classify based on general similarities that often are non-metric? How does one measure similarities between complex objects characterized by heterogenous features? To answer the first qestion, I am developing similarity discriminant analysis (SDA), a generative framework for similarity-based classification. Research articles and research-grade Matlab software for SDA are available. To answer the second question, I am researching flexible heterogeneous similarity measures based on information theory and psychology.

Robust underwater acoustic communication schemes - Acoustic communications waveforms propagating in the underwater channel can be severely distorted in both time and frequency, causing corruption of the received data. Variations in water depth, bottom type, sound speed profile, and source and receiver location can cause a wide range of multipath interference, which results in time spreading of the transmitted signals and consequent intersymbol interference (ISI) in the received bit stream. Different wind and surface wave conditions, in combination with platform motion, can result in varying degrees of Doppler shift and spreading. Robust modulation schemes, used in conjunction with signal processing algorithms, can limit the deleterious effects of the doubly-spread underwater acoustic channel, thus enabling low-error communications between multiple platforms.

Underwater acoustic signal processing and classification - Environmental complexities radically influence the behavior of sound waves in water, and thus affect the parameters measured by sonars. Signal processing algorithms help mitigate the environmental effects on the measured acoustic signals, and contribute to the successful development of signal detection and classification techniques. I use the Sonar Simulation Toolkit (SST), CASS, and Matlab for sonar research.

Blind demodulation and automatic modulation classification - Blind demodulation techniques, that is techniques that do not rely on a priori knowledge of signal parameters, are combined with modulation classification methods to automatically demodulate unknown signals. Blind techniques include blind equalization with the constant modulus algorithm (CMA), blind symbol rate estimation, carrier and bandwidth estimation, residual carrier and phase compensation. Modulation classification is achieved by automatic classifiers operating on features extracted from partially demodulated signals.



Home Software Publications CV
APL web site EIS Dept. web site Personal web site Contact me Email: lucagc_at_u.washington.edu Address: 1013 NE 40th Street, Seattle, WA, 98105 Campus Address: Henderson Hall, Box 355640.		Applied Physics Laboratory, University of Washington CALL FOR PAPERS - SPECIAL SESSION Learning from pairwise relationships: non-metric proximities, indefinite kernels, and graphs I am an engineer in the Environmental & Information Systems Department. I am also an affiliate assistant professor with the Dept. of Electrical Engineering. I work on various projects developing algorithms in machine learning, signal processing, digital communications, and acoustics. A sample of my projects is below. Similarity-based statistical learning - Basic research into new statistical learning architectures that can recognize patterns based on the pairwiase similarity of samples and effectively fuse many disparate data sources. This research seeks answers to two main questions: How does one classify based on general similarities that often are non-metric? How does one measure similarities between complex objects characterized by heterogenous features? To answer the first qestion, I am developing similarity discriminant analysis (SDA), a generative framework for similarity-based classification. Research articles and research-grade Matlab software for SDA are available. To answer the second question, I am researching flexible heterogeneous similarity measures based on information theory and psychology. Robust underwater acoustic communication schemes - Acoustic communications waveforms propagating in the underwater channel can be severely distorted in both time and frequency, causing corruption of the received data. Variations in water depth, bottom type, sound speed profile, and source and receiver location can cause a wide range of multipath interference, which results in time spreading of the transmitted signals and consequent intersymbol interference (ISI) in the received bit stream. Different wind and surface wave conditions, in combination with platform motion, can result in varying degrees of Doppler shift and spreading. Robust modulation schemes, used in conjunction with signal processing algorithms, can limit the deleterious effects of the doubly-spread underwater acoustic channel, thus enabling low-error communications between multiple platforms. Underwater acoustic signal processing and classification - Environmental complexities radically influence the behavior of sound waves in water, and thus affect the parameters measured by sonars. Signal processing algorithms help mitigate the environmental effects on the measured acoustic signals, and contribute to the successful development of signal detection and classification techniques. I use the Sonar Simulation Toolkit (SST), CASS, and Matlab for sonar research. Blind demodulation and automatic modulation classification - Blind demodulation techniques, that is techniques that do not rely on a priori knowledge of signal parameters, are combined with modulation classification methods to automatically demodulate unknown signals. Blind techniques include blind equalization with the constant modulus algorithm (CMA), blind symbol rate estimation, carrier and bandwidth estimation, residual carrier and phase compensation. Modulation classification is achieved by automatic classifiers operating on features extracted from partially demodulated signals.