|
Traditional signal classification typically involves some classifier operating on features of a known signal. Consider, however, the task of classifying a signal that has been corrupted through an unknown linear time-invariant (LTI) system. We are currently researching well-posed alternatives to blind deconvolution that involve joint estimation of a clean signal as well as a class label. This research has a variety of applications, including passive acoustic tomography and classifying marine mammals vocalizations in shallow water. In previous work, we showed how an LTI model and cepstral deconvolution could be used to reconstruct information from THz signals (see the THz project page). This work is funded by the 2007 ONR Young Investigator Award Program. Publications: "Joint Deconvolution and Classification with Applications to Passive Acoustic Underwater Multipath," Hyrum Anderson and Maya R. Gupta, Journal of the Acoustical Society of America , vol. 124, no. 5, 2973-2983, 2008. "QDA classification of multipath-corrupted observations using uncorrupted training features," Hyrum Anderson and Maya R. Gupta, Proc. Intl. Symp. on Underwater Reverberation and Clutter, 2008. "Joint deconvolution and classification for signals with multipath," Maya R. Gupta, Hyrum S. Anderson, and Yihua Chen, Proc. of the IEEE ICASSP Conf. , 2007. "Maximum likelihood signal classification using second-order blind deconvolution probability model," Maya R. Gupta and Hyrum S. Anderson, Proc. of the IEEE Statistical Signal Processing Workshop, 2007. Related Work: |
|
 Back to Projects |
||