3:30, SAM-P8.1
CLASS-BASED IDENTIFICATION OF UNDERWATER TARGETS USING HIDDEN MARKOV MODELS
N. DASGUPTA, P. RUNKLE, L. CARIN
It has been demonstrated that hidden Markov models (HMMs) provide an effective architecture for classification of distinct targets from multiple target-sensor orientations. In this paper, we present a methodology for designing class-based HMMs that are well suited to the identification of targets with common physical attributes. This approach provides a means to form associations between existing target classes and data from targets never observed in training. After performing a wavefront-resonance matching-pursuits feature extraction, we present an information theoretic tree-based state-parsing algorithm to define the HMM state structure for each target class. In training, class association is determined by minimizing the statistical divergence between the target under consideration and each existing class, with a new class defined when the target is poorly matched to each existing class. The class-based HMMs are trained with data from the members of its corresponding class, and tested on previously unobserved data. Results are presented for simulated acoustic scattering data.

3:30, SAM-P8.2
UNIFORMLY MOST POWERFUL CYCLIC PERMUTATION INVARIANT DETECTION FOR DISCRETE-TIME SIGNALS
F. NICOLLS, G. DE JAGER
The uniformly most powerful invariant (UMPI) test is derived for detecting a target with unknown location in a noise sequence. This test has the property that for each possible target location it has the greatest power of all tests which are invariant to cyclic permutations of the observations. The test is compared to the generalised likelihood ratio test (GLRT), which is commonly used as a solution to this detection problem. Monte-Carlo simulations show that the powers of the two tests are comparable, thereby justifying near-optimality of the GLRT.

3:30, SAM-P8.3
TIME AND SCALE EVOLUTIONARY EVD AND DETECTION
N. ERDOL, S. KYPEROUNTAS, B. PETLJANSKI
Optimal detection of a known signal in nonstationary noise requires tracking the eigenvalue decomposition (EVD) of the noise data over time. To take advantage of information in the long-term, as well as short-term, correlation lags we turn to EVD over wavelet subspaces. In this paper, we develop a multirate EVD updating method over multiresolution subspaces and find maximum detectability nodes on wavelet binary full-tree structures. We use theoretical analysis to justify the effectiveness of a hyperspectrum for noise based on time and scale evolutionary EVDs. We also show results obtained with simulated 1/ f noise and noise collected by hydrophones of an underwater sonar communication system. Initial results are encouraging as they clearly indicate many subspaces, where detectability is significantly higher than in the original space prior to wavelet decomposition.

3:30, SAM-P8.4
GEOMETRIC LINEAR DISCRIMINANT ANALYSIS
M. ORDOWSKI, G. MEYER
When it becomes necessary to reduce the complexity of a classifier, dimensionality reduction can be an effective way to address classifier complexity. Linear Discriminant Analysis (LDA) is one approach to dimensionality reduction that makes use of a linear transformation matrix. The widely used Fisher's LDA is "sub-optimal" when the sample class covariance matrices are unequal, where sub-optimal means that there exists a better linear transformation that minimizes the loss in discrimination power. In this paper, we introduce a geometric approach to Linear Discriminant Analysis (GLDA) that can reduce the number of dimensions from n to m for any number of classes. GLDA is able to compute a better linear transformation matrix compared to Fisher's LDA for unequal sample class covariance matrices and is equivalent to Fisher's LDA when those matrices are equal or proportional. The classification problems we present in this paper demonstrate and strongly suggest that Geometric LDA can generate the "optimal" classifier in a lower dimension.

3:30, SAM-P8.5
LEAST SQUARES DETECTION OF MULTIPLE CHANGES IN FRACTIONAL ARIMA PROCESSES
M. COULON, A. SWAMI
We address the problem of estimating changes in fractional integrated ARMA (FARIMA) processes. These changes may be in the Long Range Dependence (LRD)parameter or the ARMA parameters. The signal is divided into "elementary" segments: the objective is then to estimate the segments in which the changes occur. This estimation is achieved by minimizing a penalized least-squares criterion based on the parameter estimates computed in each segment. The optimization problem is then solved using a dynamic programming algorithm. Simulation results on synthetic data are reported.

3:30, SAM-P8.6
IMPROVED POWER-LAW DETECTION OF TRANSIENTS
Z. WANG, P. WILLETT
Recently, a power-law statistic operating on DFT data has emerged as a basis for a remarkably robust detector of transient signals having unknown structure, location and strength. In this paper we offer a number of improvements to the original power-law detector. Specifically, the power-law detector requires that its data be pre-normalized and spectrally white; a CFAR and self-whitening version is developed and analyzed. Further, it is noted that transient signals tend to be contiguous both in temporal and frequency senses, and consequently new power-law detectors in the frequency and the wavelet domains are given. The resulting detectors offer exceptional performance and are extremely easy to implement. There are no parameters to tune, and they may be considered "plug-in" solutions to the transient detection problem.

3:30, SAM-P8.7
ACTIVITY DETECTION IN UNKNOWN NOISE ENVIRONMENT
E. FISHLER, H. MESSER
In many applications there exists an array of cells (or bins), each containing either an activity (signal) plus noise, or noise only. A common problem is to identify the active bins, assuming that the noise level in the array is unknown. In this paper we present a novel approach for solving this problem. The approach is based on two steps. In the first, we estimate the noise level and in the second we perform a sequential test to decide, for each bin, whether it is active or not. We show that the proposed algorithm collapses to well known special cases. The performance of the proposed algorithm is analyzed analytically and is demonstrated via simulation results.

3:30, SAM-P8.8
ON-LINE MODEL SELECTION OF NONSTATIONARY TIME SERIES USING GERSCHGORIN DISKS
P. MICHEL, J. TOURNERET, P. DJURIC
The paper proposes a method for on-line model selection of nonstationary time series. The method is based on computation of the covariance matrix of the data, transformation of the matrix by Housholder's tridiagonalization, and application of a clustering algorithm that can separate the Gerschgorin disks of the transformed covariance matrix into disks that correspond to the signals and noise, respectively. The method is applied to on-line estimation of the the number of harmonic signals in noise. Simulation results are presented that show the performance of the proposed method.

3:30, SAM-P8.9
WAVELETS IN THE FREQUENCY DOMAIN FOR NARROWBAND PROCESS DETECTION
Z. WANG, P. WILLETT, R. STREIT
Detecting signals that are long, weak, and narrowband is a well known and important problem in acoustic signal processing. In this paper an ad hoc scheme is developed: its stages include the DFT, a multiresolution decomposition in the frequency domain, and a GLRT. The computational load is light, and the performance is remarkably good. This is so not just in the original narrowband situation, but also, due to an inherent adaptivity to the data, in the detection of signals that are relatively broadband in nature. Generalizations are given to CFAR operation in both prewhitened and unwhitened cases, and to the detection of multi-band signals. As regards the last, it is discovered that there is little loss from over-estimating the number of bands.

3:30, SAM-P8.10
BLIND DETECTION OF INDEPENDENT DYNAMIC COMPONENTS
L. HANSEN, J. LARSEN, T. KOLENDA
In certain applications of independent component analysis (ICA) it any of interest to test hypotheses concerning the number of components or simply to test whether a given number of components is significant relative to a ``white noise'' null hypothesis. We estimate probabilities of such competing hypotheses for ICA based on dynamic decorrelation. The probabilities are evaluated in the so-called Bayesian information criterion approximation, however, they are able to detect the content of dynamic components as efficient as an unbiased test set estimator.

3:30, SAM-P8.11
ON THE USE OF MATCHING PURSUIT TIME-FREQUENCY TECHNIQUES FOR MULTIPLE-CHANNEL DETECTION
A. PAPANDREOU-SUPPAPPOLA, K. GHARTEY, D. COCHRAN
In situations where the presence of a signal is to be detected in several noisy channels, often one channel will have higher signal-to-noise ratio (SNR) than the others. When the SNR on one channel is sufficiently high that the signal can be extracted from that channel, it may be possible to use the extracted signal to aid in detecting the presence of the signal on the other channels. In this paper, the matching pursuit time-frequency method with matched signal dictionaries is used to extract a chirp signal from a noisy channel. The extracted signal is used in one channel of a generalized coherence (GC) detector with the goal of detecting the presence of the signal on other, even noisier,channels. This approach is compared via simulation to a GC detector that does not pre-process the highest SNR channel to extract the signal. Detector performance is shown to be significantly enhanced by matching pursuit signal extraction prior to coherence estimation.