Chair: Cliff Carter, NUWC, New London (USA)
Brigitte Colnet, CRIN- INRIA CNRS (FRANCE)
Jean-Claude Di Martino, CRIN- INRIA CNRS (FRANCE)
In this paper we present a neuromimetic approach to bearing estimation issue. The proposed method is based on time-delay neural networks. This kind of network is well suited to take into account constraints encountered in signal processing: it deals with the dynamic nature of signal and discovers acoustic and temporal features. According to the propagation model of plane waves, the network has to relate the delays between sensors to enable source localisation. The time-delay neural network approach is encompassed in a successive refinement method. Thus, accuracy is increased while the number of networks to look at the whole horizon is reduced.
Amlan Kundu, NCCOSC (USA)
George C. Chen, NCCOSC (USA)
In this paper, we have presented an integrated hybrid neural network and hidden Markov model (HMM) classifier that combines the time normalization property of the HMM classifier with the superior discriminative ability of the neural net (NN). In the proposed integrated hybrid HMM and neural net classifier, a left-to-right HMM module is used first. The HMM module segments the observation sequence belonging to every exemplar into a fixed number of states starting from the left. After this segmentation, all the frames belonging to the same state are replaced by one average frame. Thus, every exemplar, irrespective of its time scale variation, is transformed into a fixed number of frames, i.e., a static pattern. The multi-layer perceptron (MLP) neural net is then used as the classifier for these time normalized exemplars. Experimental results using two different feature extraction schemes clearly demonstrate the superiority of the hybrid integrated classifier.
LT. Charles W. Victory, U.S. Navy/NCCOSC
Richard Trueblood, Orincon Corporation (USA)
Robust detection and classification for active sonar processing in acoustically complex environments is a difficult and challenging problem. Complex bathymetry and propagation effects may cause multipath spreading of the transmitted signal before it arrives back at the receiver. Correlating with a replica of the transmitted signal may thus severely degrade the performance of a system. This paper explores the use of a multilayer perceptron to compensate for channel and other medium effects in a acoustically complex environment. It is shown that adaptation to the environment in such scenarios can lead to significant processing gain and that a multilayer perceptron is capable of implementing this type of processing.
Joseph N. Maksym, Defence Research Establishment Atlantic (CANADA)
This paper explores the use of an artificial neural network to distinguish between echoes from a constellation of acoustic reflectors representing a target and similar echoes produced by other reflectors, e.g. reverberation. The network was both trained and tested with simulated data. Wide band linear frequency modulation was used in order to resolve the highlights of the target.
K. C. Ho, Royal Military College of Canada
A. E. Scheidl, Royal Military College of Canada
R. J. Inkol, Department of National Defence (CANADA)
A new approach to the identification of a constant amplitude signal with frequency/phase modulation is investigated. We model the incoming signal phase as a linear combination of a set of orthogonal vectors and use the significant coefficients as features for identification. Because of the data compression ability of orthogonal transform, a few coefficients are sufficient for signal representation, thereby reducing the processing time and system complexity. The choices of transforms and feature size are discussed. The performance of the new identifier is studied through simulations.