Feedforward layered perceptron neural networks seek to capture a system mapping inferred by training data. A properly trained neural network is not only capable of mimicking the process responsible for generating the training data, but the inverse process as well. Neural network inversion procedures seek to find one or more input values that produce a desired output response for a fixed set of synaptic weights. There are many methods for performing neural network inversion. Multi-element evolutionary inversion procedures are capable of finding numerous inversion points simultaneously. Constrained neural network inversion requires that the inversion solution belong to one or more specified constraint sets. In many cases, iterating between the neural network inversion solution and the constraint set can successfully solve constrained inversion problems. This paper surveys existing methodologies for neural network inversion, which is illustrated by its use as a tool in query-based learning, sonar performance analysis, power system security assessment, control, and generation of codebook vectors.

No abstract

The Yellow Sea represents a natural laboratory for the study of shallow water acoustic propagation as influenced by a strong summer thermocline, energetic internal wave fields, and a seabed considered relatively range independent. In this paper results of transmission loss measurements obtained in the Yellow Sea, during the joint China-U.S. Yellow Sea ’96 experiment, conducted in August 1996 are presented. Explosive sources were deployed at ranges 1–30 km, and acoustic data were recorded on a vertical array that spanned the entire water column depth of ∼75 m. The data are in 1/3-octave bands with center frequencies 70 to 700 Hz. A simple geoacoustic model is derived for the seabed that shows consistency with data at ranges of O(1 km), where raylike properties are exhibited, and of O(10 km), where modelike properties are exhibited. With summer conditions prevailing, the sound speed profile was described by two isovelocity layers linked by a layer with a linear gradient. Internal tide activity modulated the depth of the linear gradient layer over the course of the experiment, and this effect is also examined using various modeling approaches.

No abstract

A bottom-mounted, wideband active sonar system designed to detect unfriendly vessels entering a closed area is described. Wepresent signal-proeessingalgorithms for target detection, classiAcation, locallzation, and tracking, and present performance predictions based on simulations. These algorithms are shown to be effective With state of the art, reallzable sonar systems.The signals are binary phase shift keyed (BPSK) with a range resolution of approximately 0.01 m and Doppler resolution of 0.5 d s . Received signals from multiple directional receive beams are basebanded and Hilbert transformed to obtain a complex representation. These signals are correlated with replicas of the transmitted signal time comprplsed or dilated to represent a number of different hypothetical target Dopplers, giving a processed signal with three dimensions: time, chauneI, and target Doppler.. . Background interference is estimated wing a blocktrunsversal Alter. The processed signal is dlvtded by the estimated backgronnd to obtain signal-to-interference ratio (SIR) and SIR threshold crossings. are detected and contiguous detections are clustered. Range Is estimatedusing arrival time.Bearing :is estimated using an amplitude comparison for adjacent receivebeam. Elevation is estimated using split beam phase between signals from elements separated vertIdly. Finally, the crossings are clustered by location and Doppler.Clusters are classified based on amplitude and estimated physical extent and shape. Clnsters with sufficiently high classification are passed to a tracking algorithm that compntm a Kalman liiter track using clusters derived from sequential traosmissions. This filter operates using linked line of sight (LOS) coonlinates, a technique long used in radar. The LOS coordinates used in this application are range, range rate, hearing, add elevation;Use of such coordinates Is advantageous in thlr appiication because the errors in the coordinates are nearly Independent of one another.Performajxe of thls 'algorithm was investigated using active acoustic signals synthesized by the Sonar Simnlation Toolset (SST), a digital program written by Robert Goddard at the AppUed Physics Laboratory. This program. generates a complex baseband representation of the acoustic signal In eacb chinneL This 'sigdal is pmcessed by the detectlou,classificatIon, and track@g algorithms. SST simulates effects of aconstic propagation ihclnding .refraction and scanering and reflection from a target. Several simulated encounters with different target geometries are presented to illustrate detection, localization, and tracking performance.