Matched Field Method in the Problem of Moving Transmitter Location with Small Wave Size Antenna
   In this paper the method is presented for processing small wave size antenna output signals, matched to frequency and time variations of spectrum power density of the signals, received from broad-band moving transmitter in acoustic wave guides.
   The method is based on the regularity of interference structure of acoustic fields in ocean wave guides which is the regularity of space and frequency distribution of field energy in the wave guide. The method enables to perform the measurements of the distance to the low intensity transmitter and increases the probability of detection of low intensity transmitter on the background of noise. For this purpose the signal is first processed with two-dimensional spectrum analysis with finite integration time. The resultant spectrum functions form matrix of spectrum power density values in the plane of time and frequency. Second, the spectrum matrix is processed with two-dimension integral transform consistent with field. It is shown that coordination to field for the first order approximation, providing required precision in distance estimation and noise-proof
   features may be performed with two-dimension Fourier analysis of the spectrum matrix.
   Theoretical analysis for the method feasibility for shallow sea, near surface channel, bottom wavequide is presented. The results of experiments in this field, performed in shallow sea and deep ocean are discussed. This work was supported by RFFI fund (Grant 9705-65920).

Determining Class of Underwater Vehicles in Passive Sonar Using Hidden Markov Model with Hausdorff Similarity Measure
    The main purpose of this paper is about detection and classification of Underwater Vehicles (UVs) using features extracted from their acoustic signals. Constructing rules to classify UVs is more difficult than surface vessels classification or even for ships/UVs problem. Indeed, our main goal (classification of UVs) plus two other above mentioned problems are at the edge of oceanic engineering research around the world.
    Why a good solution is hard to find? Answer is : those acoustic signals ;
   >have rapid variation of characteristics with both frequency and time;
   >are highly nonstationary and impulsive;
   >show variations in Signal to Noise Ratio (SNR) due to multipath propagation.
   Therefore; classic algorithms as : template matching or correlation methods are not suitable.
    We have proposed in our paper an algorithm for classification of UVs acoustic signals based on Hidden Markov Model (HMM) with Hausdorff Similarity Measure (HSM). The HMM is a proper stochastic model for speech recognition and classification of spoken wards. We used this model to recognize UVs acoustic signals from other environmental noise and therefore would be able to classify some UVs. The HMM is trainable and in all applications; we must train it by features extracted from recorded sounds to the sufficient amount. Therefore, a collection of sounds for all desired UVs is needed. In other wards; by a trained HMM we can recognize related reference sound, only. So, at least one HMM is necessary for sound recognition of one UV.
    We considered three classes of UVs and simulated their sounds thereafter constructed three HMM with optimal number of states. An important note in the proposed algorithm is using the HSM for abetter similarity measure of features. Some extracted features are used in train section of algorithm and others are used for test and method evaluation. Measurement of similarity is done by HSM instead Euclidean measure, in train or test section. Promising results are:
   >using HSM as a new method increases robustness of algorithm against fluctuations of SNR in received signals.
   >at low and fix SNR (<0dB) has a better performance in recognition of sounds; for instance, we obtained 7.3% increase in performance at SNR + -16dB.
   almost the same performance is resulted at high and fix SNR. Meanwhile, new method need more computations and therefore faster processors.
   In brief, we used a new method in HMM-based recognition algorithm, moreover implemented it in new area that maybe increases our insight about underwater world.

Chaotic Model of Airborne Laser Radar Background Signal and It's Application to Target Detection
   In this paper airborne laser radar background signal is modeled with chaos (here the background signal means that received signal is back from sea surface and sea bottom without target such as submarine in the signal path). And based on the chaotic model we presented a deep sea target detection method. Airborne laser radar is a faster and more efficient techniques for hydrography and target detection than acoustic echo sounder. In shallow water, target information is included in the strong backscatter signal. But the signal path is very bad, the effective attenuation coefficient of backscatter of laser pulse in seawater increases with water depth in the exponent scale. Thus the target detection in deep seawater becomes very difficult. However the strong initial condition of received signal to laser beaming angle activates our premise that the background signal is the result of a chaotic dynamical system. Under the confirmation of that conclusion, we present a new efficient target detection method based on the premise that the background signal is chaotic.
    An introductory review of chaos is presented at first. Chaos definition of Newhouse is introduced in this paper and it is tested by classifying some kinds popular signal. Based on the chaos definition we show in a convincing way that the background signal is chaotic by presenting the results of a detailed experimental study using real-life data collected from different location: 1) the background signal has a finite correlation dimension which lies in 6-8; 2) the background signal has at least one positive Lyapunove exponent; 3) the background signal is local predicable. Most importantly, we show that both the correlation and the largest Lyapunove exponent are essentially invariant to the laser beaming angle and geography location.
    A feedforward neural network with two hidden layers is applied to examine the local predictivity of the background signal. To reconstruct the underlying dynamical system of the background signal, the geometry information of the background signal and a kind of pruning algorithm are adopted to determinate the structure of the neural network. That work reduces the flexibility of reconstruction procession and improves neural network approximation performance. After the neural network is trained with little training and prediction error, it is used as a prediction model in the chaos based method that is applied to problem of detection of submarine in deep seawater,where one-prediction method is applied for received signal and the prediction error is processed using correlation detection method. When the prediction error exceed the threshold, a target is declared to exist. The performance of this new detection method is shown satisfying for real-life data sets and submarine in deep seawater is detected successfully. Submarine detection performance and results are given in the paper.

Real-Time Adaptive Array Processing for Underwater Acoustic Channel
   The establishment of a reliable communication link is of vital importance for high speed underwater data communication. Multipath effect in a time-varying underwater acoustic channel usually causes substantial signal fading. Therefore, adaptive beamformer based on an array of sensors is often employed to reduce noise and interfering signals, while preserving desired signal from the direction of interest. In this paper, we investigate the performance of Frost adaptive beam-former [1] under multipath Rayleigh and Rice fading channels [2] through computer simulations. The measures of beampattern, array gain, normalized mean square error, and bit error rate are utilized for performance evaluation. Simulation results have shown that the performance of Frost beamformer is degraded under Rayleigh or Rice fading channels, as compared to an ideal nonfading channel. However, performance improvement caused by Frost beamformer is still significant.
    For the implementation of a 4-element Frost adaptive beamformer, a TMS320C30 digital signal processor (DSP) board is used in a receiver architecture. The direction of arrival (DOA) of the reference signal needed in the beamformer is provided by the DOA estimation algorithm of far-field approximation [3], which is implemented on another TMS320C30 DSP board. An experiment is also constructed in a water tank to verify the feasibility and validity of the developed adaptive array processors.
   [1] O.L.Frost, "An algorithm for linearly constrained adaptive array processing". Proceedings of the IEEE, vol. 60, Aug. 1972.
   [2] X.Geng and A.Zielinski, "An eigenpath underwater acoustic communication channel model", OCEANS'95 MTS/IEEE
   [3] J.-H. Lee, Y.-M. Chen and C.-C. Yeh, "A covariance approximation method for near-field direction-finding using a uniform linear array", IEEE Trans. Signal Processing, vol. 43, May 1995.

Implementation of Adaptive Processing in Integrated Active-Passive Sonars Deploying Cylindrical Arrays
   n modern sonar systems, it is important that all possible active and passive modes of operation be exploited. Similarly, the implementation of computationally intensive adaptive beamformers in sonar systems is of equal practical importance. This is because the non-conventional processing schemes can provide improved array gain performance for signals embedded in partially correlated noise fields. Adaptive beamformers, however, have a major practical disadvantage of requiring long convergence periods. Previous investigations by the authors of this paper have introduced adaptive beamforming schemes with near-instantaneous convergence for match filter processing. The formulation of the problem, however, had been directed for line towed array applications.
    The concept for implementing successfully adaptive schemes in 3-dimensional (3-D) arrays of sensors, such as, cylindrical arrays, is similar to that of line arrays. In particular, the basic step is to minimize the number of degrees of freedom associated with the adaptation process. This step will minimize the adaptive scheme's convergence period and achieve near-instantaneous convergence.
    Thus, the research effort has been centered on the definition of a generic beamforming structure that decomposes the beamforming process of 3-D sensor arrays into sub-sets of coherent processes. The approach is to separate the computationally intensive multi-dimensional beamforming into two simple modules, which are line and circular array beamformers. Thus, the multi-dimensional beamforming process can now be divided into coherent sub-processes which lead to efficient implementation in real-time sonar systems. Furthermore, the application of spatial shading to reduce the side-lobe structures can now be easily incorporated. Moreover, the new approach makes the implementation of adaptive schemes in multidimensional sensor arrays practically achievable.
    The proposed adaptive processing concept has been implemented in an integrated active-passive real-time sonar deploying a cylindrical array. Real data results from the adaptive and conventional beamforming outputs of the cylindrical array sonar system demonstrate the superior performance of the adaptive beamformer in suppressing the reverberation and cluttering effects in active sonar applications. Moreover, for passive sonar applications, the adaptive processing provides substantially improved angular resolution performance as compared with that of the conventional beamformer. Both these two performance improvements for a cylindrical array sonar are of particular importance for mine hunting operations.
    The investigation reported in this paper includes also the definition of an ROV deploying a cylindrical array sonar for underwater robotics operations, such as mine-hunting. The various aspects of robotics arrangements for the ROV will be discussed in association with the sonar performance characteristics provided by the integrated functionality of the cylindrical array sonar including an adaptive beamformer.

Shallow Water Bathymetric Data Improvement in Seabeam 2100 Multibeam Systems
    Seabeam 2100 is a new generation multibeam sonar capable of mapping the ocean floors with swath coverage ranging from +-65 degrees to +-45 degrees for the deepest ocean floor. Since the main design goal was to attain full ocean depth capability, a low acoustic frequency of 12kHz was chosen to accomplish that goal. As a result, the array sizes were large (3 to 4 meters) and the near-field distance of the distance before which beams can be formed properly increased limiting the performance of the systems in shallow waters.
    Improvements in the shallow water data quality of deep water multibeam systems can be attractive for a number of reasons. First of all it can avoid the need for a separate system for mapping the shallow waters. This is specially true for the key issue of mapping the ocean floors within the Exclusive Economic Zone (EEZ) which can go as deep as thousands of meters for a number of countries. Another factor is the potential for robust performance. The deep water systems are mounted on bigger boats with deeper drafts and therefore they are capable of providing good data even in foul weather conditions.
    This paper addresses the issues involved in obtaining more accurate data in shallow waters using high-resolution signal processing techniques. Although a number of theoretical techniques exist, application feasibility will be examined in light of the SeaBeam 2100 system architecture, real time processing requirements and design trade-offs. Results are presented both from simulated data and real world experiments.

Application of Neural Network for Reat-Time Underwater Signal Classification
    This paper discussed the application of a modified supervised neural network to the problem of classification of underwater acoustic signals in real time. Concepts related to Analysis, feature extraction, detection and classification of underwater acoustic signal. A new network model which derived for this application is presented, it can be used to analysis and classify an underwater acoustic signal which may encounter in the area of ocean Engineering and environment measurement, such as under sea localization, under sea communication and under sea environment survey.
    A great deal of efforts have been made on frequency domain feature extraction and the real time neural network classification algorithm. In order to detect the task-relevant part of underwater signal from background noise and other interference, a fast and accurate spectral estimation procedure is designed, which lead up to a close study of signal preprocessing, segmentation, windowing, frequency estimation and post-processing. Then by using of the self-learning and adaptive process of Neural Network, the parameters of the network model can be justified, these parameters are utilized to organize the network model which gives very good classification results. In order to meet the real time processing requirement, a neural network with two steps organization is proposed.
    The functional behavior of proposed model is verified by simulation, the input signals for performance evaluation are measured from ocean. Simulation results show that applying the modify algorithm developing in this paper can increase the classification rate and decrease the convergence time. Results obtained by simulation demonstrate the effectiveness of the approach. Following the analysis of the results, conclusions and recommendations will be provided.