Application Of Wavelet In Detection Dispersive Channel And Feature Extraction Of Wideband Echo

Higher capabilities were put forward for detection,identification and anti-interference of torpedo in future naval battle. Compared with the narrow-band processing,wide-band processing is a new direction for underwater signal processing for its advantages including abundant information of wide-band echo and little correlation of reverberation. On the other hand,the resolution of wavelet is adapted in time-frequency plane and the structure of wavelet transform is similar to that of wide-band correlation processing. So wavelet transform is an important tool in future underwater signal processing. This dissertation is devoted to systemic research the two projects encountered in wide-band homing in theory and from experiment.Firstly,the theories of wide-band signal processing and wavelet are introduced in detail. Three involved concepts of wide-band signal/system are three-type conditions,echo model and Wide-band Ambiguity Function (WAF). And the resemblance is pointed out between Continuous wavelet transform (CWT) and WAF. In the course of computing WAF,the scaling replica is very important. Based on the properties of CWT,three approaches have been proposed for the scaling of a wavelet that has no analytic form. One method is by multirate signal processing with high complexity. Another is using the cross wavelet transform also with high computationally intensive. The third is by sampling theorem,which can be adopted in engineering due to its fast algorithm.Secondly,detection of two important dispersive channels was studied in theory and from experiment. Two approaches to improve the robustness of the detector,each based on a different theirs distortion mechanism. One is based on the modeling as fast fading distortion (FFD) whose optimum detector is the segmented replica correlator (SRC). The second approach is replica correlation integration (RCI) for time spreading channel (TSD). At the same time,the optimum detector of FFD and TSD in wavelet domain have been investigated according to properties of CWT. They are wavelet-domain-segmented-replica-correlator (WDSRC) and wavelet-domain-replica-correlation-integration (WDRCI). And we have proved that their performances are consistent with those in time domain. The numeric simulation also shows their capability is superior to that of wavelet-domain replica correlator (WDRC). On the other hand,we found that for exploiting the full potential of RCI,we must overcome a serious lack of matching. Then we propose a multi-hypothesisRCI that minimized lack of matching. Simulation results for a numerical example show preliminarily that this method for estimating spreading time of TSD channel with a proposed multi-hypothesis RCI is simple,convenient,and feasible. Then the experiment is carried through in tank for verifying the results including detection capability of SRC and RCI and feasibility of proposed multi-hypothesis RCI for estimating spreading time of TSD.At last,the subject of feature extraction and identification from wide-band echo was discussed deeply and systemic. The process of target identification is formulated,which involved feature extraction,dimension reduction and classification. For solving the overlap of each subspace of a wavelet library in frequency,there are two approaches that are Best Basis (BB) and Local Discriminant Bases (LDB). Their measure and searching algorithm were researched. Seven approaches of feature were proposed. They are scale-wavelet power spectrum,time-wavelet power spectrum,coefficient of MRA,power spectrum in subspace of MRA,coefficients of wavelet packet (WP),cost function of each WP's subspace and power spectrum of each WP's subspace. We also defined four discriminant measures. Each extraction and reduction was applied for wide-band echoes of bottom sediments of Geneva Lake. During identification,simple rule was designed for determined the decomposed level of MRA and WP in classification. Results for experimentation data show:(1) Feature dimension is very high when using time-wavelet power spectrum,coefficient...