Fundamental Studies On Target Identification Based On Torpedo Warning Sonar

The essence of signal processing is finding differences between signal and disturbance. Different objects' signals bring different information. Finding invariant characteristics is the key of target classification and identification.Task of torpedo warning sonar is to determine whether the target is torpedo or not. In the dissertation some fundamental researches have been done on this, mainly on time and frequency character of torpedo signals. Line spectra is the emphasis of the dissertation, which include that of torpedo's radiating noise representing target's rhythms and mono-component or multi-component frequency varying self-guided signals of torpedo, such as LFM and time-frequency hop signals and so on. The paper is composed of four parts. The first part is a brief introduction. The second part is about how to get DEMON from target's radiating noise and how to make it cleaner. The third part is about high precision frequency estimation under stationary signal assumption and the end is about frequency tracking under nonstationary assumption. With the above researches we hope to hold sufficient knowledge of line spectra and get more character as possible as we can to improve torpedo indentification possibility.Firstly, methods of getting DEMON are studied. Propeller's running modulates wide-band noise and generate amplitude modulation (AM) signal. This AM signal is also wide-band so it sufferes little from multi-path channel. By means of demodulation we can obtain target rhythms such as propeller speed and so on. This is the widely used DEMON analysis. By DEMON we can get more and stabler line spectrum information than by LOFAR, which is to make power spectrum analysis of signal directly. Several demodulation methods are introduced and compared. Such as demodulation by absolutevalue, square, Hilbert transform,and line spectra generation property of cyclostationary signal. Square demodulation is proved to be a practical and efficient method.Approaches on how to clean DEMON are researched. Two present methods are introduced which are adaptive line enhancer (ALE) and 1— spectrum methodusing high order cumulant. Three new methods are provided, which are demodulation by means of envelop periodogram, multi-path signal and vector signal. These methods use different information and theory and have different effects, but they all compute simply and are fit for application in engineering. Trail results show that 1 — spectrum method does better than the other generally.Under stationary signal assumption high precision frequency estimating technology is researched. Emphasis is put on frequency estimator based on two-order adaptive notch filter (ANF). Firstly, property of ANF is analyzed, convergence time and band width and processing gain of the filter are given. Then property of the frequency estimator based on ANF is researched, and influence of the difference between the reference and desire signals is analyzed. Lower bound of estimation variance of this adaptive frequency estimator is derived. Frequency auto-tracking adaptive frequency estimator (FATAFE) is proposed. Simulation and theoretical results show that under the same conditions FATAFE obviously decreases estimated bias and variance relative to the present ones. FATAFE can realize unbiased estimation and the estimated variance approaches theoretical lower bound unless signal-to-noise ratio is very low. Results on trail data of target's radiated noise show that FATAFE can estimate propeller speed (frequency of fundermental line-spectra) with high precision.Under nonstationary signal assumption, estimation of instantaneous frequency of line spectra or narrow band signal, that is, frequency tracking, has been researched. Tow adaptive methods are introduced. One is based on adaptive linear prediction, which is also called adaptive measurement of frequency (AMF) and the other is based on constrained adaptive notch filter (CANF). The later isproved to be superior to the former on computing speed, estimating precision and tracking ability. Based on CANF, forgetting factor self-adjusting method (CANF-FFSA) is presented as an improvement. Simulation shows that CANF-FFSA can realize better harmonization between frequency estimating precision and tracking ability according to the system state.