Study On Detection And Track Of Target Signal In Time-Frequency And Time-Space Domain

Detection of signals is widely used in spectrum supervision,sonar,and radar,which can be processed in many dimensions,such as time,frequency and space domain.The detection of acoustic signal in passive sonar system is one of important method to discover the underwater target.There are two types of detection in passive sonar.One is the detection in time-frequency domain,which is executed on the LOFAR(Low frequency analysis and recordings)and DEMON(Detection envelope modulation on noise).The other is the detection in time-space domain,which is executed on the bearing-time record(BTR).Several signal detection problems in the passive sonar system are studied from the view of time-frequency and the time-space respectively in this thesis.First,for the problem that the background noise floor is unflat and the data amount needed to be processed is large in the LOFAR,two background noise estimation algorithms in the time frequency domain are proposed,and a signal broadband blind detection algorithm based on Hidden Markov Model(HMM)is proposed.Second,for the detection of weak signal from the BTR in passive sonar,a track before detect(TBD)algorithm of single and multi-sensors are studied.A TBD algorithm based on HMM is proposed,which is more robust to the impulse interference than the conventional detect before track.The experiment of using the real sonar data shows the effectiveness of the algorithm.Specifically,the main contributions of this thesis are summarized as follows:1.The classical background noise estimation algorithm does not satisfy the need of the real-time burst signal broadband detection because of the unflat background noise floor and the large data amount.Two low-complexity and low memory-consumption background noise estimation algorithms are proposed.The effectiveness of the algorithms is theoretically analyzed and verified by the real passive sonar data.These two algorithms are also developed to detection of the satellite wideband frequency hopping signal.The results show that the performances of the algorithms are closed to the FCME-CA algorithm,while their computation complexity and memory consumption is much lower than that of the FCME-CA.2.For the problem of complex background environment,such as pulse interference and unflat background noise floor,in the signal broadband detection on the LOFAR,a burst signal broadband detection algorithm based on HMM is proposed.This algorithm models the frequency line in the time-frequency spectrum as HMM,and realizes the detection of the signal and the estimation of the start time and the termination time of the burst by solving the HMM.The detect probability is higher than that of the wavelet transform algorithm and power detector while its false alarm probability is smaller.The algorithm is verified by the real passive sonar data.The simulation and the experiment of the real very high frequency(VHF)wideband data also show the effectiveness of the algorithm.3.For the problem that the oceanic acoustic environment noise doesn't follow Gaussian distribution completely,the distribution of the measurements in the BTR is analyzed by using the real sonar data from the experiment on the sea.The coincidence between the distribution of the real sonar data and the Gaussian distribution,alpha stable distribution and the asymmetrical Laplace distribution are compared and the conclusion is that the asymmetrical Laplace distribution is suitable for the measurements in the BTR.An estimation algorithm based on Expectation Maximization(EM)algorithm and asymmetrical Laplace distribution mixed model is proposed,which can estimate the parameters of the noise distribution and give some prior information of the noise to the signal detection algorithm in the BTR.4.For the problem of detect the multi-target bearing-only trajectory in the BTR,a single target TBD algorithm based on HMM is proposed.Compared with the conventional detect before track algorithm,this algorithm is much more robust to the impulse interference and the detect probability is higher.This algorithm is developed to the muti-target case later.The simulation and the experiment of the real sonar data show the effectiveness of the algorithms.5.For the target signal detection in a certain region by using multiple passive sonars,a multi-target detection and track algorithm based on Bayesian theory is proposed.The realization of the algorithm based on grid probability method and particle filter is also given out.The characteristic of the algorithm is that the the probability of target exist is binded into each grid and the clustering algorithm is used to decide whether a new target appears or not.This characteristic avoids the re-initial of the target and the target is splited.