Investigations On Space-Time-Frequency Joint Detection And Tracking Technology Of Weak Line-Spectrum

The narrow-band line-spectrum signals in the radiated noise of ships are characterized by high strength,good stability,and low propagation loss.Therefore,the detection of the linespectrums is of great importance for the detection,tracking,and classification of low noise and quiet underwater targets.The line-spectrum is used to detect and track underwater weak targets and two detection and tracking algorithm are designed for the passive sonar system under different operating modes in this dissertation.And the main contributions are as follows:1.The ocean ambient noise and the basic methods of sonar array signal processing are analyzed and summarized.And the basic model for line-spectrum detection and tracking based on Hidden Markov Models(HMM)is established.2.An algorithm based on 1-D HMM with dynamic transition probability is proposed for the passive sonar working mode based on direction estimation.The block processing framework is adopted and the change of frequency state is modeled as a first-order Markov process in the lofargram.And the Viterbi algorithm is used to dynamically search the maximum a posterior time sequence of frequency states in each time-frequency block.During the iteration,the first derivative estimates of the sequence are obtained in real-time to dynamically adjust the state transition probability matrix in HMM.And a power spectrum accumulation method based on dynamic sliding window is proposed when estimating the birth and death of line-spectrum.The dissertation also designs a method of merging similar line-spectrum trajectories in adjacent regions based on data association theory.The processing results of simulation and actual data show that the proposed method can effectively detect and track the frequency state of complex time-varying line-spectrums and has excellent processing performance under low SNR.3.A time-space-frequency joint detection and tracking method based on 2D HMM is proposed for the passive sonar working mode of multi-directional pre-beamforming.The changes of the frequency and azimuth states of a line-spectrum signal are modeled as two independent first-order hidden Markov processes and the Viterbi algorithm is adopted to estimate the time sequence of frequency-azimuth states with the maximum posterior probability in the 2-D HMM state space.A novel background equalization method is proposed in the array signal preprocessing stage to remove the trend of broadband noise in Frequency-Azimuth(FRAZ)spectrum while preserving the spectral intensity distribution of line-spectrum components in frequency and azimuth dimensions.Aiming at the computational problem of dynamic search in the 2-D HMM state space,the block processing framework is also used to divide the time-space-frequency region,and the line-spectrum trajectories are extracted by subregion.An extended dynamic A matrix and a three-dimensional adaptive sliding window are designed in the process of trajectory dynamic extraction to adapt to the line-spectrum signals with rapid changes in azimuth and frequency.In addition,the method of merging similar line spectrum trajectories in adjacent sub-regions introduces the azimuth dimension to expand.The performance of the existing line-spectrum detection methods in the two dimensions of frequency and azimuth is compared,and the validity and correctness of the proposed method is verified by processing results of actual data.