Acoustic Vector-Sensor Array Processing Based On Sparse Decomposition Theory

Acoustic vector sensor technology is a benchmark advance attracted many countres’attention, and one of the most prospective branches in underwater acoustic field. Sparsedecomposition theory provides a simple, flexible and self adaptive representation method ofsignals. Through sparse decomposition theory, it can essentially reduce the cost of signalprocessing and improve the compression efficiency. Compressive sensing theory is afundamental revolution in modern information theory derived from blind source separationand sparse decomposition theory. The compressive sensing theory is a new informationacquisition theory and breaks the restriction of traditional sampling theorem. This paperapplied the sparse decomposition and compressive sensing theory to underwater acousticsignal processing, the main contents and conclusions are as follows:1. DOA estimation based on acoustic vector sensor arrayAccording to the sparsity of targets azimuth in angle search space and using the form ofarray response vector to build-up dictionary, it can construct DOA estimation model ofacoustic vector array based on sparse decomposition theory. The DOA estimation with highresolution can be obtained in small snapshot number and low SNR case using single snapshotand multi-snapshot algorithm. The performance of algorithm is analyzed using Song HuaLake experimental data, and compared with conventional beam-forming, beam-forming basedon four-order cumulants, MVDR and MUSIC. The DOA estimation results approved theexcellent performance of the algorithm in small number snapshots and low SNR case, andsuitable for DOA estimation of moving targets.Through compression of array data using projection metrix based on compressivesensing theory, it can reduce the computational cost of subsequent processing greatly. Thedata compression based compressive sensing theory makes MVDR algorithm effective whenthe snapshots number is less than array elements.2. Frequency estimationBased on the sparsity of frequency, several frequency estimation models are constructedaccording to signal form, time-domain manifold vector and time window model respectively.With numerical simulation and experiment data processing, the performance of eachalgorithm was analyzed detailedly. Comparisons with periodogram and MUSIC show that thealgorithm has obviously advantanges under the condition of low SNR, especially smallsnapshots. The compressive sensing theory can compress the sampling size and reduce the computation cost of subsequent processing greatly.3. Frequency and azimuth joint estimation based on single acoustic vector sensorWith the vector signal analysis, the joint estimation algorithm based on single acousticvector sensor was proposed. From the simulation results, the influence of targets number,SNR, snapshots number to algorithm performance was analyzed.4. Optimization of dictionary constitutionThere is a problem of heavy computational cost because of higher dimentional spacesearch. The problem was resloved by the changeable step-size and dictionary dynamicconstitution method.