Research On Interference Suppression Based On Subspace Estimation

The underwater sonar systems face energy attenuation for long-distance propagation and the clutter from the surface carriers,marine organisms,and marine environment,resulting in reduced signal to interference and noise(SINR)of the received data.Restricting the further improvement of the sonar system's performance is one of the unavoidable and urgent problems in the high-precision operation of the deep sea.For cooperative sonar systems,SINR is an essential factor affecting system performance.Suppressing interference and noise without affecting the amplitude and phase characteristics of the desired signal is the leading research goal to improve system performance.The linear phase of the subspace projection features provides the possibility to achieve this goal.We establish the data model under the subspace theory based on the cooperative sonar system.Introduced the theory of subspace and subspace projection;proved the feasibility of suppressing interference and noise through spatial prediction;analyzed the cause of"spatial entanglement"when interference exists and its influence on existing methods;deduced expressions of the two typical projection methods and the relationship between them.Aiming at the problem that the conventional subspace dimensionality estimation method based on the information criterion only considers the observed probability density,the algorithm's performance is reduced under the conditions of low signal-to-noise ratio,low snapshot,etc.A Generalized Bayes Information Criterion(GBIC)based method for subspace dimension estimation is proposed in this dissertation.This method considers the probability density of the observation and the joint probability density of the corresponding eigenvalue and the eigenvector.The expressions of GBIC are given for both deterministic signal and stochastic signal,respectively.The algorithm is verified by simulation and a tank test.The simulation results show that the performance of GBIC is always better than the conventional information-based algorithms.The tank test results show that GBIC can still effectively estimate the subspace dimension when the signal-to-noise ratio is-10d B when the existing methods are entirely ineffective.Aiming at the problem of subspace entanglement when interference exists.We propose a time-invariant subspace estimation algorithm so-called Matched Generalized Likelihood Ratio Test(M-GLRT).The algorithm uses GLRT as the detection statistic and uses the known reference signal to estimate the eigenvector that best matches the signal subspace.Then a projection matrix is constructed to linearly project the received data to suppress interference and noise.Simulation and tank test results show that the algorithm can effectively overcome subspace entanglement caused by interference and accurately estimate the desired signal subspace.Moreover,the algorithm has a good angular resolution.The tank data results show that the algorithm can still locate the target sound source even when the conventional processing methods are almost ineffective,especially when the correlation peak between the interference and the reference signal is higher than the correlation peak between the expected signal and the reference signal.Aiming at the time-varying subspace tracking problem under substantial interference,the matched constrained subspace tracking(MCST)algorithm is proposed.The algorithm first uses M-GLRT to estimate the inaccurate signal subspace and project the data onto the subspace.It then converts the subspace tracking problem into an unconstrained optimization problem computed by recursive least squares(RLS)solution.Compared with the conventional algorithm based on eigenvalue decomposition,this algorithm's computational complexity is reduced from O(N~3)to O(Nr) while ensuring the orthogonality of the eigenvectors of the recursive process.Simulation and tank test results show that the MCST algorithm can accurately estimate the time-varying signal subspace under the conditions of low SINR,substantial interference,and correlated interference.It results in suppressing the interference and noise without affecting the phase characteristics of the signal through projection operations.The target positioning results in the tank indicate that in the environment of strong interference and time-varying signals,effective target positioning cannot be achieved.Compared with the un-processing results,the MCST algorithm increases the positioning error from 2.869m to 0.163m.