Research On DOA And Beamforming Methods Based On Signal's Time-Space-Frequency Properties

The goal of array signal processing is to estimate the desired information via exploiting the signals impinging on the sensor array.It is well known that direction-of-arrival(DOA)and beamforming are the two main branches of array signal processing.DOA estimation is a problem of significance since it provides prior information for beamforming.Beamforming is the key technique for the array to extract the desired signal while suppress the interferences and background noise terms.Unlike the single antenna case,the sensor array can use the geometry and beampattern information of the sensors,resulting in more accurate estimates,higher output gains and estimating larger number of sources.However,in practical applications,the demanding circumstances can lead to severe performance degradations for the conventional DOA estimators and beamformers.Moreover,it is difficult to improve the estimation accuracy only by exploiting the array spatial properties.To further enhance the performances of the DOA estimators and beamformers,we developed several solutions by exploiting the time-space-frequency properties of the arriving signals,which can lead to the output performance improvment to a certain degree.The main work and contributions of this paper are concluded as follows:Firstly,we studied the DOA and beamforming techniques via exploiting the signal properties in the time d omain.It is shown that the conventional signal processing algorithm becomes suboptimal when the time domain properties of the sources are not fully exploited.Moreover,it is noted that the NC signals have been widely used in many modern communication systems.The performance of the conventional algorithms can be improved by exploiting this particular temporal property.To this end,we devised a noncircular(NC)deterministic maximum likelihood(NC-DML)method.To reduce the computational complexity,we presented a novel NC alternating projection approach for computing the NC-DML estimator.Moreover,by using the structure of the three-axis crossed array along with the NC properties of the arriving signals,we proposed a structured least squares based NC-ESPRIT method.As for the adaptive beamfromer,we devised a shrinkage widely linear complex-valued least mean squares method,which exploits the NC properties of the signal-of-interest and offers a variable step size.As a result,the proposed solution yields faster convergence speed and smaller steady-state errors.Secondly,we investigated the DOA and beamforming methods via exploiting the signal properties in the space domain.The emerging technique of compressive sensing(CS)has gained a lot attention in array signal processing area since the exploitation of sparsity of the incoming signals helps to improve the performance of DOA estimator and beamformer especially in the demanding scenarios,e.g.,low SNR,small snapshots number and coherent sources.In practice,the CS based techniques suffer from grid mismatch issue as well as the problem of choosing hyper-parameters.To circumvent these problems,we proposed an iteratively reweighted method for joint dictionary parameter learning and sparse signal recovery,where the DOA estimation problem was addressed from a superresolution CS perspective.To handle the coherent signal problem,we devised a robust relaxation algorithm.The block successive upper-bound minimization is employed as a base framework to update each source component.Moreover,we developed an adaptive beamforming technique,where the channel vectors were modeled as being drawn from a Gaussian mixture model.Our design formulation entails minimizing the outage probability subject to transmit power constraints.Finally,we discussed the DOA and beamforming algorithms via exploiting the signal properties in the frequency domain.To handle the wideband signal processing issue,the discrete Fourier transformation can be employed to divide the wideband signal into several narrowband ones.As a result,many wideband signal processing techniques based on frequency domain are proposed in the literature.Besides,we notice that the underdetermined DOA estimation has gained a lot of attention recently.To deal with the underdetermined wideband DOA estimation issue,we proposed a sparse learning approach,which is developed by exploiting the fact that the signals across different frequency bins share the same spatial property.It is well known that the performance of beamformer depends essentially on the availability of the signal-free training snapshots.Therefore,we started from the single frequency signal case and devised a reconstructed interference-plus-noise matrix based approach.Then we generalized this algorithm to the wideband scenario to discuss the associated beamforming techniques in frequency domain.Simulation results revealed the effectiveness of our proposed solutions.