Sequential Detection And Fast Direction Estimation Of Array Signals

The spatial electromagnetic environment is becoming more and more complex,mainly due to the fast development and widespread use of radar,communication and other electronic devices,and the spatial electromagnetic signals are becoming spectrally,temporally and spatially overlapped at the same time.Array signal processing techniques are able to deal with multiple simultaneous and spatially adjacent signals,they have been attracting the attention of academic and industrial communities in the past decades.In order to track the variations of dynamical spatial spectra and provide information for applications such as electromagnetic spectrum management and dynamic spectrum access,it is essential to carry out in-depth research in array signal detection and direction finding methods with short delays and high processing effiencies.Specially,on space-borne and unmanned airborne array platforms that have very limited computation resources,it is even more important to save computational loads of array processing,or follow sequential processing flowchart to disperse the computational load into the whole sampling period.This thesis studies sequential signal detection and fast direction-of-arrival(DOA)estimation problems with sensor arrays,and focuses on improving the timeliness of array processing from two aspects.One is processing the array outputs sequentially to recover dynamic spatial spectra,and trying to obtain the signal directions simultaneously.The other is proposing fast DOA estimation methods to obtain direction estimates using as few computations as possible based on array snapshots collected previously.In order to deal with the two problems,this thesis proposes sequential signal detection and DOA estimation methods,together with one-and two-dimensional fast DOA estimation methods.Successful signal detection is the precondition for effective snapshot collection and array processing.This thesis begins with this problem,and proposes a sequential signal detection method by introducing in the subspace tracking technique.The new method skips over the covariance matrix decomposition procedure,which is widely used in previous signal detection and enumeration methods,and contains only computationally efficient and parallelism friendly linear correlations,and thus is much faster than its counterparts.The array snapshots are processed one by one,and the signal energy is accumulated with improved timeliness,the detection delay is then no longer constrained by a particular temporal window as that in the covariance matrix-based methods.Theoretical analyses are carried out to reveal the performance of the proposed method in detection probability,detection delay and the corresponding DOA estimation precision.Variant methods with modifications are also proposed based on the original method to adapt to circumstances with known noise variance or multiple incident signals.The thesis then studies the problem of fast one-dimensional DOA estimation based on the array outputs collected after successful signal detection.Two methods are proposed in the circumstances of multiple snapshots and a single snapshot,respectively.In the case of multiple snapshots,the propagator method is introduced to estimate the signal subspace of the array outputs,and the subspace estimate is then exploited to establish a low-dimensional characteristic equation with its dimension equaling the signal number.The signal directions are finally estimated by rooting this equation.This method skips over the computationally demanding procedures of covariance matrix decomposition and spatial scanning used in traditional subspace-based DOA estimation methods,and contains only low-dimensional computations and numerical calculations,which are computationally very cheap,and thus the proposed method is computationally much more efficient than the previous ones.In the case of a single snapshot,the techniques of matching pursuit and alternative optimization are introduced in and a two-step process is followed for fast DOA estimation.The matching pursuit technique is used to obtain coarse DOA estimates for the incident signals first,and the coarse estimates are taken as initial values which are then refined with the alternative optimization technique to finally obtain high-precision DOA estimates.The proposed method provides a computationally efficient way for solving the one-snapshot DOA estimation problem,which has been considered to be very difficult for years.The problem of fast two-dimensional DOA estimation is then solved using the widely studied L-shaped arrays.Computationally efficient one-dimensional DOA estimation methods are proposed first to obtain directional variables along two orthogonal dimensions,which use the outputs of the two linear arrays on the two arms of the L-shaped array separately.And then two methods,a covariance matrix-based one and a signal variance-based one,are proposed to pair the DOA estimates along the two dimensions.The two pairing methods can be synthesized to further improve the performance.Rigorous analyses on computational complexity are carried out to demonstrate the predominance of the proposed method against existing counterparts.In cases when the computational resource of the array processor is somewhat less limited,a spatial sparsity-based sequential detection and DOA estimation method is proposed to reconstruct the dynamic spectrum snapshot by snapshot.The new method brings into play the predominance of the sparse Bayesian learning(SBL)based array processing methods in noise inhibiting,and introduces structural optimizations and mathematical simplifications to the original reconstruction procedure.A computationally more efficient sequential sparse reconstruction technique is then proposed to extract the signal components buried in array outputs,and finally recover the whole dynamic spatial spectrum.The sequentially recovered spectrum contains information about signal number and directions at each time instant,and the lasting period of each signal.As far as my knowledge goes,the sequential sparse reconstruction algorithm and the recovered dynamic spectrum have never been reported previously.The proposed method can be extended from narrowband signals to wideband ones with moderate modifications.