Research On DOA Estimation Based On Compressive Sensing

The direction-of-arrival (DOA) estimation is a very important research direction in arraysignal processing field, which finds wide applications in many fields, such as radar,communication, seism, sonar and other national and military area. However, the classical spatialspectrum estimation methods require a great deal of independent identically distributed samplingdata, which is difficult to carry out in some field, and have a poor estimation under lowsignal-noise ratio case or multiple highly correlated signals circumstance. CompressiveSensing(CS) has been a new method in the research area of signal processing, and hasrepresented its advantages in many applications. To overcome shortcomings of the traditionalalgorithms, this paper conducts a study of CS and its application on DOA estimation. The mainwork and results are summarized as follows:1. The theoretical basis and the latest development about DOA estimation are summarized.The fundamental concepts and theories of array signal processing are discussed summarily, andthe models of narrow and wideband array signal processing are set up. Some classical algorithmsof DOA estimation which are applied widely are introduced.2. CS theory is deeply studied in this paper, including three core points: the sparserepresentation of signal, the design of measurement matrix and the reconstruction of signal. Tosolve the sparse reconstruction problem of multiple measurement vectors(MMV), where both theobservations and the dictionary are noisy, an alternating descent algorithm is proposed based onthe algorithm of Regularized-FOCal Underdetermined System Solver to Multiple MeasurementVectors(RM-FOCUSS). The simulation results show that the algorithm significantly improvesthe reconstruction quality of signals and has better performance than the RM-FOCUSS.3. Based on the model of joint-sparse representation of the array signal, this paper presents anovel Spatial Compressive Sensing(SCS) method for DOA. Firstly, it established anovercomplete atom dictionary according to the array geometry. Then it completed SingularValue Decomposition (SVD) of the array output to abtain lower dimensional signal subspace.Finally the sparse solution corresponding to minimization objective function is deduced usingthe iteration algorithm; it can obtain high-resolution estimation of DOA. Compared with thetraditional high-resolution methods, the proposed method is very preponderant with a smallnumber of snapshots and shows less sensitivity to the prior information of source number,besides it offers higher resolution and estimation accuracy under low signal-noise ratio case andmultiple coherent signals circumstance.4. The computational complexity of the SCS method is increasing with the source number，a novel method combing the CS theory and Khatri-Rao product dictionaty is proposed. Thedictionaty of this approach is using the self Khatri-Rao product of the dictionaty which is builtby array manifold and has better performance. Compared with the SCS methods, it can handlemore sources with the same sensors and has higher estimation performance under lowsignal-noise ratio case, besides its computational complexity is not increased with the sourcenumber. 5. The DOA estimation method using compressive sampling array(CSA) is studied. Thispaper proposes a novel SVD-CSA algorithm and CSA-MUSIC algorithm. Compared with thetraditional array architecture，the hardware complexity is significantly reduced because of muchsmaller number of frontend circuit chains in the CSA architecture, resulting in reducing the datewhich needs to be stored, transferred and processed. The random sample array whichcorresponds to the sparse matrix as the measurement matrix is proposed. This random arrayprovides a design frame which can easily implement the CSA or reduce the sensor numberpractically. Compared with the SCS methods, SVD-CSA algorithm is more robust to noise withthe same quantity of dada. CSA-MUSIC algorithm offers better performance than MUSIC, andhas a lower computational complexity.