Research On Parameter Estimation Algorithm For Local Scattered Distributed Sources

In recent years, characteristics such as open channel, complexity and diversity of environment, and random mobility of communication users have made distributed source a popular topic in studying the phenomena of scattered multipath, corresponding to the high-order signal model. The statistical parameters like the central direction of arrival (DOA) and angular spread have found their applications in mobile communication for local scattered source, low angle radar tracking system for echo sources and troposphere or ionosphere radio transmission for scattered source. There, distributed source has been a popular topic in array signal processing.In this thesis, we analyze the cause of distributed source and its characteristic. Based on the summary of research status, pivotal problems and main algorithms, this thesis has studied several algorithms for parameter estimation in local scattered environment. A propagator method has been proposed for estimating parameters jointly. A novel parameter estimator is suggested, which is very robust in case of large angular spread. In addition, the algorithms based on rotation invariance have been proposed for localization of distributed sources. Compared with the existing methods, our approach has reduced computational cost and can give effective parameter estimation in local scattered environment.Most of the existing subspace-like joint parameter estimation methods require an expensive computation in eigen-decomposing the covariance matrix of received signals. We have suggested a joint parameter estimation algorithm based on propagator method. In this method, the noise subspace is estimated by propagator method when the steering vector is expanded using its first-order Taylor series. Since there is no eigen-decomposition involved, the computational load is greatly reduced. In addition, we have also proposed a beamspace propagator method based on Schur-Hadamard product for joint parameter estimation. The parameters are estimated by beamspace data instead of sensor array processing with reduced dimension, which is suitable for more sensors processing fewer signals. To tackle the problem of large angular spread, we have suggested another parameter estimation algorithm based on Schur-Hadamard product beamforming. The integral steering vector is deduced to be Schur-Hadamard product between the steering vector of the point source and a real vector, which can avoid the integral in peak-finding searching. Since angular spread parameter is included in the constraint, the proposed algorithm is robust and can be applied to scenarios of large angular spread.Because of the two-dimensional joint searching manipulation in the algorithms shown above, the computational complexity is still high. We then suggest a decoupled algorithm based on the pre-estimation of the central DOA. By decoupling the computation of the central DOA and angular spread, we can simplify the two-dimensional joint searching to one-dimensional searching, which has reduced the computational complexity. Furthermore, the central DOA estimator doesn't depend on any pre-knowledge of angular signal intensity, which is robust to the distribution of scatters.In scenarios of local scattered multipath, rotation invariance between signal subspaces is destroyed. In solving the DOA estimation problem, we can relate the bases of two signal subspaces by utilizing a rotation matrix based on the approximate form of the steering vector expressed as Schur-Hadamard product. The central DOA can therefore be estimated by TLS-ESPRIT. The resulting DOAs don't depend on any assumption on the spatial distribution of the source and are hence robust to mismodeling. Thus it is applicable to distributed sources with different forms of angular signal intensity.The classical ESPRIT needs two eigen-decomposition, which is different in manipulation. We have derived another algorithm based on P-ESPRIT, for estimating the central DOA of coherently distributed sources. The rotation matrix between two subarrays is proven to hold based on the steering vector expressed as Schur-Hadamard product. Then we estimate the rotation matrix using propagator method, which leads to the closed form solution to the central DOA. As such, the peak-finding searching and eigenvalue decomposition are avoided, which significantly reduced the computational complexity compared with classical algorithms.To estimate the two-dimensional DOAs of local scattered signals, we have suggested a novel algorithm based on L-shape array for coherently distributed sources. By taking advantage of the array configuration, we can first obtain the elevation angle. And then estimate the azimuth angle by the second-order statistics based on Schur-Hadamard product. Since the proposed second-order statistics is not sensitive to white noise, we can obtain an improved performance. As such, the two-dimensional joint manipulation has been simplified as a two-step one-dimensional estimation problem, which has also reduced the computational complexity by a large scale. In addition, we also propose a two-dimensional DOA estimation algorithm using three subarrays X, Z and W. We first construct a second statistics with data collected at subarray X and then estimate the rotational invariance matrices by speculating the relationships between subarray X and Z, X and W using the propagator method. The closed solution to the azimuth and elevation angles can be obtained from the proposed second statistics and the rotational invariance matrices. Most importantly, because we don't apply any peak-finding searching and eigenvalue decomposition in this algorithm, we can significantly reduce the computational complexity.