Utilização de filtragem espacial e otimização numérica em um método de estimação de direção de chegada em arranjos de sensores

Abstract

The Improved SEAD estimator has a robust direction-of-arrival (DOA) estimation performance of closely-spaced signal sources. However, it presents some difficulties on the DOA estimation of widely-spaced sources and high SNRs. Furthermore, it shows a computational effort of exponential order. This is due to the algorithms used on estimate generation and refinement procedures. So, this work proposes changes to that estimator. The first one is to remove the amplitude threshold in the pre-selection stage that divides signal peaks from noise peaks. In that way, one can lower the probability of disregarding signal peaks on generation of initial estimates. The second proposal is to replace the discrete algorithm based on Branch-and-Bound in the refinement stage with a local optimization numerical method. That discrete algorithm has a high computational effort and produces a lower bound on estimation performance higher than the Cramér-Rao Bound (CRB). Thus, the Improved SEAD is not an asymptotically efficient estimator. In that sense, the Newton’s Method and a quasi-Newton method were developed, allowing for a significant reduction of that computational effort, besides producing DOA estimates in a continuous interval, effectively removing the lower bound on estimation performance at high SNRs. Finally, the technique of maximum spatial eigenfiltering recently presented allows for significant gains on the DOA estimation performance of closely-spaced sources. On the other hand, this eigenfilter may considerably attenuate widely-spaced sources when the SNR decreases. Then, the third proposal is applying spatial filtering to the Improved SEAD. Additionally, this work proposes two new filters that overcome the maximum spatial eigenfilter’s limitation on the DOA estimation of widely-spaced sources and present a good estimation performance considering closely-spaced sources. The result is the Modified SEAD, an estimator that has a competitive DOA estimation performance for both closely and widely-spaced sources and has a considerably lower computational effort to that of the Improved SEAD and another DOA estimator proposed in the literature.