| Abstract: | Efficiently performing high-resolution direction of arrival (DOA) estimation under low signal-to-noise ratio (SNR) conditions has always been a challenge task in the literatures. Obvi-ously, in order to address this problem, the key is how to mine or reveal as much DOA related in-formation as possible from the degraded array outputs. However, it is certain that there is no per-fect solution for low SNR DOA estimation designed in the way of winner-takes-all. Therefore, this paper proposes to explore in depth the complementary DOA related information that exists in spa-tial spectrums acquired by different basic DOA estimators. Specifically, these basic spatial spec-trums are employed as the input of multi-source information fusion model. And the multi-source in-formation fusion model is composed of three heterogeneous meta learning machines, namely neural networks (NN), support vector machine (SVM), and random forests (RF). The final meta-spec-trum can be obtained by performing a final decision-making method. Experimental results illus-trate that the proposed information fusion based DOA estimation method can really make full use of the complementary information in the spatial spectrums obtained by different basic DOA estim-ators. Even under low SNR conditions, promising DOA estimation performance can be achieved. |
