POWER EQUALIZATION MODEL BASED ON NEURAL NETWORKS FOR UNEQUAL POWER DOA ESTIMATION

In the case of unequal power sources, the traditional DOA estimation methods suffer from performance degradation in terms of resolution, especially when the signal angle separation is small. A non-end-to-end power equalization model based on neural networks is proposed to improve the quality of the covariance matrix when unequal power signals are incident, making it look like a power equalization signal. Based on that, we can obtain better estimation accuracy through DOA estimation methods like MUSIC. The numerical experiments demonstrate that the proposed method surpasses the original MUSIC algorithm in terms of success rate under conditions of unequal power sources with small angle separation and exhibits approximately the same estimation accuracy with the other subspace method like ERNS and IPNS. In addition, the method maintain good performance under low number of snapshot and significantly reduces processing time which makes it possible to use in real-time applications.