A high-speed DOA estimation method for low SNR underwater acoustic echo signal

Deep neural network (DNN) is widely used in direction-of-arrival (DOA) estimation due to its powerful non-linear data fitting ability and relatively fast real-time performance. However, due to the severe underwater background noise, this method has poor estimation accuracy in DOA estimation. In this paper, we propose a method based on matched filters and deep neural networks (MF-DNN). It utilizes matched filtering and data dimensionality reduction techniques to process the data features, which are then used as inputs to the model. A DNN is employed to construct the classification model. Matched filtering enhances signal-to-noise ratio, while a triangular matrix-based dimensionality reduction reduces input data dimension, lowering computational complexity and improving model generalization, and an adaptive matrix estimation algorithm within the DNN updates network model weights, reducing network size. The simulation results indicate that under low signal-to-noise ratio conditions, the proposed method significantly enhances azimuth estimation accuracy without compromising computational speed.