Robust DOA Estimation Using Hybrid 1-D Convolutional Neural Network and Long Short-Term Memory Deep Learning Algorithm

This paper introduces a novel hybrid deep learning architecture combining a one-dimensional convolutional neural network (CNN) with a long short-term memory (LSTM) network to achieve reliable direction of arrival (DOA) estimation in single-input multiple-output (SIMO) systems. A key advantage of this deep learning-based approach is its resilience to array imperfections, multipath interference, and weak signal conditions, which often challenge traditional physics-based methods. The hybrid CNN–LSTM model is rigorously evaluated against other deep learning architectures, including deep neural networks (DNNs), LSTMs, and gated recurrent units (GRUs), to benchmark its performance. The method leverages the symmetric properties inherent in the covariance matrix of the received SIMO signals to construct a synthetic feature set, which is then utilized as input for the deep learning models under supervised learning. The results demonstrate superior performance of the proposed hybrid model, significantly enhancing DOA estimation accuracy. At a DOA estimation interval of 1∘, the hybrid CNN–LSTM achieves excellent performance compared to other architectures such as DNN, GRU, and LSTM. Comprehensive simulations further validate the robustness and precision of the proposed hybrid approach, illustrating its effectiveness in addressing key challenges in DOA estimation. The findings underscore the potential of this method for advancing the state-of-the-art in this domain.