E-STNet: A Non-Ideal Array DOA Estimation Method Based on Enhanced Spatio-Temporal Features

To address the challenge of degraded DOA estimation performance under array errors and low signal-to-noise ratio conditions, this paper proposes an Enhanced Spatio-Temporal Network (E-STNet). This network adopts a dual-source input architecture. By integrating multi-scale pooling and a hybrid Long Short-Term Memory-Transformer (LSTM-Transformer) encoder, the network jointly refines spatial feature representations and captures multi-granularity temporal dependencies. Simulation results demonstrate that, under challenging scenarios such as array errors, low Signal-to-Noise Ratio (SNR), and closely spaced sources, E-STNet achieves higher estimation accuracy and stronger robustness than conventional algorithms and existing deep learning methods, providing an effective solution for DOA estimation in complex environments.