Decoding motor imagery based on electroencephalography (EEG) is limited by high data noise and high model computational complexity. Starting from EEGNet, this study achieves high-accuracy decoding through three steps. First, frequency domain analysis was performed to reveal the frequency modeling patterns of deep learning models. Utilizing prior knowledge from brain science regarding the key frequency bands for motor imagery, we adjusted the convolution kernels and pooling sizes of EEGNet to focus on effective frequency bands. Subsequently, a residual network was introduced to preserve high-frequency detailed features. Finally, temporal convolution modules were used to deeply capture temporal dependencies, significantly enhancing feature discriminability. Experiments were conducted on the BCI Competition IV 2a and 2b datasets. The 2a dataset includes multi-channel data with 22 channels, while the 2b dataset contains low-channel data with only 3 channels, reflecting significant scenario differences. Our method achieved average classification accuracies of 86.23% and 86.75% respectively, surpassing advanced models like EEG-Conformer and EEG-TransNet. Meanwhile, the Multiply–accumulate operations (MACs) were 27.16 M, a reduction of over 50% compared to the comparison models, and the Forward/Backward Pass Size was 14.33 MB. This significantly reduced computational complexity and memory footprint. This paper employs the simplest and most fundamental techniques in its design, highlighting the critical role of brain science knowledge in model development. The proposed method demonstrates broad application potential.

We propose a hybrid Convolutional Graph Neural Network (C-GNN) for direction-of-arrival (DOA) estimation in sparse sensor arrays under low-snapshot conditions. The C-GNN architecture combines 1D convolutional layers for local spatial feature extraction with graph convolutional layers for global structural learning, effectively capturing both fine-grained and long-range array dependencies. Leveraging the difference coarray technique, the sparse array is transformed into a virtual uniform linear array (VULA) to enrich the spatial sampling; real-valued covariance matrices derived from the array measurements are used as the network’s input features. A final multi-layer perceptron (MLP) regression module then maps the learned representations to continuous DOA angle estimates. This approach capitalizes on the increased degrees of freedom offered by the virtual array while inherently incorporating the array’s geometric relationships via graph-based learning. The proposed C-GNN demonstrates robust performance in noisy, low-data scenarios, reliably estimating source angles even with very limited snapshots. By focusing on methodological innovation rather than bespoke architectural tuning, the framework shows promise for data-efficient DOA estimation in challenging practical conditions.

Accurate direction-of-arrival (DOA) estimation with high resolution is fundamental to many array sensing applications. In practice, however, sparse arrays with missing sensors and snapshot-limited observations often lead to incomplete and noisy second-order statistics, which substantially degrades the performance of conventional eigendecomposition-based estimators. In this paper, we propose a two-stage Transformer framework for sparse-array DOA estimation that explicitly separates correlation recovery from angle inference. The first stage operates in the correlation domain and learns to reconstruct a clean and complete correlation vector from partially observed measurements using masking-aware tokenization and global-context modeling. The recovered representation can be further converted into a structured covariance matrix, providing an interpretable interface to classical signal processing back-ends. Based on the recovered features, the second stage adopts a Transformer regressor to directly predict multi-source DOAs. Extensive simulations on a large-scale dataset with SNRs from −5 to 10 dB and various snapshot numbers demonstrate that the proposed method delivers robust accuracy and improved stability in low-SNR and snapshot-limited regimes, while maintaining competitive performance at higher SNRs. Additional evaluations with an ESPRIT back-end further confirm that the recovery-based covariance yields more reliable DOA estimation than conventional difference–coarray processing, with particularly evident gains under challenging noise conditions.

Direction-of-arrival (DOA) estimation is a fundamental task in array signal processing and is widely used in radar, sonar, wireless communications, and acoustic localization. Although classical methods such as MUSIC and ESPRIT can achieve high resolution under favorable conditions, their performance often degrades in challenging scenarios involving low signal-to-noise ratios, limited snapshots, and closely spaced sources. To address these difficulties, this paper proposes a Bayesian off-grid DOA estimation framework for close-angle and multi-source scenarios. The proposed method combines multi-measurement-vector evidence learning, diversified candidate construction, and multi-start joint continuous-manifold refinement so that multiple plausible close-angle hypotheses can be preserved and further optimized on the exact angular manifold. In this way, the proposed framework alleviates the source merging caused by high steering-vector coherence and improves estimation robustness in challenging conditions. Experimental results under close-angle, well-separated, varying-snapshot, and three-source settings demonstrate that the proposed method achieves competitive and, in many difficult cases, superior estimation accuracy compared with several representative baseline methods, confirming its effectiveness for robust close-angle DOA estimation.

Off-grid direction-of-arrival (DOA) estimation based on sparse Bayesian learning (SBL) can alleviate angular discretization mismatch, but its practical performance may be affected by unreliable posterior relevance statistics, sensitivity of effective error precision learning, and unstable offset correction. This paper proposes a reliability-guided stabilized off-grid SBL method for multisnapshot DOA estimation. The method is developed within the standard first-order multiple-measurement-vector Bayesian model and introduces three stabilization modules. First, a confidence-guided MAP-type shrinkage relevance update is introduced to suppress weak and non-dominant posterior components through reliability-controlled non-expansive shrinkage. Second, a posterior-concentration-guided damped noise update is introduced to stabilize scalar effective error precision learning when the sparse support is uncertain. Third, a trust-region cubic-regularized Newton refinement is formulated to obtain bounded active-support off-grid corrections from the posterior expected reconstruction error. Simulation results under off-grid deviation, varying SNRs, varying snapshot numbers, different source separations, and random-angle scenarios show that the proposed method achieves competitive and stable estimation performance compared with representative classical and sparse Bayesian baselines.