Background/Objectives: Wearable affective human–computer interaction increasingly relies on sparse-channel EEG signals to ensure comfort and practicality in real-life scenarios. However, the limited information provided by sparse-channel EEG, together with pronounced inter-subject variability, makes reliable cross-subject emotion recognition particularly challenging. Methods: To address these challenges, we propose a cross-subject emotion recognition model, termed TSCL-LwF, based on sparse-channel EEG. It combines a multi-scale convolutional network (TSCL) and an incremental learning strategy with Learning without Forgetting (LwF). Specifically, the TSCL is utilized to capture the spatio-temporal characteristics of sparse-channel EEG, which employs diverse receptive fields of convolutional networks to extract and fuse the interaction information within the local prefrontal area. The incremental learning strategy with LwF introduces a limited set of labeled target domain data and incorporates the knowledge distillation loss to retain the source domain knowledge while enabling rapid target domain adaptation. Results: Experiments on the DEAP dataset show that the proposed TSCL-LwF achieves accuracy of 77.26% for valence classification and 80.12% for arousal classification. Moreover, it also exhibits superior accuracy when evaluated on the self-collected dataset EPPVR. Conclusions: The successful implementation of cross-subject emotion recognition based on a sparse-channel EEG will facilitate the development of wearable EEG technologies with practical applications.

Time-interleaved high-speed data acquisition (HS-DAQ) systems based on three-dimensional network-on-chip (3D NoC) architectures require efficient transmission of multi-channel sampled data. However, existing general-purpose NoC architectures are not well suited for the deterministic multi-source aggregation communication pattern of HS-DAQ systems, which may result in increased long-distance transmission overhead and localized traffic congestion. To address these issues, this paper proposes a 3D aggregation backbone network-on-chip architecture (3D AB-NoC) for HS-DAQ systems. This architecture is based on a 3D Mesh structure and enables the rapid transmission of long-distance aggregated data by establishing coarse-grained aggregation backbone subnets between defined aggregation routing nodes. Furthermore, a Backbone-Boosted XYZ (BB-XYZ) routing algorithm is proposed, which selects the next hop based on the remaining transmission distance of both the basic mesh path and the backbone path, thereby achieving coordinated routing for both local transmission and long-distance aggregation. Simulation results based on the Access Noxim platform demonstrate that the 3D AB-NoC maintains stable basic communication performance under general communication scenarios. In HS-DAQ communication scenarios, the proposed architecture reduces average latency by up to approximately 44%, improves throughput by approximately 29.3%, and significantly suppresses the growth of maximum latency under high-load conditions.

Background: Major depressive disorder (MDD) is a common psychiatric disorder. Electroencephalography (EEG) provides physiological information for depression detection, but many existing methods rely on dense multichannel recordings, limiting their use in lightweight EEG screening. Frontal sparse-channel EEG reduces acquisition burden but provides limited spatial information, requiring effective within-window representation learning and cross-window temporal modelling. Methods: We propose FSC-BiMamba, a dual-branch bidirectional Mamba network for subject-independent window-level MDD classification from frontal sparse-channel EEG. For each window, a Time–Frequency Map Encoder (TFME) learns local time–frequency patterns, while a Frequency-Domain Statistical Descriptor Encoder (FSDE) encodes complementary frequency-domain statistical descriptors. An Adaptive Dual-Token Fusion (ADTF) module integrates the resulting tokens through feature-wise gating to form a unified window representation. Representations from eight consecutive windows are then processed by a bidirectional Mamba (BiMamba) module to capture cross-window context. Finally, a softmax classifier converts each contextualized representation into a window-level class prediction. Results: Performance was evaluated using stratified subject-independent five-fold cross-validation. FSC-BiMamba achieved window-level accuracies of 82.70 ± 2.37% and 84.55 ± 8.01% on MODMA and Mumtaz2016, respectively. Together with its compact architecture, these results indicate a favourable balance between classification performance and model size. Conclusions: FSC-BiMamba effectively integrates complementary window-level representations with cross-window temporal context. It achieved the highest accuracy among the compared models on both datasets, demonstrating a favourable performance–size trade-off for lightweight EEG-based MDD screening.