Auditory Attention Decoding from Low Spatial Resolution Ring-Ear Wearable Electrodes: Implications for Intelligent Hearing Aids

High-density cap-EEG has achieved strong performance in auditory attention decoding (AAD), but its bulkiness and complexity limit its use in wearable applications such as daily-use hearing aids. Ring-ear wearable electrodes (cEEGrid) offer a promising alternative. However, their low spatial resolution poses significant challenges for reliable decoding, especially under short decision windows. In this study, we investigate whether auditory attention can be directly decoded from low-resolution periauricular electrophysiological signals recorded by cEEGrid, without relying on external audio information. To this end, we propose a lightweight neural network, cEEGridNet, designed for ring-ear recordings, and adopt an attention guidance protocol that better reflects realistic listening conditions. With a 1000 ms decision window, the proposed approach achieved left-right spatial auditory attention decoding accuracies of 64.90% and 62.92% under cross-trial and cross-subject validation, respectively. cEEGridNet outperformed the baseline models under both validation settings, but its generalization to unseen trials and unseen subjects remained limited. Challenges also remain in timbre-based attention decoding, and possible gaze-related contributions cannot be excluded.