Real-Time On-Chip Machine Learning Based Wearable Behind-The-Ear Electroencephalogram Device for Emotion Recognition

We propose a compact, low‑power, behind‑the‑ear (BTE) wearable electroencephalogram (EEG) device with an on‑chip machine learning (ML) pipeline for real‑time emotion recognition. The device integrates multichannel dry electrodes, low‑noise analog front end (AFE), microcontroller with on‑chip neural accelerator, and optimized signal processing and classification models to perform affective state estimation at the edge. This paper describes the hardware architecture, embedded signal preprocessing, feature extraction, lightweight deep learning model design, model compression and quantization methods, and an evaluation strategy using publicly available and in‑house datasets. We report design trade‑offs for power, latency, and accuracy, and present a roadmap for clinical and consumer applications. Results from an implementation prototype (hardware emulation and software-in-the-loop) demonstrate promising classification accuracy with sub‑200 ms end‑to‑end latency and sustained operation on a small battery for multiple hours. We conclude that on‑chip ML in BTE‑EEG wearables is a viable pathway toward private, low‑latency emotion‑aware applications.