Miniaturized Wearable System for Multimodal EEG/ECG/EMG Sensing and Real-Time Physiological Monitoring

Figure 1. Overview of the miniaturized wearable multimodal biosensing platform. (a) Body-centric electrode placement with the assembled PCB (5.6 × 3.8 cm) and wireless data link to a host computer. (b) Photographs of the experimental setup showing EEG acquisition via scalp-mounted electrodes, EMG recording from the forearm, and ECG capture from the chest, alongside representative waveform previews for each modality. (c) System block diagram of the signal chain from the electrode header through input RC filtering, programmable gain amplification, digital decimation, and high-resolution ADC to the microcontroller and wireless communication module, with integrated power management for charging, battery supply, and voltage regulation.

Figure 2. ECG signal validation of the proposed wearable platform. (a) Continuous single-lead ECG recording over 50 s. (b) Progressive zoom-in: upper panel shows a 3 s segment with clearly resolved cardiac cycles; lower panel further magnifies two consecutive heartbeats with annotated P, Q, R, S, and T wave components and R–R interval. (c) Beat-to-beat instantaneous heart rate derived from R–R intervals, with a smoothed trend line indicating stable cardiac rhythm throughout the recording. The blue line with circular markers represents the beat-to-beat instantaneous heart rate, and the red line represents the smoothed trend line indicating stable cardiac rhythm throughout the recording.

Figure 3. Comparative analysis of EMG signals under active contraction (red) and resting (blue) conditions. (a,b) Raw EMG waveforms showing high-amplitude burst activity during contraction versus low-amplitude baseline at rest. (c) Overlaid root-mean-square (RMS) envelopes of both states. (d) Power spectral density comparison, revealing broadband elevation during muscle activation. (e,f) Time–frequency spectrograms showing periodic broadband energy bursts during contraction versus a uniformly low-energy background at rest.

Figure 4. Resting-state EEG spectral characterization. (a) Raw multi-channel EEG waveforms from four bipolar derivations (P3–O1, Fp1–F7, F7–T3, T3–T5), with an inset showing an enlarged segment of the Fp1–F7 channel. (b) Overlaid power spectral density (PSD) estimates across channels. (c) Time-resolved band-power curves showing dynamic spectral fluctuations over the recording period. (d) Band-power distribution for delta, theta, alpha, and beta frequency bands by channel.

Figure 5. CNN-based five-class epileptic state classification on the Bonn EEG dataset. (a) Schematic of the processing pipeline: EEG signals are segmented and converted to time–frequency representations via STFT, then fed into a CNN comprising convolutional, fully connected, and softmax layers. (b) Example spectrograms for each class: Z (healthy, eyes open), O (healthy, eyes closed), N (interictal, hippocampal), F (interictal, epileptogenic zone), and S (ictal), warmer colors indicate higher spectral power and cooler colors indicate lower spectral power. (c) Overall confusion matrix aggregated across five folds, blue color intensity represents the number of correctly classified samples, whereas red color intensity indicates misclassified samples. (d) Per-fold classification accuracy with a mean of 86.60% ± 2.97%.