Understanding Neural Signals:
Non-Invasive Interface Engineering
Investigating the computational principles of electroencephalography (EEG), high-fidelity biological signal acquisition, artifact rejection, and real-time neural decoding.
The Nature of Electroencephalography
Electroencephalography (EEG) records the macroscopic summation of postsynaptic potentials generated by synchronized pyramidal neurons in the cerebral cortex.
When millions of pyramidal neurons fire synchronously, their ionic currents generate extracellular voltage fluctuations that propagate through brain tissue, skull, and scalp. Non-invasive EEG measures these minute electrical potentials—typically ranging from 10 to 100 microvolts (µV)—using surface electrodes.
The primary advantage of EEG lies in its extraordinary temporal resolution. Neural dynamics operate in the millisecond regime, enabling instantaneous observation of cortical states that cannot be captured at equivalent timescales by metabolic imaging modalities like fMRI or PET.
However, scalp EEG presents a fundamental engineering challenge: extreme attenuation, spatial volume conduction through the cranium, and low signal-to-noise ratios (SNR).
Signal Acquisition & Transduction
High-precision telemetry demands careful analog front-end design to preserve biological fidelity prior to digitization.
10-20 Standard Placement
Standardized geometric placement across frontal (F), central (C), parietal (P), occipital (O), and temporal (T) regions ensures anatomical spatial correspondence and reproducible sensor coordinates.
Contact & Skin Interface
Maintaining low inter-electrode impedance (<5 kΩ) is essential to minimize thermal noise and eliminate capacitive voltage division at the stratum corneum skin boundary.
Common-Mode Rejection
Differential bio-amplifiers with high Common-Mode Rejection Ratios (CMRR >110 dB) and active driven-right-leg (DRL) circuits suppress pervasive 50/60Hz electromagnetic mains interference.
Digital Signal Processing Pipeline
Transforming corrupted multi-channel raw voltage data into orthogonal spatial and temporal features.
Signal Acquisition
High-impedance multichannel microvolt telemetry.
- 10-20 Standard Montage
- 24-bit ADC Sampling
- Active Shielding
Artifact Removal
De-noising & physiological noise suppression.
- 50/60Hz Notch Filtering
- 0.5–50Hz Bandpass
- ICA Ocular Rejection
Feature Extraction
Decomposing spatial & temporal spectral power.
- Wavelet Transforms
- Common Spatial Patterns
- Power Spectral Density
Neural Decoding
Mathematical mapping to user intentionality.
- Latent State Estimation
- Spatial Covariance
- Low-Latency Classification
Interface Synthesis
Real-time computational command generation.
- Deterministic Output
- Event Triggering
- Closed-Loop Feedback
Filtering, Artifact Removal & Feature Extraction
Algorithmic separation of true cortical dynamics from physiological contamination.
Blind Source Separation & Noise Rejection
Scalp potentials are invariably corrupted by high-amplitude non-cerebral sources: ocular blinks (EOG, up to 500 µV), scalp muscle clenching (EMG), and cardiac activity (ECG).
We employ Independent Component Analysis (ICA) to linearly unmix multichannel sensor arrays into statistically independent latent sources, identifying and projecting artifact components to zero without attenuating overlapping cortical waveforms.
Common Spatial Patterns & Wavelets
To isolate event-related desynchronization (ERD) and synchronization (ERS), spatial filtering algorithms such as Common Spatial Patterns (CSP) construct spatial filters that maximize the variance of one cognitive condition while minimizing it for another.
In parallel, Continuous Wavelet Transforms (CWT) provide adaptive time-frequency localization, avoiding the rigid time-bandwidth trade-offs of standard Fourier analysis.
Neural Decoding & Intent Synthesis
Mapping high-dimensional feature manifolds into deterministic computational directives.
Neural decoding translates extracted feature vectors into actionable intent. Because scalp EEG covariance matrices lie on symmetric positive-definite (SPD) Riemannian manifolds, Riemannian geometry approaches and spatial regularized classifiers offer enhanced robustness against session-to-session non-stationarity.
Exploiting Riemannian distance metrics on SPD matrices to overcome inter-session distribution shifts.
Streamlined numerical linear algebra pipelines optimized for execution on embedded DSP coprocessors.
Establishing sub-100ms real-time feedback loops to allow co-adaptation between human cortex and decoding algorithms.
Higgsion Research Direction
Qualitative exploration boundaries and our long-term engineering philosophy.
Our exploratory work in neurotechnology focuses on the foundational software and hardware hurdles preventing non-invasive neural interfaces from operating robustly outside controlled laboratory environments.
Edge Signal Processing: Investigating low-power digital signal processing pipelines capable of performing spatial decomposition and artifact rejection locally on wearable hardware before wireless transmission.
Adaptive Spatial Covariance: Exploring algorithmic frameworks that continuously track and compensate for baseline electrode drift and impedance changes during continuous operation.
Minimal Sensor Topologies: Researching channel-selection optimization techniques to achieve high decoding fidelity with sparse, ergonomic electrode arrays rather than cumbersome full-cap montages.