Technical writing
Signal insights.
Technical articles on BCI signal processing, motor rehabilitation, and clinical deployment — written for engineers and clinicians who need to understand what's actually happening in the data.
Why Raw EEG Data Fails in Clinical BCI Workflows
Electrode impedance drift, scalp movement artifacts, and session-to-session brain pattern variability make raw motor-imagery EEG nearly unusable outside a controlled research lab.
Read
Riemannian Geometry for Motor-Imagery Classification: A Practical Introduction
Covariance matrices of EEG lie on a curved symmetric positive definite manifold, not Euclidean space. Computing the mean and distance in Riemannian space dramatically improves classifier robustness.
Read
Three Approaches to Electrode Drift Compensation in Real-Time BCI
We compare re-referencing strategies, online Riemannian covariance updating, and Euclidean alignment for maintaining decode quality through a two-hour therapy session.
Read
Common Spatial Pattern Filter Banks: What They Do and When They Break
CSP works brilliantly when your training and test data come from the same session. Here is what happens when they do not — and why filter bank CSP helps.
Read
Mu Rhythm, Beta Rebound, and What They Mean for Motor Rehabilitation
Event-related desynchronization in the mu band and beta rebound are the neural signatures Synaptiq decodes. Understanding their physiological origin clarifies why a reliable classifier must track spectral changes in real time.
Read
IEC 62304 for BCI Software Teams: What It Actually Requires
A practical walkthrough of IEC 62304 for small BCI teams: software safety classes, unit test structure, change control, and software bill of materials.
Read
Online Adaptation vs. Offline Calibration: Trade-offs in Clinical BCI Deployment
Should your BCI decoder recalibrate once per session or adapt continuously? Both approaches have operational consequences for clinical workflow and regulatory documentation.
Read
Integrating a BCI Decoder with a Rehabilitation Exoskeleton: An Engineering Guide
Command latency requirements, safety interrupt protocols, and the electrical interface between a BCI software layer and an exoskeleton actuator stack.
Read
Navigating FDA 510(k) for Non-Invasive BCI Software: Predicate Devices and Software Level
For a BCI signal decoding component sold as a software accessory to a Class II device, the 510(k) pathway requires identifying a predicate with substantially equivalent intended use.
Read
The Inter-Session Non-Stationarity Problem: Why Your BCI Breaks on Tuesday
A classifier trained on Monday morning data routinely fails by Friday afternoon. Here is what is actually happening in feature space when covariance structure shifts.
Read
Breaking Down the Latency Budget in a Real-Time BCI Pipeline
100 milliseconds is the practical ceiling for perceptible motor intent delay. We account for every millisecond: amplifier buffer, USB, preprocessing, classification, command encoding.
Read
Euclidean Alignment: A Simple but Effective Session-to-Session Normalization for EEG
He et al.'s 2019 Euclidean alignment method re-centers each session's covariance matrix around the identity before Riemannian classification. A deep dive into why it works.
Read