Neural Signal Processing

The course begins with the origin of a biosignal itself: the resting membrane potential, the action potential, and how this activity is expressed differently depending on the scale at which it is recorded. It then follows a signal through the complete acquisition chain of electrodes, amplification, and sampling including the electrical safety considerations any real recording system must satisfy. A substantial part of the course addresses noise and artifacts, since no real recording is clean; participants learn to identify physiological and environmental sources of contamination and apply detection and removal methods that preserve the underlying signal.

The course then moves into frequency-domain analysis: the Fourier transform, the Nyquist sampling principle, and the standard neural frequency bands, extended into nonstationary methods — the short-time Fourier transform, spectrograms, wavelet analysis, and event-related potentials for signals whose character changes over time. Filtering is treated as a practical design decision throughout, with low-pass, high-pass, band-pass, notch, and moving-average filters each tied to the specific problem they solve.

The course closes by applying these methods across the major recording modalities in neuroscience — EEG, ECoG, EMG, EOG, spikes and local field potentials, and MEG — and assembling them into a complete signal-processing pipeline. By the end of the course, participants will be able to take a raw neural recording, understand its contents, clean it, analyze it in both time and frequency, and extract information from it with confidence.