Scalp to Source
Progress

Paths › path-realtime

path-realtime

Real-time and low-channel systems

For: Neurofeedback / wearable developers. 12 steps · ~12 h.

Why this order

A real-time system has two constraints an offline analysis does not: there is no future, and there is a deadline. This path is ordered so that each lesson tightens one of them.

What the hardware can deliver (L0.3–L0.5). Sampling rate, hardware bandwidth and trigger timing set the ceiling on everything downstream, and L0.3’s distinction between the output rate and the usable bandwidth is the one most often ignored on wearable devices. L0.4 and L0.5 build the ability to look at a live trace and say what is wrong with it, which is the only debugging tool an online system has.

Filtering, which is where causality becomes concrete (L1.5, L1.6). Offline filtering is zero-phase because it can use the future; online it cannot, so the filter delays the signal and that delay is part of the latency budget. L1.5 is the lesson to read twice. L1.6 matters because a notch is often the largest single contributor to that budget.

Cleaning that can run causally (L2.7). ASR and its relatives are here, and ICA is not, because a rolling ICA fit on a handful of channels is neither stable nor fast.

Feature extraction that fits in a buffer (L4.3, L4.4). Filter-Hilbert and multitaper give band amplitude with known latency, and L4.4 is where baseline normalisation — the step that decides what a feedback signal is relative to — is defined.

Then reporting and application (L6.6, L7.3, L7.5, L7.6). L6.6 comes before the applied lessons deliberately: neurofeedback and wearable work have a reporting problem before they have a technical one, and the standards are easier to accept before you have a system to defend. L7.3 assembles the latency budget, L7.5 says what a consumer-grade signal can and cannot support, and L7.6 covers the reliability question that any longitudinal or biomarker claim runs into.

The path skips the ERP and inference chains almost entirely; the entries below say which of them are worth returning for.

Paths do not add content; they filter and order the ladder (§7). Lesson pages keep their own prerequisite links.

Lessons (12; about 11 h 40 min of lesson time)

  1. L0.3 Amplifiers, sampling and recording · 45 min
  2. L0.4 Reading raw traces · 60 min
  3. L0.5 The artifact atlas · 75 min
  4. L1.5 Filters: FIR, IIR, and what they do to your data · 75 min
  5. L1.6 Line noise · 40 min
  6. L2.7 ASR, SSP and alternatives · 45 min
  7. L4.3 Multitaper and filter-Hilbert · 50 min
  8. L4.4 Baseline normalization and ERD/ERS · 60 min
  9. L6.6 Reporting standards · 40 min
  10. L7.3 Real-time processing and neurofeedback · 90 min
  11. L7.5 Mobile and consumer EEG · 60 min
  12. L7.6 Resting-state and biomarkers · 60 min

Skipped prerequisites (optional reading)

This is the most aggressively filtered of the four paths, so the list is long. Each entry is a prerequisite of a lesson the path includes; they are optional here, not removed, and the lesson pages keep their prerequisite links.

  • L0.2 Electrodes, montages and the 10-20 system — a prerequisite of L0.4, and the reference for what a four- or eight-electrode headset is and is not covering.
  • L1.2 Time and frequency domains — a prerequisite of L1.3; the Fourier grounding under every band-amplitude feature.
  • L1.3 Power spectral density — a prerequisite of L1.4 and L1.6, and where the variance of a short-window power estimate is explained — the reason an online band-power readout is noisy.
  • L2.6 EOG regression and ICA — a prerequisite of L2.7. Worth reading for the comparison: it is what ASR is an alternative to, and it explains what you give up by not using it online.
  • L4.2 STFT and Morlet wavelets — a prerequisite of L4.3 and L4.4; the window-length against resolution trade-off is the same trade-off as buffer length against latency.
  • L4.6 Is it really an oscillation? — a prerequisite of L7.6, and the lesson that decides whether a feedback signal is tracking a rhythm or a slope.
  • L6.1 The multiple-comparisons landscape — a prerequisite of L6.4 and L6.6; needed as soon as you evaluate a protocol across sessions, electrodes or bands.
  • L6.2 Mixed models and trial-level data — a prerequisite of L6.6; the natural model for repeated sessions within participant.
  • L6.3 Effect sizes, power and precision — a prerequisite of L6.6 and L7.6; the chapter behind any claim that a protocol works.
  • L6.4 Circularity and analytic flexibility — a prerequisite of L6.6, and the standard failure mode of neurofeedback evaluation.
  • L6.5 Decoding as inference — a prerequisite of L6.6; read it if the system classifies rather than thresholds.

Lessons in order

  1. L0.3 · Amplifiers, sampling and recording ~45 min
  2. L0.4 · Reading raw traces ~60 min
  3. L0.5 · The artifact atlas ~75 min
  4. L1.5 · Filters: FIR, IIR, and what they do to your data ~75 min
  5. L1.6 · Line noise ~40 min
  6. L2.7 · ASR, SSP and alternatives ~45 min
  7. L4.3 · Multitaper and filter-Hilbert ~50 min
  8. L4.4 · Baseline normalization and ERD/ERS ~60 min
  9. L6.6 · Reporting standards ~40 min
  10. L7.3 · Real-time processing and neurofeedback ~90 min
  11. L7.5 · Mobile and consumer EEG ~60 min
  12. L7.6 · Resting-state and biomarkers ~60 min