Learning center

EEG spectra: density, numerical tables and saved sessions

Explore an EEG recording, understand spectral-density units and inspect numerical tables. The local spectral fits do not establish parity with the full specparam or FOOOF methods.

Explore an EEG recording, understand spectral-density units and inspect numerical tables. The local spectral fits do not establish parity with the full specparam or FOOOF methods.

1. Verify the recording and selected scope

Use the PhysioNet EEG Motor Movement/Imagery Dataset v1.0.0 S001R01 eyes-open EDF recording. The reviewed source contains 64 channels, 9,760 samples per channel at 160 Hz: 61 seconds. Its SHA-256 is [recorded file identifier]. Preserve channel labels, order, calibrated samples, source units and attribution with the original EDF.

Open Spectral Parameterization in Advanced EEG/MEG. For the recorded full-source scope, set Max channels to 64, Seconds to 61 and Sampling Hz to 160. A lower channel limit selects the first channels in source order; a shorter duration selects the beginning of the recording. It does not select a random sample, arbitrary channel set or later time window. Check the log’s actual channel/sample counts before comparing results.

The input must be a finite, regularly sampled signal matrix with declared channel/sample orientation and matching channel names. This app requires at least two channels and 64 samples. Event lists, spike times and connectivity matrices are not signal inputs. Missing or nonfinite samples require explicit preprocessing; constant channels cannot be standardized. If the source supplies a sampling rate, changing the control to a conflicting value is refused: resample explicitly instead of relabeling time.

2. Interpret standardized density, not absolute voltage power

Each selected complete channel is mean-centered and divided by its population standard deviation. A one-sided periodogram uses a symmetric Hann window, with no additional detrending inside the density calculation. The plotted density is therefore in 1/Hz, not volts squared/Hz or microvolts squared/Hz. Standardization changes amplitude interpretation even though the EDF loader preserves calibrated input.

For this exact 160 Hz, 9,760-sample example, the retained frequency range is 1–72 Hz and contains 4,332 bins. The bin spacing is 160/9760 = 1/61 Hz. More generally, the code retains 1 Hz through min(80 Hz, 0.45×sampling rate), with a positive-frequency fallback if that range contains fewer than eight bins. Do not assume every input has this example’s range or resolution.

The one-sided density doubles the interior positive-frequency bins, not DC or the even-length Nyquist bin. Zero density remains zero; a 10⁻¹⁸ floor is applied only for logarithmic fitting. Inspect the saved frequency and linear-density columns when reading small values rather than estimating them from a log plot.

3. Read local fit components and residuals

The recorded workflow selected eight descriptive routes. Use Fit spectrum, Aperiodic fit, Peak detection and Goodness of fit to orient your review, and retain the exact selected route list with the run. Do not assume every default-selected app route is covered by the recorded evidence.

The local two-pass descriptive fit separates a fixed aperiodic component and up to four qualifying peaks. Peak widths are full widths at half maximum using nearest bins. A separate bounded knee fit uses b − log10(k + fexponent). These implementations are not official specparam/FOOOF algorithm parity; a matching vocabulary or literature citation does not establish matching estimates.

Read residuals and fit diagnostics alongside the model line. An exponent is not a direct excitation/inhibition measurement, and a qualifying peak is not physiological ground truth. Electrical interference near 60 Hz, artifacts and preprocessing can affect the example. A no-peak channel must not be interpreted as having a measured zero-Hz oscillation. The independent density checks do not validate physiological peak identity or the fitted model parameters against specparam.

4. Inspect complete numerical tables

TableWhat to check in this recorded example
Channel spectral parameters64 rows in selected source order. The index is zero-based. Read channel name, exponent, intercept, R², RMSE, peak count, peak frequency, log-height, FWHM and corrected alpha power together.
Mean spectrum numeric data4,332 rows with Frequency Hz, Mean PSD (1/Hz), Aperiodic log10 PSD, Model log10 PSD and Residual log10 PSD. Linear density and logarithmic model columns have different meanings.
Fit Spectrum summaryVerify 64 channels, 9,760 samples, 61 seconds and 160 Hz, plus the explicit source-channel list, standardization, density convention and model boundary. Summary formatting is not the complete precision of the structured tables.

The first recorded mean-spectrum row is 1 Hz with density approximately 0.5077977251 in 1/Hz; the last is 72 Hz. Those are checks for this source and preparation, not expected values for another EEG recording. The saved numerical rows, rather than the number visible in the table viewport, define completeness.

5. Export tables and save a viewer session

  1. Use Export PNG for a selected figure, Export All Tables (CSV) for the complete table container and Export Report for the text report. Preserve all three kinds of output; a picture cannot replace numeric data.
  2. The all-tables CSV uses Table,Metric,Value. For a structured table, __columns__ identifies its columns and each row entry stores a JSON array in the Value cell. It is a multi-table container, not a flat single-table spreadsheet. Select the named table and decode its row arrays before counting or calculating with values. Text cells beginning with spreadsheet formula characters are escaped with an apostrophe: the recorded summary mean appears as '-1.02262e-06. This escape belongs to the CSV text representation; it does not change the complete numeric arrays stored in structured rows or the session.
  3. Use Save Session and then Load Session in the viewer to review saved results. The recorded session restored nine plots and sixteen tables, including all 4,332 mean-spectrum rows. A session preserves result content; it does not replace the source EDF, settings or a fresh reproducibility run.
  4. Open a PNG outside Quoneon and inspect the exported table container separately. Compare table names, column names, row counts and numerical values with the saved session. Avoid spreadsheet conversions that silently reinterpret numbers or labels.
  5. Retain a source/output checksum record when checking file identity. The manifest used in the recorded walkthrough was created by the external verifier; it is not a native product export feature. Do not claim that clicking Save Session produces that manifest.

6. Keep the evidence boundary with the results

The recorded workflow’s 209 checks cover source identity, EDF calibration, standardization and density, including all retained per-channel bins, ten independently computed scalar DFT bins and the actual Run mean-density table. This is one eyes-open recording, not a full-dataset study, diagnosis, motor-imagery classifier or whole-product accuracy claim.

Read the written guide’s source and evidence inventory, the recorded comparison report and the continuous recorded workflow. PhysioNet’s dataset page supplies the citation and Open Data Commons Attribution License v1.0. No new scientific run, video capture or deployed acceptance is claimed by this guide.

Worked examples describe particular inputs, methods and recorded results. Check current availability and input requirements before planning an evaluation.

← Back to the learning center