Demo data with known answers

An 8-channel LabChart-style export, 300 s at 1000 Hz. Channel 6 is the stimulus: 5 V pulses of 5 s every 30 s from t = 30 s (9 pulses). Channel 8 is the LDF signal:

LDF(t) = 120 + 5·sin(2πt/400) + 3·sin(2π·0.13·t) + 1.5·sin(2π·6·t) + noise(SD 2)
         + Σ_onsets 30·g((t − onset)/4),   g(x) = x³·e^(3(1−x)) for x > 0, else 0

That is a ~120 PU baseline, a slow drift (period 400 s), vasomotion at 0.13 Hz, a 6 Hz cardiac ripple and white noise, plus a response of +30 PU peaking 4 s after every onset. The response shape is a gamma-variate function, a common model of slow haemodynamic responses^[12].

Cropped file

20–280 s of the same recording (stim, LDF, t, Fs); onsets at 10, 40, … 250 s.

Trials file

The 8 onsets whose −5 to +20 s window fits, block-averaged to 10 Hz (segmentedLDF, segmentedTime, Fs).

Ground truth

onsets, stimulus duration 5 s, response peak delay 4 s, amplitude 30 PU, baseline 120 PU.

Other formats

core/demo/demoLDFFormats.m writes the same recording at 100 Hz (every 10th sample) as the other files Extract LDF reads: a LabChart text export (channels Stimulus, Blood pressure and LDF, a comment at every onset), a PeriSoft-style table (semicolons and decimal commas), a table without a time column (the rate, 100 Hz, must be given), a Spike2 MATLAB export (an LDF waveform channel and a Stim event channel) and an EDF+ file (LDF and Stimulus, a Stim annotation at every onset). Each opens with the same answer.

Raw speckle images of a rodent cortex, 80 × 64 px (width × height), 10 frames per second for 90 s (900 frames), camera exposure T = 5 ms. Each pixel is an independent speckle: its intensity follows a gamma distribution with mean I₀ and contrast K (SD / mean = K exactly), with the exposure model of Bandyopadhyay et al. and Boas & Dunn^{[34, 33]}:

K² = β·(exp(−2x) − 1 + 2x) / (2x²),   x = T / τc,   β = 1
flow(t) = baseline · (1 + 0.25·g(u)),   g(u) = u³·e^(3(1−u)) for u > 0,   u = (t − onset) / 4 s

• Parenchyma: τc = T/20 (K = 0.221). Vessel (vertical band at x = 14–25 px): τc = T/200 (K = 0.071). Static region (bone / tape, x ≥ 68 and y ≥ 52): K = 0.7, no flow.
• Activated area: a disk at (x, y) = (52, 28), radius 12 px, whose flow (1/τc) rises by 25% after each stimulus, peaking 4 s after onset; the flow everywhere else does not change.
• Stimuli: 5 s pulses (5 V, variable stim) at 10, 30, 50 and 70 s.
• Illumination: Gaussian fall-off, mean intensity about 1300–2000 counts; a camera dark level of 100 counts is added; uint16 frames.
• ROIs (roiMasks, roiNames): Activated area (a disk of radius 7 px inside the activated disk, so 7 × 7 contrast windows centred in it stay inside), Control cortex (a disk of radius 8 px at (40, 50)) and Vessel (x = 18–21, y = 8–56).
• Ground truth (truth): frame rate, exposure, β, dark level, onsets, the baseline T/τc and K of every pixel, the activated mask, the response (25%, peak at 4 s), the true flow ratio in the activated area over time and the same seen through 1/K².

Expected results are on the laser speckle section.

30 s, 8 channels 100 µm apart, whisker stimulus pulses of 20 ms every 2 s from 1 s (15 stimuli). The raw channels are at the TDT rate 24414.0625 Hz; the LFP and stimulus are at 1017.25 Hz (24× decimation).

LFP

ERP(x) = −120 µV·exp(−(x − 15 ms)²/(2·(5 ms)²)) + 60 µV·exp(−(x − 40 ms)²/(2·(12 ms)²))
LFP_c(t) = profile_c·Σ ERP(t − onset) + shared(t) + local_c(t) + white noise (1 µV)
profile_c = exp(−(depth_c − 300 µm)²/(2·(150 µm)²))   → largest on channel 4

The background shared(t) is the same on every channel (an AR(1) process with coefficient 0.98, scale 3 µV), so it cancels in the CSD; local_c(t) is small independent AR(1) noise (0.9, 0.3 µV). Ground truth: onsets, sink channel 4, N1 at 15 ms, P2 at 40 ms, spacing 100 µm. The LFP file also stores the spacing as lfp_spacing_um (100), which fills Spacing (µm) in LFP Analysis.

MUA

Spike trains are Poisson with a 2 ms refractory period; each unit is seen on neighbouring channels with a gain of exp(−|distance|/0.8) (skipped below 0.05). Noise is white, 10 µV. The MUA file holds channels 3–5. Ground truth: every unit's spike times, rates and home channel.

Oscillation demo (core/demo/demoLFPOscillations.m)

The same LFP plus a 6 Hz theta rhythm (40 µV, identical on all channels, frequency wandering with SD 0.5 Hz, so not phase-locked to the stimuli) and a 40 Hz burst (10 µV, 50–250 ms after each stimulus with 25 ms raised-cosine ramps, channels 3–5, same phase at every stimulus).

Basic stack (DemoData.imagingStack)

96 × 96 px, 150 frames at 10 Hz. A dark vertical vessel at x = 60 whose diameter is 12 + 3·sin(2π·0.2·t) px (9–15 px, period 5 s); a bright red blood cell (Gaussian spot, sigma 2 px) moving down the upper part of the vessel at 2 px per frame (it never reaches the diameter line at y = 70); a cell at (24, 30), radius 6 px, with calcium transients of ΔF/F = 1 at 3, 7 and 11 s (rise 0.1 s, decay 0.8 s) added on top of the tissue background; sensor noise SD 0.02. roiMask marks the cell.

Advanced stack (core/demo/demoImagingAdvanced.m)

• Every frame translated by a sub-pixel shift that follows a smooth random walk, zero mean, at most 3 px in each direction (the true shifts are stored).
• Three cells: (22, 24) r = 6 px, events at 4.0, 8.5, 13.0 s; (26, 78) r = 5 px, events at 5.5, 10.5 s; (82, 30) r = 7 px, events at 7.0, 12.0 s (ΔF/F amplitudes 1.0, 0.8, 1.2).
• The vessel at x = 60 (12 ± 3 px, period 5 s) and a red blood cell moving 3 px per frame down the whole image, crossing the diameter line (35, 64)–(85, 64) about every 3 s.

Faults stack (DemoData.imagingFaults, demo_imaging_faults.mat)

A calcium imaging movie with the faults of the ROI Analysis checks; see Recordings with faults.

• Scalp: an oddball study of 8 participants, 32 channels at 250 Hz, trials from −200 to 800 ms, already cleaned; Standard (40), Target (15) and Novel (15) trials, 5 rejected per participant. P300 at 350 ms, largest at Pz: Target 10 > Novel 6 > Standard 2 µV; P1 +2 µV at 60 ms (Oz), N1 −5 µV at 100 ms (Cz), 10 Hz alpha over O1 / Oz / O2 in random phase; after Target the alpha amplitude halves from 350 to 650 ms (power −75%, −6 dB), a known answer for time–frequency that leaves the ERPs unchanged on average.
• Rodent: a continuous 60 s recording from 4 skull screws at 1000 Hz with 30 light flashes; a visual evoked potential over V1 (−40 µV at 50 ms, +25 µV at 100 ms), 30% of that over M1.
• Raw (eeg/raw, Try raw demo): 3 participants of the same oddball design as BrainVision Recorder files (float32, 500 Hz, about 100 s each), the same 32 channels recorded against FCz (the online reference, not in the data), markers S 1 / S 2 / S 3 = Standard (40) / Target (15) / Novel (15), 1.2–1.5 s apart from 2 s. The same P1 / N1 / P300 / alpha (and its decrease after Target) as the scalp case, plus electrode offsets (±400 µV) and linear drift (±1 µV/s), 50 Hz line noise of 5–15 µV per channel, a noisy T7 (150 µV white noise) and blinks (150 µV at Fp1 / Fp2, 50 ms SD) in 8 trials per participant and 6 more between trials. A 100 µV peak-to-peak rejection after a 0.1–30 Hz band-pass and the average reference (without T7) removes exactly the 8 blink trials. Ground truth: the blink trials and the N1 at Cz against FCz (about −2 µV) and against the average of the good channels (about −4 to −5 µV).
• The scalp and rodent data are written as EEGLAB (numbers inside, or in a .fdt file), FieldTrip, BrainVision and plain .mat files; the rodent recording also as EDF+, BDF (with a Status channel), XDF and an EEG-BIDS dataset. The scalp positions lie on an idealised spherical head; the rodent screws are in mm from bregma in the EEGLAB, FieldTrip and BIDS files. Scalp maps use them: the P300 over Pz, the N1 over Cz and, on the skull, the visual evoked potential over V1. Expected results are on the EEG page.
• Faults (eeg/faults/sub-01_faults.set, copied as demo_eeg_faults.mat): one participant with the faults the EEG checks look for; see Recordings with faults.

24 LDF trial files: the same 8 animals in three conditions (Control, Stimulated, Drug), 8 trials each (−5 to 20 s at 10 Hz), response peaking 4 s after onset on a 110–130 PU baseline.

amplitude(animal, condition) = trueMean(condition) + animal effect (SD 3 PU)
                              + animal × condition scatter (SD 2.5 PU)
trueMean = 18, 30, 24 PU (Control, Stimulated, Drug);  each trial adds jitter (SD 2 PU)

Population effects: Stimulated − Control = 12 PU, d_z = 3.27 (paired), Cohen's d = 3.02 (unpaired), ANOVA η² = 0.60. The realised per-file amplitudes are stored as well.

Faults study (groups_faults/, demoGroups(folder, struct('Faults', true))): 18 files, the same 6 animals in the three conditions, with the faults of the statistics checks; see Recordings with faults.

Seven demo files and one folder contain known faults, so each blood-flow, electrophysiology, imaging and statistics check can be seen to fire (open them from the demo folder; expected Checks are on the LDF, laser speckle, LFP and MUA, EEG, imaging and statistics pages).

• demo_ldf_faults.mat (DemoData.ldfFaults): the cropped LDF demo with the baseline rising linearly from 120 to 160 PU, a 0.3 s jump of +400 PU at 72 s (inside trial 3, onset 70 s) and 1 s at 0 PU at 31 s, between trials 1 and 2 (probe lifted).
• demo_lsci_faults.mat (demoLSCI(struct('Faults', true))): the speckle demo with each pixel the mean of 2 independent speckles (contrast 1/√2 of the model: cortex K = 0.16), the whole field (vessel, activated area, static corner, illumination) shifted 4 px to the right from 45 s, the illumination falling linearly by 20%, an exposure of 25 ms written in the file, and a fourth ROI, Thinned skull, on the static corner (K = 0.49).
• demo_perfusion_faults.mat (core/demo/demoPerfusion.m): perfusion images as an imager exports them, 80 × 64 px (width × height), one image per second for 90 s, kind 'perfusion', units PU: cortex 1000 PU, the vessel at 3400 PU clipped to 3000 PU (a typical export range), the activated disk +25% after each stimulus (10, 30, 50, 70 s), speckle-like noise (gamma, SD 10%).
• demo_lfp_faults.mat (DemoData.lfpFaults): the LFP demo (30 s, 8 channels 100 µm apart, 15 stimuli at 1, 3, … 29 s, the same noise) with three faults: a stimulus artefact, the same on every contact, of +800 µV at each onset decaying with a 3 ms time constant (still about 150 µV at 5 ms, where the N1 window starts); the response centred on the deepest contact (8) instead of 4, so the CSD sink is at the edge of the probe; and no lfp_spacing_um in the file, so the CSD uses the default 100 µm.
• demo_mua_faults.mat (DemoData.muaFaults): MUA like the demo (channels 3–5 at 24414 Hz, 30 s, a stimulus every 2 s from 1 s, the same spike shapes) with one fault per channel. Channel 3: noise of 40 µV instead of 10 over units of about 90 and 110 µV, so the spikes barely cross the threshold (low signal-to-noise). Channel 4: a unit of about 70 µV firing at random at 30 spikes/s without a refractory period, a quarter of its spikes followed 0.8–0.95 ms later by a second, smaller spike, next to a clean unit of about 90 µV (2 ms refractory period). Channel 5: the 110 µV unit shrinking steadily to half its size over the recording (amplitude drift).
• demo_eeg_faults.mat and eeg/faults/sub-01_faults.set (core/demo/demoEEG.m; the .mat is a copy of the EEGLAB dataset): one participant of the oddball design (Standard 40, Target 15 and Novel 15 trials in random order, 250 Hz, −200 to 800 ms, 32 channels recorded against FCz, the same P1 / N1 / P300 / alpha as the scalp demo) with three faults. Blinks (150 µV at Fp1 / Fp2) in 9 Target, 2 Standard and 1 Novel trials: a 100 µV peak-to-peak rejection removes 60% of Target but 5% of Standard trials and leaves 6 Target trials. 8 of 32 channels bad: FC6, T7, T8, TP9 and TP10 noisy (40 µV white noise), F8, PO9 and PO10 flat (0.2 µV); Suggest flags all 8 and nothing else, and unless they are marked bad every trial exceeds 100 µV. Pz interpolated: its data are the mean of P3, P4, CP1 and CP2, and the EEGLAB history ends with pop_interp of Pz. The history also lists the 0.1–30 Hz band-pass, the cut into trials and the baseline removal (−200 to 0 ms).
• demo_imaging_faults.mat (DemoData.imagingFaults): 96 × 96 px, 150 frames at 10 Hz, saved as a 12-bit camera saves them (uint16, 0–4095), with three cells 10 px across (soft-edged disks of radius 5 px; ROIs roiMasks / roiNames Cell 1–3 at the cells' mean position) on a textured background of about 600 counts, calcium transients of ΔF/F 0.8 (Cell 1 at 3 and 9 s, Cell 2 at 4 and 10 s, Cell 3 at 6 and 12 s) and noise of 20 counts. Motion: the whole field slides sideways from −7 to +7 px over the 15 s (14 px, more than a cell is wide). Bleaching: everything is multiplied by 0.65 + 0.35·exp(−t / 6 s), down to about 65%. Saturation: Cell 2 is 3400 counts above the background, so its transients go above 4095 and are clipped.
• groups_faults/ (demoGroups(folder, struct('Faults', true)), seed 20260926, trial jitter SD 1 PU): 18 LDF trial files like the group demo, the same 6 animals in Control, Stimulated and Drug, with set peak amplitudes. Control 18, 16, 19, 17, 20, 18 PU; Stimulated = Control + 12, 11, 13, 12, 12 and 40 PU (animal 6 responds three times as much: the values are not normal); Drug = Control + 6, 7, 5, 6, 7, 6 PU (the same rise in every animal while Stimulated − Control varies: sphericity violated).