WinEEG magic header

📂 Mitsar / WinEEG File Notes

EEG stack is 3 parts:

.EEG = raw stream (INT16 LE, multiplexed frames)

.ERD/.VHDR = meta, offsets, chans, fs, calibration pointers

.EVT/.VMRK = event / marker table

Diff from BrainVision: Mitsar doesn’t keep all calibration in the sidecar. Some of it lives inside the .EEG header itself.

🧩 .EEG Layout (reverse-engineered)

Header (0x000–0x3FF)

Magic words: 0x0F0E, 0xB7CC, 0x0065 → loader sanity check

Channel count, fs, frame size values

Calibration table: 1 byte per chan at 0x0326–0x0338

EEG chans: 0x0C = 12 µV/bit (default sensitivity)

Reduced sets: 0x03 = 3 µV/bit

Marker chans untouched (0x0A, 0x00)

Some garbage/patient/reserved blocks we don’t care about

Data section

Multiplexed INT16 LE

Frame = n_chans * 2 bytes

Alignment is strict: (filesize – header – trailer) % frame_size == 0

Trailer

Tiny footer/pad/CRC-ish block

Size varies (2–34 bytes observed)

LB EO EEG → 34 bytes, NP EC → 2 bytes

Wrong trailer = loader chokes (seek fail)

⚡ Calibration Hack

Breakthrough = those single-byte µV/bit values.

Stock = 0x0C (12 µV/bit)

Low-sens = 0x03 (3 µV/bit)

Ratio checks out (12/3 = 4×).

WinEEG multiplies sample × calib byte → µV.

If you don’t patch → WinEEG thinks everything is ~0 amplitude → blank grid.

Patch EEG chans to 0x01 (1 µV/bit) → traces appear, scaling knob works.

🔧 EDF → Mitsar .EEG Conversion Flow

Parse EDF

Pull chan count, fs, sample count

EDF must be uniform fs across chans (Mitsar doesn’t handle mixed rates)

Export raw INT16 LE

Scale EDF float → int16

Interleave channels (frame-major)

Splice into Mitsar template

Take a known-good Mitsar .EEG (same chan count)

Copy header/trailer verbatim

Overwrite middle frames w/ EDF samples

Keep head+tail frames if paranoid

Patch calibration bytes

EEG chans → 0x01 (1 µV/bit)

Marker chans (0 + 18) untouched

Validate trailer alignment

(filesize – 1024 – trailer) % (n_chans*2) == 0

Auto-detect trailer size if needed

✅ Rules of the Road

Header magics intact → otherwise WinEEG throws seek failed

Trailer correct → else misaligned, loader dies

Calibration patched → else invisible traces

Data is raw int16 LE multiplexed → big endian or planar = nope

Sampling interval exact (µs precision) → even tiny fs mismatch = bugged

🎯 EDF → Mitsar Recipe (final)

Load EDF + convert → raw int16

Grab template .EEG (match chan count)

Copy header + trailer unchanged

Replace data block (keep markers intact)

Patch 0x0327–0x0337 = 0x01 (EEG chans)

Save, check alignment, done

👉 TL;DR:
Mitsar .EEG is a template-driven container:
magic header + 1-byte calibration table + raw INT16 frames + fixed-len trailer.

Calibration patch was the last lock. Once you flip those bytes, EDF → EEG works and WinEEG plays nice.