WFDB Signal Data Files

WFDB signal files contain the raw waveform data for one or more signals. Each file is typically in a binary format optimized for efficient storage and processing.

Signal files are referenced in the corresponding header file (.hea), which specifies the format used for each signal.

---

Overview of Formats

WFDB supports multiple signal encoding formats. These differ in how sample amplitudes are stored, and whether data is compressed or packed.

The key formats include:

Format	Description
8	First differences stored as signed 8-bit integers.
16	16-bit two’s complement integers (little-endian).
24	24-bit two’s complement integers (little-endian).
32	32-bit two’s complement integers (little-endian).
61	16-bit two’s complement integers (big-endian).
80	8-bit offset binary (unsigned 8-bit, subtract 128 to recover).
160	16-bit offset binary (unsigned 16-bit, subtract 32,768 to recover).
212	Packed 12-bit two’s complement samples (compact format, common in PhysioBank).
310	Packed 10-bit two’s complement samples (legacy format).
311	Alternative packed 10-bit samples (different packing from 310).
508, 516, 524	Signals compressed with FLAC (8, 16, or 24 bits per sample).

---

Commonly Used Formats

• Format 16 is a simple, widely supported format: each sample is stored as a 16-bit signed integer, little-endian.
• Format 212 is highly space-efficient: two 12-bit samples are packed into three bytes. Many PhysioNet records use this format.
• Formats 508/516/524 use FLAC compression, providing lossless compression for signals (with some restrictions).

---

Multiplexed Signals

When multiple signals are stored in the same file (multiplexed format), samples are interleaved:

sample 1 of signal 0
sample 1 of signal 1
sample 2 of signal 0
sample 2 of signal 1
...

This multiplexing order ensures synchronized playback and analysis.

---

Important Notes

• Byte Order:
  • Formats 16, 24, 32 store data little-endian (least significant byte first).
  • Format 61 stores data big-endian (most significant byte first).

• Scaling:
  The raw integer values are scaled to physical units (e.g., millivolts) using the gain and baseline values provided in the header file.

• Compression:
  FLAC-compressed formats (508, 516, 524) require specialized handling.
  They impose limits on the number of signals (maximum 8) and require synchronized sampling rates.

---

Summary

To correctly interpret a WFDB signal file:

• Read the corresponding header file to determine the format.
• Parse the signal data according to the specified encoding.
• Apply scaling factors to convert sample values to physical units.