History of format changes in WFDB

[bemoody]

These are the changes I'm aware of, relating to the header and signal file formats and how they are interpreted by the WFDB library. (Annotation files are another story.)

Most of these changes are "bug fixes" of some sort, but since older applications are generally capable of reading the newer files (incorrectly), these can also be regarded as changes to the format. For example if you need to produce a database that can be read by WFDB 10.5.24, then you can't use gap segments in multi-frequency records.

10.0.1:

• (new feature) Format 311

10.2.0:

• (enhancement) Removed limit of 32 signals per record
• (enhancement) Removed limit of 4 samples per frame

10.4.0:

• (new feature) Variable-layout records
• (enhancement) Support for magic "#wfdb" comment
• (semantics) Minimum sample value treated as an invalid sample

10.5.0:

• (new feature) Formats 24, 32

10.6.0:

• (enhancement) Unpadded signal files (3n+2 samples) in format 311
• (enhancement) Variable-layout works with multi-frequency
• (implementation) Variable-layout sample scaling uses rounding rather than truncation

10.7.0:

• (new feature) Formats 508, 516, 524
• (enhancement) Removed limit of 254 characters per line
• (enhancement) Multi-segment works with byte offsets
• (enhancement) Variable-layout works with skews
• (implementation) Removed limit of 2^31-1 samples on 32-bit systems
• (implementation) Low-res sample averaging uses rounding rather than truncation

[cx1111]

Very useful, thanks! Semi related question: do 24 bit and 32 bit formats still represent NAN using -32768?

[bemoody]

To be clear, there's no such thing as "NaN" in the WFDB format (which concerns itself with integer sample values) nor in the libwfdb API (which concerns itself with "WFDB_Sample", an integer type.) "NaN" is a concept from IEEE 754 that has no direct relevance and no precise equivalent. The libwfdb API considers WFDB_INVALID_SAMPLE to represent an "invalid sample". In particular, the API needs to return some value when you're reading a variable-layout record and the signal in question is absent. WFDB_INVALID_SAMPLE is currently defined as -32768, and it would be difficult to change that. But as far as the signal file formats are concerned, -32768 is a perfectly ordinary sample value. Invalid samples within a segment should be represented by the minimum possible value, which is -8388608 in format 24, and -2147483648 in format 32.