Hawkynt

tANS (Table-based Asymmetric Numeral Systems)

Table-driven ANS entropy coder over a 2048-state table. Symbols are spread with Duda’s precise initialization (slots ranked by the keys (2k+1)/(2f)) rather than the FSE pseudo-random walk this collection’s FSE implementation uses, and renormalization emits a whole precomputed bit group per symbol instead of peeling single bits. Order-0 model: the normalized frequency table is transmitted in the header.

Properties

Property	Value
Category	Compression Algorithms
Sub-category	Entropy Coding
Security status	🎓 Educational Only
Complexity	Advanced
Inventor	Jarek Duda
Year	2013
Origin	Not specified
Source	algorithms/compression/tans.js

Security

Status: 🎓 Educational Only

No vulnerabilities are recorded for this implementation.

Documentation

• Asymmetric Numeral Systems (original paper)
• ANS with applications to data compression
• ANS on Wikipedia (tANS section)

References

• Jarek Duda homepage
• Finite State Entropy (the FSE spread, for comparison)
• ANS discussion thread

Test vectors

8 vectors ship with this algorithm and run in the test suite. Byte values are hexadecimal.

Vector 1 — Empty input - length header only

Field	Value
input	(empty)
expected	00000000

Vector 2 — Single distinct symbol - whole table, zero bits emitted

Field	Value
input	41414141
expected	040000000b0100410008000000000000

Vector 3 — Two symbols, equal counts - interleaved spread

Field	Value
input	4142
expected	020000000b020041000442000400000200000040

Vector 4 — Two symbols, reversed order - pins the final state field

Field	Value
input	4241
expected	020000000b020041000442000401000200000000

Vector 5 — Natural text round-trip

Source: Regression test for table desync

Field	Value
input	74686520717569636b2062726f776e20 666f78206a756d7073206f7665722074 6865206c617a7920646f672e20
expected	(empty)

Vector 6 — All 256 byte values round-trip

Source: Regression test for table desync

Field	Value
input	000102030405060708090a0b0c0d0e0f 101112131415161718191a1b1c1d1e1f 202122232425262728292a2b2c2d2e2f 303132333435363738393a3b3c3d3e3f 404142434445464748494a4b4c4d4e4f 505152535455565758595a5b5c5d5e5f 606162636465666768696a6b6c6d6e6f 707172737475767778797a7b7c7d7e7f 808182838485868788898a8b8c8d8e8f 909192939495969798999a9b9c9d9e9f a0a1a2a3a4a5a6a7a8a9aaabacadaeaf b0b1b2b3b4b5b6b7b8b9babbbcbdbebf c0c1c2c3c4c5c6c7c8c9cacbcccdcecf d0d1d2d3d4d5d6d7d8d9dadbdcdddedf e0e1e2e3e4e5e6e7e8e9eaebecedeeef f0f1f2f3f4f5f6f7f8f9fafbfcfdfeff
expected	(empty)

Vector 7 — Long run round-trip

Source: Regression test for zero-bit renormalization

Field	Value
input	5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a 5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a 5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a 5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a … (512 bytes; the full value is in the source)
expected	(empty)

Vector 8 — Uneven frequencies round-trip

Source: Regression test for apportionment

Field	Value
input	414141424343444545454546
expected	(empty)

---

← All algorithms

This site is open source. Improve this page.