IPR / prior-art notes — Draft Generation 1
An engineering risk register for prior-art neighborhoods and intellectual-property uncertainty in Draft Generation 1.
View authoritative sourceScope and risk register
This is an engineering risk register, not legal advice or a patent clearance opinion. Avelune does not claim to be patent-free.
The project prefers old/openly described mechanisms where performance is comparable. If a technically superior mechanism with material IPR uncertainty is ever retained, that fact and the cost of avoiding it belong here rather than being hidden in implementation code.
| V1 mechanism | Prior-art neighborhood | V1 decision / risk note |
|---|---|---|
fixed 8x8 Walsh-Hadamard residual transform |
classical Hadamard/Walsh signal transforms |
retained; simple, old mathematical transform; no codec-specific patent clearance claimed |
translational block motion compensation |
decades of predictive video coding |
retained in a deliberately simple half-sample bilinear form; broad technique is old, but this register is not a freedom-to-operate opinion |
three spatial predictors (DC/left/top) |
longstanding image/video prediction |
retained; deliberately minimal |
static byte rANS |
ANS/rANS literature (Jarek Duda and subsequent work) |
retained after implementation testing; exact V1 model/syntax independently specified |
exact ⇐4-color block palette |
longstanding indexed/palette image coding and screen-content coding |
retained with a simple literal per-block syntax |
immutable explicit reference IDs |
project container/codec organization |
retained; avoids mutable reference-slot semantics |
reversible lifting Haar audio transform |
classical Haar/lifting construction |
retained; q=1 gives exact integer reconstruction |
reversible mid/side stereo |
longstanding lossless stereo decorrelation |
retained |
CRC-32C |
standardized checksum family |
retained for corruption detection, not cryptographic integrity |
Pre-freeze ideas not in V1
Pre-freeze ideas that are not in V1 include intra-block-copy, patch dictionaries, motion meshes/lattices, lapped transforms, DCT/DST alternatives, normative loop filters, LPC/pitch speech coding, and MDCT audio. Their earlier appearance in research notes does not make them V1 dependencies.