Crate oxideav_h261 

Pure-Rust ITU-T H.261 video codec (decoder + encoder).

Scope (ITU-T Rec. H.261 03/93):

• Picture header — PSC (20 bits: 0000 0000 0000 0001 0000), TR (5), PTYPE (6) with source format bit (0 = QCIF, 1 = CIF), PEI/PSPARE loop.
• GOB header — GBSC (16 bits), GN (4), GQUANT (5), GEI/GSPARE loop.
• Macroblock layer — MBA VLC (Table 1), MTYPE VLC (Table 2), MQUANT (5), MVD VLC (Table 3), CBP VLC (Table 4).
• Block layer — TCOEFF VLC (Table 5) + EOB + 20-bit (escape | run | level) escape, zigzag scan per Figure 12, dequantisation per §4.2.4, 8x8 IDCT, INTRA DC as 8-bit FLC per Table 6.
• INTRA / INTER / INTER+MC / INTER+MC+FIL macroblock types.
• Integer-pel motion compensation (range +-15 pels). Chroma MVs are the luma MV halved then truncated toward zero (§3.2.2).
• Loop filter (FIL bit of MTYPE) — separable 1/4, 1/2, 1/4 filter, §3.2.3.
• Output is YUV 4:2:0 in an oxideav_core::VideoFrame. Two picture sizes: QCIF 176x144 and CIF 352x288.
• Encoder: I + P pictures, integer-pel MC (spiral+diamond ME), per-GOB MQUANT rate control, FIL loop-filter RDO, full §4.2.3.4 MV-pred.
• BCH (511,493) error-correction framing (§5.4): the bch module wraps / unwraps the outer FEC multiframe layer (alignment pattern, Fi bit, BCH parity computation and per-frame syndrome check), and implements the spec-mandated t = 1 single-bit error correction via bch::locate_single_error / bch::decode_multiframe_with_correction. Sweep-tested across all 511 protected bit positions per frame.
• Hypothetical Reference Decoder buffer model (§5.2 + Annex B): the hrd module exposes the per-picture bit cap (256 kbits CIF / 64 kbits QCIF) and the HRD buffer-occupancy walk used to verify a coded sequence won’t underflow a conforming decoder’s receiving buffer at a given channel rate.
• RTP payload-format wrapping (RFC 4587): the rtp module packs an elementary stream into a sequence of GOB-aligned H.261 RTP payloads (4-byte header + bitstream slice) and unpacks them back, plus the §4.2 RECOMMENDED MB-level fragmentation (packetize_mb_fragmented) that parses the Huffman layer to split oversize GOBs on macroblock boundaries with the full §4.1 GOBN/MBAP/QUANT/HMVD/VMVD context. The RtpPacketizer glue wraps each payload in a full RFC 3550 §5.1 RTP fixed header (V/P/X/CC/M/PT/seq/timestamp/SSRC) so callers can hand pack_frame output straight to UDP / DTLS / SRTP.
• RTCP report packets (RFC 3550 §6.4): the rtcp module builds and parses Sender Report (SR, PT=200) and Receiver Report (RR, PT=201) packets — the control-channel companions to the rtp data path. RtpPacketizer tracks the sender’s running packet / octet counts so a conformant SR can be emitted directly from its session state.
• RTCP SDES / BYE / APP packets (RFC 3550 §6.5 / §6.6 / §6.7): the rtcp module also builds and parses Source Description (SDES, PT=202), Goodbye (BYE, PT=203), and Application-Defined (APP, PT=204) packets, plus compound-packet (§6.1) concatenation and walking. APP carries a 4-byte ASCII name + 32-bit-aligned application-dependent data for application-layer extensions that do not warrant their own IANA-registered packet type.
• SDP media-type signalling (RFC 4587 §6.1.1 / §6.2): the sdp module maps the video/H261 media type and its three optional parameters (CIF, QCIF, D) to the SDP a=rtpmap / a=fmtp attribute lines, with the §6.2.1 offer/answer helpers (per-format MPI frame-rate bound, the “at least one picture size” invariant, the RFC-2032 QCIF fallback).
• Annex D still-image transmission (§D.2 + §D.3): the annex_d module exposes the annex_d::SubImageIndex mapping to / from the 5-bit TR field, the still-image dimensions (4× the currently transmitted video format), and the Figure D.1 2:1 × 2:1 sub-sample / reassemble transform between a full-resolution still image and its four sub-images.

Out of scope:

• Multi-bit (weight ≥ 2) BCH (511,493) error correction. The generator polynomial has minimum distance d = 3, so the spec’s t = (d − 1) / 2 = 1 correction capability is the upper bound; weight-2 or denser error patterns are flagged (uncorrectable_frames in bch::DecodedMultiframe) but cannot be uniquely resolved by this code on its own — the inner H.261 video VLC layer handles those via GOB resync.

No runtime dependencies beyond oxideav-core, oxideav-codec, and oxideav-pixfmt.

Modules

annex_d

H.261 Annex D — Still image transmission.

bch

H.261 §5.4 — BCH (511, 493) forward error-correction framing.

block

H.261 block-layer decoding — §4.2.4 of ITU-T Rec. H.261.

decoder

H.261 decoder front-end — walks start codes, dispatches picture / GOB / MB parsing.

encoder

H.261 encoder — Baseline (I + P pictures, integer-pel MC, FIL loop filter).

fdct

Forward 8x8 DCT for H.261 (inverse of idct.rs).

gob

H.261 GOB layer parser — §4.2.2 of ITU-T Rec. H.261.

hrd

H.261 §5.2 + Annex B — Hypothetical Reference Decoder (HRD) buffer model.

idct

8x8 IDCT for H.261 (Annex A of ITU-T Rec. H.261).

mb

H.261 macroblock-layer decoding — §4.2.3 of ITU-T Rec. H.261.

picture

H.261 picture header parser — §4.2.1 of ITU-T Rec. H.261 (03/93).

quant

Forward quantisation for H.261 (inverse of the dequantisation in block::dequant_ac).

rtcp

RTCP control-channel packets — RFC 3550 §6.

rtp

H.261 RTP payload-format packing and unpacking — RFC 4587.

sdp

H.261 SDP media-type and fmtp/rtpmap parameter mapping — RFC 4587 §6.1.1 and §6.2.

start_code

Start-code scanning for H.261.

tables

VLC tables from ITU-T Rec. H.261 §4.2 (transcribed directly from the PDF).

Constants

CODEC_ID_STR

The canonical oxideav codec id for ITU-T H.261 video.

Functions

register

Unified registration entry point: install the H.261 codec factories into the codec sub-registry of a RuntimeContext.

register_codecs

Register the H.261 decoder and encoder with a codec registry.