v0.0.7

Other

• ac4 round 306 — encoder-side aspx_hfgen_iwc_1ch/_2ch writers
• ac4 r299: multi-envelope ASPX body writers (num_env > 1) consuming the r292 packer
• ac4 round 292 — encoder-side TIME-direction ASPX envelope DPCM packing
• ac4 round 285 — real per-band β₃ for the 5_X ASPX_ACPL_3 encoder
• ac4 round 279 — decision-driven SAP-coded ASPX_ACPL_1 residual layer
• ac4 round 271 — SAP-coded alpha_q decision driver (select_alpha_q_for_pair)
• ac4 r263: build_chparam_info_none + select_ms_used_for_pair
• ac4 r260: encoder-side ChparamInfo builders — duals of extract_sap_abcd
• ac4 r257: SAP-aware ASPX_ACPL_1 residual-layer writer
• drop release-plz.toml — use release-plz defaults across the workspace
• ac4 r246: encoder-side Table-181 SAP residual extractor
• ac4 r243: encoder-side chparam_info() / sap_data() builders
• ac4 r240: encoder-side HF QMF energy aggregator (dual of Pseudocodes 90 + 91)
• ac4 r234: encoder-side ASPX envelope extractor (inverse of P82/83 + P80/81 DPCM)
• ac4 r226: write_aspx_data_{1,2}ch_real_envelope() builders
• ac4 r219: ASPX envelope value-emitting helpers (sig/noise F0/DF/DT)
• ac4 r215: real per-band γ₁ / γ₂ / γ₃ / γ₄ extraction in 5_X ASPX_ACPL_3 encoder
• ac4 r208: real per-band γ5 / γ6 extraction in 5_X ASPX_ACPL_3 encoder
• ac4 r202: real per-band α + β extraction in 7.0/7.1 ASPX_ACPL_2 encoder
• ac4 r196: real per-band α1/α2 extraction in 5_X ASPX_ACPL_3 encoder

Added

• Round 306 — encoder-side aspx_hfgen_iwc_1ch() /
  aspx_hfgen_iwc_2ch() writers (crate::encoder_acpl3). The exact
  duals of the decoder's aspx::parse_aspx_hfgen_iwc_1ch /
  parse_aspx_hfgen_iwc_2ch (ETSI TS 103 190-1 §4.2.12.6 / §4.2.12.7,
  Tables 55 / 56). Until now every encoder body writer emitted this
  HF-generation / interleaved-waveform-coding element as the all-zero
  compact form (aspx_tna_mode[*] = 0, all three presence bits 0),
  even though the decoder fully parses real inverse-filtering modes,
  additive harmonics (add_harmonic), frequency-interleaved coding
  (fic_used_in_sfb) and time-interleaved coding (tic_used_in_slot).
  New encoder_acpl3::write_aspx_hfgen_iwc_1ch /
  write_aspx_hfgen_iwc_2ch take real per-SBG tna_mode (2 b, masked
  to 0..=3) plus per-SBG / per-timeslot flag vectors via the public
  encoder_acpl3::AspxHfgenIwc1ChPayload /
  encoder_acpl3::AspxHfgenIwc2ChPayload payloads, and auto-derive
  every gate from the payload (*_present / *_left / *_right set
  iff the slice has an active flag in range; the 2ch TIC path uses the
  compact aspx_tic_copy = 1 form when both channels carry the same
  active pattern). Under aspx_balance = 1 only channel-0 tna_mode
  is written (decoder mirrors it); short caller slices zero-pad. The
  existing write_aspx_data_1ch_minimal HFGEN block is refactored to
  route through the new 1ch writer with a default payload — output
  stays byte-identical. Eight integration tests in
  tests/round306_aspx_hfgen_iwc_writers.rs pin the bit-exact
  round-trip through the decoder parsers (all-zero compact form, real
  flags, padding + masking, balance-mirror, distinct-tna, TIC-copy,
  TIC-right-only, full multi-field stress).

• Round 292 — encoder-side TIME-direction ASPX envelope DPCM packing
  (crate::encoder_acpl3). The dual of the direction_time == true
  branch of the decoder's aspx::delta_decode_sig /
  aspx::delta_decode_noise (ETSI TS 103 190-1 §5.7.6.3.4 Pseudocode
  80 / 81). The prior round-219/226/234/240 envelope-coding chain only
  emitted the FREQ direction (freq_dpcm_encode_qscf); the decoder also
  accepts a per-envelope direction flag and walks a TIME branch
  reconstructing qscf[sbg][atsg] = prev[sbg] + delta·values[sbg]
  (with prev the previous envelope's row, or qscf_prev_last for the
  first envelope). New encoder_acpl3::time_dpcm_encode_qscf inverts it
  exactly (values[sbg] = (qscf[sbg] − prev[sbg]) / delta), with
  zero-extend-short-prev and ±1-step semantics matching the decoder
  (delta = 0 treated as 1 for totality). New
  encoder_acpl3::dpcm_encode_qscf_envelopes packs a full
  qscf[sbg][atsg] matrix into per-envelope
  encoder_acpl3::AspxEncodedEnvelope { values, direction_time } rows,
  selecting the cheaper direction per envelope by minimising
  Σ|values[sbg]| (FREQ wins ties; force_freq reproduces the legacy
  single-direction scaffold). Twelve integration tests
  (tests/round292_aspx_time_direction_dpcm.rs) pin the bit-exact
  round-trip through both delta_decode_sig and delta_decode_noise,
  step/totality edges, short-prev zero-extension, the min-L1 policy,
  force_freq parity with freq_dpcm_encode_qscf, and empty inputs.
  Total tests 941 (was 929).

• Round 285 — real per-parameter-band β₃ extraction for the 5_X
  SIMPLE/ASPX_ACPL_3 encoder (crate::encoder_acpl3 +
  crate::encoder_ims). Closes the round-215 "β₃ stays at the
  round-95 zero-delta scaffold" deferral. Per ETSI TS 103 190-1
  §5.7.7.6.2 Pseudocode 118 steps 8-10, β₃ is the gain on the third
  decorrelator output y₂; step 10 + step 11 give the centre channel
  a wet contribution C_wet = −√2 · 0.5 · β₃ · y₂ carrying energy
  0.5 · β₃² · E[y₂²]. y₂ is decoder-side decorrelator state and
  unobservable at encode time, but its energy is not: the
  decorrelator + ducker chain is energy-preserving in steady state,
  so E[y₂²] ≈ E[v₃²] with the third-Transform drive
  v₃ = (γ₁+γ₃+γ₅)·x0in + (γ₂+γ₄+γ₆)·x1in (Pseudocode 118 step 2)
  fully determined by the carrier spectra and the quantised γ matrix
  the encoder is already emitting. New
  encoder_acpl3::extract_beta3_q_per_band_centre_residual energy-
  matches that wet contribution against the per-band least-squares
  remainder of the round-208 centre dry fit
  E_res = Σ (C − K·(γ₅·L + γ₆·R))² (K = 1 + √(1/2), using the
  quantised γ₅ / γ₆ the decoder will apply), giving the encoder
  decision β₃ = √(2 · E_res / E[v₃²]) — a non-negative magnitude,
  quantised per §5.7.7.7 Table 207 (beta3_q = round(β₃ / beta3_delta)
  with beta3_delta = 0.125 Fine / 0.25 Coarse and the symmetric
  ±cb_off clamp at ±8 / ±4 — half the BETA3 F0 codebook length
  per the staged ETSI table file §A.3 Tables A.46 / A.47). New BETA3
  value writers write_acpl_beta3_f0_value / write_acpl_beta3_df_value
  mirror the round-208 γ writers (symbol_index = q + cb_off
  addressing); a new full acpl_data_2ch() emitter
  write_acpl_data_2ch_real_alpha_beta_full_gamma_beta3 lifts the β₃
  entropy layer from zero-delta scaffold to real FREQ-direction DPCM
  codewords. New public builder
  encoder_acpl3::build_5_x_acpl3_body_from_pcm_spectra_real_alpha_beta_full_gamma_beta3
  is a drop-in over the round-215 full-γ builder with an extra
  beta3_scale decision knob, and new caller-facing entry points
  encoder_ims::Ac4ImsEncoder::encode_frame_pcm_5_0_acpl3_real_alpha_beta_full_gamma_beta3
  / _5_1_ accept [L, R, C, Ls, Rs] / [L, R, C, Ls, Rs, LFE] PCM.
  beta3_scale = 0.0 reproduces the round-215 byte stream exactly
  (the all-zero β₃ row emits exactly the zero-delta scaffold
  codewords). Four new unit tests pin the Table-207 quant grid +
  clamp, the BETA3 F0/DF writer round-trip through
  parse_acpl_huff_data + Pseudocode-121 accumulation, the
  zero-residual ⇒ β₃ = 0 / uncaptured-centre ⇒ β₃ > 0 decision
  split, and builder byte-equality at beta3_scale = 0. Six
  integration tests (tests/round285_5_x_acpl3_real_beta3.rs) pin
  5.0 → 5-channel and 5.1 → 6-channel decoder round-trips, the
  decode-side recovery of the exact per-band beta3_q row through
  parse_5x_audio_data_outer + differential_decode, IMS
  byte-equality with the round-215 entry at beta3_scale = 0,
  wire-liveness of the β₃ layer for an uncaptured centre, and
  bit-determinism. Total tests 919 → 929.

• Round 279 — decision-driven SAP-coded ASPX_ACPL_1 residual layer
  (crate::encoder_acpl3 + crate::encoder_ims). Wires the
  round-271 select_alpha_q_for_pair decision driver into the encoder
  proper, per ETSI TS 103 190-1 §5.3.4.3.2 / Table 181 + §5.3.2
  Pseudocode 59. New
  encoder_acpl3::select_acpl1_residual_chparam_pair runs the
  least-squares alpha_q decision per target (L, Ls) / (R, Rs)
  pair over the residual layer's single-window-group
  [max_sfb_master] layout — the residual layer's two
  chparam_info() payloads drive two independent 2x2 SAP systems
  mapping the transmitted (sSMP_A, sSMP_3) / (sSMP_B, sSMP_4)
  tracks to the preliminary front/surround pairs — and materialises
  the rows via the round-260
  build_chparam_info_sap_data_from_alpha_q builder, falling back to
  the header-only SapMode::None row when no band raises
  sap_coeff_used. The picked alpha_q is clamped to [-30, +30] so
  the pair-major DPCM deltas Pseudocode 59 accumulates stay within the
  HCB_SCALEFAC-codable [-60, +60] range on a worst-case sign flip.
  New encoder_acpl3::build_5_x_acpl1_body_from_pcm_spectra_sap_auto
  (+ caller-facing
  encoder_ims::Ac4ImsEncoder::encode_frame_pcm_5_0_acpl1_sap /
  _with_max_sfb) additionally closes the round-257 deferred carrier
  side: the two_channel_data() payload now carries the Table-181
  matrix-input carriers (sSMP_A, sSMP_B) recovered through
  invert_sap_table_181 — on a SAP-coded band the transmitted pair is
  (M, S − g·M) (mid + side prediction residual) rather than the raw
  L/R preliminaries the round-257 builder still emitted, so the
  decoder's apply_sap_table_181 forward mix reproduces the requested
  (L, R, Ls, Rs) preliminaries exactly (up to sf_data quantisation).
  Measured: for Ls = κ·L correlated surround the optimal projection
  g* = (1 − κ) / (1 + κ) collapses the transmitted residual to
  near-silence — SAP residual energy < 5 % (unit, synthetic spectra) /
  < 10 % (full PCM → MDCT → encode → decode integration) of the
  identity path's raw-Ls residual — while a no-benefit input
  (`L...

v0.0.6

Other

• ac4 r190: close ASPX_ACPL_1 desync — fix aspx_framing() FIXFIX prefix
• ac4 round 187 — pin ACPL_1 residual / α-β desync follow-up
• round 181 — close r128 alpha_q desync at parser indexing + aspx_data_2ch SIGNAL band count
• round 174 — fix ALPHA / BETA3 F0 cb_off (latent #1121 desync)
• round 144 — real per-band α + β extraction for 5_X ASPX_ACPL_2
• round 139 — 7.1-with-LFE ACPL_1 real per-band α + β
• round 135 — real per-band α + β extraction for 7_X ASPX_ACPL_1
• round 132 — real per-band β extraction in ACPL_1 5.0 encoder
• round 128 — real per-band α extraction in ACPL_1 5.0 encoder
• round 125 — 7.0 (3/4/0) SIMPLE/Cfg3Five multichannel encoder
• round 118 — 7.0/7.1 SIMPLE/ASPX_ACPL_1 multichannel encoder
• Round 114: 7.1 (3/4/0.1) SIMPLE/ASPX_ACPL_2 multichannel encoder (LFE)
• Round 107: 7.0 SIMPLE/ASPX_ACPL_2 multichannel encoder
• round 103 — 5_X SIMPLE/ASPX_ACPL_1 multichannel encoder path
• round 100 — 5_X SIMPLE/ASPX_ACPL_2 multichannel encoder path
• ac4 round 95: 5_X SIMPLE/ASPX_ACPL_3 multichannel encoder path
• ac4 round 91: 7.1 (3/4/0.1) SIMPLE/Cfg3Five encoder (7 SCE + LFE)
• ac4 round 80: 5.1 SIMPLE/Cfg3Five encoder (5 SCE + LFE) + decoder LFE PCM render
• ac4 round 74: 5.0 SIMPLE/Cfg3Five multichannel forward analysis (5 SCE)
• ac4 round 52: joint M/S CPE (Path B, b_enable_mdct_stereo_proc=1)
• ac4 round 51: stereo SIMPLE/ASF split-MDCT (Path A: 2x SCE) encoder

Fixed

• Round 190 — close the 5_X ASPX_ACPL_1 desync the r187 tests
  pinned. Two minimal A-SPX writers
  ([crate::encoder_acpl3::write_aspx_data_2ch_minimal] and
  [write_aspx_data_1ch_minimal]) emitted aspx_int_class = FIXFIX
  as the wrong prefix code: 0b11 (2 bits) instead of 0b0 (1 bit)
  per ETSI TS 103 190-1 Table 126. The decoder's
  [crate::aspx::AspxIntClass::read] correctly walks the prefix —
  0 → FixFix, 10 → FixVar, 110 / 111 → VarFix / VarVar — so
  the writer's 11 start signalled the parser to read the VarFix /
  VarVar branch instead. For our config that put the parser in the
  VarFix branch with b_iframe = 1: it then read
  var_bord_left (2 b), num_rel_left (2 b — num_aspx_timeslots = 15 > 8 makes Note-1 fields 2-bit wide), and tsg_ptr (2 b).
  Net: parser consumed 9 bits in the framing where the writer
  only emitted 3 bits, a 6-bit upstream drift that the
  silence / L-only / Ls-only test paths masked (α / β quantised to 0
  ⇒ the acpl_data_1ch body shape was constant minimum-cost on each
  side and the num_param_sets_cod bit positions on both sides
  sampled 0 within the long run of zero codewords). With non-zero
  α / β the codewords shift, the pair-1 num_param_sets_cod bit
  position lands on a 1, and pair1 reads num_param_sets = 2 —
  the r187 symptom. Fix is one line per writer: emit
  bw.write_bit(false) for the FIXFIX prefix, matching Table 126.
  • The r187 test #4 (acpl1_combined_l_and_ls_pair1_currently_misaligns)
    was renamed to acpl1_full_round_trips_with_aligned_pair_lengths
    and its assertion flipped from assert_eq!(n1, 2) (pinned
    misalignment) to assert_eq!(n1, 1) (post-fix). All four
    combinations now round-trip with
    pair0.num_param_sets = pair1.num_param_sets = 1.
  • Total tests 784 (unchanged from r187 — r190 fixed the third pin
    in place rather than adding new ones; the bit-level diagnosis is
    carried in the test file's module doc-comment instead).

Added

• Round 187 — characterisation tests pinning the remaining
  5_X ASPX_ACPL_1 residual / α-β-writer desync the r181 follow-up
  flagged. Four end-to-end pinning tests in
  tests/round187_acpl1_residual_desync_characterization.rs sweep
  the encoder's encode_frame_pcm_5_0_acpl1_real_alpha_beta across
  four input combinations and assert the decoder's recovered
  acpl_data_1ch_pair[0/1].framing.num_param_sets so the next round
  can iterate on the residual-layer / α-β writers without regressing
  the aligned silence / L-only / Ls-only paths.
  • Silence (all-zero PCM) → both pair slots resolve
    num_param_sets = 1.
  • L-carrier-only (Ls = Rs = 0) → both pair slots resolve
    num_param_sets = 1; the write_two_channel_data carrier writer
    is exercised non-trivially while α / β stay quantised to 0
    (correlation Σ L · Ls = 0 ⇒ α extractor returns 0; surround
    energy E[Ls²] = 0 ⇒ β extractor returns 0).
  • Ls-residual-only (L = R = 0) → both pair slots still resolve
    num_param_sets = 1; the write_acpl_1_residual_layer joint-MDCT
    residual writer is exercised non-trivially with max_sfb_master
    non-zero band budget but α / β stay 0 because carrier energy is 0.
  • Combined L-carrier + Ls-residual (L = 0.5, Ls = 0.05) →
    pair0 still resolves num_param_sets = 1, but pair1 drifts to
    num_param_sets = 2. The pin captures this as the currently
    expected behaviour so the next round's residual-layer fix can
    flip the assertion back to 1 once aligned.
  • Diagnostic narrative in the test file's module doc-comment
    triangulates the bug surface: the writer→parser pairs for
    write_acpl_data_1ch_real_alpha_beta ↔ parse_acpl_data_1ch
    are bit-exact in isolation (pinned by
    round181_alpha_desync_fix::standalone_*); back-to-back
    invocations into the same BitWriter without byte alignment
    between them also round-trip cleanly. The drift therefore sits
    upstream of pair0 — either in the joint-MDCT residual writer
    (write_acpl_1_residual_layer) vs the inline residual walk
    inside parse_aspx_acpl_1_2_inner_body's ASPX_ACPL_1 branch, or
    in the two_channel_data() L/R carrier writer vs
    parse_two_channel_data — when L and Ls are simultaneously
    non-trivial. Total tests 784 (was 780).

Fixed

• Round 181 — A-CPL acpl_huff_data() Pseudocode-121 indexing +
  aspx_data_2ch() SIGNAL band count. Closes the user's "alpha_q
  desync" follow-up the round-174 ALPHA / BETA3 F0 cb_off fix
  deferred. Two distinct layers were involved.

  • Layer 1 — §4.2.13.7 Table 65 / §5.7.7.7 Pseudocode 121 parser
    indexing. Pre-r181 [crate::acpl::parse_acpl_huff_data] packed
    the (num_param_bands - start_band) Huffman-decoded values into
    a vector starting at index 0. Per the spec the same array is the
    input to Pseudocode 121's for (i = 0; i < num_bands; i++)
    accumulation — so positions [0..start_band) are zero (the
    encoder did not transmit those bands) and the F0 codeword lands at
    values[start_band]. The packed-from-0 layout silently shifted
    the §5.7.7.7 DIFF_FREQ accumulation by start_band parameter
    bands for the 5_X SIMPLE/ASPX_ACPL_1 PARTIAL path
    (acpl_qmf_band > 0, start_band > 0). The r181 fix rewrites
    [crate::acpl::parse_acpl_huff_data] to return a length-
    num_param_bands vector indexed by full param-band number — the
    spec-aligned acpl_<SET>[ps][i] shape Pseudocode 121 reads. The
    [AcplHuffParam`] doc comment now spells out the new indexing
    contract.
  • Layer 2 — §4.2.12.4 Table 52 aspx_data_2ch() SIGNAL band
    count. Per ETSI TS 103 190-1 §4.3.10.4.9 (Table 124 NOTE 3)
    aspx_ec_data(SIGNAL, …) reads num_sbg_sig_lowres SIGNAL bands
    when the corresponding aspx_freq_res[env] bit was emitted as 0
    and num_sbg_sig_highres when it was 1 or absent (when
    freq_res_mode != Signalled the encoder writes no in-band
    aspx_freq_res bit and the decoder's
    freq_res.get(env).copied().unwrap_or(true) fallback selects
    high-res). Pre-r181 [crate::encoder_acpl3::write_aspx_data_2ch_minimal]
    hard-coded num_sbg_sig_lowres regardless — so for the
    encoder's default freq_res_mode = DurationDependent config
    (20-band high-res vs 10-band low-res, no in-band freq_res bit)
    the writer emitted 10 SIGNAL F0+DF codewords per channel while
    the parser read 20. The 20-vs-10 mismatch buried every
    subsequent acpl_data_1ch() α / β codeword in trailing
    zero-padding, recovered as length-num_param_bands all-zero
    rows. r181 keys the SIGNAL band count off cfg.signals_freq_res()
    matching the writer's own freq_res-bit gate.
  • 4 new round-181 unit / integration tests in
    tests/round181_alpha_desync_fix.rs:
    [standalone_alpha_writer_round_trips_through_parser] (Layer 1,
    standalone writer→parser→differential_decode),
    [parser_values_are_indexed_by_full_param_band_number] (Layer 1,
    structural — confirms values[0..start_band) == 0 and
    values[start_band..] carries the F0 + DF accumulation),
    [end_to_end_acpl2_asymmetric_surround_recovers_nonzero_alpha]
    (Layer 2 — encode 5.0 ASPX_ACPL_2 with asymmetric L/Ls energy,
    decode, assert acpl_data_1ch_pair[0/1].alpha1[0].values
    carries non-zero entries),
    [end_to_end_acpl2_silence_still_round_trips] (Layer 2 regression
    guard — silence input still yields all-zero α/β). Total tests
    780 (was 776).
  • The r132 acpl_data_1ch_real_alpha_beta_round_trips_byte_exact
    test is re-shaped to apply the spec-aligned Pseudocode 121 DIFF_FREQ
    accumulation directly on the parser's length-num_param_bands
    output (instead of cumulating the pre-r181 packed
    (num_bands - start_band)-length slice).
  • The 5_X SIMPLE/ASPX_ACPL_1 PARTIAL end-to-end path retains a
    separate joint-MDCT residual-layer alignment issue (the residual
    sf_data writer and the decoder's decode_asf_long_mono_body_with_max_sfb
    appear to read different total bit counts on non-trivial inputs)
    that the r181 ACPL_2 end-to-end tests do not exercise. Tracking
    as the remaining "alpha_q desync" follow-up — the structural
    Layer 1 + Layer 2 fixes are independent of it and land here.

• Round 174 — ALPHA / BETA3 F0 codebook cb_off corrected per ETSI
  TS 103 190-1 §A.3 Tables A.34 / A.35 (ALPHA) and A.46 / A.47 (BETA3).

  • Pre-fix cb_off = 0 for ALPHA / BETA3 F0 conflicted with the §5.7.7.7
    Pseudocode 121 ...

v0.0.5

Other

• ac4 round 50: section-boundary DP optimiser + SNF emission
• ac4 round 49: HCB1..11 codebook-selection optimiser + parameterised max_sfb
• ac4 round 48: forward MDCT + ASF entropy encoder for arbitrary PCM
• ac4 round 47: IMS bitstream_version=2 TOC parser + mono SIMPLE/ASF tone encoder
• round 46: AC-4 IMS encoder scaffold + ACPL_1 surround Ls/Rs spec audit
• ac4 round 45: stereo-CPE M=2 synced companding for ACPL_3 surround pair
• ac4 round 44: companding sync_flag=1 cross-channel exact synchronisation
• ac4 round 43: companding sync_flag=1/avg branches + ACPL_1 sb0 hookup
• ac4 round 42: cfg0/cfg1/cfg3 trailer-aware ASPX + §5.7.5 companding
• ac4 round 41: 5_X cfg2 ASPX trailers + Table 181 SAP for ACPL_1
• ac4 round 40: SAP a/b/c/d (Pseudocode 59) + Table 183 + Ls/Rs walker
• ac4 round 39: 5_X cfg0/cfg1/cfg3 dispatch + 7_X additional-channel pair
• ac4 round 38: LFE body decoder + cfg2_back_mono end-to-end + ACPL_3 centre
• ac4 round 37: wire 7_X ACPL_1/_2 dispatch + cfg0 centre end-to-end decode
• ac4 round 36: wire 5_X ASPX_ACPL_1 / ACPL_2 Pseudocode 117 into decoder
• Round 35: extend ETSI validation suite to float reference tables
• drop dead linkme dep
• round 35 — EMDF payloads_substream parser + DRC PCM gain application
• cargo fmt pass after round 34
• update round 34 status (SNF + FIXVAR/VARFIX/VARVAR atsg + ACPL_3)
• ac4 round 34: FIXVAR/VARFIX/VARVAR atsg + SNF inject + 5_X ACPL_3 wiring
• auto-register via oxideav_core::register! macro (linkme distributed slice)
• unify entry point on register(&mut RuntimeContext) (#502)

Added

• Round 50 — Section-boundary DP optimiser + Spectral Noise Fill
  (SNF) emission (TS 103 190-1 §5.7.4 section_data + §5.7.6 SNF +
  Pseudocodes 100 + 105 + Table 39/42 + Table SCFB):

  • encoder_asf::dp_optimise_sections(cost_band_cb, max_sections) —
    dynamic-program over scale-factor bands that finds the globally
    cheapest sequence of (start, end, cb) sections, paying the
    per-section header cost (4 + 3 * (floor((L-1)/7)+1) bits per
    Table 39) against each band's per-codebook bit cost. Section
    count capped at 16 per ETSI Table SCFB. Supersedes the round-49
    greedy run-length codebook-merge optimiser.
  • encoder_asf::section_overhead_bits(len) — closed-form per-section
    overhead in bits matching the spec's n_sect_bits=3 / esc=7
    long-frame layout: 7 bits for L ∈ 1..=7, 10 bits for L ∈ 8..=14,
    13 bits for L ∈ 15..=21, etc.
  • encoder_asf::build_band_codebook_cost_table(natural_q_per_band)
    — precomputed per-band per-codebook bit cost (rows of length 12;
    cb=0 cost 0 only for all-zero bands; HCB1..11 costs from
    bit_cost_for_band; u32::MAX for codebooks that can't represent
    the band's natural quant magnitudes). Drives the DP via O(1)
    prefix sums where every band is feasible for the codebook.
  • encoder_asf::build_sections_from_dp(sections, max_sfb) — lowers
    the DP-derived (start, end, cb) triples into an AsfSections
    suitable for the existing write_section_data /
    write_spectral_data_sections emitters.
  • encoder_asf::compute_snf_dpcm_for_zero_quant_bands(coeffs, sfb_offset, max_sfb, sfb_cb, max_quant_idx) — for each band that
    quantises to all-zero (cb == 0 || mqi == 0), estimates the band's
    RMS energy from the original MDCT coefficients and picks the
    HCB_SNF index whose snf_gain = 2^((idx*1.5 - 84)/4) best matches.
    Returns Some(per_band_idx) when at least one zero-quant band
    has measurable energy; None for fully-silent input.
  • encoder_asf::write_snf_data(bw, snf, sfb_cb, max_quant_idx, max_sfb)
    — emits b_snf_data_exists (1 bit) plus per-zero-quant-band
    HCB_SNF Huffman codewords per Table 42 / Pseudocode 105. Round-trips
    cleanly through the existing parse_asf_snf_data decoder path
    (round 36+).
  • encoder_asf::measure_greedy_vs_dp_bits(transform_length, max_sfb, coeffs) — diagnostic helper returning (greedy_bits, dp_bits)
    for any input spectrum so callers can quantify the section-boundary
    optimiser's contribution to total frame size.
  • Ac4ImsEncoder::encode_frame_pcm_with_max_sfb now drives the DP
    optimiser + SNF emission internally; existing call sites get the
    new path automatically with no API change. White-noise spectral
    SNR holds at 27.5 dB (round-49 baseline) with section overhead
    reduced and SNF emission turned on for high-frequency zero-quant
    content.

• Round 49 — Codebook-selection optimiser (HCB1..11) + parameterised
  max_sfb (TS 103 190-1 §5.7 + Pseudocodes 17 + 19 + 20 + Annex A.0
  huff_codes + Table SCFB):

  • encoder_asf::pick_best_codebook_for_band — per-band codebook
    optimiser sweeping HCB1..11 and choosing the lowest-bit-cost
    codebook whose q_max covers the band's natural quantised range.
    Anchor scalefactor targets peak quant ≈ 12 (HCB9/10's q_max → 3×
    more quantisation levels per band than the round-48 HCB5-only
    baseline). HCB11 always qualifies via its Pseudocode 20 escape so
    very-high-energy bands don't clip.
  • encoder_asf::bit_cost_for_band — precise bit-counter modelling
    HCB1..11 codeword length + sign bits for unsigned codebooks +
    per-Pseudocode-20 n_ext extension bits for HCB11 escapes.
    Mirrors the encoder emitter's exact bit shape (inline magnitude
    saturates at 16 for HCB11, sign bit per non-zero post-saturation
    line for unsigned codebooks).
  • encoder_asf::build_sections_from_per_band_cb — collapses runs
    of consecutive same-codebook bands into a single AsfSections
    entry so the emitted asf_section_data() honours the spec's
    grouping pseudocode without spurious cb-switch overhead.
  • encoder_asf::write_section_data + write_spectral_data_sections
    — multi-section asf_section_data + asf_spectral_data emitters.
    Per-section emission walks sect_start..sect_end bins with the
    section's codebook, handles cb == 0 silent bands, and writes
    Pseudocode 20 escape bits for HCB11 outliers.
  • Ac4ImsEncoder::encode_frame_pcm_with_max_sfb(frame, max_sfb) —
    new public entry point parameterising max_sfb (round-48 default
    was hard-coded to 40 → ~6.4 kHz at tl=1920 / 48 kHz). Pad target
    scales with max_sfb (2KB / 4KB / 8KB tiers). encode_frame_pcm
    keeps the round-48 default of max_sfb=40 for backwards
    compatibility.
  • White-noise round-trip SNR jumps from 13.6 dB (round-48 HCB5-only
    baseline) to 27.5 dB (round-49 HCB1..11 optimiser, q_target=12,
    max_sfb=50) — measured spectrally against the encoder's own MDCT
    coefficients pre/post quantisation. 1 kHz tone reconstruction at
    max_sfb=55 preserves >100% of input energy (vs ~40% at the
    round-48 max_sfb=40 default).

• Round 48 — Forward MDCT analysis + ASF entropy encoder for arbitrary
  PCM input (TS 103 190-1 §5.5 MDCT + §5.7 SIMPLE + §5.8 ASF +
  Pseudocodes 17-19 + Annex A.0 huff_codes):

  • encoder_mdct::mdct_naive + EncoderMdctState — forward MDCT
    direction complementing the decoder's mdct::imdct. Naive
    O(N²) direct-summation cosine basis (correctness-first; encoder
    isn't on a hot path). Sign convention + scaling matched against
    the decoder's IMDCT through a Princen-Bradley TDAC round-trip
    test (constant-signal recovery in steady-state middle frame
    within 1% error). EncoderMdctState carries the previous-frame
    N PCM samples for cross-frame 50% TDAC overlap.
  • encoder_asf::quantise_coeff + pick_scalefactor_for_band +
    encode_pair + write_sect_len_incr + write_scalefac_data +
    build_mono_simple_asf_body_from_pcm_spectrum — closed-form
    forward ASF entropy encoder for the long-frame, single-window-
    group, mono SIMPLE channel case. Per-band scalefactor selection
    via the closed-form solve sf_min = ceil(100 + 4*log2(max_abs/q_max^(4/3)))
    keeping every quantised line within HCB5's ±4 magnitude bound
    after q = round(sign(c)*|c/sf_gain|^(3/4)). Single-section
    HCB5 emission across 0..max_sfb; reference scalefactor
    • DPCM-coded per-band deltas via HCB_SCALEFAC; b_snf_data_exists = 0.
  • Ac4ImsEncoder::encode_frame_pcm(input: &[f32]) — public entry
    point taking arbitrary float PCM input (range [-1.0, 1.0])
    and emitting a structurally-valid IMS v2 frame end-to-end:
    forward MDCT analysis → per-band scalefactor + quantisation →
    HCB5 entropy coding → wrap in v2 IMS TOC + audio_size header.
    Lazily initialises a per-encoder EncoderMdctState on first
    call; bumps sequence_counter modulo 1024. Mono / 48 kHz / 24 fps
    by default (frame_len = 1920 samples, max_sfb = 40 covering
    bins 0..600 ≈ 7.5 kHz).
  • 13 new unit tests across the three modules cover: forward MDCT
    zero-in-zero-out + linearity + Princen-Bradley constant-signal
    • sine-wave SNR > 40 dB; quantise/dequantise round-trip;
      pick_scalefactor q_max bound; encode_pair signed/unsigned
      round-trip via huff_decode + split_qspec;
      build_mono_simple_asf_body_from_pcm_spectrum end-to-end parse
      via the existing ASF decoder; full encode → decode round-trip
      for 1 kHz tone (peak amplitude > 1000 i16), multi-tone
      250+500+1000 Hz (SNR > 10 dB on steady-state frame), and
      silence (peak < 50 i16); encode_frame_pcm bumps
      sequence_counter per call.
  • Codebook-selection optimiser (try HCB1..11 per section, pick
    min-bits), section-boundary optimiser (split bands by
    codebook), spectral noise fill, and stereo / multichannel
    forward analysis remain deferred for round 49+.

• Round 46 — AC-4 IMS encoder scaffold + ACPL_1 surround Ls/Rs
  ASPX-extension spec audit (TS 103 190-2 §6.2.1.1 / §6.3.2.5,
  TS 103 190-1 §4.2.6.6 Table 25):

  • encoder_ims::Ac4ImsEncoder — Auditor-mode AC-4 IMS encoder
    skeleton. Emits a structurally-valid raw_ac4_frame() payload
    with the IMS-flavo...

v0.0.4

Other

• land §5.2.5.2.2 Heuristic Scaling (round 33)
• env_prev tracking + walker state hoisting (round 32)
• land §5.2.3-5.2.7 PCM synthesis chain (round 31)
• land Tables 43-46 bitstream walker (round 30)
• land Annex C tables + arithmetic decoder core (round 29)

Added

• Round 33 — §5.2.5.2.2 Heuristic Scaling (Pseudocodes 27/28/29/30)
  (TS 103 190-1 §5.2.5.2.0 selector + §5.2.5.2.2):

  • New map_db_to_lin_q10() (Pseudocode 29) and map_lin_to_db_q10()
    (Pseudocode 30) — Q.10 fixed-point dB↔linear converters using the
    Annex C.14 SLOPES_DB_TO_LIN / OFFSETS_DB_TO_LIN /
    SLOPES_LIN_TO_DB / OFFSETS_LIN_TO_DB LUTs (already shipped in
    ssf_tables). Out-of-range inputs clamp to the spec's 100 << 10
    (dB→lin) and 40 << 10 (lin→dB) ceilings.
  • New heuristic_scaling() (Pseudocode 28) implements the full
    HeuristicScaling(iRfu, env_in, ...) -> int_weights_dB[] chain:
    dynamic-range compression on env_in[] when the spread exceeds
    the 40 Q.10 threshold, sort-descending of env_local[],
    Map_dB_to_Lin per band, weighted sum scaled by iRfu², reverse
    water-filling to find iTCurrLev, and a final per-band
    Map_Lin_to_dB(iTCurrLev) - env_local[band] weight that's clamped
    to [0, 15 << 10].
  • New apply_heuristic_scaling() (Pseudocode 27) wraps Pseudocode 28
    with the env_in = 3 * env_alloc pre-multiply, the LF-boost
    threshold (i_w_dB[0] knocked down by 3), and the
    env_alloc_mod = (env_alloc - i_w_dB).clamp(ENV_MIN, ENV_MAX) +
    f_gain_q = pow(10, 1.5 / 20 * f_w_dB) post-processing. Returns
    (env_alloc_mod[band], f_gain_q[band]).
  • synthesize_granule() now dispatches the §5.2.5.2.0 selector:
    when f_rfu > 0 && !variance_preserving AND the SSF bandwidths
    table is available, the heuristic-scaling branch fires and
    inverse_heuristic_scale() consumes the resulting f_gain_q[]
    instead of the previous all-1 stub. The variance_preserving
    block also correctly skips the inverse-scale call per
    §5.2.5.2.0 step 5. Pre-r33 the synth crashed (well, silently
    bailed) on f_pred_gain != 0 blocks; now they decode all the way
    through.
  • 11 new lib unit tests (470 → 481):
    map_db_to_lin_zero_input / map_db_to_lin_out_of_range_clamps /
    map_db_to_lin_monotone_within_table /
    map_lin_to_db_zero_input / map_lin_to_db_out_of_range_clamps /
    heuristic_scaling_zero_envelope_yields_zero_weights /
    heuristic_scaling_clamps_to_max /
    apply_heuristic_scaling_short_circuits_on_empty /
    apply_heuristic_scaling_clamps_env_alloc_mod /
    synthesize_granule_runs_with_heuristic_scaling_branch /
    synthesize_granule_variance_preserving_skips_heuristic.

• Round 32 — SSF SHORT_STRIDE env_prev tracking + walker state
  hoisting (TS 103 190-1 §5.2.3.0 Note 2, §5.2.3.0b Pseudocode 4b,
  §4.3.7.4.2 Pseudocodes 54-57):

  • SsfSynthState gains a new env_prev: Vec<i32> field that
    synthesize_granule() latches at the end of each granule with the
    resolved envelope (post-decode_envelope δ-chain), not the raw
    delta symbols. SHORT_STRIDE P-granules now use this latch as the
    env_prev[] interpolation input when the caller doesn't supply
    one — interpolate_envelope no longer degrades to a flat-zero
    envelope across frame boundaries on real P-frame streams.
    Ac4Decoder::run_ssf_channel drops its zero-vector
    state_idx_env_prev stub and passes an empty slice; the synth
    pulls from state.env_prev automatically.
  • Ac4Decoder adopts Vec<SsfChannelState> (one per channel,
    grown on demand) keyed ssf_walker_state. New
    walk_ac4_substream_stateful() and the matching
    _stateful variants of parse_mono_audio_data_outer,
    parse_stereo_audio_data_outer, parse_stereo_data_body, and
    parse_aspx_acpl1_mdct_body thread an
    Option<&mut [SsfChannelState]> through the SSF body parses so
    the walker's dither / noise RNGs (Pseudocodes 54-57) and
    prev_pred_lag_idx / last_num_bands / env_prev (raw symbol
    snapshot) persist across frames. The original public functions
    keep their pre-r32 signatures and delegate with None so the
    sibling repos / test fixtures stay binary-compatible.
  • 4 new lib unit tests (466 → 470):
    synthesize_granule_latches_env_prev (verifies state.env_prev
    holds the post-decode_envelope chain after each granule),
    short_stride_p_frame_uses_state_env_prev (proves a P-granule
    interpolates against the latched I-granule envelope, not zero),
    synthesize_ssf_data_chains_env_prev_across_granules (two-granule
    end-to-end), and
    walk_ac4_substream_stateful_persists_ssf_walker_state
    (round-trip through the substream walker leaves the channel-0
    state's last_num_bands / last_n_mdct / env_prev populated).

• Round 31 — Speech Spectral Frontend (SSF) PCM synthesis chain (TS
  103 190-1 §5.2.3 / §5.2.4 / §5.2.5 / §5.2.6 / §5.2.7 +
  §5.2.8.1 — Pseudocodes 4a / 4b / 4c / 4d / 4e / 26 / 31 / 32 / 33 /
  34 / 35 / 36 / 37 / 38 / 39):

  • New ssf_synth module turning the per-block indices on
    crate::ssf::SsfData into n_mdct spectral lines per block.
    Functions: decode_envelope (Pseudocode 4a — δ-decode chain over
    env_curr[]), interpolate_envelope (Pseudocode 4b — SHORT_STRIDE
    fixed-point linear interpolation between env_prev[] and the
    current granule's env[]), decode_gains (Pseudocode 4c —
    pow(10, gain_idx * 0.1) per block, LONG_STRIDE clamp to 1.0),
    refine_envelope (Pseudocode 4d — band-≥2 gain application + the
    round(2 * gain_idx / 3) allocation tweak with [-64, 63] clamp).
  • decode_predictor (Pseudocode 4e) reconstructs f_pred_gain from
    PRED_GAIN_QUANT_TAB and f_pred_lag = 640 * 2^((idx - 509)/170),
    with i_prev_pred_lag_idx carried forward on SsfSynthState.
  • compute_helpers (Pseudocode 26 — f_rfu,
    i_alloc_dithering_threshold, adaptive noise gains) +
    build_alloc_table (Pseudocode 31 — no-rfu path:
    env_alloc_mod = env_alloc).
  • inverse_quantize_block (Pseudocode 32) implements all three
    branches: i_alloc == 0 noise-RNG path with the
    variance-preserving band > 1 branch, dithered branch via
    Idx2Reconstruction + POST_GAIN_LUT[i_alloc - 1] + the
    f_post_gain_var_pres = sqrt(post_gain) * f_adaptive_noise_gain_var_pres
    rule, and the no-dither MMSE branch via mmse_laplace
    (Pseudocode 33).
  • inverse_heuristic_scale (Pseudocode 34) — currently a no-op
    because the no-rfu path leaves f_gain_q == 1.
  • build_c_matrix (Pseudocode 39) reconstructs the per-tab_idx
(2*Rf+1, 65, Rt) prediction-coefficient matrix from the
    quantized bytes in crate::ssf_pred_coeff using the
    1.1787855 * (q - 146) / 128 reconstruction formula and the
    spec's s = (-1)^(k+1) mirror rule for negative-η.
  • SubbandPredictorState::run (Pseudocodes 35 / 36 / 37) maintains
    f_spec_buffer[NUM_SPEC_BUF=5] + f_env_buffer[NUM_ENV_BUF=4]
    histories, runs the model-based extractor (f_period, k_s,
    tab_idx, Z-matrix even-reflection, the per-bin
    Σ_{ν,k} s * C[ν][f][k] * Z[bin+ν][k] summation), then applies
    Pseudocode 37's per-band shaper (f_envelope * f_pred_gain)
    with the I-frame integer_lag = 0 clamp.
  • inverse_flatten (Pseudocode 38) sums f_spec_res + f_spec_pred
    and multiplies by the per-band signal envelope.
  • synthesize_granule() runs the chain across every block in one
    granule; synthesize_ssf_data() runs it across both granules of
    one frame, threading env_prev[] between them.
  • Ac4Decoder adopts Vec<SsfSynthState> (one per channel) and
    consumes tools.ssf_data_primary / tools.ssf_data_secondary
    after the existing ASF/A-CPL pipeline: each granule's
    num_blocks * n_mdct spectrum is split per-block, fed into the
    per-channel KBD-windowed IMDCT + overlap-add, then truncated /
    padded to the frame's sample count. SSF substreams now emit real
    PCM instead of silence.
  • 16 new lib unit tests (450 → 466) covering: empty/empty-tail
    envelope decode, low-band-no-gain refinement, allocation-table
    clamping (min + max), zero-RFU + unit-window helpers, predictor
    presence/absence + delta-lag carry, full C-matrix dimensions
    across all 37 tab_idx values, the negative-η mirror rule, the
    subband-predictor zero-gain pass-through and finite-output smoke
    tests, the per-band envelope-gain inverse-flattening test, and a
    LONG_STRIDE I-granule synthesis end-to-end smoke (synthesize_granule
    on a synthetic granule). Plus one decoder-level integration test
    (ssf_synth_long_stride_iframe_end_to_end) that walks a synthetic
    SSF bitstream through parse_ssf_data then synthesize_ssf_data
    and verifies finiteness + zero-padding past num_bins.
  • Note: §5.2.5.2.2 Pseudocodes 27 / 28 / 29 / 30 (full Heuristic
    Scaling) are deferred — when f_pred_gain == 0 the spec
    short-circuits to env_alloc_mod = env_alloc + f_gain_q = 1,
    which is the no-rfu path landed here. Synthesis of streams that
    enable the predictor across many bands at once with
    variance_preserving == 0 will lose the heuristic envelope
    spreading until a follow-up round.

• Round 30 — Speech Spectral Frontend (SSF) bitstream walker (TS 103
  190-1 §4.2.9 / §4.3.7 + §4.3.7.5 + Tables 43-46 / 111-113):

  • New ssf module with the four-table walker family:
    parse_ssf_data (Table 43 — b_ssf_iframe gate plus 1 / 2
    granules per frame_length >= 1536), parse_ssf_granule
    (Table 44 — stride_flag, I-frame num_bands_minus12, per-block
    predictor_presence_flag / delta_flag loop), parse_ssf_st_data
    (Table 45 — env_curr_band0_bits, I-frame
    env_startup_band0_bits, per-block gain_bits /
    predictor_lag(_delta)_bits / variance_preserving_flag /
    ...

v0.0.3

Other

• skip etsi_table_validation when docs sibling absent
• replace never-match regex with semver_check = false
• migrate to centralized OxideAV/.github reusable workflows
• round 28 — mono / stereo short-frame sf_data(ASF) walker
• round 27 — 7_X channel-element walker (immersive 7.0 / 7.1)
• round 26 — add per-codebook decode roundtrip sweeps
• round 25 — ASPX_ACPL_1 / ASPX_ACPL_2 inner body walker
• round 24 — grouped multichannel sf_data(ASF) walker + ASPX_ACPL_3 inner body walker
• round 23 — multichannel sf_data(ASF) Huffman codebook table walk
• round 22 — ASPX_ACPL_1/2 multichannel wrapper (Pseudocode 117) + 5_X-walker glue
• round 21 — ASPX_ACPL_3 transform synthesis (Pseudocodes 118/119)
• round 20 — ETSI Huffman table audit + 5.X cfg0/1/2 + sf_info_lfe
• round 19 — design 5_X channel-element walker family
• round 18 — wire ASPX_ACPL_1 joint-MDCT residual layer
• round 17 — wire A-CPL synthesis into Ac4Decoder
• adopt slim AudioFrame shape
• land §5.7.7 A-CPL QMF synthesis math (round-16)
• outer §4.2.14.1 metadata() walker + §5.7.7.2 sb_to_pb
• A.4 Huffman codebooks + dialog_enhancement parser
• A.3 Huffman codebooks + acpl_data_*ch parser
• A.5 Huffman codebook + drc_frame parser
• implement complex-covariance TNS (chirp + α0 + α1) — round-11
• land §5.7.6.4.2.2 A-SPX limiter (P72 + P96..101) — round-10
• pin release-plz to patch-only bumps

Added

• Round 28 — mono / stereo short-frame sf_data(ASF) walker (TS 103
  190-1 §4.2.8.3-6 Tables 39-42, §4.3.6.2.6 Pseudocodes 2/3/5):

  • New spec-correct _grouped payload parsers in asf_data.rs —
    parse_asf_section_data_grouped(),
    parse_asf_spectral_data_grouped(),
    parse_asf_scalefac_data_grouped(),
    parse_asf_snf_data_grouped() — each takes per-group transform-
    length and max_sfb arrays and walks the spec's outer
    for (g = 0; g < num_window_groups; g++) loop. Critically:
    asf_scalefac_data() consumes a single 8-bit
    reference_scale_factor at the head with first_scf_found shared
    across groups (DPCM state is continuous over the whole frame), and
    asf_snf_data() consumes a single 1-bit b_snf_data_exists
    gate at the head. This matches Tables 41 / 42 verbatim.
  • New helpers in asf.rs:
    derive_per_group() / derive_per_group_with_max_sfb() resolve
    per-group (transf_length_idx, transform_length, max_sfb) from
    (AsfTransformInfo, AsfPsyInfo) per Pseudocodes 2 (get_transf_length)
    and 5 (get_max_sfb), including the b_different_framing
    half-frame split (Pseudocode 3's grouping-bit shift +
    num_windows_0 - 1 boundary injection).
  • New body decoders:
    • decode_asf_grouped_mono_body[_with_max_sfb]() — wraps the four
      _grouped payload parsers; returns the per-group dequantised
      spectra concatenated group-major.
    • decode_asf_grouped_stereo_joint_body() — joint-MDCT residual
      layer with shared section, two independent spectral bodies (L/M
      then R/S), shared scalefactors (band-wise max_quant_idx over
      both channels), per-group ms_used[g][sfb] flag arrays, then
      snf. Inverse M/S applied per-group: L = M+S, R = M-S for bands
      with ms_used set.
    • decode_asf_mono_body_dispatch() /
      decode_asf_mono_body_for_max_sfb() — long-frame vs grouped
      dispatch wrappers used by all per-channel call sites.
  • Wired into the four mono / stereo call sites:
    • parse_mono_audio_data_outer() — mono SIMPLE / ASPX path.
    • parse_aspx_acpl2_mdct_body() — single-channel ASPX_ACPL_2
      MDCT residual.
    • parse_aspx_acpl1_mdct_body() joint + split — ASPX_ACPL_1
      joint-MDCT residual layer (two independent mono bodies with
      max_sfb_0 / max_sfb_side_0) and the split case.
    • parse_stereo_data_body() joint + split — stereo CPE body
      with both joint MDCT (shared section + ms_used) and split MDCT
      (two independent mono bodies).
  • Real Dolby AC-4 mono / stereo streams that include short-frame
    sf_data(ASF) (i.e. the encoder picks short-window sub-frames)
    now decode end-to-end without bailing at the
    num_window_groups != 1 guard. The grouped multichannel walker
    in mch.rs from r24 (per-group interleaved
    section + spectral + scalefac + snf) is left untouched — its
    pinned tests continue to pass.
  • 9 new tests: 4 in asf_data.rs (grouped section / scalefac
    reference-once / scalefac DPCM-state-carries / snf gate-once) and
    5 in asf.rs (decode_asf_grouped_mono two-group + truncated;
    parse_mono_audio_data_outer SIMPLE short-frame; parse_stereo_data_body
    split + joint short-frame). 425 tests (414 lib + 5 + 6
    integration), up from 416.

• Round 27 — 7_X channel-element walker (immersive 7.0 / 7.1)
  (TS 103 190-1 §4.2.6.14 Table 33 + §4.3.5.7 Table 98):

  • New parse_7x_audio_data_outer() walker in mch.rs plus a
    SevenXCodecMode enum (Table 98 — 2 bits, 4 codepoints: SIMPLE /
    ASPX / ASPX_ACPL_1 / ASPX_ACPL_2; no ASPX_ACPL_3 in 7.X). The
    walker mirrors the 5_X SIMPLE/ASPX path's coding_config selector
    but with the 7.X-specific shape:
    • 2-bit 7_X_codec_mode (vs 3-bit for 5_X — no Reserved values).
    • LFE mono_data(1) gated on channel_mode == "7.1" (mapped from
      the parent substream's channel count: 7 → 7.0, 8 → 7.1).
    • companding_control(5) for ASPX_ACPL_{1,2} only — SIMPLE/ASPX in
      7.X have no leading companding (different from 5_X where ASPX
      gets companding_control(5)).
    • Cfg0 body: 2ch_mode + two_channel_data + two_channel_data (no
      centre mono inside the switch).
    • Cfg2 body: four_channel_data only (no surround mono inside the
      switch). Both centre / surround monos move out to a single
      trailing mono_data(0) call gated on coding_config in {0, 2},
      placed after the additional-channel block.
    • SIMPLE/ASPX-only additional-channel block: 1-bit
      b_use_sap_add_ch gating optional chparam_info()×2, then a
      mandatory two_channel_data() for the front-extension /
      back-surround pair beyond the 5.X core. Lands in new
      tools.seven_x_b_use_sap_add_ch,
      tools.seven_x_add_chparam_info and
      tools.seven_x_additional_channel_data slots.
    • ASPX_ACPL_1-only joint-MDCT residual layer (max_sfb_master +
      chparam_info×2 + sf_data×2) — same shape as the 5_X path,
      n_side_bits derived per the Table 33 NOTE from the largest
      signalled transform length across all preceding
      two_channel_data / three_channel_data / four_channel_data /
      five_channel_data (including the additional-channel one when
      it's the largest).
    • Trailers: aspx_data_2ch×2 + aspx_data_1ch for any non-SIMPLE,
      plus an extra aspx_data_2ch for ASPX (covering the additional
      pair); acpl_data_1ch×2 for ASPX_ACPL_{1,2} landing in
      tools.acpl_data_1ch_pair[0/1] (shared with the 5_X
      §5.7.7.6.1 pair walker).
  • walk_ac4_substream now dispatches channels == 7 (7.0) and
    channels == 8 (7.1) into the new walker. Previously these
    channel counts fell through to the catch-all that just records
    channel_mode_channels and bails — real Dolby AC-4 streams using
    a 7_X_channel_element now parse end-to-end without hitting the
    catch-all.
  • Walker is try-and-bail with the same contract as the 5_X
    walker: any inner Huffman / parse miss surfaces Ok(()) to the
    caller, leaving already-populated tools.* slots intact. The
    deeper aspx_data / acpl_data trailers are gated on
    b_iframe && tools.aspx_config.is_some().
  • 11 new lib tests (394 → 405 total): SIMPLE Cfg3 (no SAP), 7.1
    SIMPLE LFE walk, SIMPLE Cfg0 (two pairs + trailing centre mono),
    SIMPLE Cfg2 (four-channel + back surround mono), SIMPLE Cfg1 (no
    trailer), SIMPLE with b_use_sap_add_ch == 1 (chparam pair
    populated), ASPX_ACPL_2 non-iframe Cfg1 (no additional-channel
    block), ASPX_ACPL_1 I-frame Cfg0 (residual layer + Cfg0 trailer),
    ASPX_ACPL_1 zero max_sfb_master bails silently, truncated
    SIMPLE five_channel_data bails silently, and
    SevenXCodecMode::from_u32 round-trip.

• Round 25 — ASPX_ACPL_1 / ASPX_ACPL_2 inner body walker
  (TS 103 190-1 §4.2.6.6 Table 25 + §5.7.7.6.1 Pseudocode 117):

  • New parse_aspx_acpl_1_2_inner_body() helper in mch.rs walks the
    bits past the existing companding_control(3) + 1-bit coding_config selector for the 5_X ASPX_ACPL_1 / ASPX_ACPL_2
    paths. The body shape (Table 25):
    two_channel_data() OR three_channel_data() →
    [ASPX_ACPL_1 only] max_sfb_master (n_side_bits) + chparam_info()×2 + sf_data(ASF)×2 joint-MDCT residual layer →
    [coding_config==0 only] mono_data(0) centre/surround trailer →
    aspx_data_2ch() + aspx_data_1ch() + acpl_data_1ch()×2.
    The two acpl_data_1ch payloads land in
    tools.acpl_data_1ch_pair[0] (D0-side) and
    tools.acpl_data_1ch_pair[1] (D1-side) per Pseudocode 117 — the
    same pair the §5.7.7.6.1 run_acpl_5x_pair_pcm() PCM driver
    consumes.
  • n_side_bits is derived per the §4.2.6.6 NOTE: largest signalled
    transform length from the preceding two_channel_data() /
    three_channel_data() (look up tables::n_msfb_bits_48 Table 106
    column 2). The joint-MDCT residual sf_data bodies reuse
    decode_asf_long_mono_body_with_max_sfb() against a synthesised
    long-frame AsfTransformInfo at the dominant transform length.
  • The walker is try-and-bail: every step returns Ok(()) to the
    outer walker on any inner Huffman / parse miss, leaving the
    already-populated tools.* slots intact (matching the round-24
    ASPX_ACPL_3 walker contract). Deeper aspx_data / acpl_data steps
    are gated on b_iframe && tools.aspx_config.is_some() —
    non-iframe paths simply consume what they can of the upstream
    channel data and stop.
  • Active acpl_config_1ch for the pair-extraction step is...

v0.0.2

Other

• fix clippy 1.95 lints
• drop oxideav-codec/oxideav-container shims, import from oxideav-core
• wire §5.7.6.4.3 noise + §5.7.6.4.4 tone into aspx_extend_pcm (round-9)
• end-to-end FFT probe for A-SPX noise + tone HF injection
• wire §5.7.6.4.2 per-envelope HF envelope adjustment (P90+P91+P95)
• land §5.7.6.4.3 noise + §5.7.6.4.4 tone generators
• land §5.7.4 QMF synthesis + A-SPX HF regen pipeline (round-7)
• add §5.7.3 QMF analysis scaffold — QWIN + single-slot transform
• derive §5.7.6.3.1 master-freq-scale and wire aspx_ec_data()
• wire aspx_delta_dir + effective qmode into substream walker
• transcribe all 18 A-SPX Huffman tables + aspx_ec_data() walker
• A-SPX Huffman scaffolding + Annex A.2 table metadata
• implement aspx_hfgen_iwc_2ch() per Table 56
• implement aspx_hfgen_iwc_1ch() per Table 55
• parse aspx_delta_dir() per-channel delta-direction bits
• wire aspx_framing into the ASF substream walker
• parse aspx_framing() end-to-end for all four interval classes
• parse aspx_config + companding_control sidecar
• stereo joint M/S test + refresh lib-level doc
• stereo CPE decoder test — different tones on L and R
• stereo CPE body decode (split + joint M/S) and per-channel IMDCT
• refresh lib-level doc for new coefficient pipeline
• wire ASF data path into decoder — real mono PCM output
• Huffman-driven ASF data parsers and dequantisation
• implement IMDCT + KBD window + overlap-add
• add sfb_offset tables for 48 kHz family (Annex B.4-B.7)
• transcribe ASF Huffman codebooks and add decoder helpers
• document sfb_offset tables (B.4/B.5/B.6) as next-up work
• parse asf_psy_info + Annex B num_sfb / Table 106 n_msfb_bits
• land ASF substream walker (ac4_substream + audio_data outer layers)
• switch workflows to master branch

v0.0.1

chore: Release package oxideav-ac4 version 0.0.1