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
  (Ls = L ⇒ zero side energy ⇒ g* = 0) encodes bit-for-bit
  identical to the round-103 identity path (strict-superset
  invariant). Five new unit tests in src/encoder_acpl3.rs pin the
  selector's per-band (1.7, 1, 0.3, −1) extraction for κ = 0.2,
  the SapMode::None fallback on equal pairs, the ±30 clamp on a
  near-anti-correlated pair, the identity byte-equality, and the
  bit-stream round trip (decoder walker recovers sap_mode = 3 rows;
  forward Table-181 mix matches all four preliminaries within 20 %
  relative L2; residual energy < 5 % of raw surround energy). Four new
  integration tests in tests/round279_5_x_acpl1_sap_auto.rs cover
  the 5-channel AudioFrame shape, the recovered SAP rows + residual
  collapse vs the identity encoder on the same tone fixture, the
  no-benefit byte-equality through the full encoder entry point, and
  sequence-counter advancement. Total lib tests 689 (was 684);
  integration suites +4.

• Round 271 — alpha_q decision driver select_alpha_q_for_pair
  (crate::asf). The SAP-coded (SapMode::SapData) analogue of the
  round-263 select_ms_used_for_pair — completes the encoder decision
  surface for the third non-reserved chparam_info() arm. Given the
  target stereo MDCT spectra (L, R) it picks the per-(group, sfb)
  alpha_q[g][sfb] index and the matching sap_coeff_used[g][sfb]
  flag per ETSI TS 103 190-1 §5.3.2 Pseudocode 59 + §5.3.3.2. The
  decoder reconstructs the output pair from the transmitted tracks via
  the SAP matrix (a, b, c, d) = (1 + g, 1, 1 - g, -1), g = alpha_q · 0.1; inverting (det = -2) gives the tracks the encoder must
  transmit: I_0 = M = (L + R) / 2 and I_1 = S − g·M with S = (L − R) / 2. SAP coding is therefore a one-tap prediction of the side
  track from the mid; the g that minimises the transmitted residual
  energy Σ (S[k] − g·M[k])² per parameter band is the least-squares
  projection g* = ⟨S, M⟩ / ⟨M, M⟩, quantised by alpha_q = round(10 · g*) and clamped to the HCB_SCALEFAC-codable range [-60, +60]
  (the offset of 60 is applied by write_sap_data, not the driver).
  sap_coeff_used is raised only when the quantised index is non-zero
  (a pure-mid band, ⟨S, M⟩ == 0, and a zero-mid-energy band both
  clear the flag so no SAP bit is spent where prediction offers no
  benefit). The decision is taken on the even (pair-leading) sfb of
  each (sfb, sfb+1) pair and copied to the odd partner, matching the
  pair-major flag-copy semantics of Pseudocode 59 and
  build_chparam_info_sap_data_from_alpha_q. New public type alias
  crate::asf::SapAlphaDecision for the (alpha_q, sap_coeff_used)
  matrix pair. Five new unit tests in src/asf.rs pin: the
  least-squares projection (S = M → alpha_q = +10, S = -M → alpha_q = -10, with odd-partner inheritance); pure-mid and
  zero-energy bands clear the flag; round-trip through
  build_chparam_info_sap_data_from_alpha_q + extract_sap_abcd
  reproduces the (2, 1, 0, -1) matrix on picked bands and identity
  on cleared bands; saturation to alpha_q = 60 for g* ≫ 6;
  multi-group independence. The returned matrices plug directly into
  the round-260 build_chparam_info_sap_data_from_alpha_q builder,
  closing the encoder path from per-group L/R MDCT spectra to a
  fully-populated SapMode::SapData ChparamInfo. Total lib tests
  684 (was 679); integration suites unchanged.

• Round 263 — build_chparam_info_none + select_ms_used_for_pair
  encoder helpers (crate::asf). Completes the
  build_chparam_info_* family with the trivial third arm
  (SapMode::None, header-only emission whose extract_sap_abcd
  reproduces identity per-sfb across any per-group bound), plus a
  per-(group, sfb) M/S-vs-L/R decision driver
  (select_ms_used_for_pair) that picks ms_used[g][sfb] per band
  using the standard joint-stereo concentration criterion:
  min(E_M', E_S') < min(E_L, E_R) over the per-band MDCT bins
  [sfb_offset[sfb], sfb_offset[sfb+1]). For a correlated pair, M'
  carries the signal and S' vanishes (min_ms = 0); for an
  uncorrelated or anti-correlated pair, M' and S' both sit near
  (E_L + E_R) / 4. Ties (zero-energy bands, no concentration
  benefit) resolve to false so the encoder doesn't spend a
  ms_used bit when joint coding offers no concentration. The
  returned Vec<Vec<bool>> plugs directly into
  build_chparam_info_ms_used and the result round-trips through
  extract_sap_abcd to the per-sfb (1, 1, 1, -1) matrix on picked
  bands and identity on the rest. Five new unit tests in
  src/asf.rs cover: SapMode::None builder extract + bit-stream
  round-trip; per-band correlated / anti-correlated / one-sided /
  zero-energy decision discrimination; round-trip through
  build_chparam_info_ms_used + extract_sap_abcd; respect of the
  per-group max_sfb bound; multi-group independence. Total lib
  tests 679 (was 674); integration suites unchanged.

• Round 260 — encoder-side ChparamInfo builders
  (crate::asf::build_chparam_info_ms_used +
  crate::asf::build_chparam_info_sap_data_from_alpha_q). Encoder-
  side duals of [crate::asf::extract_sap_abcd] (§5.3.4.3.2 /
  Pseudocode 59) for the two non-trivial SapMode arms.

  • build_chparam_info_ms_used wraps a per-(group, sfb) ms_used
    flag matrix into a ChparamInfo with sap_mode = 1; feeding
    the result into extract_sap_abcd reproduces the per-sfb
    (1, 1, 1, -1) vs identity (1, 0, 0, 1) mix the input
    describes, and a write_chparam_info → parse_chparam_info
    round-trip recovers the same row.
  • build_chparam_info_sap_data_from_alpha_q is the real
    workhorse: starting from per-(group, sfb) alpha_q indices
    (range [-60, +60] — the HCB_SCALEFAC raw-symbol offset of 60
    is applied by the writer, not the builder) plus per-pair
    sap_coeff_used flags, it computes the pair-major DPCM
    dpcm_alpha_q[g][sfb] deltas Pseudocode 59 accumulates back
    into alpha_q[g][sfb]. Odd sfbs leave the dpcm slot at zero
    (decoder inherits from the pair-mate); even sfbs compute
    cur - prev with the code_delta policy mirrored exactly
    from extract_sap_abcd — code_delta == 1 when g > 0,
    max_sfb_per_group[g] == max_sfb_per_group[g-1], and the
    caller-supplied delta_code_time is set, with the reference
    being alpha_q[g-1][sfb]; otherwise the reference is
    alpha_q[g][sfb-2] for sfb > 0 and zero for sfb == 0. The
    fully-uniform "all set" matrix is detected and sap_coeff_all
    is raised so the bitstream elides the per-pair flag array.
    delta_code_time is normalised to false on single-group
    payloads (Table 48 doesn't transmit the bit there).
  • Round-trip guarantees pinned by five new unit tests in
    src/asf.rs: extract_sap_abcd reproduces the original
    alpha_q row on set bands and identity on cleared bands
    (build_chparam_info_sap_data_pair_major_round_trip +
    ..._unused_bands_pass_through); the cross-group
    delta_code_time path delivers the expected dpcm_alpha_q
    deltas (..._delta_code_time_cross_group); single-group
    delta_code_time = true input is normalised to false on
    emit (..._single_group_drops_delta_code_time); and
    write_chparam_info → parse_chparam_info recovers the same
    SAP body which extracts to the original alpha_q
    (..._round_trips_through_bitstream).
  • Slots into the round-257 SAP-aware residual-layer writer: an
    IMS encoder that runs a psychoacoustic decision per
    parameter-band (M/S vs. alpha-driven SAP joint stereo) can now
    materialise the ChparamInfo pair from its decision matrix
    instead of hand-crafting the inner SapData body — the same
    bytes the decoder's parse_chparam_info walks back into the
    apply_sap_table_181 pipeline. Total lib tests 674 (was 667);
    integration suites unchanged.

• Round 257 — SAP-aware ASPX_ACPL_1 residual-layer writer
  (write_acpl_1_residual_layer_sap + body-builder wrapper
  build_5_x_acpl1_body_from_pcm_spectra_sap). Pairs the round-246
  Table-181 inverse with the existing round-243
  [crate::encoder_asf::write_chparam_info] emitter so the IMS
  encoder's §4.2.6.6 Table-25 case ASPX_ACPL_1: residual layer can
  now express any of the three SAP coefficient families produced by
  [crate::asf::extract_sap_abcd] — identity (sap_mode = 0), M/S
  (sap_mode = 1) and SAP-coded alpha_q (sap_mode = 3) — rather
  than being hard-pinned to the identity row by the round-103
  [write_acpl_1_residual_layer].

  • The new private helper write_acpl_1_residual_layer_sap takes
    (coeffs_l, coeffs_r, coeffs_ls, coeffs_rs) preliminary
    spectra plus an Option<&[ChparamInfo; 2]> and (1) emits the
    chparam_info() pair via write_chparam_info with
    max_sfb_per_group = [max_sfb_master], (2) recovers the
    joint-MDCT residual (sSMP,3, sSMP,4) via
    [crate::asf::invert_sap_table_181] driven by the same chparam
    pair, and (3) writes the two sf_data(ASF) bodies for the
    recovered residual spectra bounded by max_sfb_master. When
    chparam_pair = None (or both rows carry sap_mode = 0) the
    body is bit-for-bit equivalent to
    write_acpl_1_residual_layer(... coeffs_ls, coeffs_rs) — the
    identity-row inverse reduces to s3 = ls, s4 = rs. The inverse's
    surround-silent convention past max_sfb_master is preserved.
  • The new public body builder
    build_5_x_acpl1_body_from_pcm_spectra_sap mirrors the round-103
    build_5_x_acpl1_body_from_pcm_spectra API with the extra
    chparam_pair: Option<&[ChparamInfo; 2]> slot wedged in between
    the surround spectra and the ASPX config. The legacy
    identity-only builder is unchanged.
  • Five new tests in encoder_acpl3::tests:
    write_acpl_1_residual_layer_sap_none_matches_legacy pins the
    bit-equivalence of the SAP-aware path with chparam_pair = None
    against the legacy emitter on identical Ls/Rs preliminaries;
    write_acpl_1_residual_layer_sap_identity_explicit_matches_default
    pins explicit identity rows == default None;
    write_acpl_1_residual_layer_sap_ms_row_roundtrips_through_decoder
    feeds the body through parse_chparam_info and asserts the
    decoder recovers sap_mode = 1 with the right ms_used rows
    on both chparam slots; build_5_x_acpl1_body_sap_none_matches_legacy
    is the body-builder analogue of the bit-equivalence test;
    build_5_x_acpl1_body_sap_ms_decoder_recovers_chparam feeds the
    full body through parse_5x_audio_data_outer and asserts
    tools.acpl_1_residual_chparam[0..1] recover the original
    chparam pair with all max_sfb_master ms_used bands present.
    Total lib tests 667 (was 662); existing integration suites
    remain green.
  • The downstream decoder pipeline that consumes this is already
    wired up: round-30 decoder.rs (lines 2661-2705) reads the
    persisted tools.acpl_1_residual_chparam and feeds it through
    apply_sap_table_181 to re-mix the L/R/Ls/Rs preliminary
    spectra before IMDCT, so an encoder building a body via the
    new SAP-aware path produces a stream that round-trips through
    the existing decoder without further changes.

• Round 246 — encoder-side Table-181 SAP residual extractor
  (invert_sap_table_181, dual of apply_sap_table_181). An IMS
  encoder that wants to populate the §4.2.6.6 ASPX_ACPL_1 residual
  layer (Table 25 row case ASPX_ACPL_1:, two trailing
  sf_data(ASF) bodies carrying sSMP,3 / sSMP,4) now has a
  closed-form 2x2-per-sfb inverse of the §5.3.4.3.2 / Table 181
  first-stage SAP matrix that recovers the joint-MDCT preliminary
  spectra (sSMP_A, sSMP_B, sSMP_3, sSMP_4) from a target
  (L, R, Ls, Rs) preliminary set and a chparam_info() pair.

  • [crate::asf::invert_sap_table_181] / new public type alias
    [crate::asf::SapTable181EncodeOutput]. Inversion splits the
    Table-181 5x5 matrix into the two independent 2x2 sub-systems
    (L, Ls) ↔ (A, s3) driven by chparam_pair[0] and
    (R, Rs) ↔ (B, s4) driven by chparam_pair[1]. Per sfb the
    inverse uses det = a*d - b*c and the closed-form
    [[d, -b], [-c, a]] / det.
  • For the three SAP coefficient families produced by
    [crate::asf::extract_sap_abcd] the determinant is always
    non-singular: identity row gives det = 1, M/S row
    (1, 1, 1, -1) gives det = -2, and the SAP-coded row
    (1 + g, 1, 1 - g, -1) with g = alpha_q * 0.1 also gives
    det = -2. The implementation tolerates a hypothetical
    det == 0 band (e.g. a future spec extension) by emitting
    silence for that band instead of panicking, mirroring the
    forward path's graceful-degradation convention.
  • Outside the SAP-coded extent (bins past
    sfb_offset[max_sfb_master]) the forward pass leaves the front
    pair at (L, R) = (A, B) and zeros the surround pair; the
    inverse mirrors this — A = L, B = R, s3 = s4 = 0 — so
    the round-trip is symmetric at the band boundary. Returns
    None when the transform_length has no entry in
    sfb_offset_48, matching the forward path's failure mode.
  • Five new unit tests in src/asf.rs cover identity-row
    inverse, M/S-row inverse, forward-then-inverse round-trip on
    both the identity and M/S rows, and the unsupported-tl None
    return. All 5 pass; the existing crate test suite remains
    green (662 lib tests pass at this commit).

• Round 243 — encoder-side chparam_info() / sap_data() builders
  (dual of parse_chparam_info / parse_sap_data, Table 47 / 48).
  Adds a reusable encoder helper covering all four sap_mode codes —
  the parser's complement for §4.2.10.1 Table 47 (chparam_info())
  and §4.2.10.2 Table 48 (sap_data()). Until this round the
  encoder's six chparam-emission sites in encoder_asf.rs open-coded
  bw.write_u32(0, 2) for sap_mode = 0 (identity SAP); now there
  is a single builder that handles sap_mode = 0 (header-only),
  sap_mode = 1 (header + per-(group, sfb) ms_used[g][sfb] bit
  array), sap_mode = 2 (reserved; header-only, mirroring the
  parser's accept-and-skip behaviour) and sap_mode = 3 (full
  sap_data() body — sap_coeff_all bit, per-pair flag array when
  sap_coeff_all = 0, delta_code_time when num_window_groups != 1,
  per-pair HCB_SCALEFAC-coded dpcm_alpha_q deltas).

  • [crate::encoder_asf::write_chparam_info] — emits the 2-bit
    sap_mode selector and dispatches to the matching payload
    branch. Half-built ChparamInfo inputs (rows shorter than
    max_sfb_per_group) zero-fill the missing entries so the writer
    stays total. A sap_mode = 3 input with sap_data = None
    emits a SapData::default() body that the parser walks
    successfully.
  • [crate::encoder_asf::write_sap_data] — emits the sap_coeff_all
    bit, the per-pair flag array (skipped when sap_coeff_all = 1),
    the conditional delta_code_time bit and the per-pair DPCM
    deltas. The DPCM map is the same delta + 60 → HCB_SCALEFAC index the round-49 [crate::encoder_asf::write_scalefac_data]
    uses, with the same [0, 120] clamp policy.
  • Round-trip is bit-exact with [crate::asf::parse_chparam_info]
    and [crate::asf::parse_sap_data] across sap_mode in {0, 1, 2, 3}, including: single- and multi-group ms_used payloads;
    sap_coeff_all = 1 single-group and sap_coeff_all = 0
    partial-pair multi-group bodies with delta_code_time = 1;
    the parser's pair-flag copy semantic (one bit drives both halves
    of (sfb, sfb+1)); asymmetric pair-flag input rows; and the
    full [-60, +60] DPCM delta range. Out-of-range deltas clamp at
    the codebook boundary (±60), matching the existing scale-factor
    writer's policy.
  • Thirteen integration tests in
    tests/round243_chparam_info_writer.rs pin: sap_mode = 0 emits
    exactly 2 bits as a header-only element; sap_mode = 2 (reserved)
    is round-trip stable as a header-only emission; sap_mode = 1
    single-group ms_used recovers entry-for-entry; sap_mode = 1
    multi-group with 3 groups of (3, 4, 1) bands recovers the full
    matrix; missing ms_used rows zero-fill on the wire; sap_mode = 3 sap_coeff_all body recovers the DPCM deltas at even-sfb
    pair starts; sap_mode = 3 partial-pair body with
    sap_coeff_all = 0 recovers both the flag array and the
    selectively-emitted DPCM entries; sap_mode = 3 multi-group
    body with delta_code_time = 1 recovers across two groups;
    sap_mode = 3 with sap_data = None emits a default body that
    parses as a sap_coeff_all = 0 all-false row; out-of-range DPCM
    deltas clamp to ±60; a full sweep of every legal delta in [-60, +60] round-trips exactly; sap_mode = 0 drops a populated
    ms_used / sap_data payload on emission; in-memory sap_mode
    values with high bits set are masked to the on-wire 2-bit field.
  • Total tests 883 (was 870). The encoder now has a single reusable
    chparam-emission helper covering every legal sap_mode, ready
    for the §4.2.10 SAP-mode decisioning work (M/S vs. independent
    vs. joint-MDCT) to feed real per-band ms_used[] / per-pair
    DPCM arrays into the existing 5_X / 7_X channel-element walkers
    in place of today's hard-coded identity-SAP literals.

• Round 240 — encoder-side HF QMF energy aggregator (dual of
  Pseudocodes 90 + 91). Closes the first half of the round-234
  remaining-work note by landing the per-(sbg, atsg) energy
  aggregator that converts an HF QMF matrix into the per-sbg
scf vector the round-234 envelope-index extractor consumes —
  completing the encoder's q_high → scf → qscf → DPCM → on-wire bytes chain for real ASPX envelope coding.

  • [crate::encoder_acpl3::aggregate_qmf_to_sbg_atsg] — aggregate
    an HF QMF matrix q_high (shape [absolute_sb][ts]) into a
    [sbg][atsg] matrix of average squared magnitudes per
    Pseudocode 90's per-subband energy reduction grouped by
    Pseudocode 91's SBG borders. Tolerates QMF rows shorter than
    tsz (entries past the bounds contribute zero), zero-span ATS
    intervals and zero-span band groups (return 0.0), and
    sbg_borders[i] < sbx (clamps upward to sbx so callers can
    pass spec-shaped absolute borders verbatim).
  • [crate::encoder_acpl3::extract_aspx_sig_envelope_scf_from_qmf]
    / [crate::encoder_acpl3::extract_aspx_noise_envelope_scf_from_qmf]
    — per-side helpers that pick the leading envelope (atsg = 0)
    column of the aggregator output, producing a per-sbg Vec<f32>
    ready to feed the round-234 envelope-index extractor.
  • New public type
    [crate::encoder_acpl3::AspxQmfEnvelopeChannel] — { q_high: &[Vec<(f32, f32)>], sbg_sig_borders: &[u32], sbg_noise_borders: &[u32] } per-channel bundle consumed by the
    QMF-driven envelope builder.
  • [crate::encoder_acpl3::build_aspx_real_envelope_channel_from_qmf]
    — convenience builder that runs the QMF aggregator + the round-234
    extract_aspx_*_envelope_indices extractors + the round-234
    build_aspx_real_envelope_channel builder end-to-end and returns
    owned (sig, noise) Vec<i32> ready to drop into the round-226
    AspxRealEnvelopeChannel { sig: &[i32], noise: &[i32] } slot.
  • Fourteen integration tests in
    tests/round240_aspx_qmf_energy_aggregator.rs pin: constant-
    energy aggregation matches the per-cell mean; per-ATSG
    partitioning recovers a [1.0, 9.0] split; per-SBG partitioning
    recovers a [1.0, 16.0] split; sub-sbx borders clamp upward;
    empty SBG / ATSG borders return empty matrices; zero-span ATSG
    cells return 0.0; the per-side helpers emit per-sbg vectors
    mirroring the aggregator; the QMF-driven convenience builder
    matches the manual aggregator + extractor + builder chain
    entry-for-entry; an integer-quant-grid input (scf = 64 and
    128 for Fine signal) hits the expected [F0 = 0, DF₁ = 2]
    DPCM payload; short QMF rows contribute partial energy without
    panicking; the QMF-driven builder is deterministic across
    repeated invocations; different QMF inputs produce different
    DPCM payloads. Total tests 870 (was 856).
  • Refs ETSI TS 103 190-1 §5.7.6.4.2.1 Pseudocodes 90 + 91.

• Round 234 — encoder-side ASPX envelope extractor (inverse of
  Pseudocodes 80, 81, 82, 83). Closes the round-226 deferral by
  landing the per-(sbg, env) envelope-index extractor that inverts
  Pseudocode 82's scf = n_subbands · 2^(qscf/a) reconstruction and
  Pseudocode 83's scf_noise = 2^(6 − qscf_noise) reconstruction so
  the round-226 write_aspx_data_{1,2}ch_real_envelope builders can
  be chained with caller-supplied envelope-energy scale factors.

  • [crate::encoder_acpl3::quantize_sig_scf] — scf → qscf for one
    signal-envelope band per Pseudocode 82. qmode_env = Fine ⇒
    a = 2 (1.5 dB step), Coarse ⇒ a = 1 (3 dB step);
    num_qmf_subbands mirrors the dequantizer's 64. Non-positive
    scf clamps to a finite quant index instead of producing
    -inf so the spec's scf[0] = scf[1] carry-through path and
    callers passing 0 for silent bands stay well-defined.
  • [crate::encoder_acpl3::quantize_noise_scf] — scf → qscf for
    one noise-envelope band per Pseudocode 83 (qscf = round(6 − log2(scf))).
  • [crate::encoder_acpl3::freq_dpcm_encode_qscf] — invert the
    FREQ-direction DPCM accumulator qscf[sbg] = sum(values[0..=sbg])
    of Pseudocode 80 / 81. Returns [F0, DF₁, DF₂, …] where
    F0 = qscf[0], DF[sbg ≥ 1] = qscf[sbg] − qscf[sbg − 1]. Empty
    input returns an empty vector.
  • [crate::encoder_acpl3::extract_aspx_sig_envelope_indices] /
    [crate::encoder_acpl3::extract_aspx_noise_envelope_indices] —
    per-channel compositions scf[] → qscf[] → [F0, DF₁, …] ready
    for the round-219 value-emitting helpers + the round-226 builder
    pair.
  • New public type
    [crate::encoder_acpl3::AspxEnvelopeScfChannel] — { sig: &[f32], noise: &[f32] } per-channel envelope-energy payload.
  • [crate::encoder_acpl3::build_aspx_real_envelope_channel] —
    convenience wrapper that runs both extractors and returns owned
    (sig, noise) Vec<i32> pairs callers wire into
    AspxRealEnvelopeChannel by slice reference.
  • Round-trip property: feeding caller scf slices through the
    extractor, then the round-226 builder, then re-parsing the body
    through parse_aspx_ec_data + the decoder's delta_decode_sig /
    delta_decode_noise + dequantize_sig_scf /
    dequantize_noise_scf, recovers the input scf vector within
    the per-band rounding of round(a · log2(scf / 64)) /
    round(6 − log2(scf)).
  • Fourteen integration tests in
    tests/round234_aspx_envelope_extractor.rs cover: forward-inverse
    identity at integer-quant grid points for both Fine and Coarse
    signal step sizes; forward-inverse identity for Pseudocode 83
    on the noise side; non-positive scf clamps to a finite quant
    index; FREQ-DPCM encoder produces [5, 2, −4, −4, 1] for
    qscf = [5, 7, 3, −1, 0] with the decoder's accumulator
    recovering the input; empty / single-band inputs pass through;
    end-to-end accumulator + Pseudocode-82 / 83 round-trip from
    caller scf through extractor through Pseudocode-{82, 83};
    build_aspx_real_envelope_channel matches direct calls; full
    encoder→decoder loop wiring build_aspx_real_envelope_channel
    into write_aspx_data_2ch_real_envelope recovers the input
    scf vectors through the decoder's full pipeline; determinism
    across repeated invocations; different inputs produce materially
    different DPCM payloads; empty per-channel slices return empty
    vectors.
  • Total tests 856 (was 842). With this round the encoder now has
    the complete scf[] → on-wire bytes chain for real ASPX
    envelope coding; remaining envelope-coding work is the energy
    estimator that turns input MDCT spectra into the per-sbg
scf vectors the extractor consumes (the inverse of Pseudocodes
    90 + 91), plus driving the new extractor + builder pair from
    the existing high-level encode entry points. β₃ extraction in
    the 5_X ACPL_3 path and real Table-181 SAP-derived residual
    content for the ACPL_1 paths remain deferred.

• Round 226 — write_aspx_data_2ch_real_envelope() and
  write_aspx_data_1ch_real_envelope() builders. Closes the second
  step of the README's "real ASPX envelope coding" deferral. The
  round-219 value-emitting ASPX-Huffman primitives
  (write_aspx_sig_f0_value / write_aspx_sig_df_value /
  write_aspx_noise_f0_value / write_aspx_noise_df_value) are now
  driven by per-channel envelope builders that emit a full
  ETSI TS 103 190-1 §4.2.12.4 Table 52 (aspx_data_2ch()) or
  §4.2.12.3 Table 51 (aspx_data_1ch()) body with caller-supplied
  F0 + signed DF quant indices.

  • New public type
    [crate::encoder_acpl3::AspxRealEnvelopeChannel] — { sig: &[i32], noise: &[i32] } per-channel envelope payload.
  • [crate::encoder_acpl3::write_aspx_data_2ch_real_envelope] —
    accepts (cfg, ch0, ch1) and writes the Table-52 body with
    aspx_xover_subband_offset = 0, FIXFIX framing (num_env = 1,
    optional aspx_freq_res = 0), aspx_balance = 1 (shared
    channel-0 framing), SIGNAL + NOISE delta-direction bits = FREQ,
    aspx_hfgen_iwc_2ch all-zero trailer, then four aspx_ec_data
    calls (ch0 SIGNAL LEVEL, ch1 SIGNAL BALANCE, ch0 NOISE LEVEL,
    ch1 NOISE BALANCE). qmode is forced Fine on FIXFIX + num_env == 1 per Table 52.
  • [crate::encoder_acpl3::write_aspx_data_1ch_real_envelope] —
    accepts (cfg, ch) and writes the Table-51 body with two
    aspx_ec_data calls (SIGNAL + NOISE, both LEVEL).
  • The SIGNAL band count keys off cfg.signals_freq_res(): low-res
    when the in-band aspx_freq_res = 0 bit is emitted (Signalled
    mode), otherwise the parser's freq_res.get(env) .copied().unwrap_or(true) fallback selects high-res (matching
    the r181 fix in write_aspx_data_2ch_minimal).
  • Caller slices shorter than the derived SBG count zero-pad the
    trailing envelope positions; F0 values outside [0, codebook_length) clamp to the codebook edge; DF values outside
    [-cb_off, +cb_off] saturate to the symmetric edge — matching
    the round-219 helper semantics.
  • Eight integration tests in
    tests/round226_aspx_real_envelope_writers.rs cover:
    deterministic 2ch envelope round-trips through
    parse_aspx_ec_data recovering caller inputs per channel;
    1ch envelope round-trip with LEVEL-only stereo_mode; short
    input slices zero-pad in place; 2ch / 1ch byte determinism;
    all-zero inputs decode to all-zero envelopes; different
    per-channel inputs produce different bytes; out-of-range DF
    saturates at the codebook's +cb_off edge (Fine/Level DF
    cb_off = 70).
  • The minimum-bit-cost write_aspx_data_2ch_minimal /
    write_aspx_data_1ch_minimal writers stay in place; no
    existing call site is touched, so every previous round's
    byte-stream expectations remain valid.
  • Total tests 842 (was 834). The remaining ASPX envelope-coding
    work is the per-(sbg, env) envelope-index extractor that
    inverts Pseudocode 82's scf = n_subbands · 2^(qscf/a)
    reconstruction so the new builders can be chained with input
    MDCT spectra. β3 extraction in the 5_X ACPL_3 path and real
    Table-181 SAP-derived residual content for the ACPL_1 paths
    remain deferred.

• Round 215 — real per-parameter-band γ₁ / γ₂ / γ₃ / γ₄ extraction
  in the 5_X SIMPLE/ASPX_ACPL_3 encoder. Layered on top of the
  round-208 real γ₅ / γ₆ (centre) + the round-196 real α₁ / α₂ +
  real β₁ / β₂ path: the γ₁..γ₄ entropy layers — previously emitted
  as the round-95 zero-delta scaffold codewords — now carry per-band
  magnitudes derived from per-band 2×2 least-squares fits of the
  (L, Ls) and (R, Rs) output channel pairs onto the (L, R) carrier
  pair. In §5.7.7.6.2 Pseudocode 118 step 5 the (L, Ls) pair is
  built by the first ACplModule2 invocation with (a = α₁, b = β₁, y = y₀), and step 11 scales Ls = √2·z1. Forming
  (L + Ls/√2) cancels the y₀·β₁ decorrelator contribution
  exactly, leaving L + Ls/√2 = (γ₁·x0in + γ₂·x1in) which expands
  to (1 + √2)·(γ₁·L + γ₂·R) via the step-1 carrier rescaling. By
  symmetry with step 6 the same fit shape gives (γ₃, γ₄) from
  (R + Rs/√2)/(1 + √2). New
  [encoder_acpl3::extract_gamma_1_2_q_per_band_surround_least_squares]
  and [extract_gamma_3_4_q_per_band_surround_least_squares]
  solve the 2×2 normal equations
  [<L,L> <L,R>; <L,R> <R,R>]·[γ; γ'] = [<L,T>; <R,T>] per
  parameter band. Bands with a degenerate Gram matrix (no L or R
  energy, or perfectly collinear L = ±R within numerical tolerance)
  keep γ = γ' = 0. New
  [encoder_acpl3::build_5_x_acpl3_body_from_pcm_spectra_real_alpha_beta_full_gamma]
  drops all six γ extractors into the acpl_data_2ch() body
  alongside the round-208 γ₅ / γ₆ extractor; β₃ stays zero-delta.
  Caller-facing
  [encoder_ims::Ac4ImsEncoder::encode_frame_pcm_5_0_acpl3_real_alpha_beta_full_gamma]
  / encode_frame_pcm_5_1_acpl3_real_alpha_beta_full_gamma wrap
  the new builder, accepting a 5- / 6-channel
  [L, R, C, Ls, Rs (, LFE)] input (vs the round-208 3- / 4-channel
  [L, R, C (, LFE)] input that could only drive the centre γ
  layer). Nine integration tests in
  tests/round215_5_x_acpl3_real_full_gamma.rs pin: 5.0 round-trip
  to a 5-channel AudioFrame; 5.1 round-trip to a 6-channel
  AudioFrame; silent surround (Ls = 0) yields γ₂_q = 0 in every
  band when probed directly; silent surround (Rs = 0) yields γ₃_q
  = 0 in every band; α/β/γ_scale = 0.0 reproduces the round-95
  zero-delta scaffold byte-for-byte; γ_scale = 0.0 reproduces
  the round-196 real-α-β bytes byte-for-byte; loud-surround vs
  silent-surround inputs produce materially different bytes (the
  round-208 path would emit identical γ₁..γ₄ codewords regardless
  of surround input); the encoder is bit-deterministic for matched
  inputs and fresh state. Total tests 822 (was 813). β₃ extraction
  (requires modelling the unobservable decorrelator output y₂)
  and real ASPX envelope coding remain deferred.

• Round 208 — real per-parameter-band γ5 / γ6 extraction in the
  5_X SIMPLE/ASPX_ACPL_3 encoder. Layered on top of the round-196
  real α₁ / α₂ + real β₁ / β₂ path: the γ5 / γ6 entropy layers now
  carry per-band magnitudes derived from a 2×2 per-band
  least-squares fit of the centre channel. In §5.7.7.6.2
  Pseudocode 118 step 7 the centre output z4 is built by the
  third ACplModule2 invocation with (a = 1, b = 0, y = 0):
  z4 = 0.5 · (γ5·x0in + γ6·x1in). Step 11 scales z4 *= √2
  before QMF synthesis; step 1 rescales the carriers
  x0in = (1 + √2)·L, x1in = (1 + √2)·R. The centre
  reconstruction (β3 = 0, ducker = 1) is therefore
  C ≈ K · (γ5·L + γ6·R) with K = √2·(1+√2)/2 = 1+√(1/2). The
  round-208 extractor solves the 2×2 normal equations per
  parameter band:

    [ <L,L>  <L,R> ] [γ5]   [ <L,C>/K ]
    [ <L,R>  <R,R> ] [γ6] = [ <R,C>/K ]
  

  for (γ5, γ6) that minimise the MDCT-bin-wise residual
  Σ (C/K − γ5·L − γ6·R)². Bands with a degenerate Gram matrix
  (no L or R energy, or perfectly collinear L = ±R within
  numerical tolerance) keep γ5 = γ6 = 0. The quantiser uses the
  Table-208 linear gamma_q = round(γ / gamma_delta) mapping with
  the symmetric ±cb_off clamp (cb_off = 20 Fine / 10 Coarse,
  table magnitude bound ±2.0). γ1..γ4 + β3 stay at the round-95
  scaffold (those parameter sets drive the (L, R, Ls, Rs)
  sub-pipeline plus the ACplModule3 cross-residual — neither of
  which has a per-side surround reference at encode time for the
  5.0 / 5.1 PCM input layouts the real-γ entry point targets).

  • [crate::encoder_acpl3::extract_gamma_5_6_q_per_band_centre_least_squares]
    — per-parameter-band 2×2 least-squares γ5 / γ6 extractor.
  • [crate::encoder_acpl3::build_5_x_acpl3_body_from_pcm_spectra_real_alpha_beta_gamma]
    — drop-in replacement for the round-196
    build_5_x_acpl3_body_from_pcm_spectra_real_alpha_beta with
    additional coeffs_c: Option<&[f32]> + gamma_scale: f32
    parameters. gamma_scale = 0.0 reproduces the round-196 byte
    stream exactly; alpha_scale = beta_scale = gamma_scale = 0.0
    reproduces the round-95 zero-delta scaffold.
  • [crate::encoder_ims::Ac4ImsEncoder::encode_frame_pcm_5_0_acpl3_real_alpha_beta_gamma]
    • ..._5_1_... — high-level entry points accepting [L, R, C]
      (5.0) or [L, R, C, LFE] (5.1) PCM.
  • tests/round208_5_x_acpl3_real_gamma.rs (8 tests) pins: 5.0
    round-trip to 5-channel AudioFrame; 5.1 round-trip to
    6-channel AudioFrame; silent-centre input produces
    γ5_q = γ6_q = 0 in every band; C = (L + R) / 2 produces
    non-zero γ_q in ≥1 tonally-active band; loud-centre vs
    silent-centre inputs produce materially different bytes (the
    round-196 path would emit identical γ codewords regardless of
    centre input); α/β/γ_scale = 0.0 matches the round-95
    scaffold byte-for-byte; γ_scale = 0.0 reproduces the
    round-196 real-α-β bytes exactly; encoder is bit-deterministic
    for matched inputs and fresh state.
  • Real γ1..γ4 extraction (the (L,R,Ls,Rs) sub-pipeline mix
    parameters) requires per-side surround references which the
    5.0 / 5.1 PCM input layout does not carry — these stay at the
    round-95 zero-delta scaffold pending a 5.1+Ls+Rs PCM input
    layout. Real β extraction for the 7_X ACPL_3 paths, real ASPX
    envelope coding, and real Table-181 SAP-derived residual
    content (for the ACPL_1 paths) remain deferred.

• Round 202 — real per-parameter-band α + β extraction in the
  7.0 / 7.1 SIMPLE/ASPX_ACPL_2 multichannel encoder. The 7_X
  (immersive) counterpart to the round-144 5.0 ACPL_2 real-α-β
  path and the real-α-β upgrade of the round-107 / 114 zero-delta
  7_X ACPL_2 encoder. ACPL_2 does not transmit the Ls/Rs surround
  pair on the wire — the decoder reconstructs the surround from
  the L/R carriers + the two acpl_data_1ch() parameter sets per
  §5.7.7.5 Pseudocode 116 + §5.7.7.6.1 Pseudocode 117:
  z0 = 0.5·(x0·(1+α) + y·β), z1 = 0.5·(x0·(1−α) − y·β).

  • [crate::encoder_acpl3::build_7_x_acpl2_body_from_pcm_spectra_real_alpha_beta]
    — real-α-β upgrade of
    [build_7_x_acpl2_body_from_pcm_spectra]; identical body
    schedule (2-bit 7_X_codec_mode = 3, optional LFE
    mono_data(b_lfe = 1), two two_channel_data() pairs, no
    joint-MDCT residual layer, trailing centre mono_data(0),
    aspx_data_2ch + aspx_data_2ch + aspx_data_1ch envelope
    trailer) with the two trailing acpl_data_1ch_minimal writers
    replaced by write_acpl_data_1ch_real_alpha_beta. D0 module
    models (L → Ls); D1 module models (R → Rs). acpl_config_1ch (FULL) carries no qmf_band → start_band = 0 so every
    parameter band participates.
  • [crate::encoder_ims::Ac4ImsEncoder::encode_frame_pcm_7_0_acpl2_real_alpha_beta]
    • _with_max_sfb — accepts [L, R, C, Ls, Rs, Lb, Rb],
      forces the 7.0 channel_mode prefix (0b1111000, 7 b — Table
      85 channel_mode 5), emits a 7-channel S16 PCM round-trip.
  • [crate::encoder_ims::Ac4ImsEncoder::encode_frame_pcm_7_1_acpl2_real_alpha_beta]
    • _with_max_sfb — accepts [L, R, C, Ls, Rs, Lb, Rb, LFE],
      forces the 7.1 channel_mode prefix (0b1111001, 7 b — Table
      88 channel_mode 6), emits an 8-channel S16 PCM round-trip
      with the LFE element written via the round-80
      write_lfe_mono_data shared emitter.
  • tests/round202_7_x_acpl2_real_alpha_beta.rs (10 tests) pins:
    7.0 / 7.1 AudioFrame round-trip; decoder resolves
    SevenXCodecMode::AspxAcpl2 with both
    acpl_data_1ch_pair[0/1] populated; loud-surround vs
    silence-surround inputs produce materially different bytes;
    silence input round-trips with β_q = 0 in every band; encoder
    is bit-deterministic for matched inputs and fresh state;
    direct body-builder probe diverges from the round-107
    zero-delta scaffold byte stream when the caller's Ls/Rs
    spectra are non-trivial.
  • The back pair Lb / Rb is accepted for layout completeness but
    not carried by the ASPX_ACPL_2 body (the decoder's 7_X ACPL_2
    dispatch populates slots 0..4 + the LFE slot 7 — slots 5/6
    stay silent), matching the round-107 documented Table 202
    channel mapping plus the round-80 LFE PCM render at decode
    time.
  • Real β extraction for the 7_X ACPL_3 paths, real γ extraction,
    real ASPX envelope coding, real Table-181 SAP-derived residual
    content (for the ACPL_1 paths), and back-pair Lb/Rb carriage
    remain deferred.

• Round 196 — real per-parameter-band α1 / α2 extraction in the
  5_X SIMPLE/ASPX_ACPL_3 encoder. Layered on top of the round-193
  real β1 / β2 path: the two ACplModule2 instances in ACPL_3 share
  the (L, R) carrier pair as their (x0, x1) input, so without a
  per-side surround reference at encode time α₁ / α₂ are driven by
  the same L↔R cross-correlation extractor — α[pb] = α_scale · ρ(L, R)[pb] with ρ = E[L·R] / √(E[L²]·E[R²]) — clamped to the
  ALPHA_DQ table magnitude bound (±2.0 Fine / ±2.0 Coarse).

  • [crate::encoder_acpl3::extract_alpha_q_per_band_carrier_correlation]
    — extracts per-band α_q from the L / R MDCT spectra. The α
    parameter modulates the front/back dry-mix balance in
    ACplModule2 (Pseudocode 119): higher α → more dry energy on the
    front pair, lower α → more on the surround pair. Mono-like
    (highly-correlated) bands push α toward +1; decorrelated bands
    stay near α = 0.
  • [crate::encoder_acpl3::build_5_x_acpl3_body_from_pcm_spectra_real_alpha_beta]
    — drop-in replacement for the r193
    build_5_x_acpl3_body_from_pcm_spectra_real_beta that lifts
    α1 / α2 from the zero-delta scaffold in addition to β1 / β2.
    β3 / γ1..γ6 still zero-delta. With α_scale = β_scale = 0.0 the
    output is byte-for-byte identical to the round-95
    build_5_x_acpl3_body_from_pcm_spectra scaffold.
  • [crate::encoder_ims::Ac4ImsEncoder::encode_frame_pcm_5_0_acpl3_real_alpha_beta]
    and encode_frame_pcm_5_1_acpl3_real_alpha_beta — caller-facing
    entry points wrapping the new builder with the same channel-mode
    forcing / MDCT analysis / TOC framing as the existing real-β
    paths.
  • tests/round196_5_x_acpl3_real_alpha_beta.rs (4 tests) pins:
    round-trip to 5-channel AudioFrame; perfect L = R correlation
    (ρ = +1) quantises to α_q = +8 (lane 24 = 1.0); perfect
    anti-correlation L = -R (ρ = -1) quantises to α_q = -8;
    α_scale = β_scale = 0.0 is byte-for-byte identical to the
    round-95 scaffold.
  • Total tests 795 (+4 over r193).

• Round 193 — real per-parameter-band β1 / β2 extraction in the
  5_X SIMPLE/ASPX_ACPL_3 encoder. The round-95 ASPX_ACPL_3 path
  emits all 11 ACPL parameter sets as zero-delta Huffman codewords:
  with α = β = β3 = γ = 0 the §5.7.7.6.2 Pseudocode 118 / 119
  synthesis collapses to a trivial mix that produces silent surround
  outputs from non-silent carrier inputs (structurally correct but
  perceptually inert). This round adds three encoder surface entries
  that lift β1 / β2 out of the zero-delta scaffold while keeping
  α1 / α2 / β3 / γ1..γ6 at the round-95 minimum-bit-cost defaults.

  • [crate::encoder_acpl3::extract_beta_q_per_band_carrier_energy]
    — extracts per-parameter-band β_q from a single carrier's MDCT
    energy distribution. The β parameter in ACplModule2 is the gain
    applied to the decorrelator output; setting it proportional to
    √E[x²] keeps the wet/dry balance bounded so the surround
    reconstruction tracks the carrier RMS per band.
  • [crate::encoder_acpl3::build_5_x_acpl3_body_from_pcm_spectra_real_beta]
    — drop-in replacement for build_5_x_acpl3_body_from_pcm_spectra
    that runs the carrier-energy extractor over the L / R inputs and
    emits real β1 / β2 codewords. Mirrors the round-95 wire layout
    everywhere else so the decoder walks the same Table 25 body.
  • [crate::encoder_ims::Ac4ImsEncoder::encode_frame_pcm_5_0_acpl3_real_beta]
    and encode_frame_pcm_5_1_acpl3_real_beta — caller-facing entry
    points that wrap the new builder with the same channel-mode
    forcing, MDCT analysis and TOC framing the existing
    encode_frame_pcm_5_0_acpl3 / encode_frame_pcm_5_1_acpl3 use.
  • With α1 = α2 = 0 and β3 = 0 the ACplModule2 synthesis at
    parameter band pb reduces to
    z0 = 0.5·(x0·g1 + x1·g2 + y0·β1),
    z1 = 0.5·(x0·g1 + x1·g2 − y0·β1),
    and analogously (z2, z3) with β2 driving the second
    ACplModule2. Non-zero β1 / β2 inject the decorrelator output y
    that gives the Ls / Rs outputs their decorrelated spaciousness.
  • tests/round193_5_x_acpl3_real_beta.rs (7 tests) pins: round-trip
    to 5- / 6-channel AudioFrame for 5.0 / 5.1; silent input →
    all-zero β_q indices; tonal carrier + non-zero beta_scale →
    at least one non-zero β_q lane; beta_scale = 0.0 is
    byte-for-byte identical to the round-95 scaffold; silent inputs
    at any beta_scale are scaffold-identical (carrier-energy
    extractor short-circuits to 0); non-silent tonal inputs at
    beta_scale > 0 diverge from the round-95 scaffold (different
    β1 / β2 codeword bit-positions) while keeping the padded
    substream length identical.
  • Total tests 791 (+7 over r190).