[mgonzalezlopezudc]

Before this change, the HT receiver made one error decision for the complete A-MPDU. That result could not describe successful MPDUs beside a failed MPDU. HT-SIG byte conversion also lost the fields that identify an aggregate.

The radio now preserves HT-SIG fields and CRC through byte conversion. The receiver checks the common header, identifies bounded delimiter and MPDU ranges, and samples each range separately. The medium cache retains these outcomes, so repeated queries consume no extra random draws.

flowchart LR
    A[HT A-MPDU] --> B{Common PHY checks pass?}
    B -->|No| C[Reception fails]
    B -->|Yes| D{PSDU contains the declared bytes?}
    D -->|No| E[DATA fails without range outcomes]
    D -->|Yes| F[Inspect delimiter and MPDU ranges]
    F --> G[Sample and cache range outcomes]
    G --> H[Copy outcomes into a PHY indication]

Loading

The protocol fixture sends MPDUs of 100, 101, and 102 octets. In its middle-loss case, only the 101-octet MPDU fails. The PHY preserves the aggregate bytes and every range outcome. MAC member acceptance and implicit Block Ack remain outside this PR. The MAC receive component, Rx, rejects aggregate containers before standalone FCS or MAC header checks, even when every MPDU succeeds.

flowchart LR
    A[One aggregate] --> B[MPDU 1 succeeds]
    A --> C[MPDU 2 fails]
    A --> D[MPDU 3 succeeds]
    B --> E[Preserved bytes and copied outcomes]
    C --> E
    D --> E
    E --> F[Rx rejects the aggregate container]
    F --> G[Drop signal retains PHY outcomes]

Loading

The supported aggregate profile covers HT mixed format at 20 MHz, NIST or Yans, both GIs, and scalar or dimensional radios. Range outcomes use independent draws and the configured scalar SNIR reduction. The model does not represent correlated FEC errors or interference changes within one MPDU. Padding adds duration without a separate outcome. Ordinary reception, random draws, and local interframe history are the main regression risks.

Commits, in review order

1. cfa0572842: Defines PHY and future MAC stages; changes plans only.
2. b0c1d3318e: Corrects VHT header identity; includes its protocol test, guide, and parent comparison.
3. 318d6c1688: Adds a bounded aggregate inspector and unit tests; prepares later callers.
4. 994ca739b9: Recovers complete MPDUs after malformed delimiters; includes its protocol test, guide, and parent comparison.
5. d221c4ce70: Preserves all 48 HT-SIG bits and CRC; includes wire tests and the parent comparison.
6. f0f27e48cb: Adds pure NIST/Yans range probabilities; prepares receiver integration without random draws.
7. f56610d3ba: Adds cached partial reception, copied outcomes, capability checks, and safe MAC rejection; includes the extended protocol fixture.
8. 7b852672b6: Adds seeded NIST/Yans receiver tests for outcomes, cache reuse, and random draws.
9. e916480688: Adds ordinary reception controls and documents model limits without production changes.
10. a0f9a35195: Records the ten commit decisions, parent comparisons, verification results, and coverage limits.

The series starts from master at f5612791b7, which includes PR #1345. The VHT factory correction is a prerequisite. It prevents a VHT mode from entering HT validation with an HT header. The bounded inspector and probability contract remain preparation commits until their production consumers appear later in the series.

The architectural surface includes the HT header, serializer, deaggregation, radio, receiver, indication tags, and MAC receive-capability validation. The optional WLAN contract, IIeee80211AmpduErrorModel, supplies mode-support queries and range probabilities. The generic error-model contract and sealed Packet API remain unchanged. This PR adds no production configuration parameter. VHT gains correct header identity without aggregate reception support.

Custom HT error models must implement the optional contract and cover advertised receive modes. Unsupported coverage causes cRuntimeError. An absent error model retains common PHY checks without error draws. Aggregate error-rate estimates use NaN with a present model. The migration guide explains the override contract and failure behavior.

Protocol regressions in the behavior commits

Each of the four behavior commits contains its protocol fixture, guide, and comparison record. The same fixture bytes compile on that commit and its immediate parent. Fresh debug builds precede both executions. Every parent fixture fails during execution at its intended assertion or production error. Every behavior commit passes its complete selected fixture.

Protocol fixture	Parent observation	Behavior commit	Comparison record
WifiVhtHeaderIdentity.test	The VHT class assertion fails. One later run is skipped.	Both GI runs pass.	VHT header identity
WifiMpduDeaggregation.test	Two complete-input controls pass. Malformed extraction errors. One later run is skipped.	All four runs pass.	Bounded deaggregation
WifiHtAmpduWire.test	The serializer writes zero bits for a 48-bit header. One later run is skipped.	Typed and raw-byte runs pass.	HT-SIG bytes
WifiHtAmpduPartialReception.test	The copied indication is absent at the Rx drop. The 31 later runs are skipped.	All 32 runs pass.	Partial reception

The wire and reception fixtures use OMNeT++ descriptors to read new fields without a compile-time dependency on missing parent APIs. Missing fields still fail at runtime. No expected-failure declaration weakens the assertions. Cmdenv stops each parent campaign after its first error. The records distinguish that error from later cases that did not execute.

The extended reception fixture covers loss of the first, middle, or last MPDU, plus loss of all delimiters. These four patterns combine with both GIs, typed or raw bytes, and whole-signal or separate-part reception. The resulting 32 cases check all six range outcomes and intact aggregate bytes at the production Rx drop. Each case also requires a later valid CTS to set NAV for 50 microseconds. These cases use scalar radios, HCF, and seed zero with deterministic probabilities.

The helper commits retain direct calculation tests because they introduce no production protocol transition. The test-only and documentation commits also need no artificial parent failure. Structural checks confirm that the amendments preserve every original production tree and the shared test support.

Evidence and limits

Verification for a0f9a35195 on 2026-10-29:

• Fresh debug and release builds pass before publication.
• The final debug protocol selection passes all four fixtures: 40 successful simulation runs without skips or expected failures.
• Eight debug build invocations prepare seven production revisions for the parent comparisons. Every build and fixture compilation passes.
• Series, description, and source-seal checks pass. Global architecture, naming, and interface checks retain the findings from pinned master without additions.
• Fingerprint and statistical baselines remain unchanged.

Retained evidence covers nine unit fixtures, fifteen module fixtures, and five fingerprint controls. The module results include 468 aggregate runs, 576 ordinary controls, and sixteen MAC boundary runs. Exact production and fixture comparisons support reuse of these results. They are retained evidence, not fresh execution for this publication. The module fixtures retain the separate dimensional-radio and DCF coverage.

Parent cases skipped after the first error supply no executed parent evidence. Release behavioral tests, feature-isolation builds, C++ lint, and a fresh ordinary-reception comparison with master remain unverified. Full inherited-source coverage, automatic dependency selection, and diagram rendering also remain unverified. These results establish no new error-accuracy or goodput claim. MAC member acceptance, upper delivery, and implicit Block Ack remain outside the implemented scope.

Exact publication commands and evidence records

The installed runner requires a temporary descriptor without its unsupported coverage-report setting. The descriptor retains the selected test settings and points to the active checkout. The repository descriptor remains unchanged.

export MAKEFLAGS=--jobserver-style=pipe
make MODE=debug -j8
make MODE=release -j8

record=/tmp/ampdu-pr1363-publish-20261029
opp_run_protocol_tests --load "$record/inet-compatible.opp" -p inet -m debug --no-build --no-concurrent \
  -w '^tests/protocol/wifi$' \
  -f 'Wifi(VhtHeaderIdentity|MpduDeaggregation|HtAmpduWire|HtAmpduPartialReception)\.test$' \
  --result-file "$record/protocol.json"

base=f5612791b773bc83f654dbfdefa26e243f756c86
doc/project/enforcement/check-architecture.sh
doc/project/enforcement/check-naming.sh --base "$base"
doc/project/enforcement/check-pull-request.sh --base "$base" --head HEAD --description-file "$record/description.md"
doc/project/enforcement/check-interfaces.sh
doc/project/enforcement/check-source-seals.sh --base "$base"

The global gates report 36 architecture findings, 24 naming findings, and 16 interface findings. The comparison with pinned master adds no finding. The gates retain their failure status for those existing findings.

Each linked comparison record gives source revisions, fixture hashes, observed parent and commit results, commands, and coverage limits. The execution record maps all ten commits to their test decisions. Publication logs reside in /tmp/ampdu-pr1363-publish-20261029/. Parent-comparison logs reside in /tmp/ampdu-protocol-parent-20261029/.