IEEE 802.11: add HT A-MPDU with implicit Compressed Block Ack

[mgonzalezlopezudc]

The MAC currently sends Block Ack DATA in separate PPDUs. This PR adds opt-in HT A-MPDU transmission with one implicit Compressed Block Ack response. Several independent MPDUs can share one PPDU, while each retains its sequence number, FCS, lifetime, and retry state.

A failed member therefore needs no retransmission of its successful neighbors. The originator can combine that retry with fresh DATA in a later aggregate. The default MPDU policy remains disabled; users select the new HT policy explicitly.

This PR depends on #1363, which supplies HT aggregate representation and cached PHY outcomes. The validated implementation base is 56b75c959af104b5db931bef022bdfe5578eead9, the current head of that prerequisite. GitHub requires an upstream branch as the PR base, so this PR targets master and includes prerequisite history. Use the feature-only comparison to review the nine commits from this delivery.

One response can acknowledge independent members

Hypothetical example: the recipient receives members 40 and 42, but member 41 fails. Its bitmap reports those receipts. The originator retains member 41 with its original deadline. A valid partial response does not require a BAR probe.

sequenceDiagram
    participant O as Originator
    participant R as Recipient
    O->>R: One HT PPDU: MPDUs 40, 41, 42
    Note over R: PHY outcome and FCS<br/>accept 40 and 42
    Note over R: Member 41 fails
    R-->>O: After SIFS: Compressed BA for 40 and 42
    Note over O: Complete 40, 42<br/>Retain 41
    O->>R: Later aggregate: retry 41 plus new member 43
    R-->>O: After SIFS: Compressed BA

Loading

Each member carries ACK policy bits 00. Aggregate reception gives those bits their implicit Block Ack meaning. The recipient constructs its response directly from the agreement and receipt state. It does not create a synthetic BAR. Response rate selection uses the actual received DATA mode.

Queue ownership changes at DATA submission

Selection first considers transmitted retries, then numbered unsent members, then new queued members. New members have independent previews. The source queue retains its original MPDUs until the actual DATA start, after any RTS/CTS exchange and IFS delay.

HCF selects one DATA mode before spacing, size reduction, protection prediction, and admission. The complete PSDU length controls RTS selection, including delimiters and spacing. Each reduction therefore checks protection again at that same mode.

flowchart TD
    Q[Queued sources and retained members] --> P[Prepare independent previews at one DATA mode]
    P --> L[Apply window, spacing, byte, and duration limits]
    L --> A{Complete exchange fits?}
    A -->|No| D[Defer; keep sources and sequence state]
    A -->|Yes| R{Complete PSDU requires RTS?}
    R -->|Yes| C[RTS and CTS]
    C -->|RTS failure| F[Update contention state; no DATA attempt]
    C -->|CTS received| I[Existing IFS timer]
    R -->|No| I
    I --> V{Agreement, deadlines, mode, and cost still valid?}
    V -->|No| D
    V -->|Yes| T[Commit new members and finalize each header and FCS]
    T --> S[Build and submit one aggregate PPDU]

Loading

Cancellation before submission causes no DATA attempt, DATA outcome, or member retry charge. Failed RTS also preserves every member's DATA retry state. Actual submission records each transmitted allocation with the agreement owner.

A zero TXOP limit permits one aggregate DATA PPDU with its protection and response. A multiple-member aggregate must fit a positive limit. The existing initial single-MPDU exception remains available for an eligible singleton retry. Admission also refuses a later exchange when Duration rounding leaves a reservation beyond the TXOP deadline.

Receipt remains a MAC decision

The PHY supplies cached common-header and bounded range outcomes. Rx recovers delimiter boundaries and applies each MPDU outcome before its FCS check. A declared-correct FCS therefore cannot accept a PHY-failed member. The shared packet core remains unchanged.

flowchart TD
    P[Cached PHY indication] --> H{Common header and content valid?}
    H -->|No| X[Refuse before standalone MAC or FCS peeks]
    H -->|Yes| D[Recover delimiters and inspect separate members]
    D --> F[Apply member PHY outcome and FCS]
    F --> C{Any correct member?}
    C -->|No| E[One corruption outcome]
    C -->|Yes| N[Suppress EIFS for this PPDU]
    N --> A{Address, profile, and agreement permit receipt?}
    A -->|No| Z[No receipt update or implicit response]
    A -->|Yes| B[Update bitmap and apply mandatory reorder release]
    B --> R[One Compressed BA after SIFS]

Loading

Correct overheard DATA and correct DATA without an agreement still suppress EIFS. They do not gain local receipt or response permission. Physical history supplies the facts for mandatory reorder release, including delimiter continuity and preceding frames. Receipt state remains with the existing Block Ack record.

An active Basic agreement cannot accept an implicit Compressed request. HCF stops with a profile-conflict diagnostic in that case. Both peers must select the same Block Ack profile; ADDBA does not negotiate that selector.

One probe exposes lost recipient state

An absent or invalid implicit response arms one explicit BAR probe before further DATA for that agreement. This probe is an INET recovery policy. The flag survives refusal or cancellation and clears only when Tx actually submits BAR.

stateDiagram-v2
    state "Ready for DATA" as DataReady
    state "Wait for implicit BA" as WaitImplicit
    state "One probe pending" as ProbePending
    state "Wait for probe result" as WaitProbe
    state "Agreement retired" as Retired
    [*] --> DataReady
    DataReady --> WaitImplicit: Submit aggregate DATA
    WaitImplicit --> DataReady: Valid BA, including zero bitmap
    WaitImplicit --> ProbePending: Missing or invalid BA
    ProbePending --> ProbePending: No legal admission or cancellation
    ProbePending --> WaitProbe: Actually submit one BAR
    WaitProbe --> DataReady: BA or probe failure
    WaitProbe --> Retired: Recipient DELBA
    Retired --> DataReady: Later accepted ADDBA

Loading

Probe failure returns to DATA retry without a second DATA outcome or another probe between those DATA attempts. A recipient without an agreement answers BAR through the existing DELBA path. The originator can then retire its stale agreement and establish another one while fresh DATA remains queued. This progress requires a responsive peer and a legal probe admission opportunity.

Supported configuration and extension changes

The transmit profile uses HT mixed format, 20 MHz, long GI, fixed DATA rates, and immediate Compressed Block Ack. The receiver also accepts advertised HT20 short-GI aggregates. HCF rejects aggregate opt-in with a rate-control submodule during initialization.

The example supplies scalar, dimensional, and zero-TXOP configurations. Production management establishes peer capabilities and receiver limits across both AP hops. The user guide explains opt-in and recovery. The migration guide lists the changed extension roles.

The architectural surface includes aggregate policies, QoS data services, Tx/Rx callbacks, recipient receipt context, frame sequences, ACK policy, and response rate selection. Each new interface method is pure virtual. Custom implementations must supply the declared operation or an explicit unsupported-capability diagnostic.

Commit order

1. Repair cyclic sequence arithmetic at zero.
2. Preserve negotiated receiver spacing.
3. Add source-preserving HT construction and bounded admission calculations.
4. Refuse failed PHY containers before standalone parsing.
5. Add independent member preparation and the HT policy.
6. Complete MAC reception, submission, response, selective retry, and probe recovery.
7. Add the infrastructure example.
8. Explain operation, migration, and the approved HT expectation change.
9. Record completion and verification boundaries.

The large integration commit keeps the exchange contracts and their consumers together. Its earlier commits isolate the parts that can stand alone. The arithmetic fix is necessary because mandatory reorder advancement crosses sequence zero.

Boundary fixture integration

The PR audit found one test integration defect. The earlier PHY boundary fixture expected Rx to drop each whole aggregate. MAC integration extracts separate members and removes their aggregate indication. The stale listener requests that indication from a member drop and aborts.

The MAC integration commit now updates that fixture. Each member has a valid computed FCS trailer and a zero Duration value. The fixture checks PHY metadata on the aggregate, accepted members from production Rx, and failed member drops. All 16 DCF/HCF, typed/byte, scalar/dimensional, and reception-part combinations pass. Every run retains malformed-delimiter recovery and the standalone CTS check for a 50 us NAV.

The related Rx boundary fixture passes one run. The physical reorder fixture passes all 176 runs. The repair changes the test; production reception behavior remains unchanged.

Evidence and limits

Debug and release compilation pass at the final head. The original nine feature commits also compile in release mode in an isolated checkout. The final implementation evidence covers:

Check	Observed result
Focused unit tests.	17 fixtures pass.
Module tests.	46 distinct fixtures pass, including 176 physical reorder/history cases and 57 admission runs.
Infrastructure exchange.	16 PHY/protection/TXOP configurations and three healthy-agreement recovery cases pass.
Complete Wi-Fi inventory.	77 distinct fixtures: 48 passes and 29 declared expected failures after bounded disk-space retries.
Ordinary fingerprints.	All 23 selected configurations pass without baseline changes.
Graphical fingerprints.	All 20 actual values match the pinned parent. Both revisions differ from 18 stored expectations.
Examples.	Scalar, dimensional, and zero-TXOP configurations reach the simulation limit without error.

The later audit covers all changed test fixtures: 15 unit fixtures and 47 module fixtures pass across audit and repair runs. The repair reruns the 16 boundary cases and both related reception fixtures in debug mode with seed zero. Final publication checks repeat debug and release builds and the three focused module fixtures. The repair remains within the MAC integration commit, so the branch retains 33 commits, including nine feature commits. The rewrite changes no other file content. Intermediate tests for the complete prerequisite series remain unverified.

The complete protocol invocation exhausted disk space near its end. Its setup and matcher errors remain in the record. Filtered retries replace only incomplete observations. The approved N_Ampdu marker removal follows actual aggregate observation, an implicit response, and exactly 60 application deliveries. VHT markers and fingerprint baselines remain unchanged.

The execution record contains commands, configurations, seeds, failed attempts, and evidence boundaries. Raw implementation artifacts remain local under out/verification/ampdu-a2-a7/. Publication checks and per-commit build logs remain local under out/verification/ampdu-pr/. These artifacts are not repository files.

Scoped WLAN architecture, changed-file naming, and source seals pass. Global architecture, naming, and interface checks report the same findings as the pinned parent. The complete link check finds absent standards files in unchanged documents. C++ lint cannot run without a compilation database. These checks do not receive inferred passes.

Adaptive aggregate rates, HT40 transmit aggregation, VHT aggregates, nested A-MSDUs, Multi-TID Block Ack, and correlated FEC errors remain outside this profile. Positive RIFS has no runtime proof under the supported consistent mode configurations. The feature establishes neither throughput results nor complete amendment conformance. The approved base exception does not close prerequisite merge gate A0.