[mgonzalezlopezudc]

HCF lacks a complete Compressed Block Ack exchange. This PR adds explicit immediate feedback and selects Compressed by default for eligible HT links. Master already contains the agreement lifecycle and transfer-completion mechanisms through #1344. Parent controls reproduce missing exchange support, incorrect HT airtime, lost RIFS permission, and the old Basic default.

Both endpoints must select the same profile because ADDBA does not negotiate it. Block Ack requires isBlockAckSupported = true. Links without usable local or peer HT state use Normal Ack. Explicit blockAckVariant = "basic" retains Basic compatibility.

The recipient retains receipt evidence separately from its reorder buffer. For example, DATA 100 and 102 arrive while DATA 101 is absent. A BAR requests status from 100. The recipient reports bits 1, 0, 1, even if delivery removes DATA 100. After DATA 101 arrives, the recipient can report all three receipts:

sequenceDiagram
    participant O as Originator
    participant R as Recipient
    O->>R: DATA 100 and 102
    O->>R: BAR from sequence 100
    R-->>O: Compressed Block Ack reports 1, 0, 1
    O->>R: Retry DATA 101
    O->>R: BAR from sequence 100
    R-->>O: Compressed Block Ack reports 1, 1, 1

Loading

Rx derives SIFS and RIFS relations from local reception endpoints, local transmission completion, and decoded frame information. HCF passes the resulting frame indication to the recipient service. Rx subscribes at the network interface. Both callbacks accept signals only from a radio directly inside that interface. The original radio remains the signal source after upward propagation.

The generic radio medium supplies no preceding-gap API. The detector retains no transmission pointer and introduces no timer, random draw, or network scan.

Normal 2.4 GHz HT duration tests use the production transmitter and configured rate selection. The 5 GHz case retains a test override to select HT MCS 0. Bitrate configuration alone selects VHT at the same 6.5 Mbps rate. A separate test path deliberately omits the signal extension. That waveform fault remains in test code.

Interval evidence controls recipient release decisions; absent evidence does not reject an otherwise valid frame. For 2.4 GHz HT, the inclusive SIFS window is 8–12 microseconds. After a local ACK, 0.5-microsecond propagation gives an accepted eleven-microsecond gap. Five-microsecond propagation gives a twenty-microsecond gap, outside that window. In the latter test, DATA 104 remains buffered behind absent sequences 101–103. A newer BAR with SSN 104 releases it once and advances the receive origin to 105.

The main compatibility risks concern custom MAC interfaces, airtime, and contention. A conflicting addressed BAR stops the simulation. Delayed Compressed policy fails initialization. Active Compressed agreements suppress QoS DATA fragmentation. Non-QoS DATA retains ordinary reception and fragment reassembly. A-MPDU, implicit BAR, delayed Compressed Block Ack, Multi-TID Block Ack, and complete MAC stop/start support remain outside scope. Complete VHT transfer conformance remains outside scope.

Read the commits in this order:

1. 2c3e709057 Local interval evidence: Rx detects SIFS from local radio observations and supplies a frame indication. This prepares the exchange in commit 9. Tests and explanation.
2. 5cd50afbf8 HT airtime: the 2.4 GHz waveform includes its six-microsecond signal extension. The receiver derives the actual extension from local duration and decoded PHY length. Tests and explanation.
3. 9100a30162 RIFS permission: management preserves advertised RIFS Mode. Rx requires BSS permission, consecutive HT receptions, the same TA and RA, and no expected SIFS response. Tests and explanation.
4. f789063968 Bitmap ownership: direct value construction removes unowned temporary bitmap allocations during decode. Tests and explanation.
5. 3864ee29e5 Wire Types: the serializer checks complete Basic and Compressed Types, lengths, and fragment fields. Tests and explanation.
6. b920df9f42 Reversible preparation: an owned QoS DATA preview preserves queued sources until commitment. This prepares admission in commit 9. Tests and explanation.
7. e670bacf1e Receipt state: separate receipt state retains positive bits after delivery. This prepares feedback in commit 9. Tests and explanation.
8. 8c3dc7078d Control rates: helpers select legal non-HT BAR rates and calculate 32-byte Compressed response airtime. This prepares admission in commit 9. Tests and explanation.
9. 6b7e0d2798 Production exchange: HCF connects explicit Compressed feedback, local interval evidence, and complete exchange admission before source removal. Tests and explanation.
10. dbc59f7bb0 Default selection: HCF defaults to Compressed, with capability fallback and explicit Basic compatibility. Tests and explanation.
11. 67434888ea Plan locations: twelve completed Block Ack and recipient NAV documents move to plan/done with identical content.
12. 4a31a7affd Document links: affected documents use the completed plan paths, current anchors, and pinned historical CSV sources.

HCF owns admission, agreement records own receipts, and the recipient service owns reorder delivery. If an aggregate exceeds the exchange bound, HCF evaluates its first ordinary MSDU. Refusal preserves queued sources, deadlines, and sequence state. General Basic admission and admission outside active Compressed agreements retain their limits.

The contracts include IRx::invalidateInterframeHistory() and the fragmentation-permission operation on IOriginatorQosMacDataService. Rx supplies Ieee80211InterframeInd; the radio supplies actual received extension through Ieee80211ModeInd. HCF invalidates Rx history when agreement or BAR state changes. Wire Types and the HCF profile default also change. The migration guide explains custom implementation obligations. Generic medium source and Ieee80211Transmission match the pinned base. No new architecture or naming exception is necessary. The packet seal remains intact.

The branch contains twelve linear commits on pinned master f13dca88fd. The rebase retains their order and messages. Master adds VHT queries beside the signal-extension query. The conflict resolution retains all three queries. Master also supplies the signal-tag include that the second commit previously added.

The first commit contains the interface subscriptions, source filters, and completion checks. The second commit contains the restricted transmitter override and its explanation. The third commit retains the shared waveform helper. The last two document commits retain the verified runtime source and tests.

The current head is 4a31a7affd. The recorded S5 verification ran at its former head 394b78c55e. Only the activation commit message changes after that verification. Every corresponding S5 commit retains its complete file tree. The message now explains why DATA admission, source commitment, and frame-sequence callbacks share one activation boundary. No source, test, configuration, document, or expectation changes. No additional S5 build or simulation runs for this message correction.

The recorded debug and release builds pass. The user requested no build or test at intermediate commits. The earlier publication tested each original commit on 7fda2e3987; those results do not establish intermediate behavior after the upstream rebase.

The final selection passes nine protocol files, seven module files, and four unit files. The local interval fixture covers seed-zero cases. The physical exchange fixture covers zero, 0.5-microsecond, and five-microsecond propagation. Local transmission completion includes rejection of interface-originated signals. HT extension and RIFS management also pass.

The current validation also covers default, fallback, migration, admission, preparation, receipt, wire, control-rate, and originator paths. The earlier implementation campaign covers protection, deferral, and Tx NAV paths.

Master changes radio metadata for Radiotap records. Two additional module files and four additional unit files pass on the rebased head. These tests cover receiver copies, Radiotap output, serialization, PHY mode queries, and radio wire identity.

The earlier implementation campaign reproduced HT airtime, RIFS permission, local RIFS, and default-profile failures against the corresponding parents. It also matched HT and VHT default tplx fingerprints, 525b-ffe4 and e882-0164, and passed two stored Basic controls. The current publication does not repeat those parent controls or fingerprint runs. No expected-failure declaration, fingerprint baseline, or statistical baseline changes.

Commit, classification, include, NED parameter, release-note, description, source-protection, and whitespace checks pass. The interface gate retains unchanged findings in IFrameSequence.h: ITransmitStep and IReceiveStep contain bodies. That file matches the pinned base. New interface operations are pure virtual.

The publication reruns the full series gate. Global architecture, naming, and interface gates return failures. Comparison with the pinned base finds identical sets: 36 architecture candidates, 24 naming findings, and 16 interface findings. No new finding appears in those sets. Source-seal and document-seal checks pass.

The series gate reports eight notes. Seven concern paragraph layout beside trailers. One asks whether the 693-line production commit can divide. The rebase preserves the original commit messages and boundaries; these notes remain open.

The tests establish selected deterministic paths. Other seeds, general propagation models, custom delayed MAC delivery, external custom implementations, and complete stop/start recovery remain unverified. The detector supports coverage class zero. Source inspection establishes constant detector work per local event, apart from existing frame checks. No performance measurement establishes a speed improvement. A full leak-detector campaign, statistical campaign, and C++ lint remain outside this evidence. Local results do not establish CI completion.

Executed commands and evidence scope

The current publication uses /home/user/omnetpp_ws/inet-compressed-blockack-s5 with retained build outputs. The directory /tmp/pr1345-master-rebase-20261029 contains pinned inputs, the patch comparison, build logs, test results, and gate results. A temporary runner file removes the unsupported coverage.report setting without a repository change. Explicit INET builds precede tests. The final checks use these commands:

make MODE=debug -j12
make MODE=release -j12
make -C tests/protocol/lib MODE=debug -j12
opp_run_protocol_tests --load /tmp/pr1345-levy.runner.opp --no-build -m debug -f '(WifiLocalInterframeInputs|WifiLocalInterframe|WifiHtSignalExtension|WifiRifsPermission|N_CompressedBa|N_BlockAck|Ac_BlockAck|Ac_BlockAckFallback|WifiBlockAckWireType)\.test$' -w '^tests/protocol/wifi$' --result-file /tmp/pr1345-master-rebase-20261029/protocol.json
opp_run_module_tests --load /tmp/pr1345-levy.runner.opp --no-build -m debug -f '(Ieee80211RxTransmissionCompletion_1|Ieee80211PreparedQosData_1|Ieee80211CompressedBlockAck(Admission|Exchange|Input|Originator|Migration)_1)\.test$' --result-file /tmp/pr1345-master-rebase-20261029/module.json
opp_run_unit_tests --load /tmp/pr1345-levy.runner.opp --no-build -m debug -f '(Ieee80211BlockAckWire_1|Ieee80211CompressedBlockAck(Record|Rate)_1|Ieee80211HtModeSet_1)\.test$' --result-file /tmp/pr1345-master-rebase-20261029/unit.json
opp_run_module_tests --load /tmp/pr1345-levy.runner.opp --no-build -m debug -f 'PcapRecorder(Radiotap|ReceiverCopy)_1\.test$' --result-file /tmp/pr1345-master-rebase-20261029/integration-module.json
opp_run_unit_tests --load /tmp/pr1345-levy.runner.opp --no-build -m debug -f 'Ieee80211(RadiotapHeaderSerializer|RadiotapPcapRecorderHelper|VhtModeQueries|RadioWireIdentity)_1\.test$' --result-file /tmp/pr1345-master-rebase-20261029/integration-unit.json

The former head 394b78c55e ran the combined protocol, module, and unit selections above. Its complete tree equals the current head. Fixtures use their declared run ranges and seed zero. The earlier implementation record remains historical evidence for its parent comparisons and fingerprint controls. The protocol test document explains that comparison method and its limits.

[levy]

why do we need a subclass? why isn't this part of the Ieee80211Transmitter?

[levy]

interframeRadio is such a weird name

[levy]

it's perfectly fine to subscribe on the network interface and not on the radio, signals propagate upwards