ieee80211: add partial HT A-MPDU PHY reception

[mgonzalezlopezudc]

An HT A-MPDU can contain a failed MPDU beside successful MPDUs. The former receiver made one error decision for the complete aggregate. It could not retain the separate PHY outcomes that later MAC acceptance needs. HT-SIG also lacked its byte representation, so aggregate identity could not survive independent byte conversion.

This PR adds partial PHY reception for HT aggregates. The radio preserves HT-SIG fields and CRC. The receiver checks the common header, finds bounded delimiter and MPDU ranges, and retains a PHY outcome for each range. A failed member leaves the aggregate bytes intact. Repeated queries reuse the same outcomes without extra random draws.

The supported receive profile is HT mixed format at 20 MHz with NIST or Yans, for long GI and short GI. The model uses independent range outcomes at the configured scalar SNIR reduction. Rx rejects aggregate containers before standalone FCS or MAC header checks. MAC member acceptance, delivery, and implicit Block Ack remain pending.

Complete comparison and dependencies

The comparison starts at master commit 675655fc3559231c56b12d651384cb8695255abc. It includes agreement lifecycle work from #1343, Basic BAR completion from #1344, and explicit Compressed Block Ack from #1345. Those dependencies remain part of the review until they enter master. This series contains 24 commits in the order below.

The dependency commits give HCF safe agreement setup and retirement, selected queue service, reversible DATA preparation, and complete explicit Compressed exchanges. The PHY commits add HT-SIG, bounded ranges, pure error-model probabilities, cached outcomes, and compatibility checks. These are distinct delivery stages. The receiver header also states its override contract.

flowchart TD
    Q["Selected extraction<br/>with normal queue effects"] --> L["Agreement lifecycle<br/>and safe retirement"]
    L --> B["Basic BAR completion"]
    T["Copied physical endpoints"] --> C["Explicit immediate<br/>Compressed Block Ack"]
    B --> C
    D["Reversible DATA preparation"] --> C
    C --> P["Partial HT aggregate<br/>PHY reception"]
    P --> R["Current Rx rejects<br/>aggregate containers"]
    R -.-> M["Later MAC member acceptance<br/>and implicit Block Ack"]

Loading

The series keeps shared mechanisms before their consumers. Corrective changes now reside with their semantic owners. Bitmap allocation repair precedes Block Ack Type support. Compressed receipt and rate helpers precede HCF activation. VHT header correction precedes HT-SIG validation. The optional probability role precedes receiver use.

1. 435fc0866d queueing: add: preserve pull effects for selected extraction.
2. 2c6b1df761 ieee80211: fix: preserve Block Ack lifecycle boundaries.
3. 832dc3672c ieee80211: refactor: let frame sequences own interruption outcomes.
4. 0c23400688 plan: add: record S3b agreement lifecycle completion.
5. 23c1939f27 ieee80211: refactor: expose DATA preview without extraction.
6. 429223850f ieee80211: fix: complete Basic BAR transfer resolution.
7. b2a1bd06c7 ieee80211: refactor: centralize originator window operations.
8. aff989c34d physicallayer: add: retain copied physical timing evidence.
9. aa9cb61cdf ieee80211: fix: include the HT signal extension at 2.4 GHz.
10. 81417fb0f5 ieee80211: fix: preserve advertised RIFS permission.
11. 9b601af370 ieee80211: fix: release temporary Block Ack bitmap allocations.
12. 1ab742b9eb ieee80211: add: serialize Basic and Compressed Block Ack Types.
13. 9a1d6ac562 ieee80211: add: prepare QoS DATA without queue extraction.
14. 30afc0aaa3 ieee80211: add: retain explicit Compressed receipt state.
15. b8048fd514 ieee80211: add: select legal Compressed control rates.
16. c00da39c26 ieee80211: add: complete explicit immediate Compressed Block Ack.
17. 4960ebc3f3 ieee80211: change: default HCF to immediate Compressed Block Ack.
18. 917471ab57 plan: add: define HT A-MPDU reception stages.
19. d2df372532 ieee80211: fix: create VHT headers for VHT signal modes.
20. 0fe439f77a ieee80211: add: preserve HT-SIG and bounded MPDU ranges.
21. dfa51d2a1a ieee80211: add: expose pure HT A-MPDU range probabilities.
22. 3678e89b11 ieee80211: add+change: retain partial HT A-MPDU PHY outcomes.
23. 2976c23355 tests: add: verify seeded NIST and Yans aggregate reception.
24. 56b75c959a tests: add: verify ordinary PHY compatibility with the parent.

Shared queue and timing contracts

Deferred DATA can hide eligible work behind a queue head. Ordinary collection removal represents disposal and omits normal pull effects. The optional ISelectivePacketCollection contract supplies a pure selection query and atomic selected extraction. Built-in queues and the priority scheduler preserve configured service order and apply normal pull effects once at each layer.

For example, a deferred head can leave a later MSDU eligible. Selected extraction retains that MSDU’s residence metadata, queue event, and departure observations. A queue without this optional capability reports unsupported selection before extraction. Ordinary head service retains its existing behavior.

QoS DATA preparation creates an owned preview while source MSDUs remain queued. Successful commitment validates source identity, eligibility, lifetime, aggregation permission, and sequence state. It transfers the same preview and assigns its sequence number. Refusal preserves source frames and the preview. Ordinary commitment uses normal dequeue or selected extraction and copies completed region metadata before source deletion.

For example, an ordinary MSDU that waits 2 ms retains that queueing time and one departure after commitment. Retransmission retains the first measurement without another pull. Aggregation retains its existing source-removal behavior. This measurement correction covers ordinary DATA, including selection behind a deferred head.

The medium supplies copied physical endpoints through IRadioMedium and the communication caches. The receiver supplies a transmission identity in its local signal-time indication. Later MAC work can query the relevant endpoints without retained reception pointers or new propagation calculations. Missing or identical identities supply no interval evidence. Const cache queries do not insert state.

The explicit Compressed recipient uses those endpoints to check SIFS or permitted RIFS relations. Its receipt record remains separate from delivery state. Positive receipts survive delivery, while absent positions remain zero. HCF checks complete DATA exchange cost before destructive preparation. Both endpoints must select the same Block Ack profile because ADDBA does not negotiate it.

Partial PHY outcomes and the MAC boundary

Consider the tested PSDU with three MPDUs of 100, 101, and 102 bytes, plus one spacing delimiter. The PSDU contains 322 bytes. Four delimiter ranges and three MPDU ranges receive outcomes. Three alignment bytes contribute duration but receive no separate outcome or draw.

If the middle MPDU fails, the receiver retains the bytes and all seven outcomes. The aggregate bit-error flag remains clear, while the result reports the failed member separately. Rx rejects the container before standalone FCS or header checks. The drop signal retains the PHY outcomes. Successful aggregates follow the same rejection path until MAC aggregate acceptance exists.

flowchart TD
    H["Common PHY checks<br/>and HT-SIG succeed"] --> O["Retain four delimiter outcomes<br/>and three MPDU outcomes"]
    O --> A["MPDU A succeeds"]
    O --> B["MPDU B fails"]
    O --> C["MPDU C succeeds"]
    A --> P["Preserve complete PSDU bytes<br/>and copy all outcomes"]
    B --> P
    C --> P
    P --> R["Rx rejects container<br/>before standalone checks"]
    R -.-> M["Later member acceptance<br/>applies delimiter and FCS rules"]

Loading

HT-SIG preserves all 48 bits through serialization and independent decode. The radio derives mode fields from the selected mode and aggregate identity from delimiter content. A singleton aggregate therefore remains distinct from a standalone MPDU. Decode preserves the received CRC so validation can detect corruption. Valid ordinary HT duration inputs remain unchanged.

The shared delimiter inspector reports PSDU-relative byte offsets and lengths. It checks delimiter CRC, signature, length, and content bounds. Bounded recovery can find later complete members after an invalid delimiter. The scan permits explicit byte reinterpretation when a delimiter candidate enters a typed MAC header. Typed and byte representations therefore retain the same recovered ranges. Existing typed delimiter error flags remain effective.

Deaggregation uses the same bounded layout. An incomplete layout can contain recovered complete members, but extraction alone grants no MAC acceptance. The PHY and future MAC therefore share one layout description without a second parser.

Failure boundaries and cache ownership

Invalid HT-SIG CRC or unsupported fields fail the HEADER decision before any range draw. A valid header whose length exceeds the PSDU instead fails DATA bounds checks. That case retains the accepted header property but supplies no ranges. Decapsulation refuses unsafe trailer removal.

Separate-part reception retains the existing PREAMBLE or HEADER abort, which supplies no container to MAC. Whole-signal reception can supply a marked failure container. Neither path supplies valid aggregate content after a failed common header.

The medium’s communication cache owns each decision. PREAMBLE calculates no HEADER or DATA outcome. DATA uses the retained common-field decisions and owns copied range values. WHOLE summarizes DATA. Direct success queries and result creation reuse those decisions.

sequenceDiagram
    participant R as Radio
    participant M as Medium and cache
    participant P as WLAN receiver
    participant E as NIST or Yans
    R->>M: Request DATA decision
    M->>P: Calculate after cache miss
    P->>M: Obtain common-field decisions
    loop Each bounded delimiter or MPDU range
        P->>E: Request success probability for this length
        E-->>P: Return existing DATA probability
        P->>P: Sample once when probability requires a draw
    end
    P-->>M: Return decision with owned range values
    M-->>R: Return retained decision
    R->>M: Repeat decision, success, or result query
    M-->>R: Reuse outcomes without new draws

Loading

The result indication retains copied values after reception objects leave their lifetime. With a present model, aggregate PER, BER, and SER are NaN. This profile supplies no aggregate probability estimate. The receiver does not calculate a whole-PSDU estimate solely for those fields. Ordinary MPDU metadata and absent-model metadata retain their existing behavior.

Error-model and capability contracts

A present error model in an HT receiver must implement the optional IIeee80211AmpduErrorModel role. NIST and Yans implement it through their shared base. The support query is pure. The range operation uses the existing DATA calculation for the requested length and scalar SNIR. The generic IErrorModel contract remains unchanged.

Before its MIB update, the MAC supplies the exact mode set, channel widths, and spatial stream limit to the receiver. The model must cover all advertised long-GI and short-GI receive modes within those bounds. An unsupported mode causes cRuntimeError before the update. Modes outside the effective stream limit do not cause rejection.

The receiver header now states that override contract. Overrides must preserve the coverage check, failure boundary, capability state, and random draw count. An absent error model needs no probability coverage. It retains acquisition and SNIR checks without error draws. Ordinary unit-disk HT operation retains its absent-model path; this PR supplies no unit-disk aggregate support claim.

The changed PHY surface includes header fields, serialization, the radio, the receiver, the optional C++ and NED role, and copied indication metadata. The MAC adds the capability check before its MIB update. Rx uses its existing drop and contention path for aggregate rejection. These changes add no recovery timer, agreement store, or probability store. They change no sealed Packet API source.

VHT modes now create their existing VHT header type. The former HT header type could enter HT validation without an HT MCS. This correction adds no VHT aggregate profile or VHT serializer implementation.

Verification and limits

The PR head is 56b75c959af104b5db931bef022bdfe5578eead9. All 24 commit trees equal the previously tested trees. The retained scoped debug results identify the earlier head 2664911ef9f14fb0080cd9f0dc9a21e770c2717d. The 19 transitions with changed compiled INET inputs also pass incremental debug builds. Other transitions retain the matching library and build their selected test executables.

Fresh debug and release builds pass for the final source tree. Compared with 458aaef56af81c998e43d4528e02014db5719272, its only source change is the receiver contract comment.

The final union passes 36 module fixtures, 10 unit fixtures, one serializer fixture, and four protocol cases. All 12 selected functional fingerprint cases pass. The five series gates, source seals, and scoped WLAN architecture checks pass. Global architecture, naming, and interface reports retain the same inherited findings as the original audit.

The original audit at 458aaef56af81c998e43d4528e02014db5719272 passes debug compilation, the focused wire fixture, and nine module fixtures. Those module fixtures contain 1,138 distinct simulation runs. The PHY and compatibility campaigns use seeds 0, 1, and 2. These results identify the original audit revision; they do not establish per-commit results for the reconstructed series.

The retained publication record also supplies debug and release builds and five aggregation and QoS functional fingerprint controls. The eight original PHY commits change no recorded fingerprint expectation. Dependency baseline changes retain their original values and causal commit ownership. This reconstruction introduces no new baseline value.

Evidence	Scope and limit
HT-SIG and delimiter wire fixture	Independent bytes, decode, singleton identity, CRC, length bounds, truncation, alignment, and typed recovery.
Probability matrix	10,752 exact comparisons across 64 HT20 modes, six lengths, seven SNIR inputs, and minimum and mean reductions.
Aggregate receiver fixture	468 runs with scalar and dimensional radios, both signal-part modes, and seeds 0, 1, and 2.
MAC boundary fixture	Sixteen runs through the production radio, MAC, and Rx. DCF and HCF cover typed and byte representations.
Independent generator checks	HEADER and range outcomes, draw counts, subsequent values, and cache reuse.
Ordinary compatibility fixture	576 records with identical inputs for the parent and feature revisions.
Receiver capability checks	Missing roles, incomplete advertised coverage, absent models, non-HT custom models, runtime mode-set changes, and effective stream limits.
Prepared ordinary DATA	Head and non-head commitment, one pull, elapsed queueing time, queue events, refusal, fallback, and retry.

The retained parent comparison uses 4fa5cc0de15ee57388bcb3f20e866cab5972a670. Its 576 records contain 498 successful receptions, 66 delivered failure containers, and twelve separate-part header aborts. The original publication compares later records with those retained parent records. A new reconstructed-head run does not imply a new parent execution.

The first abstraction uses independent range draws at the current scalar SNIR. It does not model correlated FEC errors or interference changes within a member. Alignment padding has no separate outcome. These checks establish model integration and compatibility, not new error accuracy or goodput. Release behavioral tests and feature-isolation builds remain outside the evidence.

The migration guide explains custom queue, timing, QoS, receiver, and error-model obligations. The PHY plan defines the receive profile and ownership. The PHY execution record retains original commands, revisions, seeds, and scope limits. Mermaid diagrams describe the design; they do not represent captured simulation traces.