LDPC as a first-class primitive: bench-derived RX caps, encoding proof, coding-gain waterfall#288
Merged
Merged
Conversation
…f on rx.txhit, per-encoding counters, coding-gain waterfall
Bench campaign (encoding-matrix devourer<->devourer cells, 8812AU reference,
ch6/ch36): every plugged chip — 8812AU, 8814AU, 8822BU, 8812CU, 8812EU —
passes HT-LDPC / HT-LDPC+STBC / VHT-LDPC both directions at full delivery.
The vendor HAL_DEF_RX_LDPC table (all-false on Jaguar1, "IOT issue" C-cut
TX disable) is 2013-era interop-advertisement policy, not hardware truth.
- AdapterCaps: ldpc_rx_ht / ldpc_rx_vht / ldpc_rx_flag — the bench-derived
LDPC RX truth table, split HT/VHT because the decoders are separate
silicon paths (the 8821A field failure is VHT-only). ldpc_rx_flag=false
on the 8814A: it decodes LDPC but the vendor wired no per-frame
indicator (rxdesc offsets 16/20 unparsed, no ldpc bit in the Jaguar1
phy-status report) — RxAtrib.ldpc reads 0 on frames it decoded fine.
- rx.txhit gains rate/bw/stbc/ldpc so an encoding-matrix pass proves what
encoding was actually decoded, not just that frames arrived.
- rx.energy gains frames_ldpc/frames_stbc (DEVOURER_RX_AGG_SA-filtered) —
the live per-encoding delivery split an adaptive controller compares.
- regress.py --modes txside:rxside filter: LDPC/STBC truth-table runs want
--modes devourer:devourer (kernel-TX rows strip those bits and double
the runtime).
- tests/ldpc_waterfall.{sh,py}: delivery-vs-TXAGC-index waterfalls for
BCC vs LDPC at the same MCS; measured on 8812AU->8822BU at MCS7/20MHz:
+3.0 dB LDPC gain at the 10%-delivery crossing, and in the near-field
saturation regime LDPC delivered ~80% where BCC rolled over to 0-19%.
Hardware-validated: caps events (8812A 1/1/1, 8814A 1/1/0), frames_ldpc
== frames on a pure LDPC feed, ldpc=1 txhits confirmed on 8822BU/8812CU/
8812AU RX. ctest 20/20.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What
Makes LDPC a low-level library primitive backed by bench facts instead of vendor folklore. Closes the loop on the "one Rocket M5 outperforms my 8812AU rig" thread: LDPC is a free ~3 dB the kernel-driver ecosystem throws away (aircrack-ng/88XXau strips the radiotap LDPC/STBC bits on TX; devourer honours them per-packet on all three generations).
Bench campaign (this rig, 2×-direction encoding-matrix, devourer↔devourer)
The vendor
HAL_DEF_RX_LDPCtable (all-false on Jaguar1; 8812A C-cut TX "IOT issue" disable, dated comments from hpfan 2013) is association-IE advertisement policy, not hardware truth — the 8812A demonstrably decodes LDPC on-air.Measured coding gain (
tests/ldpc_waterfall.sh, 8812AU→8822BU, MCS7/20 MHz, TXAGC-index sweep): +3.0 dB at the 10%-delivery crossing. Bonus: in the near-field saturation regime BCC MCS7 rolls over and dies (0–19%) while LDPC keeps delivering ~80% — LDPC also rides through distortion, not just noise.Changes
AdapterCaps.ldpc_rx_ht/ldpc_rx_vht/ldpc_rx_flag— bench-derived RX truth table, split HT/VHT (separate decoder silicon; the 8821A failure is VHT-only). Emitted inadapter.caps.rx.txhitgainsrate/bw/stbc/ldpc— an encoding-matrix pass now proves what encoding was decoded (a pass withldpc:0= TX fell back to BCC).rx.energygainsframes_ldpc/frames_stbc— live per-encoding delivery split for adaptive controllers.tests/regress.py --modes txside:rxside— truth-table runs use--modes devourer:devourer(kernel rows strip the bits and double runtime).tests/ldpc_waterfall.{sh,py}— the coding-gain measurement harness.docs/logging.mdschemas,tests/README.md, CLAUDE.md.Validation
1/1/1, 8814A1/1/0;frames_ldpc == frameson a pure LDPC feed.ctest20/20.🤖 Generated with Claude Code