ADR-081: Implement 5-layer adaptive CSI mesh firmware kernel (#404)

* ADR-081: adaptive CSI mesh firmware kernel + scaffolding

Introduces a 5-layer firmware kernel that reframes the existing ESP32
modules as components of a chipset-agnostic architecture and authorizes
adaptive control + a compact feature-state stream as the default upstream.

Layers:
  L1 Radio Abstraction Layer  — rv_radio_ops_t vtable + ESP32 binding
  L2 Adaptive Controller      — fast/medium/slow loops (200ms/1s/30s)
  L3 Mesh Sensing Plane       — anchor/observer/relay/coordinator (spec)
  L4 On-device Feature Extr.  — rv_feature_state_t (magic 0xC5110006)
  L5 Rust handoff             — feature_state default; debug raw gated

Files:
  docs/adr/ADR-081-adaptive-csi-mesh-firmware-kernel.md  (new)
  firmware/esp32-csi-node/main/rv_radio_ops.h            (new)
  firmware/esp32-csi-node/main/rv_radio_ops_esp32.c      (new)
  firmware/esp32-csi-node/main/rv_feature_state.{h,c}    (new)
  firmware/esp32-csi-node/main/adaptive_controller.{h,c} (new)
  firmware/esp32-csi-node/main/main.c                    (wire L1+L2)
  firmware/esp32-csi-node/main/CMakeLists.txt            (add 4 sources)
  firmware/esp32-csi-node/main/Kconfig.projbuild         (controller knobs)
  CHANGELOG.md                                           (Unreleased)

Default policy is conservative: enable_channel_switch and
enable_role_change are off, so behavior matches today's firmware
unless an operator opts in via menuconfig. The pure
adaptive_controller_decide() is exposed for offline unit tests.

Reuses (does not rewrite): csi_collector, edge_processing (ADR-039),
swarm_bridge (ADR-066), secure_tdm (ADR-032), wasm_runtime (ADR-040).

* ADR-081: implement Layers 1/2/4 end-to-end + host tests + QEMU hooks

Turns the ADR-081 scaffolding into a working adaptive CSI mesh kernel:
Layer 1 radio abstraction has an ESP32 binding and a mock binding; Layer 2
adaptive controller runs on FreeRTOS timers; Layer 4 feature-state packet
is emitted at 5 Hz by default, replacing raw ADR-018 CSI as the default
upstream.

New files:
  firmware/esp32-csi-node/main/adaptive_controller_decide.c  (pure policy)
  firmware/esp32-csi-node/main/rv_radio_ops_mock.c           (QEMU binding)
  firmware/esp32-csi-node/tests/host/Makefile                (host tests)
  firmware/esp32-csi-node/tests/host/test_adaptive_controller.c
  firmware/esp32-csi-node/tests/host/test_rv_feature_state.c
  firmware/esp32-csi-node/tests/host/esp_err.h               (shim)
  firmware/esp32-csi-node/tests/host/.gitignore

Modified:
  adaptive_controller.c         — includes pure decide.c; emit_feature_state()
                                  wired into fast loop (200 ms = 5 Hz)
  rv_radio_ops_esp32.c          — get_health() fills pkt_yield + send_fail
  csi_collector.{c,h}           — pkt_yield/send_fail accessors (ADR-081 L1)
  rv_feature_state.h            — packed size corrected to 60 bytes
                                  (was incorrectly 80 in initial commit)
  main.c                        — mock binding registered under mock CSI
  CMakeLists.txt                — rv_radio_ops_mock.c under CSI_MOCK_ENABLED
  scripts/validate_qemu_output.py — 3 new ADR-081 checks (17/18/19)
  docs/adr/ADR-081-*.md         — status → Accepted (partial);
                                  implementation-status matrix; measured
                                  benchmarks (decide 3.2 ns, CRC32 614 ns);
                                  bandwidth 300 B/s @ 5 Hz (99.7% vs raw);
                                  verification section
  CHANGELOG.md                  — artifact-level entries

Tests (host, gcc -O2 -std=c11):
  test_adaptive_controller:  18/18 pass, decide() = 3.2 ns/call
  test_rv_feature_state:     15/15 pass, CRC32(56 B) = 614 ns/pkt, 87 MB/s
                             sizeof(rv_feature_state_t) == 60 asserted
                             IEEE CRC32 known vectors verified

Deferred (tracked in ADR-081 roadmap Phase 3/4):
  Layer 3 mesh-plane message types, role-assignment FSM, Rust-side mirror
  trait in crates/wifi-densepose-hardware/src/radio_ops.rs.

* ADR-081: Layer 3 mesh plane + Rust mirror trait — all 5 layers landed

Fully implements the remaining deferred pieces of the adaptive CSI mesh
firmware kernel. All 5 layers (Radio Abstraction, Adaptive Controller,
Mesh Sensing Plane, On-device Feature Extraction, Rust handoff) are
now implemented and host-tested end-to-end.

Layer 3 — Mesh Sensing Plane (firmware/esp32-csi-node/main/rv_mesh.{h,c}):
  * 4 node roles: Unassigned / Anchor / Observer / FusionRelay / Coordinator
  * 7 message types: TIME_SYNC, ROLE_ASSIGN, CHANNEL_PLAN,
    CALIBRATION_START, FEATURE_DELTA, HEALTH, ANOMALY_ALERT
  * 3 auth classes: None / HMAC-SHA256-session / Ed25519-batch
  * Payload types: rv_node_status_t (28 B), rv_anomaly_alert_t (28 B),
    rv_time_sync_t (16 B), rv_role_assign_t (16 B),
    rv_channel_plan_t (24 B), rv_calibration_start_t (20 B)
  * 16-byte envelope + payload + IEEE CRC32 trailer
  * Pure rv_mesh_encode()/rv_mesh_decode() plus typed convenience encoders
  * rv_mesh_send_health() + rv_mesh_send_anomaly() helpers

Controller wiring (adaptive_controller.c):
  * Slow loop (30 s default) now emits HEALTH
  * apply_decision() emits ANOMALY_ALERT on transitions to ALERT /
    DEGRADED
  * Role + mesh epoch tracked in module state; epoch bumps on role
    change

Layer 5 — Rust mirror (crates/wifi-densepose-hardware/src/radio_ops.rs):
  * RadioOps trait mirrors rv_radio_ops_t vtable
  * MockRadio backend for offline tests
  * MeshHeader / NodeStatus / AnomalyAlert types mirror rv_mesh.h
  * Byte-identical IEEE CRC32 (poly 0xEDB88320) verified against
    firmware test vectors (0xCBF43926 for "123456789")
  * decode_mesh / decode_node_status / decode_anomaly_alert / encode_health
  * 8 unit tests, including mesh_constants_match_firmware which asserts
    MESH_MAGIC/VERSION/HEADER_SIZE/MAX_PAYLOAD match rv_mesh.h
    byte-for-byte
  * Exported from lib.rs
  * signal/ruvector/train/mat crates untouched — satisfies ADR-081
    portability acceptance test

Tests (all passing):
  test_adaptive_controller:   18/18   (C, decide() 3.2 ns/call)
  test_rv_feature_state:      15/15   (C, CRC32 87 MB/s)
  test_rv_mesh:               27/27   (C, roundtrip 1.0 µs)
  radio_ops::tests (Rust):     8/8
  --- total:                 68/68 assertions green ---

Docs:
  * ADR-081 status flipped to Accepted
  * Implementation-status matrix updated; L3 + Rust mirror both
    marked Implemented
  * Benchmarks table extended with rv_mesh encode+decode roundtrip
  * Verification section updated with cargo test invocation
  * CHANGELOG: two new entries for L3 mesh plane + Rust mirror

Remaining follow-ups (Phase 3.5 polish, not blocking):
  * Mesh RX path (UDP listener + dispatch) on the firmware
  * Ed25519 signing for CHANNEL_PLAN / CALIBRATION_START
  * Hardware validation on COM7

* Add test_rv_mesh to host-test .gitignore

Fixes an untracked-file warning from the repo stop-hook: the compiled
binary was built by make but the .gitignore update was missed in
8dfb031. No source changes.

* Fix implicit decl of emit_feature_state in adaptive_controller

fast_loop_cb calls emit_feature_state() at line 224, but the static
definition is at line 256. GCC treats the implicit declaration as
non-static, then the real static definition conflicts, and
-Werror=all promotes both to hard build errors.

Add a forward declaration above the first use. Unblocks ESP32-S3
firmware build and all QEMU matrix jobs.

Co-Authored-By: claude-flow <ruv@ruv.net>

---------

Co-authored-by: Claude <noreply@anthropic.com>
This commit is contained in:
rUv
2026-04-20 10:38:23 -04:00
committed by GitHub
parent b816292ead
commit 5a7f431b0e
27 changed files with 3946 additions and 14 deletions
@@ -87,6 +87,89 @@ menu "Edge Intelligence (ADR-039)"
endmenu
menu "Adaptive Controller (ADR-081)"
config ADAPTIVE_FAST_LOOP_MS
int "Fast loop period (ms)"
default 200
range 50 2000
help
Period of the fast control loop. The fast loop reads radio
health and edge-derived motion/presence/anomaly scores and
updates the active capture profile. Default 200 ms matches
the ADR-081 spec.
config ADAPTIVE_MEDIUM_LOOP_MS
int "Medium loop period (ms)"
default 1000
range 200 30000
help
Period of the medium control loop. The medium loop is where
channel selection and role transitions happen (when
enable_channel_switch / enable_role_change are on).
config ADAPTIVE_SLOW_LOOP_MS
int "Slow loop period (ms)"
default 30000
range 1000 300000
help
Period of the slow control loop. The slow loop publishes
HEALTH messages and may request CALIBRATION_START on
sustained drift.
config ADAPTIVE_AGGRESSIVE
bool "Aggressive adaptation"
default n
help
When enabled, the controller reacts to motion / anomaly
sooner and uses a tighter cadence in SENSE_ACTIVE. Default
off matches today's conservative behavior.
config ADAPTIVE_ENABLE_CHANNEL_SWITCH
bool "Allow controller to change WiFi channel"
default n
help
When disabled, the controller never calls set_channel() —
channel hopping (ADR-029) and channel override (ADR-060)
remain in charge. Enable only after Phase 3 follow-up
work has wired the channel-plan mesh message.
config ADAPTIVE_ENABLE_ROLE_CHANGE
bool "Allow controller to change mesh role"
default n
help
When disabled, the controller never advertises a different
role to the swarm bridge. Enable after the mesh-plane
ROLE_ASSIGN protocol is in place.
config ADAPTIVE_MOTION_THRESH_PERMIL
int "Motion threshold (per-mille)"
default 200
range 1 1000
help
Motion score above which the controller transitions to
SENSE_ACTIVE and selects RV_PROFILE_FAST_MOTION. Expressed
in per-mille (200 = 0.20).
config ADAPTIVE_ANOMALY_THRESH_PERMIL
int "Anomaly threshold (per-mille)"
default 600
range 1 1000
help
Anomaly score above which the controller transitions to
ALERT. Per-mille (600 = 0.60).
config ADAPTIVE_MIN_PKT_YIELD
int "Minimum packet yield before DEGRADED (pps)"
default 5
range 0 100
help
CSI callback rate (per second) below which the controller
falls back to DEGRADED mode and pins the radio to
RV_PROFILE_PASSIVE_LOW_RATE. 0 disables the degraded gate.
endmenu
menu "AMOLED Display (ADR-045)"
config DISPLAY_ENABLE