sim: ADAPTIVE flood routing with density-based relay selection and TX power saving

[Heilo27]

Summary

This PR adds a full native-C++ simulation harness for MeshCore, along with a new ADAPTIVE routing strategy that has been empirically validated across multiple topologies. Everything in the sim/ directory is simulation-only — no firmware code is touched.

What's new

ADAPTIVE routing strategy (sim/src/RoutingStrategies.h)

Hash-based relay gate selects a deterministic subset of nodes to relay each flood, with the subset size adapted to local neighbour density:

Tier	Neighbours	Relay %	Rationale
SPARSE	≤ 4	100%	Every relay counts; no redundancy to spare
MEDIUM	5–14	25%	Sufficient redundancy, reduced collisions
DENSE	≥ 15	15%	7–8 relayers covers full mesh; -70% collisions

Selection: hash(packet_seed XOR node_seed) % 100 < relay_pct

• Zero coordination — each node independently makes the same decision for a given packet
• Load balanced — different packets select different relay subsets
• Relay suppression — a node cancels its queued TX if it hears another node already relaying the same flood (hash-matched)

DensityEstimator (sim/src/DensityEstimator.h)

Passive neighbour density measurement from normal traffic — no extra protocol messages. Counts unique direct senders (hop_count == 1) over a 60-second sliding window. Thresholds are compile-time overridable.

Adaptive TX power saving (sim/src/SimNode.h)

Opt-in feature (power_save_enabled). ADAPTIVE-selected relay nodes reduce TX power in dense areas:

• DENSE tier: configurable, default 10 dBm (vs 20 dBm full power) — saves ~75% TX current
• MEDIUM tier: configurable, default 14 dBm — saves ~60% TX current
• SPARSE tier: always full power — marginal links need every dB

Motivation: at a high-density event (festival, conference), nodes running at full power 100% of the time waste battery and raise the local RF noise floor. Reducing power when you have 15+ neighbours: (a) saves battery, (b) shrinks interference radius via the LoRa capture effect, (c) doesn't hurt local delivery since nearby nodes still receive well above sensitivity.

Validated: 100% delivery maintained at MODERATE (-10 dBm DENSE) across all topologies tested.

Simulation infrastructure

File	Role
sim/src/SimBus.h	Virtual RF medium — packet delivery, collision model, channel models
sim/src/SimNode.h	Per-node MeshCore instance with metrics and routing hooks
sim/src/SimRadio.h	Radio abstraction — airtime estimation, LoRa capture effect (6 dB threshold)
sim/src/SimMetrics.h	Flood delivery rate, latency, per-flood aggregation
sim/src/SimRuntime.h	Deterministic clock and RNG for reproducible runs

Channel models available: FullMeshModel, ChainModel, GridModel, PositionalModel.

Scenario suite

Scenario	What it validates
scenario_adaptive	ADAPTIVE vs DEFAULT vs PATH_SNR_HYBRID across FM50/Chain/Grid
scenario_concurrent	2–8 simultaneous flood sources
scenario_longchain	20-hop chains at marginal SNR (6 dB)
scenario_mixed	ADAPTIVE + legacy DEFAULT firmware interoperability
scenario_dutycycle	EU 1% duty cycle enforcement
scenario_relay_pct_sweep	Grid sweep to find optimal DENSE%/MEDIUM% operating point
scenario_txpower	TX power saving vs delivery rate vs energy — festival use case

Key empirical results

• FM50, ADAPTIVE vs DEFAULT: equal delivery rate, -65% airtime, -70% collisions
• Chain-20: ADAPTIVE stays at SPARSE tier throughout — identical behaviour to DEFAULT. No regression on sparse topologies.
• TX power MODERATE (-10 dBm DENSE, -6 dBm MEDIUM): 100% delivery maintained, ~35% energy reduction
• Legacy interop: zero delivery regression when ADAPTIVE and DEFAULT nodes coexist
• Duty cycle: ADAPTIVE uses ~40% fewer TX budget slots than DEFAULT at the same load

Bug fixes (found during development)

• DensityEstimator: replaced seen[64] fixed array with std::unordered_set — was silently capping unique neighbour count at 64
• SimBus::onTransmit: added len > 255 guard before memcpy into 255-byte buffer
• SimNode p_forward gate: >= → > — was silently suppressing relays at exactly p_forward=1.0

Building

cd sim
mkdir build && cd build
cmake ..
make -j4
./scenario_adaptive
./scenario_txpower

Requires a C++17 compiler. No dependencies beyond what's already in the repo (uses lib/ed25519, deps/arduinolibs/Crypto).

Not in this PR

This PR is simulation-only. The ADAPTIVE strategy would need firmware integration work (replacing the node_seed with a pub-key-prefix hash, hooking getRetransmitDelay()) before it could run on hardware. That's a separate conversation — wanted to get the algorithm validated first.