A Real Elevated-Relay Hidden-Node Hub: Stone Mountain
Status: Real-data observation (live capture). Date: 2026-06-09.
Reproduce: python3 tools/analyze_stone_mountain.py (re-runnable as the MQTT capture grows).
UPDATE 2026-06-10 (~16 h sample,
noon-eval-20260610.md): the preliminary numbers below firmed up 10× — now 275 direct receptions, 29 nodes heard directly (24 positioned), ~93% mutually hidden, median SNR −4.5 dB. Conclusion unchanged, basis much stronger.
The project's central technical claim (hidden-node-analysis.md) is that an elevated relay makes
the hidden-node problem worse — by gathering many transmitters that cannot hear each other into
a single collision domain. Here that claim is observed in the wild, at a named high site.
Nano Newt 0b3b (!67930b3b) is a PORTDUINO router atop Stone Mountain — one of the highest
points in metro Atlanta — that gateways the local mesh to MQTT. Subscribing to its stream
(mqttmesh.internetlabs.com, scoped by ACL to exactly this gateway) gives us what Stone Mountain
hears. Filtering to direct receptions (hop_start == hop_limit, i.e. the originator heard
firsthand) and joining to node positions:
What the data shows (first ~1 h sample)
| Metric | Value |
|---|---|
| nodes heard directly (positioned) | 12 |
| distance from Stone Mtn | median 22.9 km, max 123 km (7 of 12 beyond 20 km) |
| median link SNR | −4.6 dB — near the decode margin |
| node-pairs too far apart to hear each other (>5 km) | ~92% |
| geographic spread of heard nodes | pairwise median 35 km, max 156 km |
The left map is the picture itself: one elevated relay (red star) with sight-lines to nodes scattered across the whole region. The right histogram is the quantified hidden-node condition — almost every pair of nodes Stone Mountain hears is tens of km apart, far beyond ground-to-ground LoRa range, so they are invisible to one another.
Why it matters for the design
This is the empirical instance of the scenario the design is built around:
- It is the real-world K mutually-hidden clients → one elevated relay topology that v3 simulates,
where CSMA/CAD cannot help (a node can't sense a carrier it can't hear) and TDMA eliminates the
collisions by construction. Stone Mountain is a concrete K≈12+ hub.
- It confirms the counter-intuitive point: the better the relay's vantage (Stone Mountain hears out
to 100+ km), the more mutually-hidden transmitters it concentrates — the elevated INFRA node
both helps coverage and worsens contention, which is exactly why it must also be the coordinator
(slot/time authority), not just a passive repeater.
Honest caveats
- Preliminary sample. ~1 h of capture → 27 direct receptions, 12 positioned nodes. The hub is clearly real, but the exact counts will firm up as the capture accumulates (the analysis script re-runs on the growing CSV). The relayed receptions (207 of 234) are nodes Stone Mountain did not hear firsthand — even more distant/hidden — and aren't counted here.
- "Mutually hidden" is a distance proxy. We infer "can't hear each other" from pairwise distance
~5 km (a conservative ground-to-ground range); terrain/elevation of individual nodes could move specific pairs either way. With a 35 km median pairwise separation, the conclusion is robust even so.
- The 123 km max link is exceptional (line-of-sight from a granite dome, possibly a high/elevated peer); it doesn't drive the conclusion.
Bottom line: the hidden-node hub isn't a thought experiment — here is one on Stone Mountain hearing a dozen-plus nodes spread over a 35 km median, ~92% mutually hidden, at a median SNR right at the decode margin. It is the exact problem the hybrid TDMA design exists to solve, observed live.