Networks
A simulated network is several things that change for different reasons, each its own file so that changing one leaves the others alone: the ground (geodata), the nodes on it (a nodeset), what they run and are told (a script, with its firmware), and what follows from ground and nodes, the loss table.
Geodata
The ground nodes stand on, one directory each under testbed/geodata/, of
one of two kinds.
Synthetic ground lies at 0°, 0°, its degrees metres by one fixed rule (a
nautical mile to the minute of arc), so a nodeset made on one synthetic
ground stands on any other. A pair’s loss is log-distance,
FSPL(1 m, f) + 10·n·log10(d), with n 2 for free space, 2.7 suburban.
A pack is ground compiled from public sources — terrain, clutter, buildings, roads and places, in a UTM zone. A pair’s loss on it is ITU-R P.1812-8 over the real profile, and the map draws the pack’s ground, roads and buildings with the notices of its sources. Build from sources… on the Geodata tab makes one over a rectangle of the map, taking the best source for each part of it:
| Where | From | |
|---|---|---|
| terrain and clutter | Berlin | its DGM1 and bDOM (1 m) |
| everywhere else | Copernicus GLO-30 | |
| buildings | Berlin | its LoD2 models |
| everywhere else | OpenStreetMap | |
| population | Germany | the Zensus 2022 grid |
| land cover, roads, places | the world | ESA WorldCover, OpenStreetMap |
A geodata goes from one machine to another as a zip, Export zip and Import zip…; a bare planner pack imports too. Nodes are never part of the ground: they belong to nodesets.
Nodesets
Which nodes stand where — latitude, longitude, height above the ground — with each one’s maximum power at the antenna connector, its antenna, its tags, and offsets: dB added to one pair’s computed loss, where a measurement says the model is wrong. A nodeset stands on any geodata whose extent holds one of its nodes, as a layer of the Nodes tab, and several layers run as one. What a node runs is not the nodeset’s: a script says it, so one nodeset runs on any firmware, or on a mix of them by tag.
Every node is one board, an SX1262: at 22 dBm or below a bare chip, above it (up to 27 dBm) an SX1262 behind a GC1109 front end, as a Heltec V4 is. A node is told its board when its station starts.
Importing makes a nodeset of the nodes inside the geodata’s extent from the MeshCore map, a PotatoMesh instance, planner sites or a deployed-network CSV, each node tagged with its source, its kind and how good its position is.
Antennas
A node carries one antenna from the catalogue — a bare quarter-wave wire, a spring helical, a fibreglass collinear, a panel, a yagi — each a pattern of five figures: peak gain, vertical beamwidth, tilt, horizontal beamwidth for a directional one, and a floor. Between two nodes the direction is the line between their antenna tips in three dimensions, the earth’s curvature taken off, so a low node under a high collinear’s narrow beam hears less of it than one on the horizon, and a yagi hears what it faces. The Antennas tab draws every pattern.
Who hears whom
There are no stated links. The level a frame arrives at is
L = P_tx + G_tx + G_rx − loss(tx → rx) − offset − 20·log10(f / f0)
the power it went out at, each antenna’s gain toward the other end, the loss table’s figure for the pair and the nodeset’s offset. A frame that arrives below the signal-to-noise ratio its spreading factor needs — −7.5 dB at SF7, down to −20 dB at SF12 — is not delivered at all; that is what “out of range” means here. Every transmission overlapping a receiver’s channel is interference there, whatever its spreading factor, and each receiver rules for itself, frame by frame, over every stretch of its air.
The loss table is every ordered pair’s path loss for one nodeset on one geodata in one band, derived and cached under a hash of exactly what it depends on: the ground and the nodes’ positions and heights. Antennas and offsets are layers put on it when the medium is given it, so changing them never recomputes it. Every pair is computed, not only those strong enough to carry a frame, because a pair far too weak to decode still adds to a receiver’s interference.
Time
A run goes in real time, on the wall clock, or in virtual time: the ether owns time and moves it only when every station is idle, so an hour of a busy network costs what the stations’ work costs, and a busy host slows a run down instead of changing what happens in it. With the same seed, two runs of one network put the same frames on the air at the same instants.