The classic rework scene: budget approved, hardware delivered, brackets drilled and mounted — power on, and the basement devices can't connect, the glass-curtain building drops packets all day. Taking it down and remounting doubles the labor cost and halves customer confidence.
With RF, "find out after installation" usually means too late. A site survey is the only step that converts "it should work" into "it measurably works" before construction begins. This article covers how to do one properly: calculate first, then measure, then decide placement. Antenna selection was issue 26; post-installation troubleshooting was issue 28 — this one covers the decision in between that decides everything: where to put it.
Walk the numbers on paper before anything goes on a roof. LoRa's link budget is generous (typically around 155 dB), but indoor losses deduct quickly. The arithmetic:
Available margin = link budget − path loss − Σ penetration losses − installation margin
Estimate path loss with the free-space formula (add 6 dB per doubling of distance; indoors, redo the estimate with a propagation exponent around 4); sum penetration losses wall by wall using the table below; the installation margin covers weather, humidity, foot traffic, and device reorientation. The hard rule: any location with less than 10 dB remaining is treated as unreliable — not "might fail," but "will fail, we just don't know when."
A common example: a terminal on the third-floor office, gateway candidate in the first-floor telecom room. Three concrete floor slabs (25 dB each) plus two brick walls (12 dB each) puts penetration above 100 dB — the proposal dies on paper. No site visit needed to test "maybe it works."
Additional losses by material (near 868 MHz, cross-referenced field values; expect ±5 dB per building):
| Obstacle | Additional loss | Notes |
|---|---|---|
| Drywall / light partitions | 3–6 dB | Minor |
| Brick wall (24 cm) | 10–15 dB | Most common |
| Reinforced concrete wall (20 cm) | 18–25 dB | Load-bearing, heavy loss |
| Concrete floor slab | 25–30 dB | The cross-floor killer |
| Metal doors / racking | 25–35 dB | Effectively shielding |
| Coated glass (Low-E curtain wall) | 20–30 dB | Most overlooked |
| Human body (close range) | 8–12 dB | Mandatory for wearables |
The most valuable word in this table is "coated." Curtain-wall glass in modern office buildings carries a metallic thermal coating — to RF it is effectively a metal mesh. The building looks transparent; the signal doesn't get in. Asking the facility manager one question — "is the glass Low-E?" — saves ten site visits.
The same logic applies vertically: budget cross-floor links at 25 dB per slab, minimum. From basement level one to the third floor, five slabs — out on paper. The answer to deep indoor coverage is never "a more powerful gateway." It is more gateways, each penetrating fewer walls.
The paper pass only eliminates obviously unfeasible designs; final placement comes from measurement. Five steps:
Step one: desk survey. Satellite imagery for the building layout; mark candidate locations; confirm three things with the facility manager — power availability, backhaul conditions (how far the Ethernet runs, what the 4G signal looks like), and access rights to rooftops and telecom rooms. A location without power and backhaul scores zero no matter how good the RF is.
Step two: temporary mounting. Set up a gateway at each candidate location on a pole or tripod, as close as possible to the final mounting height — the antenna issue's conclusion still holds: height beats gain. Raising the same gateway two meters routinely outperforms an antenna ten times the price.
Step three: the walk test. Carry an off-the-shelf terminal (a standard temperature or battery-reporting node works) along the planned device locations, pausing at representative points to log RSSI and SNR. The platform's real-time data page does the recording: every uplink shows the received signal quality at the gateway side; photograph each point and the survey data has provenance.
Step four: interpreting the numbers. Three thresholds: RSSI below −120 dBm is the warning zone; SNR approaching the demodulation limits at low data rates (−20 dB at SF11, −7.5 dB at SF12) means one fade from packet loss; and check uplink and downlink separately — a working uplink does not guarantee a working downlink, because gateway downlink power is regulator-capped (14 dBm EIRP on EU868 downlink). Many "uplink full bars, downlink lost" projects die on exactly that line. Confirm ADR convergence at every point during the walk test; do not stop at first join.
Step five: the report. Placement decision, per-point RSSI/SNR records, photos, and the paper budget, assembled into one survey report. Recommend attaching it to the contract — later coverage disputes get settled with data, not with recollections of "someone said three kilometers."
Once measurements are in, converge the decision with five rules:
Misjudgment one: good RSSI means a good network. Issue 28 documented the real case: RF survey green across the board, random packet loss six months after launch. RSSI is only physical-layer reception strength. ADR state, downlink windows, and ACK retransmissions are protocol-layer indicators — read them during the walk test, not after go-live.
Misjudgment two: one test represents the whole year. Seasonal vegetation (outdoors), HVAC ducting and occupancy density (indoors), and device reorientation all consume margin at the 5 dB level. That is why the paper budget keeps a 10 dB hard line — one sunny walk test cannot measure the rainy season.
Misjudgment three: budgeting with "maximum range." The vendor's fifteen kilometers is open-field line-of-sight at the lowest data rate. Urban coverage is typically estimated within a kilometer or two; indoor coverage is estimated in walls, not meters. The budgeting unit should be "margin at the worst point" from day one, never "average distance."
A survey costs little — one gateway, one terminal, half a day of labor. Rework costs a lot — disassembly, downtime, and customer trust. Three things worth taking away: the 10 dB margin hard line, the Low-E glass question, and testing uplink and downlink separately.
This is issue 36 of the LoRaWAN series. Antenna gain and mounting: issue 26. Protocol-layer troubleshooting: issue 28. Link budget and physical-layer parameters: the dedicated LoRa PHY article. Penetration-loss figures are cross-referenced field values; verify against an actual walk test per building. This article involves no USD amounts and no currency conversion.