ManThink LoRaWAN Engineering Handbook · Article 8 · 2026-09-04 Topic: Geolocation
Welcome back to the ManThink LoRaWAN engineering series. In the previous seven articles we covered joining procedures (OTAA/ABP), adaptive data rate, regional parameters, physical-layer SF/BW/CR, device classes, concentrator chip generations, and gateway host-controllers. Those questions all answer "how does the signal arrive?". Geolocation is a different question: you already received an uplink frame somewhere in the city — can you reverse-engineer where that packet originated, and how close can you get?
This article zooms in on that problem. LoRaWAN defines two standard geolocation methods — TDOA and RSSI — both built on the same physical premise: a single uplink is received by three or more gateways simultaneously. The difference is which physical quantity each gateway reports back, and which geometric model the network server uses to extract location from those reports.
Two of the canonical LPWAN use cases — metering and tracking — share similar power budgets (5–10 years on battery) but want vastly different location accuracy:
| Use case | Required location accuracy | Business consequence |
|---|---|---|
| Remote metering (electricity / water / gas) | None | One uplink ends the job |
| Logistics tracking, pet finders, valuable-asset management | Decameters to hundreds of meters | Minimum "recover it" granularity |
Off-the-shelf embedded positioning has two established routes:
LoRaWAN takes a third route: let the gateways do the measurement; keep the end node untouched. The terminal transmits normally — same power, same bandwidth — and geolocation happens entirely on the gateway + network-server side. The terminal BOM, firmware, and battery are unchanged.
The question this article answers: how accurate can this approach be, and what does it depend on?
Both TDOA and RSSI rely on the same fact: a single uplink is received by ≥ 3 gateways at once. This means once a city's LoRa network is dense enough (tens to hundreds of gateways), geolocation comes "for free" over existing traffic.
What differs is:
That single fork is the TDOA-vs-RSSI boundary.
Each gateway attaches a high-resolution timestamp to every received uplink — the wall-clock "nanosecond N" when the packet arrived. As long as three or more timestamps share a common time base (typically a GPS PPS reference), the server writes down multiple Δt = 0 hyperbola equations; the intersection is the terminal's position. That is TDOA.
Semtech's classic analogy: drop a stone in a lake; ripples spread out at the speed of light. Three receivers (buoys) at known positions record "the moment the ripple reached me". The pair-of-differences closest to zero is the one nearest the terminal. Three hyperbolas in two unknowns give a point fix.
TDOA accuracy depends critically on timestamp resolution. A 100-meter path difference corresponds to a 333 ns propagation delay — for 100-meter positioning, timestamps must reach nanosecond-class. Standard MCU software interrupts cannot deliver that consistently.
Semtech's SX1303 concentrator embeds Fine Timestamp: 64-bit, PPS-relative, nanosecond resolution, recorded in hardware at packet arrival. The "event precision" becomes a hardware fact, not a software post-processing problem. This is precisely why article 6 emphasized that the SX1302→SX1303 jump is fundamentally about geolocation, not demodulation.
Semtech's official accuracy for Fine Timestamp TDOA: 75–150 meters, depending on channel conditions and multipath.
"All gateways share one time base" sounds trivial but defines the ceiling of TDOA accuracy:
| Sync method | Typical error | Deployment cost |
|---|---|---|
| GPS PPS (satellite 1 Hz) | ~10 ns | Each site needs a sky-visible GPS antenna |
| Network PTP / NTP (LAN) | ~1–100 μs | Only suitable where Ethernet infrastructure is pre-deployed |
| Operator base-station (LTE / 5G) | 100 ns – a few μs | Depends on carrier timing services |
GPS is the industry reference; the antenna must see the sky at rooftop/tower-top/external-wall — many indoor deployments make this difficult. That's why TDOA works outdoors but rarely goes deep indoors.
| TDOA | |
|---|---|
| Accuracy | 20–200 m outdoors (LoRa Alliance data) |
| Terminal cost | Zero (no extra packets required) |
| Multipath robustness | Strong (envelope-arrival time is comparatively NLOS-tolerant) |
| Indoor feasibility | Weak (GPS sync hard to deploy; metal/concrete attenuates heavily) |
| Network requirements | ≥ 3 simultaneous receiving gateways + time sync |
TDOA is suitable for: city-scale asset tracking, logistics fleet dispatch, smart livestock (cow/sheep locating), wide-area agricultural machinery monitoring — the classic "outdoor, long-range, low-power, ~100-m accuracy" profile.
Each gateway reports received signal strength (dBm) of the packet. Given a path-loss model, free-space RSSI attenuates as the square of distance; the urban log-distance model gives:
RSSI(d) = A - 10·n·log10(d)
where A is RSSI at 1 meter and n is the path-loss exponent (2 in free space, 2.5–4 in typical urban). Three gateways each give one RSSI→distance estimate; the three circles intersect at the terminal.
The headline problem with pure-RSSI models is multipath: different reflected paths yield different RSSI estimates; variance grows with distance.
Reported values from the academic literature:
LoRa Alliance official number: 1000–2000 m (1–2 km) — coarse positioning.
| RSSI | |
|---|---|
| Accuracy | Coarse — 1–2 km naïve; ~100 m with refined models |
| Terminal cost | Zero |
| Deployment requirements | Just ≥ 3 receiving gateways; no clock sync needed |
| Multipath sensitivity | High (vegetation, buildings, weather all bias RSSI) |
| Indoor feasibility | Medium (LoRa penetrates walls weakly but the inverse — indoor reflections — is manageable) |
RSSI is suitable for: pocket-level tracking, pet activity fences, fleet "in/out of city" determination — applications where "did the asset enter/exit region X" is the business question. The value is not raw accuracy but zero synchronization infrastructure.
| Dimension | TDOA | RSSI |
|---|---|---|
| Outdoor accuracy ceiling | 20–200 m (LoRa Alliance / Semtech) | 1–2 km naïve / ~100 m with refinement |
| Time synchronization | Required (GPS / PTP) | None |
| Hardware requirement | Concentrator with Fine Timestamp (SX1303-class) | Any LoRaWAN concentrator |
| Terminal-side cost | Zero | Zero |
| Multipath sensitivity | Low | High |
| Indoor feasibility | Weak | Medium |
| Network-server compute | Moderate (hyperbola intersection) | Lower (circle intersection) |
| Typical application | Asset tracking, livestock, fleet routing | Region-level "in / out" detection |
Bottom line:
| Application | Recommended approach | Why |
|---|---|---|
| Urban delivery fleet | TDOA | Outdoor, good sky view, block-level precision required |
| Shared-bike regulation | TDOA + RSSI combo | City-wide, 24/7, power-sensitive, 10k+ scale |
| Smart livestock (sheep / cattle) | TDOA | Open range, water-point 200-m virtual fences |
| High-value asset (telecom cabinets) | TDOA | Rooftop 200-m fix triggers geofence |
| Consumer pet finder | RSSI | "Within 1 km of home" UX is sufficient |
| Movable asset (construction site / warehouse) | RSSI (presence-only) | Binary "in / out" state |
| Underground parking vehicle | TDOA + geomagnetic fusion | Pure LoRa insufficient; hybrid needed |
| Smart water / gas meters | No location needed | Meter ID is uplink; location irrelevant |
The GD61x open-source gateway (ESP32-S3 + SX1302) supports 8 concurrent LoRa receive channels — more than enough for typical TDOA solve. However, SX1302 timestamp resolution does not match the SX1303 Fine Timestamp; for ±100 m accuracy, the common ManThink field recipe is:
…which gives approximate TDOA without rebuilding the hardware.
The ThinkLink platform provides:
ManThink's next-generation commercial gateways will adopt SX1303 to deliver Semtech-spec 75–150 m accuracy at the hardware level out of the box — that's the engineering baseline for fine-grain geolocation.
TDOA and RSSI are the two LoRaWAN-standard geolocation methods. They convert the time and signal strength observable in the same uplink into a position fix — each with its own accuracy ceiling, infrastructure requirement, and deployment boundary:
Geolocation accuracy is governed as much by engineering — gateway density, site calibration, time sync discipline — as by physics. LoRa Alliance's geolocation whitepaper and IEEE field studies both point to the same conclusion: accuracy improves with network densification and scenario calibration, not by physics alone.
Written by the ManThink technical team — focused on LoRaWAN gateways, protocols, and positioning tech.
Up next: "Stand Up a Private LoRaWAN Network in 5 Minutes with ChirpStack".