1. Start with a common misconception

When roaming comes up, people who have run terrestrial LoRaWAN projects tend to assume two things: first, that roaming is a device feature the device must "support"; second, that it works like cellular roaming — automatic connectivity everywhere. Both are wrong.

LoRaWAN roaming is purely a network-side mechanism: the device software does not change by a single line, gateways stay oblivious packet forwarders, and all roaming intelligence lives between Network Servers (NS). It is also not automatic like cellular — for two LoRaWAN networks to exchange each other's device traffic, both sides must sign a commercial roaming agreement and configure trust in their NS. Without an agreement, a visited network may hear your device's radio packets and will not forward them.

So why does roaming deserve an article? Because it answers two real problems: cross-border and cross-operator asset tracking (the truck leaves the country, the container leaves the service area, and the device cannot re-provision or reflash), and large multi-network deployments (a solution provider ships devices into ten countries and does not want to negotiate activation with ten networks one by one). Issue 32 covered satellite direct-to-device — the extension of LoRaWAN into places with no gateways at all. Roaming is the extension into places where someone else's gateways exist. Together they complete the coverage-boundary picture.

This article walks through the architecture (TS002 Backend Interfaces), the three roaming models, the trust model, and the current ecosystem — ending with a decision checklist.

2. The ownership model: the Home network is always the sole owner

The key to understanding LoRaWAN roaming is its ownership model: an end device always belongs to exactly one Home Network, logically and permanently.

The Home network holds the device identity (DevEUI), the custody of root keys (via the Join Server), the device profile, frame-counter state and MAC-layer logic, and finally delivers application payloads to the application server. When the device moves into another operator's footprint, that operator becomes a Visited Network — it provides radio access and forwarding, but never takes ownership of the device.

This strict separation is deliberate and central to LoRaWAN's security model. There is no such thing as the device "associating" with the visited network: it transmits uplinks as usual and opens RX1/RX2 windows as usual; the complexity hides between Network Servers. Compare cellular roaming, where mobility was a design assumption from day one — LoRaWAN roaming is an extension layered onto a star topology originally designed for mostly static devices. That lineage determines what it does well (infrequent small uplinks, intermittent connectivity) and what it does not (dense downlink, real-time control).

3. The backend architecture: three NS roles and one Join Server

The specification vehicle is the LoRa Alliance's TS002, "LoRaWAN Backend Interfaces" (version 1.1.0, November 2020; the 1.0 framework dates to 2017 — Actility led its development and was the first to implement it). TS002 defines the message flows, authentication, and roaming procedures between Network Servers.

It splits the NS role into three parts; a single NS plays different roles in different scenarios:

The third key role is the Join Server (JS), which manages OTAA activation. In the Join-Request, the device signals via JoinEUI (called AppEUI in LoRaWAN 1.0/1.0.2) which JS to consult. The JS stores, for every device it owns: DevEUI, AppKey, home NS identifier, application server identifier, and the device's LoRaWAN version. It derives session keys — handing the network session key to the NS and the application session key to the AS.

Here lies the key discipline that runs through the whole system: root keys (AppKey/NwkKey) exist only on the JS and the device, and are never sent to the NS or the AS. That is the structural reason roaming can work without leaking secrets — the visited network helps traffic cross its territory but never holds a key that decrypts anything.

A NS is identified by an NSID (a unique IEEE EUI64) and may configure one or more NetIDs (24-bit network identifiers). The NwkID prefix inside the DevAddr is one of the routing clues a visited network uses to decide where a device's traffic belongs.

4. Three roaming models: passive, handover, activation

TS002 defines three roaming arrangements. The variables that drive selection are downlink requirements and the depth of commercial trust.

Passive roaming — the baseline, and the most deployed. The visited network's fNS receives an uplink from a foreign device and forwards it, per protocol, to the home side's sNS/hNS. All MAC-layer decisions (ADR, MAC commands, downlink scheduling) stay with the home network. From the device's perspective, nothing changes. The cost: downlinks must detour back through the hNS before being transmitted by the visited network's gateway — RX1/RX2 timing is tight, and any network-to-network latency eats into the downlink success rate. TS002 lets the fNS declare whether it can support a downlink for this uplink (the DLAllowed flag), so the home side knows whether opening a window is worthwhile. When the same uplink is heard by both local and foreign gateways, the best gateway is elected by SNR. Passive roaming applies to both LoRaWAN 1.0 and 1.1 devices.

Handover roaming — lending out control. The home network authorizes the visited network to take over selected MAC functions during the roaming period: local downlink generation, frame-counter management, ADR based on local radio conditions. Downlinks no longer detour; responsiveness improves. The cost is a deeper trust relationship: session state transfers between NSs, and any inconsistency in synchronization causes problems. Consequently, handover roaming is far less deployed than passive — justified only where devices move frequently and downlink reliability matters.

Activation roaming — first join happens in someone else's network. When a device is first deployed, reset or reactivated outside its home coverage, the Join-Request routes through the visited network to the device's home Join Server for authentication. This solves an operations problem: field teams shipping devices abroad do not need to haul them back into home coverage "for a first join." Once activated, subsequent traffic follows passive or handover roaming.

One-line summary: passive = borrow base stations, not control; handover = borrow control; activation = borrow the join path.

5. The trust model: the roaming agreement is a commercial document and a technical precondition

LoRaWAN uses unlicensed spectrum, and any Alliance member can register a NetID and own a home network — "technically receivable," but the specification enforces "selectively forwarded" through an explicit trust model: inter-network interfaces run over secure transport on IP with mutual certificate authentication; which networks may exchange traffic, whether downlinks are allowed, and how traffic is billed are all defined by roaming agreements and configured as policy rules in the NS.

The design cleanly separates "technically possible" from "commercially agreed": the standard handles the former; operators negotiate the latter. The side effect is that roaming is never automatic — whether a device you bought can use a public network in some country depends on whether the device's home network operator has signed an agreement with the local operator. When selecting a network service provider, the roaming partner list is a commercial clause as important as the coverage map.

6. Ecosystem: from bilateral trials to a peering hub

Three milestones mark roaming's path to production:

7. Boundaries and a decision checklist

Roaming is not a universal answer. Three boundaries first:

  1. It does not solve coverage gaps. Roaming presumes someone else has a network where yours doesn't. Where no network exists, build one or go to satellite (issue 32).
  2. It does not relax downlink physics. Passive-roaming downlinks remain bound by Class A window timing, with an extra network hop on top. Real-time control is a bad fit everywhere, roaming or not.
  3. It is not a device feature. "Roaming-capable device" in marketing means the device follows the standard protocol with standard key custody. Whether roaming actually works depends on signed agreements on the network side.

Scenarios where roaming deserves serious evaluation: cross-border logistics and fleet tracking; solution providers shipping into multiple countries (one integration, multi-country activation); municipal assets moving across multiple operators' footprints.

Scenarios where you can ignore it: single-campus or single-city private networks; utility metering where devices never leave the service area; downlink-heavy control applications.

Three questions to ask any network provider: which operators they have formal roaming agreements with (demand the list, not the verbal assurance); which roaming model they support (passive is the baseline; handover requires written confirmation); whether activation roaming goes through a peering hub (this determines the logistics cost of multi-country deployment).

8. The platform view

Roaming is an operator-grade, NS-to-NS interface; it has little to do with self-run private LoRaWAN networks. A campus or factory deployment on ThinkLink keeps every device on its own hNS from cradle to grave — no roaming procedures involved. The value of this topic for such users is procurement literacy: when the business expands to multi-site, multi-city or cross-border operation, knowing that "a private network can reach public coverage through passive roaming" exists as a path — and that the home operator's roaming partner list is a commercial term — prevents locking a device fleet into a network it can never roam out of.

ThinkLink is a terrestrial LoRaWAN platform supporting 1.0.2 / 1.0.3 / 1.0.4. Its official documentation makes no roaming-interface capability claims; statements here stay conservative pending official confirmation.

One sentence to take away: the essence of LoRaWAN roaming is "traffic crosses networks, ownership never does, and keys never leave the two ends they belong to." Transparent to devices, a business between operators, and a cost lever for solution providers — keep the three apart and the selection process stays out of trouble.


References: LoRa Alliance TS002-1.1.0 "LoRaWAN Backend Interfaces" (Nov 2020), Actility ThingPark Exchange materials and press releases, Netmore acquisition announcement (Jan 2026), operator documentation (ThingPark Wireless, Dragino Learn, etc.). Roaming availability and pricing subject to formal operator agreements.