Read enough LoRaWAN project post-mortems and a pattern emerges: indoor gateway failures mostly trace back to software configuration and networking, while a significant share of outdoor gateway failures come from power and lightning. Weak signal can be fixed by adjusting the antenna, but a gateway that loses power entirely takes the whole coverage area's data down with it — and these incidents tend to surface six months after commissioning, because the first three months of testing never stress the system hard enough.
The previous two articles covered antenna selection and site survey. This one completes the deployment trilogy with the last piece: where the power comes from, and where the lightning goes.
A LoRaWAN gateway runs 24/7, year-round, and — more demanding than a cellular base station in one respect — it usually sits in places nobody watches daily: weak-current shafts, rooftops, well rooms. That means three realities:
The third point is the most overlooked. Article 36 made the point that the best coverage locations (roof edges, towers, rooftops) are often the hardest places to get power to. Power selection and site survey must happen in the same design pass — not "pick the spot first, figure out electricity later."
One boundary to draw: this article covers gateway-side power. End-device battery sizing (lithium-thionyl-chloride passivation, inrush surges and the like) is covered in full in the device power consumption and battery life article — those are two completely different ledgers.
Using ManThink's product line as the example: the GD61x indoor gateway supports three power options, spelled out clearly in the datasheet — Type-C from a stable 5 V source; a DC terminal accepting 5–24 V; and PoE on the G configuration (the 4G variant) only, via a matching PoE-capable Ethernet cable into the POE port. The GDI51 has an integrated PoE module, and outdoor models such as the GDO51 and GD62 also support PoE passthrough — power enters from the switch and exits from the gateway's POE interface to feed downstream equipment.
These are not interchangeable choices; each maps to a class of deployment:
Type-C 5V: commissioning and desktop scenarios only. The upside is that power is available anywhere; the downsides are that USB cable does not suit long runs and the connector is not built for industrial environments. In production projects treat it as a service spare, never the primary feed.
DC terminal 5–24 V: the workhorse for indoor engineering. Wide-voltage tolerance means a standard 12 V or 24 V switching supply works. But wide input does not mean "connect anything," and three things must be pinned down during construction:
PoE: first choice for outdoor and ceiling deployments. One cable carries power and data simultaneously — for buildings where the rooftop has exactly one weak-current conduit, it is the only viable answer. Outdoor models like the GDO51 build PoE passthrough into the enclosure: the IN port connects to the switch for data, the OUT port feeds the gateway's own POE interface. One fewer injector exposed to the weather means one fewer failure point and one fewer water ingress point.
The "plug and play" reputation of PoE misleads people. On real projects, three ledgers must be settled.
First: class and power matching. IEEE 802.3 defines four power tiers: Type 1 (802.3af) delivers 15.4 W at the power-sourcing equipment, about 12.95 W usable at the device; Type 2 (PoE+, 802.3at) is 30 W / 25.5 W; Type 3 and Type 4 (PoE++, 802.3bt) go higher. A LoRaWAN gateway typically draws only a few watts, so Type 1 suffices — but the manuals repeatedly stress using power equipment matched to the delivered specification, because some models need headroom for the passthrough port and the 4G module. Before energizing, confirm which PoE class the gateway manual requires, then confirm the injector or switch outputs the same class.
Second: cable and distance. PoE's 100-meter limit comes from the Ethernet specification itself, not from power running out. What matters more is cable construction: 100 meters of Cat6a has a DC loop resistance of roughly 12.5 ohms, a drop of about 7.5 V at Type 2 levels. Copper-clad aluminum (CCA) cable has far higher resistance, and the far-end device can brown-out and reboot. Two rules for the field: Cat5e minimum, pure oxygen-free copper preferred, CCA banned; for runs beyond 60 meters, go straight to pure-copper Cat6.
Third: heat. When many links in one cable bundle run high-power PoE, copper loss becomes heat — in the extreme case of 48 Cat6a links each carrying 60 W, the bundle dissipates 15–20 W per meter, effectively a space heater inside the cable tray. LoRaWAN gateway loads are far lower and rarely trigger this, but if you concentrate a row of PoE devices in one weak-current shaft, leave room for heat dissipation.
Remote well sites, agricultural parks, river and lake monitoring points — solar is the default answer in these places. Sizing follows a four-step algorithm, and skipping any step causes trouble.
Step 1: daily energy consumption. Sum "voltage × current × operating hours" for every load. On the gateway side, distinguish steady-state draw from transmit peaks: a complete unit with SX1302 concentrator plus backhaul module typically sits in the 3–10 W range; over 24 hours that is 72–240 Wh per day. This dwarfs the milliwatt-class end sensors, which can be ignored next to the gateway.
Step 2: days of autonomy. Set this from the local historical longest run of overcast or snow days. Five days is a reasonable figure for the middle-lower Yangtze region; seven days for northern winters. Being conservative here costs little — one fewer day of autonomy means roughly 20% more battery.
Step 3: battery capacity, with two factors that must not be dropped. The formula:
Nominal capacity = daily Wh × autonomy days ÷ DoD ÷ low-temperature derating
The two factors: LiFePO4 usable depth of discharge is about 0.8 (the charge controller will cut off earlier for protection); the temperature factor follows the site's winter minimum — LiFePO4 retains roughly 85% capacity at 0 to −10 °C and roughly 75% at −10 to −20 °C. A worked example: 120 Wh daily load, 7 days autonomy, DoD 0.8, temperature factor 0.75 gives 120 × 7 ÷ 0.8 ÷ 0.75 = 1400 Wh, or about 117 Ah on a 12 V system. Skip both factors and the naive division yields 70 Ah — that is exactly how fleets go dark collectively in the second half of winter nights.
One hidden constraint: most lithium cells cannot accept charge below 0 °C. Sites in severe cold need either self-heating battery packs or a charge controller with low-temperature charge blocking, otherwise winter sunshine goes to waste.
Step 4: size the panel by December, not by the annual average. A solar panel's rated wattage is a Standard Test Conditions number that a real site never achieves. Around 40° N, winter peak sun hours run 50–60% below summer — possibly only 1.5–2 hours per day; 45° N is worse. Design the panel on annual-average insolation and the system will go offline on schedule every winter. The correct approach: look up December peak sun hours for the location (NREL PVWatts or an equivalent), and size the panel so that "worst-month generation ≥ daily Wh × 1.3."
Lightning is the most violent way an outdoor gateway dies — and a direct strike is not required. Induced surges traveling in on antenna feeds and power lines can take out the Ethernet port, the PoE injector, and the switch in one stroke. The deployment playbook that recurs across vendor manuals organizes into four lines of defense:
Line 1: air terminal, so direct strikes detour. On outdoor pole deployments, mount a lightning rod at the top of the pole, and keep all equipment (antennas, gateway, feed connectors) inside its 45° protection cone. Note that the antenna itself must not double as the air terminal — a direct strike carries peak currents well over 100 kA, which no antenna survives.
Line 2: antenna feed surge arresters. Fit a coaxial surge arrester on the N-type connector of every antenna (LoRa, 4G, Wi-Fi, GPS), and bond the arrester to the building's common ground with 10 AWG (about 5.3 mm²) or heavier wire. Vendor field data keeps repeating one point: even well-grounded tower equipment can see 10 kA-class surges at the connector side — that is why feed arresters are non-negotiable. Also, GPS antennas need arrester specifications different from ordinary RF arresters; do not mix them.
Line 3: bond the gateway enclosure. With another 10 AWG-or-heavier conductor, connect the gateway enclosure's ground screw to the common ground point. A frequently asked detail: why must the enclosure and the arresters share the same ground? Because at strike time the whole system's potential rises together; if the enclosure, arresters, and Ethernet SPD each bond to separate "grounds," the surge travels along the potential difference between devices — using your equipment as the conductor. Equipotential bonding decides survival more than earthing itself does.
Line 4: Ethernet SPD. Install an Ethernet surge protective device where the network cable enters the building, so the PoE injector and switch both sit under its protection. This line is the most commonly skipped — the gateway rides out the strike, then the switch and upstream equipment burn, and the post-mortem reveals the surge came in on the network cable.
Three supplements:
Outdoor models' IP67 label is often misread as "throw it anywhere and it survives." IP67 does mean protection against temporary immersion, but the seal is maintained by workmanship, not by the enclosure alone:
Temperature is part of the environmental ledger too. ManThink's official operating-temperature figures: GD61x/GD62 at 25 °C to 60 °C, with the GDI51/GDI52 and GDO51 delivered to the wide-temperature specifications in their respective manuals (confirm exact ranges against the official datasheets). Below the specified floor, lithium batteries cannot discharge and LCDs fail — which is why cold-region sites need industrial low-temperature cells and insulation.
Power and lightning protection are not finished at installation. Three things after go-live:
Document everything. After installation, record: antenna height, cable types and lengths, grounding locations and conductor sizes, arrester and SPD model numbers and positions, gateway serial numbers. Take photos — six months later nobody remembers how the rooftop cabling ran. This archive pays for itself after storm season: when performance suddenly degrades, technicians can trace whether the problem sits on the feed, the Ethernet path, or the power side.
Post-storm-season inspection. Check grounding connections for looseness, arrester status windows, enclosure seals and drip loops, cable jacket cracks. SPDs with replaceable modules have the lowest inspection cost.
Verify power-fail self-healing. At commissioning, deliberately cut power once and confirm the gateway reconnects to the network server and restores the backhaul link automatically. This five-minute test during commissioning prevents the three-months-later discovery that "power recovery needs a site visit." For critical sites, add a small UPS or choose a dual-power design.
Finally, the whole selection compressed into two lists.
Which power to use:
Three things never to cheap out on:
Power and lightning protection take up little of the BOM, but they decide how long the coverage lasts. Article 36's line applies here too: before you start, prove the site can be built — and can survive.