InfiIoT Labs/Field insight 02
Radio architecture report · 20 minute read

One meter, three radios: NB-IoT, BLE and LoRa

The radio module price is the least interesting number in a metering rollout. The expensive choices hide in network ownership, basement coverage, wake-up behavior, gateway maintenance, and the truck roll that replaces a battery two years early.

InfiIoT LabsPublished 16 July 2026Standards reviewed: 9Worked example: one gas meter
Public cellular

NB-IoT

Direct wide-area path

Local 2.4 GHz

Bluetooth LE

Collector must be nearby

Sub-GHz LPWAN

LoRaWAN

Gateway or network required

Asking which radio uses the least power is like asking which pump uses the least electricity without naming the head, flow, or duty cycle. The answer can be calculated only after the work is defined.

We have used all three paths around metering architectures. Each can support a credible product. Each can also fail expensively while the radio itself performs exactly as specified. NB-IoT can spend most of its battery on a difficult network attachment. Bluetooth LE can move that cost into hundreds of collectors. LoRaWAN can look inexpensive on the first site until slow data rates and confirmed messages consume the airtime budget.

The decision equation
radio choice = deployment topology + energy per useful report + downlink contract
+ coverage evidence + five-year operating cost

Module cost belongs in the equation. Topology and field service usually deserve more weight.

01 · The topology pays the bill

A radio is also a decision about who runs the network

NB-IoT removes the site gateway from the architecture because the mobile operator has already built the access network. That is useful for a sparse fleet spread across towns. It also means roaming, coverage extensions, paging behavior, SIM lifecycle, and feature support depend on the operator. The GSMA deployment guide devotes separate sections to PSM, eDRX, paging, coverage extension, power class, and firmware upgrade support for a reason: a standards-compliant radio still depends on network configuration.[1]

Bluetooth LE puts the collector close to the meter. That collector may be a phone during a technician visit, a mains-powered gateway in a building, or an embedded concentrator in another product. Bluetooth LE was designed so that the better-powered central device can carry more of the communication work.[3] This is excellent when a collector already exists. Building and maintaining a collector fleet solely to save money on meter radios is a different business case.

LoRaWAN uses a star-of-stars topology. Gateways relay packets between end devices and the network server, and more than one gateway may hear the same uplink.[6] A private campus can own this network. A utility can subscribe to a public one. The choice determines who handles gateway placement, backhaul, network-server upgrades, keys, and failed-site diagnosis.

NB-IoTMeter → mobile operator → cloudSIM/eSIM, data plan, carrier operationsCoverage policy and network timers sit outside the product team
Bluetooth LEMeter → phone or local gateway → cloudGateway power, backhaul, software, and site supportA cheap meter radio can create an expensive gateway estate
LoRaWANMeter → one or more gateways → network serverGateway/backhaul operations or network subscriptionAirtime, regional parameters, and gateway placement set capacity
Architecture test

Delete the radio module from the cost sheet

Price the five-year system without the end-device radio. Include gateways, installation, backhaul, SIM operations, network-server hosting, key provisioning, coverage surveys, and field support. Add the module only after that subtotal is visible. This simple order of work prevents a low-cost transceiver from disguising a high-cost topology.

02 · A fair comparison

Give all three radios the same job

Radio comparisons become fiction when one candidate sends a tiny daily counter and another maintains a live, remotely commanded session. Our worked example is a battery-powered gas meter. It records consumption every fifteen minutes, sends a 96-byte application report every hour, raises tamper and valve alarms without waiting for the next batch, and stores seven days locally. A remote valve command must carry an identifier, expiry time, acknowledgement, and observed valve state.

The example is deliberately awkward. Hourly reporting exposes cellular session overhead. An alarm path exposes the cost of sleeping. A command exposes downlink reachability. Seven-day storage separates radio failure from data loss. The same payload schema, security requirement, antenna volume, and battery derating apply to every candidate.

Measurement

15-minute consumption intervals; local total remains authoritative.

Routine uplink

One 96-byte application payload per hour, plus protocol overhead.

Exception

Tamper, reverse flow, low battery, and valve state are event-driven.

Command

Valve actuation expires safely and reports the observed final state.

Storage

Seven days of readings survive a radio or gateway outage.

Service life

Battery model includes temperature, self-discharge, ageing, and field margin.

Report age matters more than packet arrival

The platform should preserve sample time, report time, and ingest time. A meter that reconnects after six hours has delivered new packets containing old measurements. Billing may accept that. Leak detection may require current data. The application must state which decisions use event time and which need current reachability.

03 · Battery arithmetic

Measure charge per useful report, including the failed ones

Peak current does not predict battery life. Neither does sleep current. The useful quantity is charge consumed across the complete state machine: wake, measure, attach or scan, transmit, receive, retry, store, and return to sleep. Multiply each state current by its duration, then add regulator loss, sensor load, battery self-discharge, temperature derating, and a field margin.

daily_charge_mAh = Σ(current_mA × seconds_in_state) ÷ 3600
field_budget = daily_charge × retry_factor + sensor + MCU + storage + self_discharge

The table below is a transparent sensitivity model. It isolates radio charge so the architectural difference is visible; it does not predict service life. The assumed NB-IoT session draws 220 mA for eight seconds. The LoRaWAN Class A report uses a 45 mA, 0.4-second uplink plus two receive windows. The Bluetooth LE meter advertises for 6 ms every five seconds at 8 mA. Sleep current is 5 µA for the cellular case and 2 µA for the local radios. Replace every number with a current trace from the intended module, antenna, SIM, firmware, battery, and network.

NB-IoT · 24 sessions/day220 mA × 8 s each≈ 11.9 mAh/dayRegistration and coverage repetition dominate
NB-IoT · 1 batched session/day220 mA × 8 s once≈ 0.61 mAh/dayBatching changes the battery case radically
LoRaWAN Class A45 mA × 0.4 s × 24 + RX windows≈ 0.21 mAh/daySlow data rate or retries increase airtime
Bluetooth LE advertising8 mA × 6 ms every 5 s≈ 0.28 mAh/dayCollector listening power sits outside the meter

The absurdly long nominal life produced by some radio-only calculations is the warning. It means the radio is no longer the limiting load, so battery shelf life, pulse capability, temperature, leakage, the metrology front end, and valve actuation take over. A ten-year label cannot be obtained by dividing amp-hours by an average current and stopping there.

NB-IoT offers a large link budget. 3GPP material gives a 164 dB maximum coupling loss target for standalone NB-IoT and identifies PSM and extended idle-mode DRX as power-saving mechanisms.[2] Deep coverage often requires repetitions, longer active time, and higher energy. Test the tail of the distribution. The median meter is rarely the one that forces a battery truck roll.

Bench test

The current-trace census

Instrument at least thirty meters across good, marginal, and failed coverage. Capture every state transition for seven days, including network registration, retries, downlink, local storage, and reconnect backfill. Report p50, p95, and the worst daily charge. A single bench trace beside a base-station emulator covers development. A fleet battery model needs the field distribution.

04 · Coverage and downlink

Range claims collapse inside cupboards, pits, and basements

A link budget belongs to the installed antenna, enclosure, frequency, wall construction, mounting height, interference, and receiver. Bluetooth SIG notes that range is a tradeoff among transmit power, receiver sensitivity, antenna gain, and the propagation environment. Typical LE Coded implementations can achieve receiver sensitivity around -103 dBm, but Coded PHY reduces application data rate.[4] The Core specification also makes LE Coded optional, so a product cannot assume every phone or gateway supports it.[5]

LoRaWAN trades data rate for transmission duration. Adaptive Data Rate can tune data rate and power for stable devices, while regional parameters define the legal and interoperable channel plans.[6][7] A slow uplink may close a difficult link, but it occupies the channel longer and spends more energy. Confirmed uplinks add downlink load and can make a weak network weaker if used as a blanket reliability setting.

Downlink is where the product contract becomes visible. An NB-IoT device in PSM is unreachable until its next wake event; eDRX trades more listening for shorter paging delay. A LoRaWAN Class A device opens two receive windows after its own uplink. Class B adds scheduled receive slots. Class C keeps the receiver open and is generally a mains-powered choice. A sleeping Bluetooth LE peripheral depends on its advertising or connection schedule and on the collector being present.

Routine reading
DesignBatch readings, preserve event time, and tolerate delayed delivery.
FailureThe billing record depends on a permanently live session.
Alarm
DesignUse an event uplink with local persistence and a stated retry budget.
FailureThe meter repeats until acknowledgement and drains itself in an outage.
Valve command
DesignDeclare maximum reachability delay, expiry, acknowledgement, and observed state.
FailureThe API accepts a command and the UI presents the physical valve as closed.
Firmware
DesignChoose a maintenance window and a radio-specific transfer and recovery plan.
FailureA large image is treated like another telemetry message.
Field test

The metal-box survey

Install the meter in the worst legal mounting position with its production enclosure and antenna. Close the cabinet. Add nearby meters until the channel is busy. Test routine uplink, alarm delivery, command latency, and recovery after a twelve-hour outage. Repeat after rotating or partly shielding the meter. The installation determines coverage. Module branding contributes no field evidence.

05 · The five-year ledger

Count the network people have to operate

Procurement usually compares module, antenna, certification, and subscription. Operations pays for the omissions: gateway surveys, SIM suspension, key recovery, failed backhaul, battery visits, replacement collectors, firmware campaigns, and support calls caused by an unclear failure boundary between the meter, local radio, gateway, carrier, and cloud path.

Five-year ownership model
TCO = devices × (radio + antenna + provisioning + certification)
+ sites × (survey + gateway + install + backhaul + maintenance)
+ devices × (network fees + battery service + support burden)
+ fleet events × (truck roll + failed command + missed data value)

NB-IoT

Usually lower site infrastructure

Recurring carrier relationship

Strong fit for sparse wide-area estates

Bluetooth LE

Low-cost end-device silicon

Collector density and support

Strong fit when phones or gateways already exist

LoRaWAN

Gateway investment can be shared

Private-network skill or provider fee

Strong fit for dense sites under one operator

Security also has an operating cost. Bluetooth LE provides pairing, bonding, encryption, and attribute permissions, but the product team must select and implement the relevant security level. Bluetooth SIG recommends authenticated LE Security Mode 1 Level 4 where the threat model requires it.[9] LoRaWAN and cellular have different key and identity lifecycles. The cost model must include secure manufacturing injection, ownership transfer, revocation, and a field recovery route that does not become a universal backdoor.

06 · Where each radio wins

Choose the failure mode you are equipped to manage

The winner changes with the organisation's network assets and tolerance for operational work. The strongest radio is the one whose dependency chain the operator can see, test, and repair.

Sparse meters across many townsNB-IoTCarrier coverage removes private gateways; verify PSM, roaming, and deep-indoor performance.
Dense campus under one facilities teamLoRaWANGateway cost is shared across many devices and coverage can be engineered locally.
Meter read or configured by a nearby technicianBluetooth LEThe phone is already the collector and supports a high-rate service session.
Basement meter rooms in managed buildingsLoRaWAN or BLE gatewayPlace the collector deliberately and use wired backhaul where radio escape is poor.
Frequent, time-bounded remote commandsNB-IoT with tested reachabilityThe wide-area path can work, but sleeping and paging policy must be part of the SLA.
Long battery life with small scheduled uplinksLoRaWAN Class ALow duty cycle works well when downlink can wait for the next uplink.

Hybrids are legitimate. A meter can use Bluetooth LE for commissioning and NB-IoT for routine service. A LoRaWAN estate can retain local optical or BLE access for diagnostics. The extra radio earns its place only when it removes a specific field burden. Otherwise it adds certification, firmware, antenna, security, and support work for the life of the product.

07 · The field acceptance test

Make the losing radio fail before procurement

A two-week radio trial can eliminate a bad architecture if it uses production enclosures, production batteries, real network accounts, and the worst installation positions. Bench range and average current are useful development measurements. They are weak procurement evidence.

Day 1–2CurrentTrace wake, measurement, attach/scan, uplink, receive, retry, command, and return to sleep.
Day 3–5CoverageMap good, marginal, and failed installations with production antenna orientation and enclosure.
Day 6–7OutageRemove carrier service, collector power, gateway backhaul, and cloud access separately.
Day 8–10DownlinkMeasure command reachability, expiry, duplicate handling, acknowledgement, and observed state.
Day 11–12ScaleCreate channel load, simultaneous reconnects, and backfill from a meter cohort.
Day 13–14OperationsReplace a battery, rotate a key, transfer ownership, swap a gateway, and diagnose one silent meter.

NB-IoT buys a carrier network. Bluetooth LE buys proximity. LoRaWAN buys control over a local LPWAN. None of them buys a working product until the battery, downlink, gateway, and support assumptions have survived the intended installation.

Research notes

Standards and original sources

The standards describe protocol capabilities and constraints. The charge figures are a worked sensitivity model for this article and must be replaced by measurements from the chosen hardware, firmware, battery, antenna, network, and installation.

  1. [1]NB-IoT Deployment Guide to Basic Feature Set Requirements, Version 3 · GSMA
  2. [2]3GPP Standards for the Internet of Things · 3GPP
  3. [3]The Bluetooth Low Energy Primer · Bluetooth SIG
  4. [4]Understanding Bluetooth Range · Bluetooth SIG
  5. [5]Bluetooth Core Specification, Low Energy Physical Layer · Bluetooth SIG
  6. [6]TS001-1.0.4 LoRaWAN L2 Specification · LoRa Alliance
  7. [7]LoRaWAN Technical Specifications and Regional Parameters · LoRa Alliance
  8. [8]TR007 Developing LoRaWAN Devices v1.0.0 · LoRa Alliance
  9. [9]Bluetooth Security and Privacy Best Practices Guide · Bluetooth SIG
Next in the lab

What if a transformer could order its own oil test?

A controlled maintenance workflow that turns a DGA trend into an evidence package, approval, laboratory sample, reconciled result, and auditable next action.

Read the next field report →
Connected-product assessment

Bring one product. Leave with a clear next step.

Use 30 minutes to test the fit around one product and one blocked decision. If there is a fit, we will outline the two-week Product Blueprint.

30-minute fit callOne productNo prepared deck