Menu

Which Radio

Which Radio? BLE vs Wi-Fi vs LoRa, The Choice That Decides Your Battery, Your Range, and Your Redesign

Notes · Radio and wirelessNovember 19, 202613 min readSenova Systems Research

BLE, Wi-Fi, and LoRa each win one column and lose another, and the wrong choice costs battery life, range, or a redesign. BLE is the cheapest way to talk to a phone (10–15 mA transmit, months to years on a coin cell). Wi-Fi is the cheapest way to move a lot of data (5–20 Mbps practical) but its 2–4 second connection costs 1.78 joules, 51,000 times the BLE reading. LoRa is the cheapest way to cross a field (2–15 km) but its spreading factor is a 23× energy lever and the EU 1% duty cycle turns a 512 kB firmware update into 21 hours. The same 24-byte reading, once an hour for a year: BLE 0.3 J, LoRa SF7 49 J, LoRa SF12 1,144 J, Wi-Fi 15,610 J, 6.3 times the whole cell.

First, a simple picture

Three ways to send a letter.

A person who wants to send a letter has three choices. They can hand it to someone standing next to them, fast, free, and it only works if the person is close. They can put it in a courier van, it carries a whole box of letters, but the van has to warm up, drive to the depot, and come back, and the van drinks fuel whether the box is full or empty.

Or they can send it by pigeon, the pigeon flies for kilometres on almost nothing, but it carries one small note, and it can only fly so often, because the law says the pigeon must rest ninety-nine times longer than it flies.

BLE, Bluetooth Low Energy, is the hand to the person next to you. Wi-Fi is the courier van. LoRa, Long Range, and LoRaWAN, the network protocol built on top of it, is the pigeon. Each one wins one column and loses another, and the wrong choice costs battery life, range, or a redesign. The radio is not a spec sheet. It is a trade between range, data rate, and energy, and the product's job decides which trade wins.

The three radios, in one table

Three radios: each wins one column and loses another

RadioRangeData rateTransmit currentSleep
BLE10 to 100 m (400 m with coded PHY)1 to 2 Mbps (about 200 kbps in practice)10 to 15 mA1 to 5 µA
Wi-Fi30 to 100 m indoors5 to 20 Mbps in practice120 to 240 mAConnection takes 2 to 4 s
LoRa / LoRaWAN2 to 15 km300 bps to 11 kbps20 to 40 mA1 to 2 µA; EU duty cycle 1%

BLE is a short-range radio designed for a phone to talk to a thing, and for things to talk to each other in the same room. Its range is 10–100 m in practice, 30–50 m indoors, up to 100 m in line of sight, and Bluetooth 5.0's coded PHY trades data rate for range: S=2 runs at 500 kbps and buys roughly twice the range, S=8 runs at 125 kbps and buys roughly four times the range, with about 12 dB of extra link budget. In the open field, coded PHY reaches 400 m to 1 km; in a real industrial building, plan for 200–500 m. The data rate is 1–2 Mbps on paper and ~200 kbps in practice, the raw rate is not the throughput, because the protocol, the connection interval, and the phone's own stack all take their cut. The transmit current is 10–15 mA, the sleep current is 1–5 µA, and a coin-cell device can run months to years. A BLE module costs $1–3. And the catch: longer range costs energy, S=8 transmits for eight times longer per packet, which means eight times the energy per packet, so the range you buy with coded PHY is paid for in battery.

Wi-Fi is the radio that already exists in every building, and it moves data faster than anything else in this article, 5–20 Mbps in practice, with the raw rate far higher. Its range is 30–100 m indoors, and it is the only one of the three that speaks the internet directly, with no gateway in between. The transmit current is 120–240 mA, ten to twenty times BLE's, and the connection overhead is 2–4 seconds: the radio has to associate with the access point, authenticate, and get an address before a single byte of your data moves. A Wi-Fi module costs $2–5. And the catch is the one the demo proves: the packet costs almost nothing, and the connection costs everything.

LoRa / LoRaWAN is the radio that crosses a field. Its range is 2–15 km, 2–5 km at SF7, 10–15 km at SF12, in rural line of sight, and it does it on a 20–40 mA transmit current and a 1–2 µA sleep current, so a battery device can run 5–10 years. A LoRa module costs $3–8. The data rate is 300 bps to 11 kbps, SF7 at 5,470 bps, SF12 at 250–293 bps, and that is the whole point of the radio: it trades data rate for range, and it trades it hard. The spreading factor (SF7–SF12) is the knob: each step up doubles the airtime and halves the data rate, while gaining about 2.5 dB of receiver sensitivity, roughly 30–40% more range per step. SF12 has 32× the airtime of SF7, 32× the energy per packet, and 32× the duty-cycle consumption. And the law is the second knob: in Europe, the 868 MHz band carries a 1% duty cycle on the default channels, the device may transmit for 36 seconds in every hour, and must stay silent 99 times longer than it transmitted. A 20-byte packet at SF7 takes 57 ms on the air and 5.6 seconds of silence after it; a 50-byte packet at SF12 takes 2.30 seconds on the air and 3.8 minutes of silence. The maximum application payload in Europe is 222 bytes at SF7 and 51 bytes at SF12, and the US replaces the duty cycle with a 400 ms dwell-time cap per uplink, which is what caps American payload sizes. LoRaWAN is the network protocol built on top of LoRa, device addressing, security, data rates, gateways, and a network server, and it is the protocol that makes the pigeon a system, not a trick.

Two more radios, named so the reader does not pick them by accident. NB-IoT and LTE-M are cellular, 1–10 km of range, 26 kbps to 1 Mbps, but a 200–500 mA transmit current with peaks of 1–2 A, and a monthly fee per SIM. They are the right answer when the product must work anywhere there is a cell tower and the customer will pay for the subscription; they are the wrong answer when the product must run for years on a coin cell.

Zigbee and Thread are mesh radios, 250 kbps, 10–100 m per hop, and the network heals itself by hopping, but the routers that carry the mesh must be mains-powered, because a router cannot sleep. They are the right answer inside a building full of mains-powered devices; they are the wrong answer on a fence post.

When each radio loses

BLE loses on range. A BLE link falls apart somewhere between 30 and 50 metres indoors, and it gets worse fast when Wi-Fi access points, microwaves, and Zigbee gateways are stomping on the same 2.4 GHz band. Coded PHY buys the range back, 200–500 m in a real building, but it costs eight times the energy per packet at S=8, and both ends must support it. BLE is the cheapest way to talk to a phone. It is not the way to cross a field.

Wi-Fi loses on energy, and the loss is not in the packet, it is in the connection. The 24-byte reading in the demo takes 0.038 ms of airtime at 5 Mbps. The connection takes 2–4 seconds at 120–240 mA. The packet is 0.002% of the energy; the connection is 99.998%. A Wi-Fi device that wakes up once an hour to send one reading is not a low-power device with a Wi-Fi problem, it is a mains-powered device pretending to be a battery device. Wi-Fi is the cheapest way to move a lot of data. It is not the way to send a little data, rarely, on a battery.

LoRa loses on data, and the loss is not the rate, it is the law. The data rate is slow, and the duty cycle is slower: at 1%, the device must stay silent ninety-nine times longer than it transmitted, and the maximum payload at SF12 is 51 bytes. The firmware update from Article #11, a 512 kB image, takes 20.5 seconds over BLE, 21.3 hours over LoRa SF7 at the duty cycle, and 16.6 days at SF12, in 2,362 packets at SF7 or 10,280 packets at SF12. Forget firmware OTA over LoRa, the pipeline's own note, and the demo proves it. LoRa is the cheapest way to cross a field. It is not the way to move a file.

Example Labs is a made-up company we use for worked examples.

DEMO

This product is made up. Call it the Example Labs field sensor hub again: the fence-post unit from Articles #6, #7, #9, and #11, the weather sensor, the event camera, the battery, the enclosure, now choosing its radio. The hub sends one reading an hour: temperature, humidity, battery, event count, timestamp, 24 bytes, 192 bits. The demo's arithmetic is the article's own, verified before writing, on the published figures cited above; the hub and its readings are the worked example, not a measured product.

Step 1, BLE. The reading is 192 bits at a practical 200 kbps: 0.96 ms of airtime. At 12 mA transmit (the middle of 10–15 mA) and 3 V, that is 34.6 µJ per reading, which matches Article #6's figure of about 30 µJ per BLE packet, because a 24-byte reading is one packet. One reading an hour for a year is 8,760 readings: 0.3 J, 0.01% of the CR2032's 2,484 joules. The radio is not the battery's problem.

Step 2, Wi-Fi. The reading is 192 bits at a practical 5 Mbps: 0.038 ms of airtime, nothing. The connection is 3 seconds (the middle of 2–4 s) at 180 mA (the middle of 120–240 mA) and 3.3 V: 1.78 J per reading, 51,000 times the BLE reading. One reading an hour for a year is 15,610 J, 628% of the whole cell. The hub dies in 58 days, on connections alone, before the sensor, the model, or the sleep current have spent a joule. The packet costs almost nothing. The connection costs everything.

Step 3, LoRa, at SF7. The reading is a 20-byte payload on the air for 56.6 ms (the Semtech AN1200.13 calculator figure, for a 20-byte payload at SF7, 125 kHz, coding rate 4/5). At 30 mA transmit (the middle of 20–40 mA) and 3.3 V, that is 5.6 mJ per reading. One reading an hour for a year is 49.1 J, 2.0% of the cell. Kilometres of range for 2% of the battery. And the duty cycle is no problem at one reading an hour: 57 ms of airtime, 5.6 seconds of silence, and 3,594 seconds to spare.

Step 4, LoRa, at SF12. The same reading takes 1,318.9 ms on the air, 23 times longer, and 130.6 mJ per reading, 23 times the energy. One reading an hour for a year is 1,144 J, 46% of the cell. The range is longer, 10–15 km against SF7's 2–5 km, and the price is nearly half the battery, for the same 24 bytes. SF12 is not always best: a device sending 24 packets a day at SF12 on the EU868 1% sub-band will consume its daily airtime budget, and the range it buys is range the product may not need. The spreading factor is a 23× energy lever, and the lever is the decision.

BLE0.3 JLoRa, SF749.1 JLoRa, SF121144.0 JWi-Fi15610.0 JLog scale. Shorter is better.
Energy to send one 24-byte reading an hour for a year, by radio (share of a coin cell in the detail). Wi-Fi's cost is the connection, not the packet.

Step 5, The farmhouse, 300 m away. The hub is on a fence post. The farmhouse, with the gateway, the router, and the mains, is 300 m away. BLE at 1M PHY reaches 50–100 m in practice: it fails. BLE coded PHY at S=8 reaches 200–500 m in a real building: it works, marginally, in the open, and it costs eight times the energy per packet. Wi-Fi reaches 30–100 m indoors: it fails. LoRa at SF7 reaches 2–5 km: it works easily, at 5.6 mJ per reading. So the product does not pick one radio. It picks three: LoRa for the link (hub to gateway, 300 m, 5.6 mJ per reading); BLE for the phone (the technician standing next to the hub, commissioning it and pushing the 512 kB firmware update in 20.5 seconds, the update from Article #11, which cannot travel over LoRa); and Wi-Fi for the backhaul (the gateway, mains-powered, in the farmhouse, talking to the router, where the 1.78 J connection is affordable because the mains never runs out). The right answer is usually not one radio. It is one radio per job, and the job that crosses the field is not the job that talks to the phone.

One product, three radios: each link gets the radio that fits it

LinkRadioWhy
Hub to gateway, 300 m across a fieldLoRa (SF7, 2 to 5 km)The only radio that reaches; 5.6 mJ a reading
Technician's phone at the hubBLECommissioning and firmware; the 512 kB update in 20.5 s
Gateway to the farmhouse routerWi-Fi1.78 J a connection is affordable on mains power

The decision test

Four questions. They sort the radios that fit the product from the ones that will cost a redesign:

  1. How far, how far apart are the device and the thing it talks to, in the real environment, with walls and metal and weather, not on a datasheet's open field? Under 100 m, BLE is in play. 300 m, coded PHY or LoRa. Kilometres, LoRa, and nothing else in this article.
  2. How much, how many bytes per message, and how many messages per day, and is there a file to move (a firmware image, a photo, a log)? Small readings, rarely: LoRa or BLE. A firmware image: BLE or Wi-Fi, never LoRa, because the duty cycle turns 512 kB into 21 hours. A photo: Wi-Fi, because the data rate is the job.
  3. How often, how many times a day does the radio wake up, and what does each wake-up cost? Once an hour: BLE 0.3 J a year, LoRa SF7 49 J, Wi-Fi 15,610 J, the connection is the cost, and the cost is multiplied by every wake-up. The radio that looks cheap once is the radio that kills the battery 8,760 times.
  4. What power, battery or mains, and if battery, which battery and how long must it last? Coin cell, years: BLE or LoRa, and LoRa only at the spreading factor the range actually needs, because SF12 costs 23× SF7. Mains: Wi-Fi is free to be expensive, and the gateway is where Wi-Fi belongs.

If the answers are far, small, rarely, battery, LoRa, at the lowest spreading factor that reaches. If the answers are near, small, rarely, battery, BLE. If the answers are near, large, often, mains, Wi-Fi.

And if the answers are far, small, rarely, battery, and there is also a firmware update and a phone in the field, the answer is all three, one per job. If any answer is "we don't know yet", the problem is not the radio. The problem is that the product's job was not written down before the radio was chosen, and the review starts by writing it down.

"But Wi-Fi is already in the building", and two more fair doubts

"But Wi-Fi is already in the building." It is, and the building is not the product. Wi-Fi's range is 30–100 m indoors, and the hub is on a fence post 300 m from the farmhouse. "Already in the building" is a reason to put Wi-Fi in the gateway, which is in the building and on the mains. It is not a reason to put Wi-Fi in the hub, which is on a fence post and on a coin cell, where the connection alone kills the battery in 58 days. The radio that is already there is the right radio for the thing that is already there.

"LoRa is too slow." It is slow, 300 bps to 11 kbps, and the slowness is the price of the range. The question is not whether LoRa is slow. The question is whether the product's messages are small enough and rare enough that the slowness does not matter: a 24-byte reading an hour is 57 ms on the air at SF7, and the duty cycle leaves 3,594 seconds to spare. If the messages are a firmware image, LoRa is not slow, it is impossible, and the article says so. If the messages are readings, LoRa's slowness is a number, not a verdict.

"BLE's range is enough." It might be, and the range that matters is the range in the real environment, not the range on the datasheet. A BLE link falls apart between 30 and 50 metres indoors, and it gets worse fast when Wi-Fi access points, microwaves, and Zigbee gateways are stomping on the same 2.4 GHz band. Coded PHY buys the range back, at eight times the energy per packet, and both ends must support it. Measure the range on the real site, with the real antenna, in the real enclosure, Article #7's enclosure is where the antenna lives, and the antenna's ground plane is Article #5's board, before promising the customer a number. The range you have not measured is the range you do not have.

Sources

  1. EmbeddedInterviewLab, "BLE vs Wi-Fi vs LoRa vs Zigbee vs NB-IoT vs LTE-M vs UWB" (embedded-ace-guide.vercel.app)
    What we used

    BLE 10–100 m / 1–2 Mbps / TX 10–15 mA / sleep 1–5 µA / months–years on a coin cell / $1–3; Wi-Fi 30–100 m indoor / 11–150 Mbps / TX 120–240 mA / connection overhead 2–4 s / hours–weeks on a battery / $2–5; LoRaWAN 2–15 km / 300 bps–11 kbps / TX 20–40 mA / sleep 1–2 µA / 5–10 years / $3–8; NB-IoT/LTE-M 1–10 km / 26 kbps–1 Mbps / TX 200–500 mA (peaks 1–2 A) / monthly fee per SIM; Zigbee/Thread 250 kbps / mesh / routers need mains; UWB 850 kbps–27 Mbps / 10–30 m

  2. Hubble, "How to Use BLE Long Range (Coded PHY)"
    What we used

    And "How to Use BLE Coded PHY for Long-Range Communication in Noisy Environments" (2026), Coded PHY since Bluetooth 5.0; S=2 at 500 kbps buys ~2× the range, S=8 at 125 kbps buys ~4× the range with ~12 dB of extra link budget; 1M PHY realistic industrial range 50–100 m, open field 100–200 m; coded S=8 realistic industrial 200–500 m, open field 400 m–1 km+; S=8 transmits 8× longer per packet, so 8× the energy per packet; a BLE 1M link falls apart between 30 and 50 m indoors, worse with Wi-Fi, microwaves, and Zigbee on the same 2.4 GHz band; Coded PHY and LoRa sit at different points on the curve, pick Coded PHY for smartphone reachability, sub-second latency, or hundreds of kbps

  3. Ezurio, "Range Testing, BL654 DVK" (application note v2.0)
    What we used

    Field test at 1.25–2.5 m antenna height: 1 Mbps BLE in a connection 625 m–1.2 km; 125 kbps coded PHY 1–1.9 km; the range doubles when going from 1 Mbps to 125 kbps coded PHY; each time the antenna height doubles at both ends, the range doubles (6 dB link-budget improvement)

  4. EmbeddedInterviewLab, "LoRa & LoRaWAN" (embedded-ace-guide.vercel.app)
    What we used

    Data rate 300 bps (SF12/125 kHz) to 11 kbps (SF7/250 kHz); EU868 mandates 1% or 0.1% duty cycle per sub-band; each SF step doubles the airtime and halves the data rate, gaining ~2.5 dB of sensitivity (30–40% more range); SF7 5,470 bps / −123 dBm / 2–5 km rural / ~36 ms airtime for a 10-byte payload; SF12 250 bps / −137 dBm / 10–15 km / ~1,155 ms; SF12 has 32× the airtime of SF7; a device sending 24 packets/day at SF12 on the EU868 1% sub-band will consume its daily airtime budget

  5. NRTechStudio, "LoRa Airtime Calculator" (Semtech AN1200.13 formulas)
    What we used

    20-byte payload at 125 kHz, coding rate 4/5, preamble 8: SF7 56.576 ms at 5,468.75 bps; SF12 1,318.912 ms at 292.97 bps; at a 1% limit, a transmitter may be on the air for 36 seconds in every hour; the 1% value is the EU ISM band rule for 868 MHz, other bands and regions use different limits, some use dwell-time limits instead

  6. Industrial Wireless, "LoRaWAN for Industrial Sites: Payload, Duty-Cycle and Range Limits"
    What we used

    LoRa Alliance Regional Parameters RP002-1.0.4: EU868 data rates DR0 = SF12/125 kHz at 250 bit/s, DR5 = SF7 at 5,470 bit/s; the three default channels (868.10 / 868.30 / 868.50 MHz) carry a <1% duty-cycle limit, at most about 36 seconds of airtime per hour; US902-928 replaces the duty cycle with a 400 ms dwell-time cap per uplink; max application payload EU868: 51 bytes at SF12, 222 bytes at SF7; range up to 15 km rural per the LoRa Alliance, with no urban or indoor figure published

  7. The Things Network, "LoRaWAN Airtime Calculator"
    What we used

    23 bytes on air at SF7/125 kHz: 61.7 ms time on air, 486 uplinks/day under the Fair Use Policy (30 s of uplink airtime per day); at SF12: 1.483 s, 51 B max payload, 20 uplinks/day; a device must stay silent for 99× its time on air on a 1% sub-band

  8. CyberDev, "LoRa up close: chirps, spreading factor, time on air and range"
    What we used

    SF7 −123 dBm / SF12 −137 dBm (a hundred times weaker than the noise, and still readable); +1 to SF means a symbol twice as long and roughly +2.5 dB of sensitivity; the main European rule is the duty cycle, typically 1%: after every transmission the device must stay silent 99 times longer than it transmitted; 20 bytes at SF7: 57 ms airtime, 5.6 s pause; 237 bytes at SF7: 374 ms, 37 s pause; 50 bytes at SF12: 2.30 s, 3.8 min pause; 237 bytes at SF12: 8.53 s, 14 min pause

More in design notes

CE markingOn Wednesday 20 January 2027, Regulation (EU) 2023/1230 applies in full and Directive 2006/42/EC is repealed, a hard cut-over, unit by unit, with no grace period for new machines built to the old Directive. The Regulation adds essential requirements the Directive never had: protection against corruption, software and update integrity, self-evolving behaviour, and AI in safety functions, and for the six Annex I Part A categories, a notified body is always mandatory. Example Labs' cobot cell costs €6,625 in testing without the AI safety function and €16,625 with it, and the notified-body route lands February–May 2027, after the deadline.21 minHuman FactorsIEC 62366-1:2015+AMD1:2020 is the international standard for usability engineering of medical devices, recognised by FDA as a consensus standard, and FDA's guidance 'Applying Human Factors and Usability Engineering to Medical Devices' (2016, revised August 2026) sets what reviewers expect. The process: specify users and use environment, identify use-related hazards, design the interface to mitigate them, validate with representative users. Formative testing uses early prototypes with 5–8 participants per user group; summative validation uses the finalised, production-equivalent device with at least 15 participants per distinct user group. On average 85% of utilisation problems are found with 5 representative users, 95% with 10. Between 2012 and 2015 there were 423 medical device recalls resulting from UI software errors. Acceptance is the correct execution of a task without use errors or difficulties, a percentage of users completing a task in a time is not a pass mark, because the users who cannot complete it are the signal, not the noise.24 minComponent deratingYour board was designed at the ratings, and now the warranty claims are coming in, slow drift, not sudden death. Here is the derating table that makes the product last: the 50% rule, NASA's table, the tantalum numbers, and what MTBF actually means.19 minEMC/EMIEvery board is a radio station. The only question is whether it knows it is one. A 60 mm clock trace on a 2-layer board fails radiated emissions at its 5th harmonic by 1.7 dB; the same trace on a 4-layer board passes with 14.4 dB of margin. This piece shows the three ways a board radiates, the return path that decides most failures, and the stackup that fixes them.25 minPCB Design for Manufacture and TestThe average PCB project needs 2.9 respins at $46,000 each, and teams that simulate first need 3.0 spins to production against 4.6 for build-and-test teams. A board that works on a bench is not the same as a board a factory can build, test, and repair. The rules that close that gap, in plain words.14 min