Industry Insights
LoRa vs narrowband FSK: what 470-510 MHz rules allow
China caps one transmission at 1 second and 50 mW e.r.p. Here is what that does to SF12, to 51-byte meter readings, and to a 17 dBm module.
LoRa vs narrowband FSK: what 470-510 MHz rules allow
One second decided the review
A water-meter design arrived at a design review with everything in order: a 17 dBm radio, a 2 dBi antenna, a datasheet that said 470-510 MHz, and a link budget that closed at four kilometres. It was rejected on one line of a regulation nobody in the room had read: a single transmission may not last longer than one second.
The meter was sending a 20-byte reading at SF12. At 125 kHz bandwidth that frame occupies the channel for 1,319 ms. The radio was inside the power limit, inside the occupied-bandwidth limit, inside the band, and still not legal for that equipment category. No amount of RF engineering fixes a number that is decided in the media-access layer.
What follows works through the three regimes that decide almost every sub-GHz meter design — China at 470-510 MHz, Europe at 868 MHz, the United States at 902-928 MHz — and turns each one into arithmetic you can check against a parameter table. Every product figure below is quoted from a shipping module's published specification. Every computed figure is labelled as computed, with its assumptions.
The rule is six numbers
China's requirements for this band come from MIIT Announcement No. 52 of 2019. Its annex, the catalogue of micro-power short-range equipment, splits devices into eight categories. The 470-510 MHz allocation sits in category four, civil metering instruments, and the category text restricts it to small-area networks: buildings, residential compounds, villages.
Requirement | Limit for 470-510 MHz metering equipment | Where it bites |
|---|---|---|
Frequency range | 470-510 MHz | A 902-928 MHz module is a different product, not a firmware setting |
Transmit power | 50 mW e.r.p. (17 dBm) | Conducted power plus antenna gain, not the number on the module label |
Power spectral density | 50 mW/200 kHz e.r.p. below 200 kHz; 10 mW/100 kHz e.r.p. between 200 and 500 kHz | Widening the channel costs 7 dB of permitted density |
Single transmission | No more than 1 second | Rules out SF11 and SF12 for normal payloads |
Occupied bandwidth | No more than 500 kHz | 500 kHz LoRaWAN profiles land in the lower density row |
Frequency tolerance | 100 x 10^-6 | About 47 kHz at 470 MHz, generous for a TCXO design |
The annex adds four conditions that are not numbers at all, and they are the ones that fail in the field. Transmission is limited to one channel at a time, which rules out the hop-during-a-packet trick some stacks use. The device must implement a pre-transmission search — listen-before-talk, channel activity detection, call it what you like — and the user must not be able to switch it off. It must not transmit on frequencies used by local sound or television broadcasters, which is a per-city constraint, not a national one. And if the equipment meets all of this, it needs no frequency licence, no station licence and no type approval, which is precisely why the band is crowded.
The category matters more than the spectrum. The same annex, two pages earlier, lists wireless remote controls at 470-566 MHz: 5 mW e.r.p., occupied bandwidth no more than 1 MHz, and a periodic transmission of at most 1 second with at least 60 minutes between transmissions. Same frequencies, one tenth the power, one packet per hour. If your product is a valve actuator rather than a meter, you are not in category four, and the design you copied from a meter will not qualify.
Europe and the United States cap different variables
Reading only one regime produces engineers who believe "duty cycle" and "airtime limit" are the same thing. They are not, and the difference changes what you optimise.
Region | Band | Power | Time rule | What the rule actually caps |
|---|---|---|---|---|
China | 470-510 MHz | 50 mW e.r.p. | Each transmission at most 1 s; LBT required | Burst length, per device |
Europe | 868.0-868.6 MHz | 25 mW e.r.p. | 1% duty cycle, or polite spectrum access | Share of the hour, per transmitter |
Europe | 869.4-869.65 MHz | 500 mW e.r.p. | 10% duty cycle, or polite spectrum access | Share of the hour, per transmitter |
United States | 902-928 MHz | 1 W conducted with 50+ hopping channels | 0.4 s of occupancy per frequency in any 20 s | Per-channel occupancy across a hop set |
The European figures are from ETSI EN 300 220-2, which lists the 868 MHz sub-bands with their duty cycles; 868.7-869.2 MHz is the strictest at 0.1%, which is the sub-band most LoRaWAN EU868 deployments avoid for exactly that reason. The American figures are from 47 CFR 15.247: a frequency-hopping system in 902-928 MHz with a 20 dB bandwidth under 250 kHz needs at least 50 hopping channels and may occupy any one of them for no more than 0.4 seconds in a 20-second period, which is 2% per channel but far more than 2% in aggregate once the hops are counted.
Three regimes, three different caps. A stack built to satisfy the European duty-cycle budget can still emit a 2.3-second transmission, and a stack built to satisfy the Chinese one-second cap can still burn 40% of an hour on a single channel in Europe. Compliance is not a property of the radio; it is a property of the configuration.
50 mW e.r.p. is not 17 dBm at the connector
The single most common error in a Chinese deployment file is writing "17 dBm" in the RF field and stopping there. The limit is effective radiated power, and radiated power includes the antenna.
P_cond(max) = 17 dBm - G_ant + 2.15 dB (e.r.p. limit, dipole reference)
P_erp = P_cond + G_ant - 2.15 dB
50 mW = 17.0 dBm e.r.p. = 19.15 dBm e.i.r.p.
Antenna gain | Maximum conducted power | Typical hardware |
|---|---|---|
-3 dBi | 22.2 dBm | Printed antenna inside a plastic meter cover |
0 dBi | 19.2 dBm | Stub antenna, no ground plane |
2 dBi | 17.2 dBm | Small dipole on a DIN-rail box |
5 dBi | 14.2 dBm | External whip on a wall bracket |
8 dBi | 11.2 dBm | Yagi aimed down a riser |
The direction of the trade is the useful part: a worse antenna buys you more permitted transmitter power, because the limit is on what leaves the box, not on what the chip delivers. A 17 dBm module with a 0 dBi antenna sits exactly on the line, with zero margin for a connector that measures a little hot. The same module behind a -3 dBi antenna has 5 dB of headroom, which is why small antennas and 22 dBm settings appear together in Chinese metering designs.
The module that does not fit this table at all is a 36 dBm part. Thirty-six dBm conducted is 19 dB above the ceiling before any antenna is attached; no antenna gain can rescue it, because gain only makes the number larger. A 36 dBm module is a legitimate product for long-range links and for markets where the limit is 1 W, but it is not a drop-in for a category-four metering deployment in China.
Airtime is the number the datasheet never prints
Every LoRa datasheet prints sensitivity and output power. Almost none prints time on air, because time on air is a function of payload, spreading factor, bandwidth and coding rate, and the datasheet would need a table rather than a line. It is also the number that decides legality in China.
def airtime_ms(sf, payload, bw_khz=125, cr=1, preamble=8, header=1, crc=1):
tsym = (2 ** sf) / (bw_khz * 1000) * 1000 # symbol time, ms
de = 1 if tsym > 16 else 0 # low-data-rate optimisation
n = 8 + max(ceil((8 * payload - 4 * sf + 28 + 16 * crc - 20 * header)
/ (4 * (sf - 2 * de))) * (cr + 4), 0)
return (preamble + 4.25) * tsym + n * tsym
Spreading factor | Symbol time | 20-byte frame | 51-byte frame | Largest payload under 1 s |
|---|---|---|---|---|
SF7 | 1.02 ms | 51 ms | 98 ms | 255 B (394 ms) |
SF8 | 2.05 ms | 93 ms | 175 ms | 255 B (697 ms) |
SF9 | 4.10 ms | 185 ms | 308 ms | 199 B (984 ms) |
SF10 | 8.19 ms | 330 ms | 575 ms | 102 B (985 ms) |
SF11 | 16.4 ms | 659 ms | 1,233 ms | 38 B (987 ms) |
SF12 | 32.8 ms | 1,319 ms | 2,302 ms | 13 B (991 ms) |
Read the last column as a specification, not a suggestion. At SF12 the largest legal frame is 13 bytes of application payload — before the LoRaWAN MAC header, frame counter and MIC, which is why the answer in practice is "SF12 cannot carry a metering frame in this band". At SF11 a 51-byte frame is 1,233 ms and fails; a 38-byte frame passes with 13 ms to spare, which is not a margin anyone should ship on.
This is where the standard and the regulation diverge, and where most compliance documents quietly go wrong. The LoRaWAN 1.0.3 regional parameters define the CN470-510 plan as 96 uplink channels from 470.3 MHz in 200 kHz steps to 489.3 MHz, 48 downlink channels from 500.3 to 509.7 MHz, and state that the band may be used provided EIRP stays below 19.15 dBm and "the transmission never lasts more than 5000 ms". Five seconds. That clause predates the 2019 announcement and was never a licence to exceed it; a device certified against the document rather than the regulation is certified against the wrong document. Newer revisions of the regional parameters restrict the band to small-scale networks and state the one-second limit explicitly, which tells you which revision your supplier is reading.
The same document contains a second operational trap: channel indices 6 to 38 and 45 to 77 are described as mainly used by China Electric Power and are to be disabled where that is true. That is 66 of the 96 uplink channels, and whether they are free depends on the city, not on the datasheet.
What the catalogue gives you
Two very different radios are sold for this band, and the choice between them is a choice about which constraint you want to spend. Three of them, with the numbers that matter here, follow; the rest of the range is on the /products page.
Module | Modulation | Transmit power | Sensitivity | Data rates | Airtime for a 51-byte frame |
|---|---|---|---|---|---|
FSK | 17 dBm, configurable | -117 dBm at 2.4 kbps, -111 at 19.2 kbps, -106 at 76.8 kbps | 2.4 / 19.2 / 76.8 kbps | 170 / 21 / 5 ms | |
S-FSK, quoted with spreading factors | 17 dBm, configurable | -140 dBm at 12 kcps, SF9 | 12 / 24 / 48 / 100 kcps | 308 ms at the SF9 operating point | |
S-FSK, quoted with spreading factors | 36 dBm at 5.5 V | -140 dBm at 12 kcps, SF9; -130 dBm at 48 kcps, SF7 | 12 / 24 / 48 / 100 kcps | 308 ms at the SF9 operating point |
The narrowband module's advantage is arithmetic, not opinion. At 19.2 kbps a 51-byte frame is 21 ms of airtime; the same frame at SF9 is 308 ms. Fifteen times less time on air, and with a transmit current of 650 mA at 5 V against 200 mA at 3.3 V, roughly four and a half times less energy per reading — 13.8 mAs against 61.6 mAs, computed from the two parameter tables and nothing else.
The price is sensitivity. Minus 140 dBm against minus 111 dBm is 29 dB, and in free space 29 dB is 28 times the distance (10^(29/20)). In a basement meter room, that 29 dB is usually the difference between a link that closes and one that does not, because the loss is concrete and steel rather than distance. That is the entire design argument in two sentences: narrowband buys throughput and battery, spread spectrum buys penetration — the same trade the 470 MHz link budget walkthrough puts in decibels, seen here in milliseconds.
The 36 dBm part in the same family is the one to be careful with. It exists because long-range links need it, and its sensitivity is identical to the 17 dBm spread-spectrum unit at the same chip rate — the extra 19 dB of output is bought with 1,950 mA of transmit current and with a regulatory position outside the metering category. Use it where the rules allow it, and do not assume that because it is sold for 470-510 MHz it is legal at 36 dBm in China.
Where designs break
Symptom | What actually happened | Fix |
|---|---|---|
Passes the power test, fails the field test with a spectrum analyser | SF12 left as the default after a range experiment | Cap the spreading factor at SF10 in firmware, and assert airtime in the join path |
36 dBm module shipped into a metering project | Conducted power compared against 50 mW, not e.r.p. | Compare e.r.p.; a 36 dBm part needs a different category or a different market |
Range collapses in one city, fine in the next | Channel mask still at the 96-channel default, transmitting under a local TV carrier | Scan 470-510 MHz at the site and blacklist what is occupied |
Duty-cycle timer implemented, transmitter still non-compliant | The EU 1% budget is not a burst-length rule; 2.3-second frames pass a duty-cycle check | Enforce both: burst duration and hourly share |
Compliance file quotes 5 seconds | Copied from a regional-parameters revision that predates the 2019 announcement | Quote the announcement and the annex, and state the revision you used |
One compound, 400 meters, run to the end
Take the case that started this article and fix it rather than reject it. Four hundred water meters across six buildings, one gateway on the roof of the tallest, a 51-byte reading per meter per hour.
Move the link to SF9 at 125 kHz. Each reading is 308 ms, below the one-second cap with room for the preamble of a retry that starts fresh — and the retry must start fresh, because two transmissions back to back without a gap are one transmission as far as a compliance test is concerned.
Sequential polling on a single channel is 400 x 0.308 s = 123 seconds per round. That is already fast enough for hourly reads, but the gateway is not limited to one channel: after blacklisting the Electric Power and broadcast carriers, assume 30 usable uplink channels. Spread evenly, that is 13 nodes per channel at 4.0 seconds per round, so a five-minute poll cycle uses under 2% of the available time and an hourly cycle uses a fraction of one percent.
Energy per reading is 200 mA for 0.308 s, which is 61.6 mAs, or 0.017 mAh. Twenty-four readings a day is 0.41 mAh per day of transmit energy — about 150 mAh a year, before receive current, which is why the receive duty cycle and not the transmit budget usually sizes the battery. The same schedule on the narrowband module is 0.092 mAh per day, 33 mAh a year, and a link that has to close through two more walls without the 29 dB of spreading gain.
The numbers do not say one technology is better. They say the constraint you have to design against is the one your regulator chose: in China it is burst length and channel agility; in Europe it is the share of the hour; in the United States it is occupancy per channel across a hop set. Pick the radio after you have picked the constraint.
Check it in the field in five steps
- Scan before you commit. Sweep 470-510 MHz in the actual building with the actual gateway antenna. Anything above the noise floor that is not yours becomes a blacklisted channel, and the list differs between a city centre and a village.
- Measure burst duration at the antenna port, not in the log. A coupler and a power meter that can resolve 100 ms is enough. Confirm the worst-case payload, including retries, and confirm that a retry waits rather than continues.
- Verify the listen-before-talk path is active and not user-disableable. In the metering category this is a requirements item, not a feature: a configuration flag that a customer can switch off puts them outside the category.
- Add up the radiated power. Conducted power plus measured antenna gain minus measured cable loss. Cable loss is negative gain, and it is the only term that works in your favour.
- Log the channel mask actually used. Compare it against the 96-channel default of the stack. The default is the single most common cause of a "works in the lab" deployment.
FAQ
Is LoRa legal in China at 470-510 MHz?
Yes, in the civil metering category of the 2019 announcement, which covers networks inside buildings, residential compounds and villages, with one channel at a time, a listen-before-talk mechanism, at most 1 second per transmission and 50 mW e.r.p. The technology is not the issue; the configuration is. A LoRa link at SF9 with a 51-byte frame is comfortably inside the rules, and the same link at SF12 is not.
Do I need SRRC type approval for a 470 MHz module?
Not if the finished device is in one of the eight micro-power categories and meets the technical requirements: the announcement waives the frequency licence, the station licence and the type approval. That is a property of the product, not of the module — a module sold as compliant can still be built into an assembly that is not, for instance by adding an amplifier or a higher-gain antenna.
Does the European 1% duty cycle apply in China?
No. China caps the duration of each transmission and requires listen-before-talk; it does not impose a percentage of time. The practical consequence is that a device designed for EU868 with a duty-cycle timer and no burst-length limiter will pass its European test and fail in China, because the two regulations limit different variables.
Which spreading factor should ship as the default?
SF9 at 125 kHz is the safest default for metering in this band: 308 ms for a 51-byte frame, 199 bytes of legal payload, and 29 dB more sensitivity than a narrowband FSK link at 19.2 kbps. Move up to SF10 only if the site demands it, and treat SF11 and SF12 as unavailable for normal payloads under the one-second rule.



