Industry Insights
HaLow vs sub-GHz camera links: bandwidth and duty cycle
A 45 KB snapshot takes 18.8 s on 19.2 kbps and 11 ms on 8 MHz HaLow. The arithmetic behind camera links, including the duty-cycle limits most plans miss.
A camera is not a sensor with a lens. A meter sends forty bytes once an hour; a camera sends forty-five kilobytes every time it decides the cat is an intruder. Both get described as "sub-GHz, kilometre range" in the same procurement meeting, and both get quoted against the same radio. That is where projects go wrong.
The two families in this article are genuinely different instruments. One is narrowband frequency-shift keying: 200 kHz of occupied bandwidth, 2.4 to 76.8 kbps, a receiver that hears −116 dBm. The other is Wi-Fi HaLow, IEEE 802.11ah, an OFDM system that runs on 1, 2, 4 or 8 MHz channels and quotes 32.5 Mbps on its best day. Put a camera on either and you will learn which number the vendor was actually selling.
This is the arithmetic for that decision, done with the standard's own rate table and the duty-cycle limits that come with the band, not with brochure numbers.
The two radios are not in the same product category
Narrowband sub-GHz exists to move small payloads a long way on very little power. Our TP5301 RS485 bridge is representative: 470–510 MHz, ≤200 kHz per carrier, 2.4 / 19.2 / 76.8 kbps, 20 ± 1 dBm, and −111 dBm sensitivity at 19.2 kbps. It draws under 0.5 W and is built to sit on a wall in a meter cupboard for a decade. The TP5510 gateway uses the same air interface with 26 ± 1 dBm and −116 dBm at 19.2 kbps, and handles up to 100 terminals.
HaLow is the opposite trade. The TP5608-FD14L20 module and the TP5618-FD13L20 board are both IEEE 802.11ah, Sub-1 GHz with a 900 MHz antenna port, offering 1 / 2 / 4 / 8 MHz channels, 32.5 Mbps at 8 MHz MCS7, and −105 dBm at 1 MHz with MCS10. They operate as AP or STA, with relay, roaming and multicast. The 5618 exposes RJ45, Type-C and a 12 V jack; the 5608 brings the same radio out to GH1.25 serial and Ethernet headers in a 96.3 × 41 × 15 mm module.
Both are legitimate sub-GHz radios. Only one of them can carry video.
Rate, computed from the standard, not from the brochure
The rate table is public: the IEEE 802.11 TGah task group published the amendment that defines it, and every module datasheet reproduces the same grid. Each 1 MHz of channel carries a fixed set of data subcarriers, each subcarrier carries a modulation, and the coding rate decides how much of it survives. With one spatial stream and the short 4 µs guard interval, the numbers look like this:
Channel | MCS0 (BPSK 1/2) | MCS4 (16-QAM 3/4) | MCS7 (64-QAM 5/6) |
|---|---|---|---|
1 MHz | 0.33 Mbps | 2.0 Mbps | 3.34 Mbps |
2 MHz | 0.72 Mbps | 4.33 Mbps | 7.22 Mbps |
4 MHz | 1.5 Mbps | 9.0 Mbps | 15.0 Mbps |
8 MHz | 3.25 Mbps | 19.5 Mbps | 32.5 Mbps |
Two entries deserve attention. The 32.5 Mbps in the TP5608 spec sheet is the bottom-right cell: 8 MHz, MCS7, 4 µs guard interval, one stream. It is a real number and it is also the least robust configuration in the table. The other entry is MCS10, which exists only at 1 MHz: BPSK at rate 1/2 with 2× repetition, 0.17 Mbps, and the −105 dBm sensitivity our spec quotes. That is the mode you actually want at the edge of coverage, and it is roughly two hundred times slower than the headline.

The scaling is almost exactly linear in bandwidth, which is what OFDM does: 234 data subcarriers at 8 MHz against 24 at 1 MHz. Halving the channel halves the rate and buys 3 dB of receiver sensitivity. There is no free lunch hiding in the table.
What a camera actually asks for
Cameras do not have one bitrate; they have three, and procurement usually quotes the wrong one.
Workload | Payload | Average rate per camera |
|---|---|---|
640×480 JPEG snapshot every 5 min (assumption: 45 KB at q70) | 45 KB = 360 kbit | 1.2 kbps |
1080p H.265 event clip, 30 s at 2 Mbps (assumption) | 7.5 MB = 60 Mbit | 0.69 kbps averaged over a day |
1080p live view at 2 Mbps | continuous | 2000 kbps |
The third row is the one that kills narrowband designs, and it is also the row that users assume is the default. A person who has used a Wi-Fi camera expects to open an app and see the yard move. That is 2 Mbps sustained, which is 104 times the entire capacity of a 19.2 kbps link — not 104 times its margin, its total capacity, with nothing else on the channel.
Even the first row is not as cheap as it looks once you count airtime rather than average rate. One 45 KB image is 360 kbit. At 19.2 kbps that is 18.75 seconds of continuous transmission. At 76.8 kbps, 4.7 seconds.

HaLow shortens that dramatically, provided you accept the mode you are using. In the robust 1 MHz MCS10 mode, the same image takes 2.1 seconds — 8.9 times faster than 19.2 kbps, not two hundred times, because the comparison is between the robust modes of each system. Move to 2 MHz MCS4 and it is 83 ms. At 8 MHz MCS7 it is 11 ms, which is the point where a camera stops being a special case and becomes ordinary network traffic.
Duty cycle multiplies whatever you have
In Europe, sub-GHz bands do not come with unrestricted airtime. ETSI EN 300 220-2 lists the limits band by band, and they are not generous:
Band | Max e.r.p. | Channel access | Restriction |
|---|---|---|---|
863–865 MHz | 25 mW | ≤0.1 % duty cycle or polite spectrum access | Audio and video limited to 300 kHz |
865–868 MHz | 25 mW | ≤1 % duty cycle or polite spectrum access | — |
868.0–868.6 MHz | 25 mW | ≤1 % duty cycle or polite spectrum access | — |
868.7–869.2 MHz | 25 mW | ≤0.1 % duty cycle or polite spectrum access | — |
869.4–869.65 MHz | 500 mW | ≤10 % duty cycle or polite spectrum access | — |
869.7–870 MHz | 5 mW | No requirement | Audio and video excluded |
Those last two columns matter more than the power column. The same document warns in its scope that some harmonised bands carry "usage restrictions such as the exclusion of video or audio use", and the 869.7–870 MHz row says it outright. A camera product designed for that band is not a product in the EU, whatever the radio can do.
Duty cycle is a multiplier applied after everything else, and it applies to fast radios too:

At 1 % duty cycle, a 32.5 Mbps HaLow link averages 325 kbps. That is still 17 times the continuous capacity of a 19.2 kbps narrowband link, so the ranking survives — but the absolute number is what a camera plan has to be built on. At 0.1 %, HaLow averages 32.5 kbps, which is only 1.7 times the narrowband link's continuous rate and far too little for live video.
This is the counter-intuitive part of the comparison: the fast radio loses more when duty-cycled, because duty cycle is a percentage of a large number that was never available continuously anyway. If your plan assumes 32.5 Mbps and the band gives you 1 %, you are planning against a number that exists for ten milliseconds in every second.
The link budget moves the other way
Sensitivity and bandwidth are not independent. A receiver listening to 200 kHz collects less noise than one listening to 1 MHz; the noise floor difference alone is 10·log10(1000/200) = 7 dB. Our narrowband gateway quotes −116 dBm at 19.2 kbps over 200 kHz. The HaLow module quotes −105 dBm at 1 MHz with MCS10. The narrowband radio is 11 dB more sensitive while listening to a fifth of the spectrum.
Add the frequency difference. 900 MHz suffers 20·log10(900/470) = 5.6 dB more free-space loss than 470 MHz at the same distance. With the same conducted transmit power at both ends, the narrowband 470 MHz link therefore closes at 16.6 dB more path loss — about 6.7 times the distance of the 1 MHz HaLow link, before either radio hits an obstacle.
That comparison carries an assumption you should check: our HaLow module datasheet does not state RF output power, while the TP5510 gateway is specified at 26 ± 1 dBm. If the HaLow side runs 20 dBm, the gap widens by another 6 dB to 22.6 dB, or roughly 13.5 times the range. Ask for the conducted power figure before you accept anyone's range claim, ours included.
The product table, with the numbers that matter
Device | Air interface | Robust mode | Fast mode | Sensitivity | Role |
|---|---|---|---|---|---|
TPUNB, ≤200 kHz | 2.4 kbps | 76.8 kbps | −111 dBm @19.2 k | RS485 bridge | |
TPUNB, ≤200 kHz | 2.4 kbps | 76.8 kbps | −116 dBm @19.2 k | Gateway, 100 terminals | |
TPUNB, 500 channels | 1.2 kbps | 76.8 kbps | −116 dBm | Transparent DTU | |
802.11ah | 1 MHz MCS10 | 8 MHz MCS7 | −105 dBm @1 MHz MCS10 | HaLow module | |
802.11ah | 1 MHz MCS10 | 8 MHz MCS7 | −105 dBm @1 MHz MCS10 | HaLow board, RJ45 |
The TP5603 is worth a note in a camera discussion because its specification includes a maximum payload of 800 bytes per packet. A 45 KB image is 57 packets before any protocol overhead. On a transparent link that is fine; on a polling network it becomes 57 opportunities for a retry.
Worked example: eight cameras at one gate
Take a site with eight cameras, each producing twenty 30-second clips a day at 2 Mbps. That is 9.6 Gbit of video per day, or 111 kbps averaged over 24 hours. It sounds small enough for a narrowband link until you divide by the right number.
Radio | Daily airtime at 60 % MAC efficiency | Share of the day |
|---|---|---|
Sub-GHz 19.2 kbps | 231 h | 965 % — impossible |
HaLow 1 MHz MCS10 (0.17 Mbps) | 26.1 h | 109 % — still impossible |
HaLow 1 MHz MCS0 (0.33 Mbps) | 13.5 h | 56 % — works, no margin |
HaLow 2 MHz MCS4 (4.33 Mbps) | 1.03 h | 4.3 % — comfortable |
HaLow 8 MHz MCS7 (32.5 Mbps) | 0.14 h | 0.6 % — video-grade |
The 60 % MAC efficiency is an assumption, not a measurement: 802.11 aggregates frames, but management traffic, ACKs and contention all take a share. In practice 55–70 % is the range to expect for bulk transfers, and it drops sharply when many stations contend.
Two conclusions fall out. First, 1 MHz HaLow is not a video link; it is a snapshot and telemetry link that happens to be faster than narrowband. Second, the honest entry point for clips is 2 MHz at a mid MCS, which is roughly 225 times the capacity of 19.2 kbps and uses a quarter of the spectrum of the 8 MHz configuration.
Now put the same eight cameras on live view and the answer changes again. 2 Mbps of payload needs 3.33 Mbps of PHY at 60 % MAC efficiency, which means 2 MHz MCS4 (4.33 Mbps) is the floor for one camera. Eight simultaneous live streams need 26.7 Mbps of PHY, which is 8 MHz MCS7 territory with nothing left over. This is why HaLow camera products ship with a stated number of concurrent streams rather than "up to N cameras".
Failure modes that only show up after install
The link is fine, the duty cycle is not. Commissioning happens on a bench with no regulator. A design that runs at 100 % duty in the lab will be illegal, or throttled by its own firmware, in the band it ships into. Check the band table before the BOM.
The robust mode was never measured. A 32.5 Mbps figure and a 1 MHz MCS10 figure describe links that differ by 190× in rate. Field range tests are usually run at the highest MCS, which fails first. Test at the MCS you will actually deploy at the cell edge.
Contention replaces the rate table. 802.11ah is still CSMA. Twenty cameras on one AP do not each get a twenty-fifth of the channel; they get a collapse in aggregate throughput and a long tail of retries. Restricted Access Window and Target Wake Time exist precisely for this, and they are configuration, not automatic behaviour.
Antenna band mismatch. A 900 MHz antenna on a 470 MHz link, or the reverse, costs far more than the 5.6 dB the path-loss formula predicts, because the mismatch loss lands on top of it. Both HaLow devices here specify a 900 MHz antenna on an SMA female connector; the TPUNB devices specify their own.
Half-duplex turnaround on narrowband. At 76.8 kbps with 800-byte packets, the packet on air lasts about 83 ms. Turnaround, preamble and ACK can add 30–50 % to that on a transparent link. Budget for it or the throughput you compute will never appear.
How to verify a link before you buy two hundred
Bench tests do not survive contact with a fence line, but they do catch the expensive mistakes. Four measurements, in this order:
# 1. Baseline UDP throughput at each width/MCS you plan to ship, 60 s each
iperf3 -c 192.168.1.20 -u -b 20M -t 60 -l 1400
# 2. The payload that actually matters: 45 KB objects, 200 times, report the median
for i in $(seq 200); do
curl -s -o /dev/null -w "%{time_total}\n" http://192.168.1.20/snap.jpg
done | sort -n | awk '{a[NR]=$1} END {print "median", a[int(NR/2)]}'
# 3. Packet loss under load, at the distance you will actually deploy
ping -c 500 -i 0.2 -s 1200 192.168.1.20 | tail -3
# 4. Re-run 1 and 3 with all cameras powered, not oneRun the same four at the far end of the site, with the antenna installed where it will live, not on a tripod at head height. Compare the median 45 KB transfer time against the arithmetic in this article; if the measured value is more than about 1.8× the computed airtime, the difference is retries or contention, and no amount of antenna gain will fix a contention problem.
For the narrowband side, the equivalent check is simpler: send the real frame your application will send, not a ping. A Modbus read of ten registers and a 45 KB image are different worlds, and only one of them is what the network will carry every day.
FAQ
Can sub-GHz narrowband carry 1080p video?
Not at any realistic quality. 2 Mbps payload against 19.2 kbps of channel capacity is a ratio of 104:1. Even 76.8 kbps leaves a factor of 26. Sub-GHz narrowband carries snapshots, event flags and scalar telemetry; video needs a wideband link.
Is Wi-Fi HaLow the same thing as LoRa?
No. LoRa is a chirp-spread-spectrum PHY aimed at very low data rates and long range. 802.11ah is OFDM with a full 802.11 MAC, which is why it can carry IP video and why it consumes more spectrum and more power. They compete for the same sub-GHz spectrum but not for the same applications. The regulatory arithmetic for the 470–510 MHz bands is covered in our narrowband versus LoRa comparison.
Does a wider channel always give more range?
The opposite. Doubling the channel width doubles the rate and adds 3 dB of noise, which costs about 30 % of the range in free space. Wider channels buy throughput; narrower channels buy distance. Choose per camera, not per site.
Which HaLow mode should a camera product ship with?
1 MHz MCS10 for battery cameras that send stills, 2 MHz MCS4 for event clips, 8 MHz MCS7 only where a mains-powered AP serves a small number of streams. And check the band's duty-cycle limit first — in a 1 % band, 8 MHz MCS7 averages 325 kbps, which changes the answer completely.
If you are sizing a link for a specific site, send us the camera count, clip length and the band you intend to use, and we will run the same arithmetic against the HaLow modules or the TPUNB gateway rather than against a brochure.



