wireless communication
802.11ah for industrial cameras: what one channel carries
A 1 MHz HaLow channel at 1 km moves 150 kbps, or one 300 kB still every 16 seconds. Here is the arithmetic for camera counts, AP plans and mast heights.
An industrial camera is a modest load on paper: 1080p H.264 at 2 Mbps, a still JPEG of 300 kB, a firmware package of half a megabyte. Put forty of them around a logistics yard and the arithmetic changes. Twenty live streams is 40 Mbps of sustained load on one access point, and the access point that carries it stands at the gatehouse while the farthest camera is 900 metres away.
That distance is where 802.11ah, sold as Wi-Fi HaLow, enters the conversation. It is Wi-Fi's answer for sub-1 GHz spectrum: OFDM with channels from 1 to 16 MHz, native IP, WPA3, and power-save machinery (Target Wake Time, Restricted Access Window) built for nodes that sleep. The catalogue part used throughout this article, the TP5608-FD14L20, publishes 32.5 Mbps at 8 MHz with MCS7 and -105 dBm at a 1 MHz PPDU with MCS10.
Those two numbers describe two different radios inside the same box, and that is the design problem. What a channel carries at 1 km is not what it carries at 200 m; the difference is a factor of a hundred in throughput, and a camera plan that ignores it fails at commissioning rather than in the datasheet review.
The standard in one page
IEEE 802.11ah is the sub-1 GHz amendment to 802.11, developed by IEEE 802.11 Task Group AH, whose sponsor ballot on draft 9.0 closed in September 2016 with a 95% approval ratio. It is the most unlike-LAN member of the family: stations are grouped to cut contention, relays extend reach, and wake/doze schedules replace the always-listening behaviour of office Wi-Fi. The PHY is 802.11a/g downclocked, with 1 MHz as the narrowest channel and 16 MHz as the widest.
The reason to care is spectrum. Sub-1 GHz travels further through walls and vegetation, and a narrower channel concentrates the same transmit power into less noise bandwidth. Wi-Fi Alliance puts the combination at 8 to 12 dB over 2.4 GHz using 20 MHz channels; add the beacon advantage, and the figure reaches roughly 20 dB, or about ten times the range (Wi-Fi CERTIFIED HaLow technology overview, November 2021).
Where that spectrum sits depends entirely on the regulator:
Market | Spectrum available to 802.11ah | Practical channel count |
|---|---|---|
United States | 902-928 MHz | 26 MHz of spectrum, so up to 26 channels of 1 MHz |
Australia / New Zealand | 915-928 MHz | 13 MHz, so up to 13 channels of 1 MHz |
Europe | 7 MHz split across the 800 and 900 MHz bands | at most seven 1 MHz channels, in non-contiguous slivers |
The European row is not a rounding error. A technology that offers 26 one-megahertz channels in the United States has seven megahertz to work with in Europe, and those seven are not contiguous. That is a hard cap on how many access points can coexist in one yard, and it is the first thing to check before a design assumes an American channel plan. At 1 km, Wi-Fi Alliance's own example is a 1 MHz channel in MCS10 with data duplication: 300 kbps raw, 150 kbps effective, the redundancy spent on link margin rather than throughput. Plan with 150 kbps and you will not be surprised; plan with 32.5 Mbps and you will be.
Do the airtime arithmetic before the topology
Every camera decision reduces to one division:
airtime_seconds = payload_bits / (phy_rate x mac_efficiency)
payload_bits = payload_kB x 8192The 8 MHz mode is specified at 32.5 Mbps. No 802.11 link delivers its PHY rate to the application; management frames, contention, acknowledgements and inter-frame gaps take a share. This article assumes 55% MAC efficiency, which is an assumption and the first number to replace with a measurement on your own hardware. That gives 17.9 Mbps of usable capacity per 8 MHz access point.

Payload | at 150 kbps (1 MHz, 1 km) | at 1.5 Mbps (mid range) | at 17.9 Mbps (8 MHz usable) |
|---|---|---|---|
300 kB JPEG still | 16.4 s | 1.64 s | 0.14 s |
500 kB JPEG still | 27.3 s | 2.73 s | 0.23 s |
512 kB firmware | 28.0 s | 2.80 s | 0.24 s |
2 Mbps live stream, one second | 13.3 s | 1.33 s | 0.11 s |
Read the last row as a ratio rather than a time: a 2 Mbps stream needs 13 times the capacity of a 1 MHz link at 1 km. No scheduling trick recovers a factor of 13. The picture that survives at that distance is a still every sixteen seconds, which is enough for a level crossing, a gate or a tank farm and nowhere near enough for a person walking past.
The 300 kB figure is an assumption too, chosen because it is a common 1080p JPEG at moderate quality. A 4K still at 1.2 MB takes about 66 s per image at 1 km, and a 25-frame burst of 300 kB stills takes close to seven minutes.
Two radios in one catalogue, and what the gap means
The two HaLow parts on this site are the same radio in two enclosures. The TP5608-FD14L20 is a 96.3 x 41 x 15 mm board-level device with a GH1.25 serial port, a GH1.25 Ethernet port and an SMA antenna; the TP5618-FD13L20 is an 80 x 80 x 30 mm box with a real RJ45, a Type-C port and a 12 V jack. Both publish the same figures: 1 / 2 / 4 / 8 MHz channels, 32.5 Mbps at 8 MHz MCS7, -105 dBm at 1 MHz MCS10, AP and STA modes with relay, roaming and multicast, 50 mA at 12 V, -20 to +70 deg C.
Set them next to the narrowband gateway family, and the trade becomes visible in two rows of a table:
TP5608 / TP5618 (802.11ah) | |||
|---|---|---|---|
Radio | 802.11ah OFDM, sub-1 GHz, 900 MHz antenna | 470-510 MHz, S-FSK | 470-510 MHz, S-FSK |
Channel | 1 / 2 / 4 / 8 MHz | 200 kHz single carrier | 200 kHz single carrier |
Rate | 32.5 Mbps at 8 MHz MCS7 | 2.4 / 19.2 / 76.8 kbps | 2.4 / 19.2 / 76.8 kbps |
Sensitivity | -105 dBm at 1 MHz, MCS10 | -121 dBm at 2.4 kbps | -121 dBm at 2.4 kbps |
Backhaul | Ethernet on the device | 4G | Ethernet |
Declared range | not stated on the specification | 1.5 km line of sight | 1.5 km line of sight |
Host interface | Ethernet plus serial, or RJ45 plus Type-C | star-network gateway | star-network gateway |
Sixteen decibels of receiver sensitivity is what the narrowband radio keeps; a factor of 62 in effective data rate is what it gives up (150 kbps effective against 2.4 kbps). That is the trade in one line: an image is worth the sensitivity, a meter reading is not. A water meter that sends forty bytes a day should not be on a 32.5 Mbps PHY, and a camera that sends 300 kB frames should not be on a 2.4 kbps one, where the same image would take seventeen minutes.
One integration detail that the summary hides: on the TP5608 the "Ethernet" is a four-pin GH1.25 harness, not an RJ45. If the camera sits in a metal enclosure, the harness, the gland and the antenna are part of your bill of materials, and the SMA connector means the antenna choice is yours as well.
Where a 4G CPE still wins
The honest comparison for a camera deployment is not HaLow against narrowband; it is HaLow against a 4G router with a SIM in it.
Criterion | 802.11ah AP | 4G CPE | Which one is cheaper to run |
|---|---|---|---|
Coverage at a site with no network | you build it, once | wherever the operator covers, usually everywhere | CPE for a one-off site |
Cameras per node | about 6 live at 2 Mbps, or 27 at one still per five seconds | dozens, limited by the uplink plan | CPE |
Recurring cost | none beyond power | data plan, per site, forever | HaLow |
Addressability | cameras are on your LAN | behind carrier NAT unless you buy a fixed IP or APN | HaLow |
Latency | queueing, milliseconds at 8 MHz | tens of milliseconds, plus jitter | CPE |
Power | 0.6 W published at 12 V | watts, plus a mains supply | HaLow |
Failure mode | your RF plan | operator coverage, SIM expiry, plan changes | neither |
The CPE wins on time to first image and on camera count; it loses on money after year one and on the fact that a camera behind carrier NAT cannot be reached from the plant network without a tunnel. HaLow wins when the site already has power and masts, when the video never leaves the site, and when the deployment has to survive a decade without a subscription. It loses when the site has no power at the camera, because 0.6 W at 12 V still needs a cable or a solar budget.
Failure modes that survive the datasheet review
The regulatory sliver. The product specification says "sub-1 GHz, 900 MHz antenna". That is not a statement about which 900 MHz. Confirm the certification of the exact model for the country before designing a mast layout, because a 26-channel plan in the US becomes a seven-megahertz plan in Europe, and the channel plan is the first thing that cannot be fixed in firmware.
The beacon sets the cell edge, not the data rate. Wi-Fi Alliance makes the point explicitly: beacons go out at MCS0, 300 kbps in a 1 MHz channel, and they are what limits range. A link that passes a throughput test at 900 m can still drop a camera that fails to hear the beacon. Test association at the far corner, not only ping.

The Fresnel zone is nine metres wide. At 1 km, midway, with 900 MHz, the first Fresnel radius works out at 17.32 x sqrt(0.25 / 0.9) = 9.1 m. A mast that clears the ground by three metres still has the path half-blocked by a container stack. At 2.4 GHz the same geometry needs only 5.6 m, which is one reason office Wi-Fi habits do not transfer.
The antenna is 83 mm, not 31 mm. A quarter wave at 900 MHz is 83 mm; at 2.4 GHz it is 31 mm. That changes the enclosure, the gland and the wind load, and it is also why 900 MHz cable loss is lower for the same cable — a detail worth checking against the cable datasheet rather than assuming.
Power save against motion. A camera that sleeps is a camera that misses the event. Target Wake Time and the Restricted Access Window are what make battery nodes possible, but a wake interval of ten seconds is a ten-second blind spot. Either the camera holds the link awake, or the trigger has to come from something else on the same node.
Contention in a yard full of metal. One-megahertz channels are narrow; a hidden node behind a container stack can transmit into the same slot as a visible one. The Restricted Access Window exists for this, and so does the relay mode the specifications list. Both need a channel plan, not defaults.
A worked example: forty cameras around one yard
Take a logistics yard with four gate cameras that stream live at 2 Mbps and thirty-six perimeter cameras that send one 300 kB still every five seconds. Stills at that cadence are 480 kbps each. On one 8 MHz access point with 17.9 Mbps usable, the load is 4 x 2.0 + 36 x 0.48 = 25.3 Mbps, or 141% of capacity. The design fails on paper before anyone climbs a mast.

Split it by function instead of by geography:
Access point | Load | Utilisation | Comment |
|---|---|---|---|
AP1 at the gatehouse, 8 MHz | 4 live x 2 Mbps + 8 stills x 0.48 = 11.8 Mbps | 66% | gates keep low latency |
AP2 on the north mast, 8 MHz | 28 stills x 0.48 = 13.4 Mbps | 75% | at the planning ceiling |
AP3, spare channel | 0 | 0% | absorbs a fifth gate camera or a firmware campaign |
Two access points carry the yard, and the third channel exists so that a one-off firmware push does not contend with the cameras. The 512 kB image takes 0.24 s on an idle 8 MHz link; at 85% utilisation, an M/M/1 approximation puts the same transfer at 0.24 / (1 - 0.85) = 1.6 s, which is still inside a two-second gate deadline but no longer inside a one-second one. The queueing figure is an estimate, not a measurement.
Now move the north camera to 1 km with no mast in between, and the same 300 kB still takes 16 s on a 1 MHz channel. That is the point where the design changes shape: either the camera moves closer, or it stops being a camera and becomes an event detector with a still attached.
Six measurements before ordering two hundred nodes
- Association at the worst point, not throughput at the best one: walk the far corner with a client and record RSSI and retry counts for ten minutes.
- Per-frame airtime, captured passively at the access point: it tells you the real MAC efficiency instead of the 55% assumption used here.
- One still every five seconds from every camera at once: the worst case is not the average, and the average is what a single-camera test measures.
- Beacon reception at the cell edge, using the same antenna and mast height as production, with the camera inside its enclosure.
- Channel plan against the certificate for the destination country, including the channel width you intend to ship with.
- Power and thermal at 12 V, since 50 mA is a published figure at room temperature and the enclosure will not be at room temperature.
If the HaLow interface lands on a Linux host rather than inside a vendor bridge, the first three are cheap to instrument:
# association quality at the far camera
iw dev wlan0 link
iw dev wlan0 station dump | grep -E 'signal|tx bitrate|tx retries'
# worst-case burst: 1500-byte frames, one camera at a time, then all at once
ping -s 1472 -c 200 -i 0.2 192.168.10.31
# capture ten minutes at the AP to compute real airtime
tcpdump -i wlan0 -w /tmp/halow-10min.pcap -G 600 -W 1If the device bridges traffic to a host instead of exposing the radio interface, the same evidence comes from the access point's station table and from a packet capture on the wired side; the numbers are still the numbers.
FAQ
Is Wi-Fi HaLow just Wi-Fi moved to 900 MHz?
No. The PHY is downclocked OFDM with a 1 MHz minimum channel, and the MAC adds grouping, relay, Target Wake Time and the Restricted Access Window. It shares the 802.11 frame format and WPA3, not the 20 MHz channel plan or the timing of 2.4 GHz Wi-Fi.
Can I run four live cameras 1 km from the access point?
Not on one 1 MHz channel. At 150 kbps effective, a 2 Mbps stream is thirteen times over budget. Four cameras at 1 km means four channels, and Europe has seven megahertz in total, so those four would consume most of the band.
Does 802.11ah replace a 4G CPE?
It replaces the subscription, not the coverage. On a site with power and masts it is cheaper after the first year and puts the cameras on your own LAN. Where there is no site network and no time, a CPE gets the first image in an afternoon.
What about ONVIF and RTSP?
They run above the link and do not care which radio carries them; what they care about is multicast handling and latency. The catalogue parts list multicast and relay support, but protocol conformance is a host-side question, so test it with the recorder you intend to use rather than assuming it from the radio specification.
How does this compare with the 470 MHz parts already on the site?
The narrowband gateway reaches 1.5 km at 2.4 to 76.8 kbps with -121 dBm sensitivity, which is the right tool for meters and the wrong one for images: 300 kB takes 32 s even at 76.8 kbps. The two families are complementary, and a mixed site usually runs both, as the product catalogue shows. If you are sizing a firmware campaign over the narrowband link instead, the arithmetic is in this article on OTA at 76.8 kbps.



