Everything Lives at 2.4

Count the radios in an ordinary house. The thermostat, the doorbell, the two speakers, the printer nobody has used since spring, the lock, the four bulbs, the sensor on the back door, the watch, the earbuds, and whatever the neighbors have on the other side of a wall that stops almost nothing. Then count the ones using 5 GHz, which is the phone, the laptop, and the television if it is recent. Everything else is at 2.4.

The usual explanation is that cheap devices got the cheap radio, and there is something to that, but it has the causation backwards. The radio is cheap because everything uses it, and everything uses it because of a decision made in a treaty document rather than a product meeting. The International Telecommunication Union designates 2400 to 2500 MHz as an industrial, scientific and medical band worldwide, in all three of its regions, and national regulators built their unlicensed rules on top of that.

Worldwide is the word carrying the weight. A company selling a sensor at fifteen dollars a unit cannot afford a separate radio, antenna and certification exercise for North America, Europe and Japan. It can afford one design that ships everywhere. There is no other band on earth where that is true, and once you see that constraint the sameness of the smart home stops looking like laziness and starts looking like the only available answer.

Which sets up the thing I actually find interesting, and it is not a complaint about any manufacturer. The properties that make this band right for one device are the properties that make it wrong for the thousandth, and every product that landed here was individually correct to land here. The ITU's own framing says the rest out loud: equipment in an ISM band has to tolerate interference from the other occupants, because nobody is coordinating them.

One radio and one certification cover every market. This is the decisive one and it is regulatory rather than physical. The band is open to unlicensed low-power use in nearly every country that regulates any, on approximately the same frequencies, which means one hardware design and one set of test reports. Everything else on this list is an engineering preference. This one is the difference between a product existing and not existing.

It gets through the wall. Lower frequencies lose less to distance and to building materials, which is why a sensor in a basement or a lock on an outside door is at 2.4 rather than 5. The gap is modest in free space and considerable through two interior walls and a floor. For a device that has to work in the one place nobody would put an access point, that difference decides whether the product functions at all.

These devices cannot pay for throughput and do not need it. A door sensor sends a few bytes an hour. Zigbee runs at around 250 kbit/s here and that is generous for the job; Bluetooth Low Energy runs 1 Mbit/s in 2 MHz channels at a maximum of 10 mW in its original specification. Range beats throughput for nearly everything in this category, and a device living two years on a coin cell cannot afford what a higher band and a wider channel would cost it.

And then volume made it self-reinforcing. Two decades of everything choosing this band produced transceivers that are mature, tiny, well documented and available from a dozen vendors. A newcomer choosing spectrum today is not comparing physics, it is comparing a part that costs a dollar and has shipped in a billion units against one that has not. The crowding made the parts cheap, and the cheap parts keep adding to the crowding.

Wi-Fi's use of the band runs from 2400 to 2483.5 MHz in most regulatory domains, which is 83.5 MHz in total. Inside it, 802.11 defines 14 channel numbers spaced 5 MHz apart: channel 1 centered at 2412 MHz, up through channel 13 at 2472 MHz, then a 12 MHz jump to channel 14 at 2484 MHz, which only Japan permits and only for the legacy mode. North America permits 1 through 11.

The trap is in the word "channel". The numbers are 5 MHz apart; the signals are not 5 MHz wide. The legacy direct-sequence transmission occupies 22 MHz, and the OFDM transmission used from 802.11g onward occupies about 16.25 MHz inside a nameplate 20 MHz slot. So one Wi-Fi channel is four to five channel numbers wide, and two networks on channels 1 and 3 are not beside each other. They are on top of each other.

Which gives the whole answer in a line of subtraction. Keeping two channels clear of one another takes five numbers of separation, or 25 MHz between centers. Start at 1, add five, add five again: 1, 6 and 11. The next would be 16 and there is no 16. At the wider legacy occupancy, channel 1 runs 2401 to 2423 MHz, channel 6 runs 2426 to 2448, channel 11 runs 2451 to 2473, and 3 MHz separates each pair. A fourth would have to end near 2498, well past the edge of the band.

Three is the North American answer rather than a law of nature, and I would rather say so. Where regulators permit up to channel 13, a 1, 5, 9, 13 plan gives four channels whose modern OFDM signals just abut, at 2402 to 2422, 2422 to 2442, 2442 to 2462 and 2462 to 2482, and it is used. It also overlaps for the older 22 MHz mode, so it buys the fourth channel by assuming nothing legacy is nearby. Either way the band fits three or four networks, and a building with more than that has already lost.

Two networks on the same channel cooperate. This is the piece that surprises people. Wi-Fi listens before it talks: a radio that hears a transmission it can decode waits for it to finish. Two networks sharing channel 6 therefore take turns. It is not free, since you are dividing airtime and everybody is slower, but it is orderly, bounded, and exactly the case the protocol was designed for. Co-channel is the good failure.

Two networks partly overlapping do not, and either way it costs more. A radio on channel 3 cannot decode a transmission on channel 1, so the cooperative mechanism never engages, and what happens instead depends on how loudly the neighbor arrives. Loud, and raw energy trips your radio's channel-busy test, so you lose airtime to a network you cannot take turns with. Quiet, and it does not stop you at all, so your frames arrive damaged, get retransmitted at a slower and more robust rate, and hold the air longer doing it.

And your one choice damages two networks. Which makes it a commons problem rather than a personal one. A network on channel 3 partly overlaps channel 1 and channel 6 at once, so one device leaves the plan and two neighbors get worse. Their equipment then sees a busier channel, moves itself, and lands on somebody else. Automatic channel selection running independently in a dozen units, each optimizing locally, can walk an entire apartment building off 1, 6 and 11.

Bonding two channels here spends two thirds of the band. The standard permits joining two 20 MHz channels into a 40 MHz one, and in this band that consumes two of the three clear channels to serve a single network. In an isolated house it may genuinely be faster. In a building it is the most antisocial setting available, and it is frequently the default, because a wider channel produces a better number on the box.

The low-rate standard underneath Zigbee and Thread, IEEE 802.15.4, defines 16 channels here, numbered 11 to 26, again 5 MHz apart, from 2405 to 2480 MHz. Four are chosen deliberately: 15, 20, 25 and 26, at 2425, 2450, 2475 and 2480 MHz, which are precisely the gap above Wi-Fi channel 1, the gap between 6 and 11, and the space above 11. A two-megahertz signal can live in a three-megahertz crack, and that is the entire design.

Bluetooth took the other approach and hops. Low Energy splits the band into 40 channels of 2 MHz and moves constantly, reasoning that if part of the band is busy at any instant a short packet elsewhere will get through. Even its three advertising channels are placed with the neighbors in mind, at 2402, 2426 and 2480 MHz, at the edges of the Wi-Fi channels rather than their centers. And then there is the occupant that is not communicating at all: a microwave oven radiates a few hundred watts near 2450 MHz, dead center, which is not a coincidence, since the band was designated for heating things before anyone sent data through it.

The contrast worth drawing is Z-Wave, which went sub-gigahertz instead, around 908 MHz in North America and 868 MHz in Europe. It gets a quieter neighborhood and better propagation at lower data rates, and pays with a different radio and a different certification per region, which is the exact cost the 2.4 crowd refused. Meanwhile the standard meant to unify all this, Matter, whose 1.3 specification was published on 8 May 2024, runs over Thread and Wi-Fi. It unifies the application layer. Underneath, it is more traffic in the same spectrum.

Things get slow, not disconnected. Congestion in a listen-before-talk protocol spends itself as waiting and retrying. The sensor still reports, four seconds late. The camera still streams, with a stutter. Nothing generates an error anybody will see and nothing lands in a log, and the complaint that eventually arrives is that the system feels unreliable, which is not a symptom you can put in a ticket.

Airtime is the shared resource, not bandwidth. The number people watch is throughput and the number that matters is time. A device transmitting slowly holds the channel longer than a fast one sending the same bytes, so one weak client at the far end of the house consumes airtime out of all proportion to what it is saying, and everything else on that channel waits behind it. The worst-connected thing on your network sets the ceiling for the rest.

The cheapest device suffers first and complains last. A phone has a good antenna, a real amplifier and a battery that tolerates retries. A twelve-dollar sensor has a trace on a circuit board for an antenna, single-digit milliwatts of transmit power, and a power budget that punishes every retransmission. It fails first in a crowded band and explains itself last, and it is usually the device doing the job you actually cared about.

Give up on throughput in this band, deliberately, as policy. That one decision settles most of the others: pick 1, 6 or 11 and stay there, refuse the 40 MHz width, and stop letting automatic selection wander, because a fixed choice on the plan beats a clever choice off it. The band cannot be won. It can only be shared more or less gracefully, and every setting that promises a bigger number takes airtime from devices that had no alternative.

Then separate the populations. The things that must be at 2.4 should not share a network with laptops and televisions that could be at 5 or 6 GHz or on a cable, because the fast devices have somewhere to go and the slow ones do not. Move everything that can move, and take the wire wherever a wire is possible, which for anything that does not need to be portable is more often than people assume once they stop treating wireless as the default.

And place the radios physically, which sounds too simple to mention until you remember that a microwave oven and a badly sited access point are both solved by moving something three feet. Distance improves every term at once: it lowers what the neighbors hear from you, lowers what you hear from them, and raises the rate your own clients sustain, which hands airtime back to all of them. Spectrum is fixed. Geometry is free.

I want to keep this in proportion. In a detached house with a handful of devices none of it is your problem, the band has ample room, and any channel you pick will work. The arithmetic only bites where density is high, which means apartments, offices, hotels, and anywhere a lot of people made the same reasonable decision within a few hundred feet of one another. If your system works, it works, and there is nothing here worth tuning.

What holds generally is the shape. This is a commons of fixed size with no owner, no admission control, and a rule saying every occupant must tolerate the others, and the tragedy is the ordinary one: every device in it was individually right to be there. There are three clean channels. There have only ever been three, and the number does not move no matter how many things get sold that need one.