Wi-Fi channel congestion

Why does every router shouting a little louder leave every network in the building slower?

Your flat's Wi-Fi is slow. You buy a router with a bigger transmit power and better antennas, and it helps — for a while. So does your neighbour, and the one below, and the landlord who fits access points on every floor. A year later everyone in the building has better hardware and everyone's connection is worse than it was.

Each individual decision was correct. Each person got a genuine improvement at the moment they made it. And the aggregate is worse for all of them. That combination is strange enough to deserve an explanation, and the explanation is not "the routers are badly made".

Begin with what is actually being shared. Two networks on the same channel are not using separate wires; they are using the same block of radio frequency in the same air. So what stops their transmissions from colliding constantly?

The answer is a rule every Wi-Fi device follows: listen first, and if the channel is busy, wait. That is carrier sense with collision avoidance, and it means co-channel networks do not really interfere — they take turns. The medium is divided in time, and the resource being divided up is airtime.

Hold on to that word, because it decides everything that follows. Ask: does raising your transmit power create more airtime? It does not. The channel still has one hundred percent of a second per second, no matter how loudly anyone shouts.

So what does the extra power buy? Two things. Your own signal reaches your devices at a higher rate, so each of your frames takes less airtime to send — that is the real improvement you felt. And your signal now reaches further, so more distant networks can hear you, which means they now defer to you when they otherwise would have transmitted freely. You did not make the pie bigger. You took a larger slice, and you took it from people who never agreed to give it.

Now let everyone do it. Every network is heard by more networks; every network defers more often; the number of devices contending for one channel's airtime grows while the airtime does not. Each person's gain came out of the commons, and once everyone has claimed their share the gains cancel, leaving only the extra contention behind. Does that sound familiar? It is the structure of a common-pool resource problem: individually rational, collectively self-defeating.

There is a second, nastier layer. Deferring politely requires that you can decode the other network's transmission well enough to recognise it. Two networks on partially overlapping channels — say 2.4 GHz channels 1 and 3, where the twenty-megahertz-wide transmissions share spectrum but are not aligned — often cannot decode each other. Then the neighbour's energy arrives not as a signal to defer to but as raw noise, corrupting frames mid-flight. This is why the 2.4 GHz band's usable arrangement is channels 1, 6 and 11: co-channel neighbours share by taking turns, whereas overlapping neighbours simply damage each other. Sharing a channel with someone is genuinely better than half-sharing one.

Now a third effect, which is the one that surprises people. When a distant, weakly-received device transmits, it falls back to a slow, robust modulation. A slow device holds the medium for far longer to send the same data. Because the medium is shared in time, everyone else waits behind it, and Heusse and colleagues documented in 2003 that a single slow station drags the throughput of fast stations down towards its own. So the neighbour whose signal you barely hear is not a small problem in proportion to their signal. Their weakness is exactly what makes them expensive.

Put the three together and the paradox dissolves. More power, wider channels, more access points — each is a way of claiming airtime that used to be someone else's, and none of them manufactures any. The honest caveat is that this is not the whole story of a slow network: your own backhaul, your device count, interference from non-Wi-Fi sources and simple distance all matter, and diagnosing which one you have requires measuring airtime utilisation rather than guessing.

The misconception worth killing is that Wi-Fi congestion is mainly about signal strength. Beyond the point where your link is fast and stable, more power gains you almost nothing and costs your neighbours something. The variable that matters is what fraction of the channel's time is already spoken for.

That reframes what a fix looks like. Anything that reduces airtime consumed is a real improvement: putting an access point closer to its clients, so frames go out at high rates and finish quickly; using the 5 GHz or 6 GHz bands, which have many more non-overlapping channels; reducing beacon and management overhead; and, in the 2.4 GHz band, staying on 20 MHz channels rather than doubling width and colliding with twice as many neighbours. Anything that merely reaches further is a transfer, not a gain.

And the collective-action shape suggests why this is not solved by good intentions. The person who turns their power down bears the cost immediately and gets a benefit that is spread over everyone in the building. That is precisely the payoff structure that produces the wrong outcome without coordination — which is why the workable answers are regulatory power limits, channel plans agreed across a building, and protocol features that make deferral cheaper, rather than appeals to neighbourly restraint.

• Why 5 GHz and 6 GHz change the arithmetic, and what they cost in range.
• Airtime fairness scheduling, and how it protects fast clients from slow ones.
• How building-wide channel plans are actually designed and enforced.

Every term the collection defines is gathered in the glossary.