Sit down in a packed coffee shop at 10 a.m., open a laptop, and the Wi-Fi crawls. The obvious explanation is that the signal is being stretched thin across too many people, like water pressure dropping when everyone flushes at once. It feels intuitive, and it is almost entirely wrong.
The radio waves reaching the laptop are just as strong as they were at 7 a.m. in an empty room. What has changed is not the signal. It is the traffic on the frequency, and the strict rulebook every device has to follow before it is allowed to speak.
Wi-Fi is a half-duplex system running on a shared medium. That is engineer-speak for something very simple: on a given channel, only one device can transmit at a time, and everyone else has to listen and wait. The access point, every phone, every laptop, every smartwatch all take turns on the same slice of the 2.4 GHz or 5 GHz spectrum. There is no separate lane for each customer. There is one lane, and thirty people trying to merge into it.
The polite protocol behind every connection
The mechanism that keeps this from turning into pure chaos is called Carrier Sense Multiple Access with Collision Avoidance, or CSMA/CA. It is baked into the IEEE 802.11 standard that defines Wi-Fi, and it works roughly like a very well-mannered dinner party. Before any device transmits, it listens to the channel. If it hears another device already talking, it waits.
When the channel finally goes quiet, no one jumps in immediately, because two devices starting at once would collide. Instead each waits a short mandatory interval, then picks a random number of extra micro-slots to count down, still listening as it does. Whoever draws the smallest number speaks first. Everyone else freezes their countdown and resumes it at the next opening.
This works beautifully with two or three devices. With thirty, the back-offs stack up. Every device spends most of its time listening, deferring, and recalculating its random wait, and the actual transmission windows shrink. A packet that would take a few milliseconds to send in an empty room might sit in the queue for tens or hundreds of milliseconds before its turn arrives. Multiply that across the thousands of packets a modern web page needs, and the tab that loaded instantly at home now hangs for five seconds on the café’s network.
When two devices talk at once
The “collision avoidance” in the name is a hope, not a guarantee. The listen-first rule makes collisions rare, but it cannot make them impossible, because two devices that both hear silence at the same instant can both start transmitting at the same instant.
It gets worse in a real room. Two laptops at opposite ends of the café can each have a clean line to the router while being completely deaf to each other. Neither hears the other begin, so both transmit, and their signals crash together at the access point. Engineers call this the hidden node problem, and it is exactly the situation a crowded, physically spread-out room produces.
Wi-Fi has an optional fix, a brief request-to-send and clear-to-send handshake that reserves the channel before a big transmission. It works, but it is not free: the handshake itself costs airtime. Every collision costs even more. A garbled frame goes unacknowledged, the sender assumes the worst, widens its random back-off window, and tries again later. A single collision therefore burns more of the shared airtime than a clean transmission would have, which is why congestion tends to feed on itself.
Why the signal-strength icon lies to you
The bars in the corner of a screen measure received signal strength, which is how loudly the access point’s radio is arriving at the antenna. They say nothing about whether the channel is free to use. A phone can show four full bars and still barely load a page, because those bars know nothing about the twenty-nine other devices currently taking turns on the same frequency.
Network engineers watch a more useful number called channel utilisation: the percentage of time the airwaves are already occupied. Once utilisation climbs past roughly 70 or 80 per cent, throughput collapses for everyone, not because any signal degraded but because the queue got too long to clear.
There is a subtler penalty on top of that. Wi-Fi negotiates a data rate with each device based on link quality, so closer and cleaner connections get faster rates and distant or noisier ones get slower rates. On a shared channel, the slowest device drags everyone down, because when its turn comes it holds the channel longer to push the same data through at a lower rate. One person in the corner with a marginal link can measurably tax the airtime for the whole room. This is the documented performance anomaly of 802.11, first analysed by Martin Heusse and colleagues at IEEE INFOCOM in 2003.
Even the idle devices are not silent. Access points broadcast beacon frames several times a second, and phones in pockets fire off probe requests as they scan for familiar networks, whether or not their owners are using them. Each of those little management frames has to take its turn in the same queue as everyone’s actual data.
The thing genuinely running out, then, is not bandwidth in the sense people imagine. It is airtime. The channel is a single conversation that only one device can hold at a time, and in a full room the scarce resource is simply the right to speak.
Interference makes the queue worse
The 2.4 GHz band, in particular, is a crowded neighbourhood. It is shared with Bluetooth headphones, microwave ovens, wireless mice, older baby monitors, and the Wi-Fi networks of every other business on the block. The band offers only three non-overlapping channels in most countries, numbered 1, 6 and 11, so in a dense area the café’s router, the bakery next door, and the apartments upstairs are often broadcasting on top of one another.
Each of those foreign transmissions counts as noise the local devices have to defer to. The polite protocol does not care whether the voice it hears belongs to your network or someone else’s. It hears the channel is busy, so it waits, and the back-off stretches further.
The 5 GHz band, and the newer 6 GHz band opened up for Wi-Fi 6E, help significantly, because they offer many more non-overlapping channels and shorter range, so networks in adjacent buildings interfere less. But inside a single crowded room, on a single access point, the fundamental problem remains: one channel, one speaker at a time.
What Wi-Fi 6 and 7 actually changed
The most recent revisions of the standard were designed less to make Wi-Fi faster in the raw sense and more to make it survive crowds. Wi-Fi 6 introduced a technique borrowed from cellular networks called Orthogonal Frequency-Division Multiple Access, or OFDMA. Instead of one device using the whole channel for one turn, the access point can slice the channel into smaller sub-channels called resource units and hand them to several devices at once. A phone checking messages, a laptop streaming audio, and a tablet loading a page can each be assigned a slice in the same instant. The queue shortens because more people speak at once, quieter and in narrower lanes, but in parallel.
OFDMA has a companion trick called multi-user MIMO. Where OFDMA divides the channel in frequency, MU-MIMO divides it in space, using multiple antennas to aim separate data streams at different devices simultaneously. The two work together: one carves the airtime into frequency slices, the other lets the access point talk to several devices in the same slice.
Wi-Fi 6 also added target wake time, letting battery-powered devices agree with the router on scheduled slots so they stop pinging the network at random moments. Fewer unscheduled wake-ups means fewer devices elbowing into the queue when they have nothing important to say.
Wi-Fi 7 pushes further with multi-link operation, which lets a single device use two bands at once, and it lifts Wi-Fi 6’s limit of one resource unit per device so a busy client can be handed several slices in a single turn.
None of it removes the need to share airtime. It just makes the sharing more efficient, a better traffic controller for the same intersection.
Why this is not the internet’s fault
One last confusion is worth clearing up. When a café’s Wi-Fi feels slow, the bottleneck is almost never the internet connection coming into the building. A modest business fibre or cable line can easily supply several hundred megabits per second, more than enough to keep dozens of laptops happy on paper.
The choke point is the last few metres, from the access point through the air to each device. That is where the CSMA/CA queue lives. The router could be sitting on a gigabit uplink and it would not help, because the packets are stuck waiting for a turn on the radio before they ever reach the wire.
This is worth remembering the next time a video call stutters in a busy hotel lobby or an airport lounge. The infrastructure is probably fine. Somewhere in the room, dozens of devices are listening to the channel, counting down random micro-second waits, and taking their turns to speak in an orderly line that nobody can see. Wi-Fi has not slowed down. It is simply being polite, and politeness, at scale, feels a lot like lag.
For more of the invisible plumbing behind everyday computing, it is worth reading how noise-cancelling headphones cancel sound by generating its mirror image, or how deleting a file only removes the pointer, not the data itself. The pattern is the same: the thing you thought was happening is not, quite, what is happening at all.
