Wireless and cellular media

Before you read. A laptop shows four bars and the word connected. Pages do not load, a file copy runs at a fraction of what it should, and moving two metres makes it better or worse for no reason anybody can see.

Nothing is broken. The access point is healthy, the cable behind it is fine, and the wired machine next to it is perfectly happy.

What is different about a link with no cable?

Everything in this track so far has assumed a wire. Two ends, one medium each, and a frame that either arrives or does not. Wireless breaks three of the assumptions that made the earlier topics simple, and almost every confusing thing about it traces back to one of those three.

Some words you will need

802.11

The IEEE family of wireless LAN standards. A family, not one document, and the exam treats it as a category.

shared medium

One channel that every device in range takes turns using. The air is always one.

half duplex

Only one end may transmit at a time. Not a configuration on wireless, a property of it.

CSMA/CA

Carrier sense multiple access with collision avoidance. Listen first, and try to avoid collisions rather than detect them.

attenuation

Signal weakening with distance, and through anything it passes through.

signal to noise ratio

How far the signal stands above the background. The number that actually predicts whether a link works.

What breaks without this

Every wireless fault gets diagnosed as the wrong thing. Connected and working are different states, and a client that reports a strong signal can be unable to pass traffic for reasons no client-side indicator shows.

Capacity gets promised that the air cannot deliver. A rated speed is what one device might achieve alone, and dividing it by the number of clients is closer to the truth than any datasheet.

Cellular and satellite get treated as slower wireless. They have different properties, particularly in latency, and one of them has a floor no engineering can move.

What a radio link actually shares

A cable gives each link its own medium. Two machines on two cables into a switch never contend for anything, which is why topic 18 could talk about full duplex as the normal case and half duplex as a fault.

The air gives that up. Every device using the same channel within range of each other shares one medium, and the rules for sharing it are the first thing that makes wireless different.

So the air is half duplex, always. Not because somebody configured it, and not as a fallback. A radio that is transmitting cannot usefully listen on the same frequency at the same time, so an access point and a client take turns, and so does every other client. That single fact is behind more wireless behaviour than anything else on this page.

The second consequence is that collisions cannot be detected the way Ethernet detected them. A transmitting station cannot hear another station transmitting, because its own signal drowns everything else out. Wired Ethernet used CSMA/CD, where the D is detection. Wireless uses CSMA/CA, where the A is avoidance: a station listens before transmitting, waits a random interval, and where necessary asks permission first. Avoidance is more expensive than detection, and the overhead is why the number on the box was never achievable.

If you already work on networks: the hidden node, and why two clients that cannot hear each other are the worst case

Carrier sense assumes that a station which listens before transmitting can hear everything it might collide with. On a wire that holds. In the air it frequently does not, and the failure has a name.

Picture an access point with a client on either side of it, far enough apart that neither can hear the other, though both hear the access point clearly. Client A listens, hears nothing, and transmits. Client B listens, also hears nothing, because A is out of range, and transmits too. Both frames arrive at the access point at once and both are destroyed. Neither client has any way of knowing why, so both retry, and under load the pair can spend more time colliding than communicating.

That is the hidden node problem, and its distinctive quality is that each client sees an excellent signal and terrible throughput. Every client-side diagnostic says the link is fine.

The mechanism 802.11 provides for it is request to send and clear to send. A station asks the access point for the medium, the access point answers with a grant that every station in its range can hear, and the stations that cannot hear each other both hear the grant and stay quiet. It works, and it costs two extra frames per transmission, which is why it is usually left off and enabled only where the problem has been diagnosed.

RFC 3819 is worth knowing about here. It is advice to people designing subnetwork layers beneath IP, it is free, and its sections on link characteristics explain why a link that loses packets for reasons unrelated to congestion is difficult for TCP specifically. Topic 09 established that TCP reads loss as congestion and collapses its window. On a wire, loss usually is congestion. In the air it usually is not, and the two disagreeing is a large part of why wireless throughput is so uneven.

One conversation at a time

Because the medium is shared and half duplex, the capacity of a channel is divided among everyone using it, and the division is worse than a straight split.

Each transmission carries overhead: the listening interval before it, the acknowledgement after it, and the retries when something goes wrong. Add clients and the overhead grows faster than the data does, so twenty clients on one access point do not each get a twentieth. They get rather less.

There is a second effect that surprises people, and it is the one worth carrying into a design conversation. A slow client slows everybody. A device at the edge of the coverage area transmits at a low data rate, because a weak signal forces a more robust and slower encoding. While it is transmitting, it holds the medium, and a frame that would take one unit of time at full rate takes many at the low one. Everybody else waits.

So one laptop in a far corner can measurably reduce throughput for a room full of people sitting next to the access point, and nothing on any of their screens will explain why.

If you already size wireless: why the slowest client sets the cost for everybody

The overhead described above is per transmission rather than per byte, which has a consequence that decides how a cell behaves under load.

Airtime is what a cell shares, and the airtime a client consumes is its data divided by the rate it is transmitting at. A client that has fallen to a low rate therefore occupies the medium for many times longer to send the same amount, and everybody else waits through all of it. One distant laptop can take most of a cell while sending almost nothing, and the clients sitting next to the access point find their throughput has collapsed without their own conditions changing at all.

That is why the client worth investigating in a slow cell is the one with the lowest rate rather than the one with the most traffic, and why a monitoring view ranked by bytes finds the wrong machine. Topic 72 does the arithmetic on a specific case.

It is also the argument against pushing coverage as far as it will go. Extending a cell to reach a distant corner adds a client that will connect at a low rate and spend airtime the rest of the cell would otherwise have used. Smaller cells with clients on higher rates carry more in total than one large cell that reaches everybody, which is the opposite of the instinct that says a stronger signal is better.

Signal strength is not throughput

Now the question at the top of this page. The client shows four bars, which is a measure of received signal strength, and it says nothing about whether the link works.

What matters is the signal to noise ratio, meaning how far the signal stands above whatever else is on that frequency. A strong signal in a noisy environment performs worse than a moderate signal in a quiet one, and the bars measure only the first half of that fraction.

The sources of noise are the reason the next topic exists. Other access points on the same channel. Neighbouring networks bleeding across channels. Microwave ovens, which occupy part of the 2.4 GHz band and are not being rude, they are operating in a band deliberately set aside for equipment that is not communications. Bluetooth devices hopping across the same band.

And moving two metres changes it because radio does not travel in a straight line only. Signals reflect off walls and metal, and the reflections arrive fractionally later than the direct path, sometimes reinforcing it and sometimes cancelling it out. A null in the pattern can be smaller than a person, which is genuinely why standing up sometimes fixes it.

So connected and working are separate states, and every client indicator reports the first one.

802.11 as a family

IEEE 802.11 is a base standard with amendments. Each amendment adds capability and carries a letter, so the letters identify amendments rather than products, and 802.11 is a family rather than one document. The Wi-Fi Alliance’s generation numbers came later and are a separate labelling scheme, which is why they do not line up neatly with the letters.

The objectives name 802.11 and no letter amendment, and no generation number either. Worth knowing before you start revising, because most material for this subject is organised around the letters. This exam asks about the band, the channel plan and the security instead, which are the next three topics.

Recognising the letters well enough to read a datasheet is useful. Memorising which one did what is not, for this.

Cellular and satellite

Two more media the objective names, and both are worth a paragraph rather than a chapter, because at this level what matters is where each one belongs.

Cellular covers the mobile network generations, and its distinguishing property for a network engineer is not speed. It is that coverage and capacity are somebody else’s, sold to you, and shared with everyone else in the cell. As a primary link it suits sites where running a cable is impractical. As a backup link it is genuinely good, because its failure modes are unrelated to those of a wired circuit: a digger through a duct does not affect it.

Satellite has one property that dominates every other. Distance imposes latency that no engineering can remove, because the signal travels at the speed of light and a geostationary orbit is roughly 36,000 kilometres up. Up and back is already a large fraction of a second before anything else happens, and that budget cannot be optimised away.

The consequence is not slowness in the sense of throughput, which can be respectable. It is that anything requiring many round trips feels terrible. Topic 09 established that TCP needs a handshake before data moves, and every one of those round trips now costs half a second. Low earth orbit constellations change the arithmetic by being far closer, which reduces the latency substantially, and the principle stands: with satellite you are buying a link whose delay is set by geometry.

If you already work on networks: why the number on the box was never achievable

Every wireless product is sold with a data rate, and the gap between that number and observed throughput is larger than in any other medium. Three things account for it, and none is dishonesty exactly.

The rated figure is a physical layer signalling rate. It describes how fast bits are put on the air under ideal conditions, and it counts everything: preambles, headers, acknowledgements, interframe spaces. Useful throughput is what is left, and roughly half is a reasonable expectation rather than a pessimistic one.

The figure assumes the best encoding, which assumes an excellent signal. Rate adaptation moves a client down to slower and more robust encodings as conditions worsen, and it does so continuously and invisibly. A client’s actual rate changes several times a minute while sitting still.

And the figure is per channel, not per client. The whole channel is shared, so the number describes the ceiling for everybody together.

The habit worth building is to divide. Take the advertised figure, halve it for overhead, then divide by the number of active clients, and treat the result as an optimistic estimate. That arithmetic done in a design meeting prevents the conversation that otherwise happens after installation.

Prove it

There is nothing to capture here. The lab in this track is Linux network namespaces, veth pairs have no radios, and simulating a radio link would produce a transcript that proves nothing about radio.

What there is instead is a document and an instrument.

IEEE 802.11. The standard is published by the IEEE and the scope statement is readable without purchase. Read it and answer one question: does 802.11 define a medium access method as well as a physical layer, and what does that tell you about why the wired and wireless access methods have different names?

Then use the instrument you are holding. Every phone can list the networks it can see, and most operating systems will report the signal strength of the one you are on. Stand somewhere with a good signal and watch the number while you walk. Note where it drops, note what you were walking past, and note whether throughput follows the number or not. That last part is the point of this topic and it takes ten minutes to observe directly.

What trips people up

1. Reading signal bars as a prediction of throughput

Bars measure received signal strength. What determines whether a link works is how far that signal stands above the noise, and the noise is not measured by anything the client shows you.

2. Expecting full duplex on wireless

The air is half duplex and cannot be otherwise. A radio transmitting on a frequency cannot usefully receive on it at the same time, which is a physical constraint rather than a setting.

3. Dividing the rated speed by the number of clients and expecting that

The division is real but the answer is optimistic, because overhead grows with the number of stations and because slow clients hold the medium for longer than fast ones.

4. Assuming a strong signal means a fast client

A client at the edge negotiates a slower encoding, and while it transmits it occupies the channel. One distant device measurably slows a room of nearby ones.

5. Learning the 802.11 letter standards for this exam

They do not appear in the objectives, and neither do the generation numbers. The band, the channel plan and the security are what is tested.

6. Treating satellite as simply slow

Its throughput can be good. Its latency is set by how far the signal has to travel, and no product can improve on the speed of light.

Work it through

The scenario at the top, taken apart in the order a fault should be.

Start by separating the two states. The client says connected, which means it associated with an access point and holds an address. That is a layer 2 and layer 3 statement and says nothing about whether frames are getting through. So the first useful question is not why is it slow, it is whether anything is arriving at all.

Then the signal number, and what it does not tell you. Four bars is received strength. If throughput is poor with a strong signal, the likely cause is on the other side of the fraction: something else is occupying the channel. That is not diagnosable from the client, which is why the next topic is about the channel plan and why a survey exists as a job.

Then the room, because two metres mattering is a clue rather than noise. Signals reflect, and reflections arriving slightly out of step with the direct path can cancel it. A null a metre across is entirely ordinary indoors, and it is why coverage is measured by walking rather than calculated from a floor plan.

Then the other clients. If the problem correlates with how many people are in the room, the medium is being shared and the answer is capacity rather than coverage: more access points on different channels, not more power on this one. Turning the power up is the intuitive fix and it usually makes things worse, because it enlarges the area over which everybody contends.

And the honest last step, which is that all of this is inference from a client. The instrument that answers it directly is a survey tool listening to the air itself, which is what the next topic’s picture shows.

Try it

Watch your own signal number change while you walk. Note where it drops and what you walked past. Walls with metal in them, lift shafts, and mirrors are the usual culprits and the effect is larger than people expect.

Count the networks your phone can see. In a flat in a city the answer is frequently more than twenty, and every one of them is sharing the same small number of channels with you. That is the single best motivation for the next topic.

Do the division. Take the rated speed of your access point, halve it, and divide by the number of devices in your home. Compare that to what you actually get. The estimate is usually closer than the box is.

Check yourself

Why is a wireless link always half duplex, and what did that force the access method to change?

Because a radio transmitting on a frequency cannot usefully receive on the same frequency at the same time. Its own signal overwhelms anything arriving.

That broke collision detection. Wired Ethernet used CSMA/CD, listening for collisions while transmitting, which is only possible if you can hear the medium while using it. Wireless uses CSMA/CA instead: listen before transmitting, wait a random interval, and where necessary reserve the medium first.

Avoidance costs more than detection, which is part of why observed throughput sits so far below the rated figure.

A client shows a strong signal and poor throughput. What does the signal indicator not tell you?

How much else is on the channel. Bars report received signal strength, and what determines whether a link performs is the signal to noise ratio, meaning how far the signal stands above the background.

A strong signal in a busy channel performs worse than a moderate signal in a quiet one. Nothing on the client measures the other half of that fraction, which is why this fault needs an instrument that listens to the air rather than a screenshot of a laptop.

How does one distant laptop reduce throughput for people sitting next to the access point?

By holding the medium for longer. A weak signal forces a slower, more robust encoding, so the same amount of data takes considerably more airtime.

The channel is shared and half duplex, so while that client transmits nobody else can. A frame that would take one unit of time at full rate takes many at the low one, and everybody waits through all of it.

This is why adding power to reach a distant client is often the wrong fix. It keeps the slow client associated rather than letting it move to a nearer access point.

Why does satellite latency resist engineering in a way that throughput does not?

Because it is set by distance and the speed of light. A geostationary satellite is roughly 36,000 kilometres up, so a round trip is already a large fraction of a second before any equipment does anything.

Throughput can be engineered: more spectrum, better encoding, more capacity. Delay imposed by geometry cannot. The practical consequence is that anything needing many round trips feels bad even when the link is fast, which includes TCP’s own handshake before a single byte of data moves.

Constellations in low earth orbit reduce the distance substantially, which is a different geometry rather than a different physics.

Why is learning the 802.11 letter standards not worth your time for this exam?

Because they are not in the objectives. 802.11 appears as a family, and no letter amendment or generation number appears anywhere.

What the objectives do test is the band, the channel plan, the regulatory limits and the security, which are the next three topics. The letters are worth recognising as amendment identifiers so a datasheet is readable, and worth no memorisation beyond that.

Across platforms

The signal number is available on every desktop operating system, and the command differs on each.

The one worth noting is macOS, which reports noise alongside signal, so the ratio this topic is about can be worked out directly rather than inferred. On the others you get the top half of the fraction and have to reason about the rest.

Signal is conventionally reported in dBm, which is a negative number where closer to zero is stronger. Around -50 is excellent, around -70 is usable, and -80 is close to unusable, which is a scale worth internalising because the percentages Windows shows are a vendor’s mapping of that number rather than a measurement in their own right.

References

• IEEE 802.11 - IEEE Standards Association, the wireless LAN standard covering both the medium access method and the physical layer. Scope readable without purchase. Accessed 2026-08-11.
• RFC 3819 - IETF, advice for subnetwork designers, and the free source on why links that lose packets without congestion are hard for TCP. Accessed 2026-08-11.
• IEEE 802.3 - IEEE Standards Association, for the wired access method this one is contrasted against. Accessed 2026-08-11.

Pictures. Freely licensed files from Wikimedia Commons, downloaded and served from this site rather than linked across to somebody else’s server. Each is resized and otherwise unaltered.

• Antenas adaptador wifi by EEIM, CC BY-SA 4.0.
• Long Distance 802.11 Wi-Fi, dish, Venezuela by Wireless Networking in the Developing World, CC BY-SA 3.0.

Where the numbers came from. Nothing on this page is captured, because the lab behind this track is Linux network namespaces and a veth pair has no radio. Simulating one would produce a transcript that proves nothing about radio, which is the same reason the cabling topics have no captures. The dBm scale in the platforms table is a widely used rule of thumb rather than a figure from a standard, and the halving rule for rated against useful throughput is an expectation rather than a specification, which is why both are described as such.

If you also work on Linux. iw is the current tool and iwconfig is the one most search results still show; the second is deprecated and reports less. iw dev wlan0 scan lists what the radio can hear, which is the closest a host gets to the survey instrument the next topic describes.