Every wired lesson in this course ends at a switch port. Most real users never touch one. They open a laptop, the laptop finds the air, and the air carries the request to the access point on the ceiling. When that air is clean, nobody thinks about it. When it is crowded, the same request that took 34 ms yesterday takes 68 ms today, and the wired network looks guilty for a crime the radio committed. This lesson is the air. I walk one request from a laptop through the radio, the access point, the WPA3 handshake, the controller, and the switch, and I show how to survey a floor, plan the channels, and prove a roaming problem with a phone in your hand.
Difficulty: Intermediate · Know this first: Lesson 184, Lesson 01
By the end of this lesson you will have done a site survey with your phone, planned the channels for one floor, and priced a wireless trip the same way the simulator does. First, a survey walk: you walk a floor with a Wi-Fi analyzer app, write down the signal strength in each room, and mark the dead spots on a sketch. Second, a channel plan: you run a small script that finds the access points sharing a channel and tells you which ones to move. Third, a trip math script that turns clients, channel width, distance, and neighbours into an airtime hop, an app hop, a signal strength, and a throughput number. You will also know the 802.11 standards by name, what WPA3 changed, and how a laptop roams from one access point to the next without dropping the call.
What you need: a phone with a Wi-Fi analyzer app and Python 3 for the two scripts. Nothing in this lesson runs against a live network from this page. The Python parts run on this machine and their outputs are shown exactly as they ran. The survey and the channel plan run on your own floor, with your own access points, and their outputs are yours to read. Your numbers will differ. The shapes will not.
A laptop on a desk has no cable. It has a radio. The radio speaks on a channel, the access point on the ceiling listens, and the air between them is shared with every other laptop, every phone, and the microwave in the pantry. The request still travels the same road as Lesson 01, but the first hop is not a wire. It is a turn to speak. That is the whole difference, and it explains almost every wireless problem I have ever debugged. The wire is private. The air is a meeting room where everyone talks at once.
The standards table below is the family tree. Read it top to bottom. Each generation is faster, and each one is a new name for the same job: move bits through the air without stepping on the neighbour.
| Standard | Wi-Fi name | Year | Band | Top speed, teaching figure |
|---|---|---|---|---|
| 802.11b | Before Wi-Fi names | 1999 | 2.4 GHz | 11 Mbps |
| 802.11a | Before Wi-Fi names | 1999 | 5 GHz | 54 Mbps |
| 802.11g | Before Wi-Fi names | 2003 | 2.4 GHz | 54 Mbps |
| 802.11n | Wi-Fi 4 | 2009 | 2.4 and 5 GHz | 600 Mbps, with MIMO |
| 802.11ac | Wi-Fi 5 | 2013 | 5 GHz | About 3.5 Gbps, with wide channels |
| 802.11ax | Wi-Fi 6 and 6E | 2019 | 2.4, 5, and 6 GHz | About 9.6 Gbps, with OFDMA |
| 802.11be | Wi-Fi 7 | 2024 | 2.4, 5, and 6 GHz | About 46 Gbps, with multi-link operation |
Two rows in that table do the most work. Wi-Fi 4 added MIMO, which is several antennas sending at once, so one radio can carry several streams. Wi-Fi 6 added OFDMA, which is one channel split into small resource units, so one access point can talk to several small devices in the same turn instead of making each one wait. Wi-Fi 6E is the same standard on a new band, 6 GHz, where the channels are wide and the neighbours are few. Wi-Fi 7 adds multi-link operation, which lets one device use two bands at once. The speeds in the table are teaching figures for the standard at its best, not promises. A real laptop in a real room gets a fraction of them, and the simulator below shows why.
Security changed in three steps, and the steps matter more than the speeds.
| Standard | Year | What it does | Why it was replaced |
|---|---|---|---|
| WPA | 2003 | TKIP encryption, a fix for WEP | TKIP was a patch on a broken cipher, and it aged out |
| WPA2 | 2004 | AES with CCMP, the long standard | Shared passwords leak, and offline guessing works against the handshake |
| WPA3 | 2018 | SAE handshake, forward secrecy, protected management frames | Current. The one to deploy on a new network |
WPA3 is the one to learn by name. Its handshake is called SAE, Simultaneous Authentication of Equals. The password never travels, and guessing it offline does not work the way it did against WPA2. Each session gets fresh keys, so a key stolen later cannot decrypt traffic from earlier. That property is called forward secrecy. WPA3 also makes protected management frames mandatory, so a forged deauthentication frame cannot kick a laptop off the network as easily as it could on WPA2. Personal mode still uses one shared password. Enterprise mode uses 802.1X, where each person logs in with their own credential and the password problem goes away.
The cast for the trip below has five parts. The client is the laptop. The access point is the radio on the ceiling. The wireless controller is the brain that configures the access points, usually over a CAPWAP tunnel, and decides the channels and the power. The switch is the wired hop from Lesson 184. The app server is the data hop, the place the answer is born. Between the access point and the app server sits the WPA3 handshake, and that order matters: the laptop proves who it is before it gets an address, and it gets an address before it asks the app anything.
Open a Wi-Fi analyzer app on your phone. Stand in the room where you work. Write down three numbers: the signal strength of your network in dBm, the channel it uses, and how many other networks share that channel. Walk to the farthest room you use and write the same three numbers down again. If the signal is weaker than -67 dBm in a room you work from, that room is a dead spot, and the fix is an access point closer to it, not a louder one where you already stand. Keep the two sets of numbers. The lab turns them into a survey.
The diagram above follows one request from a laptop to the app server. These four paths are the same idea drawn four ways. One request, answered on a clean floor. One request, stopped at the handshake. One request, slowed by a crowded floor. One laptop, walking between two access points. Run each path. One run takes 7.2 seconds across the path, so you can watch each hop light up in turn.
The return rule: the green dot only starts moving after the orange dot reaches the data hop. A return with no request is a lie. The diagram never draws one. The failure replay at the end of Run all four shows the other side of the rule. Run it and watch the orange dot stop at the WPA3 handshake and never come back.
One request walks the design from the story. Laptop, airwaves, access point, WPA3 handshake, controller, switch, app server. The handshake passes, the signal is strong, and the app server answers in 8 ms.
Not run yet.
Same laptop, wrong password. The air is quiet and the access point hears the request fine. The WPA3 handshake fails, the laptop never gets an address, and the app server never sees the request. The answer never exists.
Not run yet.
The same request on all-hands day. Eighty clients, a 20 MHz channel, six neighbours on the same channel, and the laptop in the far room. The hops add up to 61 ms before the verdict, and the airtime hop and the app hop carry almost all of it.
Not run yet.
A laptop walks from the lobby to the far room while a call is running. The controller sees the signal fading, the laptop picks the far room access point, and the fast roam moves the session without a full handshake. The call stays up. Without the fast roam, the laptop would re-authenticate from scratch and the call would drop.
Not run yet.
Every hop charges rent in milliseconds. The bar below is a worked example for one request from a laptop to the office app, not a measurement from a live system, but the shape is honest. The airtime hop and the app hop are the moving parts. On a clean floor they cost 4 ms and 8 ms, and the trip totals 34 ms. On a crowded floor they cost 18 ms and 22 ms, and the trip totals 68 ms. When the handshake fails, four hops are skipped and the trip totals 29 ms with no answer. Move a simulator slider and watch the airtime hop and the app hop move with it.
The budget bar above shows three fixed stories. This simulator lets you write your own. It answers the trade-off at the heart of this lesson: the air is shared, so more clients, a narrower channel, a longer distance, and more neighbours all cost time and speed. You set how many clients share one access point, how wide the channel is, how far the laptop sits from the access point, and how many neighbours share the channel. Move a slider and watch the signal, the throughput, the airtime hop, the app hop, and the budget total change. The numbers are a teaching example, built from the formulas below, not live measurements. The shape is the lesson: distance kills signal, neighbours steal airtime, and width is speed.
How many devices share one access point. One client has the air to itself. Eighty clients take turns, and every turn costs time.
How wide the channel is: 20, 40, 80, or 160 MHz. A wider channel carries more at once. It also needs clean spectrum, which the 6 GHz band has and the 2.4 GHz band does not.
How far the laptop sits from the access point, in metres. Signal falls with distance, and a weak signal costs retries.
How many nearby access points use the same channel. Each one is another radio waiting to speak, and each one steals airtime from yours.
Try this: (1) At the defaults, 25 clients, 80 MHz, 8 m, 1 neighbour, the airtime hop is 4 ms and the app hop is 8 ms. The signal is -46 dBm and the throughput is 456 Mbps. The budget total is 34 ms. (2) Drop to 5 clients, 160 MHz, 3 m, 0 neighbours. The airtime hop falls to 2 ms and the app hop to 6 ms. The signal is -36 dBm and the throughput is 1,152 Mbps. The budget total is 30 ms. (3) Push to 80 clients, 20 MHz, 28 m, 6 neighbours. The airtime hop grows to 11 ms and the app hop to 36 ms. The signal is -86 dBm and the throughput is 15 Mbps. The budget total is 69 ms. From the defaults, the crowded far corner costs 35 ms more per request and loses 441 Mbps of throughput. That is the trade, priced.
Formulas, in plain terms. Airtime hop = 2 ms plus 0.04 ms per client plus 1 ms per neighbour, rounded to the nearest millisecond. App hop = a width base (20 MHz is 14 ms, 40 MHz is 10 ms, 80 MHz is 8 ms, 160 MHz is 6 ms) plus 1 ms per metre past 10 m plus 4 ms when clients pass 50. Signal strength = -30 dBm minus 2 dBm per metre. Throughput = link rate (20 MHz is 150 Mbps, 40 MHz is 300 Mbps, 80 MHz is 600 Mbps, 160 MHz is 1,200 Mbps) times 1 minus 0.008 per client, times a signal factor (1.0 within 10 m, 0.7 within 20 m, 0.4 past that), times 1 minus 0.05 per neighbour, rounded to the nearest Mbps. Budget total = 0 plus airtime plus 3 plus 6 plus 2 plus 4 plus 5 plus app plus 2. The rates are teaching figures for one stream, not promises. Your room and your clients will differ. The shape will not.
Do this now, on your own floor and on your own machine. You will walk a site survey with your phone, check a channel plan with a script, and run the trip math with the same formulas the simulator uses. The Python parts run on this machine and their outputs are shown exactly as they ran. The survey runs on your floor, with your access points, and its numbers are yours. Never write down a signal reading you did not measure. A survey with invented numbers is a map of a floor that does not exist.
Open a Wi-Fi analyzer app on your phone. Stand in each room you work from and write down three numbers: the signal strength of your network in dBm, the channel it uses, and how many other networks share that channel. Use this table as your sheet. Fill in your own numbers. The example row shows the shape, not a reading. Do not copy it as your result.
| Room | Signal (dBm) | Channel | Other networks on it | Verdict |
|---|---|---|---|---|
| Desk (example shape) | -46 | 36 (5 GHz) | 0 | Good: above -67 dBm |
| Your desk | ||||
| Your farthest room | ||||
| Your meeting room |
Read the verdicts from the numbers. A room weaker than -67 dBm is a dead spot. A room with three or more neighbours on the same channel is a crowded room. A dead spot gets an access point closer to it. A crowded room gets a channel plan, which is step 1. A room that is both gets both, and the access point goes in first, because coverage beats channels when the signal cannot be heard at all.
Save this file as channel_plan.py. It takes the access points on one floor and flags the 2.4 GHz problems: a channel that is not 1, 6, or 11 overlaps its neighbours, and two access points sharing one channel steal airtime from each other. The bad floor below is the floor from the story. The fixed floor is the same floor after the plan.
"""Channel plan checker for Lesson 185.
Give it the access points on one floor: name, band, channel. It counts
co-channel pairs on 2.4 GHz, where only channels 1, 6, and 11 do not
overlap, and it suggests the fix. 5 GHz channels do not overlap at
20 MHz, so they are listed but never flagged.
"""
def check_floor(aps):
problems = []
seen = {}
for name, band, channel in aps:
if band == "2.4":
if channel not in (1, 6, 11):
problems.append(f"{name}: channel {channel} overlaps its neighbours, use 1, 6, or 11")
if channel in seen:
problems.append(f"{name} and {seen[channel]} share channel {channel}, move one of them")
else:
seen[channel] = name
return problems
BAD_FLOOR = [
("AP-lobby", "2.4", 6),
("AP-hall", "2.4", 6),
("AP-far-room", "2.4", 6),
("AP-meeting", "2.4", 3),
("AP-lobby", "5", 36),
("AP-hall", "5", 44),
]
FIXED_FLOOR = [
("AP-lobby", "2.4", 1),
("AP-hall", "2.4", 6),
("AP-far-room", "2.4", 11),
("AP-lobby", "5", 36),
("AP-hall", "5", 44),
]
if __name__ == "__main__":
for label, floor in [("bad floor", BAD_FLOOR), ("fixed floor", FIXED_FLOOR)]:
problems = check_floor(floor)
print(f"{label}: {len(problems)} problem(s)")
for p in problems:
print(f" [FIX] {p}")
if not problems:
print(" no co-channel clash on 2.4 GHz")
Run it:
python3 channel_plan.py
Expected output, exactly as it ran on this machine:
bad floor: 3 problem(s) [FIX] AP-hall and AP-lobby share channel 6, move one of them [FIX] AP-far-room and AP-lobby share channel 6, move one of them [FIX] AP-meeting: channel 3 overlaps its neighbours, use 1, 6, or 11 fixed floor: 0 problem(s) no co-channel clash on 2.4 GHz
Now change it on purpose. Put the far room access point back on channel 6 and run it again. The script flags the share. That is the check doing its job. The fixed floor in the script has only three 2.4 GHz access points, one on each clean channel. A fourth 2.4 GHz radio on the same floor has to reuse a channel, so the real fix for a four-radio floor is to move the radios that can use 5 GHz onto 5 GHz, where the channels do not overlap at 20 MHz. Put your own floor into the two lists and run it. Your access point names and channels go in. The script tells you what to move.
Save this file as wifi_math.py. It is the same math the simulator uses: the airtime hop, the app hop, the signal strength, the throughput, and the budget total. Run it and check the three lines against the simulator try-this points.
"""Wi-Fi trip math for Lesson 185.
Teaching model only, not a measurement. The formulas here are the same
ones the in-page simulator uses, so the numbers in the lesson, in the
simulator, and in this script all agree.
"""
import math
def air_ms(clients, neighbours):
"""Airtime hop: base 2 ms, plus 0.04 ms per client on the AP,
plus 1 ms per neighbouring AP on the same channel."""
return round(2 + clients * 0.04 + neighbours)
WIDTH_BASE = {20: 14, 40: 10, 80: 8, 160: 6}
LINK_RATE = {20: 150, 40: 300, 80: 600, 160: 1200}
def app_ms(width, dist, clients):
"""App hop: wider channel is faster, distance past 10 m costs
1 ms per metre, more than 50 clients adds 4 ms of contention."""
base = WIDTH_BASE[width]
dist_cost = max(0, dist - 10) * 1
crowd_cost = 4 if clients > 50 else 0
return base + dist_cost + crowd_cost
def rssi(dist):
"""Signal strength in dBm: -30 dBm at the AP, minus 2 dBm per metre."""
return -30 - 2 * dist
def throughput(width, dist, clients, neighbours):
"""Effective throughput in Mbps: link rate, less a share per client,
less a signal factor for distance, less 5 percent per neighbour."""
if dist <= 10:
signal = 1.0
elif dist <= 20:
signal = 0.7
else:
signal = 0.4
rate = LINK_RATE[width] * (1 - clients * 0.008) * signal
rate = rate * (1 - neighbours * 0.05)
return round(rate)
def trip_total(clients, width, dist, neighbours):
"""Whole trip: laptop 0, airtime, AP 3, WPA3 6, controller 2,
switch 4, DHCP and DNS 5, app, verdict 2."""
air = air_ms(clients, neighbours)
app = app_ms(width, dist, clients)
total = 0 + air + 3 + 6 + 2 + 4 + 5 + app + 2
return air, app, total
def report(label, clients, width, dist, neighbours):
air, app, total = trip_total(clients, width, dist, neighbours)
print(f"{label}: clients={clients} width={width}MHz dist={dist}m neighbours={neighbours}")
print(f" airtime hop: {air} ms")
print(f" app hop: {app} ms")
print(f" RSSI: {rssi(dist)} dBm")
print(f" throughput: {throughput(width, dist, clients, neighbours)} Mbps")
print(f" budget total: {total} ms")
if __name__ == "__main__":
report("defaults", 25, 80, 8, 1)
report("quiet room", 5, 160, 3, 0)
report("crowded far corner", 80, 20, 28, 6)
Run it:
python3 wifi_math.py
Expected output, exactly as it ran on this machine:
defaults: clients=25 width=80MHz dist=8m neighbours=1 airtime hop: 4 ms app hop: 8 ms RSSI: -46 dBm throughput: 456 Mbps budget total: 34 ms quiet room: clients=5 width=160MHz dist=3m neighbours=0 airtime hop: 2 ms app hop: 6 ms RSSI: -36 dBm throughput: 1152 Mbps budget total: 30 ms crowded far corner: clients=80 width=20MHz dist=28m neighbours=6 airtime hop: 11 ms app hop: 36 ms RSSI: -86 dBm throughput: 15 Mbps budget total: 69 ms
Now break it on purpose. Change the defaults call to 80 clients, 20 MHz, 28 m, 6 neighbours and run it again. The signal reads -86 dBm and the throughput reads 15 Mbps. That is the crowded far corner from the simulator, and two different paths agree on it. When two paths agree, the number is worth trusting. Change it back when you are done.
Take your survey sheet from step 0. Find the weakest room. Stand in it, open the analyzer, and read the channel again. If the channel is crowded, note the clean channel you would move to. If the signal is weak, note where the nearest access point would go. Write one line for that room: the fix, and the number you expect after it. Then go and check the number. A survey that ends on paper is a survey half done.
Nothing to destroy. The two scripts live on your machine, and you can run them again on the next floor. The survey sheet is yours to keep. Do not change any access point channel until the plan is written down and the people on that floor know a change is coming. A channel change at midday is a meeting interrupted, and the survey is not worth that.
Three questions. No score is saved anywhere. This is just you against the material.
Running a wireless network is not one skill. It is four levels of the same job, and every team I have worked with sits on one of them. The levels are not about how many standards you can name. They are about whether the floor works when people actually use it. Read them in order. Each level fixes the failure that hurt the most at the level below it.
At the basic level, you walk one floor with your phone and you write the numbers down. Signal strength in each room, channel, neighbours on that channel. The failure that ends this level is the second floor. One surveyed floor feels like a network. It is not. It is a start.
Do this: run step 0 of the lab. Fill in the survey sheet for your desk, your farthest room, and your meeting room. Mark every room weaker than -67 dBm. One sheet, three rooms.
At the intermediate level, the channel plan is a script and a date. You run channel_plan.py against your floor every month, and the plan is written down where the team can see it. A new neighbour on your channel gets a ticket the same week. The 2.4 GHz radios sit on 1, 6, and 11, and the clients that can use 5 GHz are steered onto it. The WPA mode is WPA3, and the shared password is gone where Enterprise is possible.
Do this: run the channel plan with your own floor, then move one access point to a clean channel and run it again. Write the date on the run. A plan with no date is a plan nobody did.
At the advanced level, the network is a system, not a set of radios. You tune roaming with 802.11k neighbour reports, 802.11v transition management, and 802.11r fast transition, so a laptop walks between access points without dropping a call. You plan capacity per access point, not per floor, and you know how many clients one radio holds before the airtime hop doubles. The controller config is versioned, and a change to it is reviewed like code.
Do this: walk between two access points on a call and watch the roam. Write down the access point you started on, the one you ended on, and whether the call dropped. If it dropped, read the roam settings on the controller. A roam you cannot measure is a roam you cannot fix.
At the master level, the floor is code and the survey is a gate. The access point placement comes from a predictive survey before the radios are bought, and a passive survey after they are installed proves the prediction or it does not. The controller config is deployed from a repository. A change that puts two adjacent access points on the same channel fails the check. The team still walks the floor every quarter. The walk reads the survey history. It does not start from zero.
Do this: write one check for one rule. Assert no two adjacent access points share a 2.4 GHz channel. Assert every SSID uses WPA3. Run the check on every config change. A plan that is only a document will drift. A plan that is a check will hold.
A laptop that walks between access points uses three standards together, and each one does a different job. 802.11k gives the laptop a neighbour report, so it knows which access point to try next instead of scanning every channel. 802.11v lets the network suggest a better access point, so a laptop stuck to a distant radio gets a nudge toward the close one. 802.11r is the fast transition: it moves the keys ahead of the laptop, so the roam does not need a full handshake. Without 802.11r, a voice call drops during the re-authentication. With it, the roam takes milliseconds and the call stays up. The three are usually switched on together on the controller, and the config below is the shape of that setting. Paste it into your own controller in your controller's own format. The names will differ. The three standards will not.
ssid: office-wifi band_steering: prefer-5ghz security: wpa3-enterprise pmf: required roaming_80211r: enabled roaming_80211k: enabled roaming_80211v: enabled channel_2ghz: auto-1-6-11 channel_5ghz: auto-36-44-149-157 min_rssi: -67 dBm max_clients_per_radio: 40
Two lines in that config do quiet work. min_rssi: -67 dBm is the line that stops a laptop clinging to a distant access point at -80 dBm when a close one is right above it. The controller drops the weak client, and the client re-associates to the close radio. max_clients_per_radio: 40 is the line that stops one radio taking the whole floor. Past 40 clients the airtime hop grows, and the simulator shows it. Both lines are the controller doing the job a human cannot do by walking the floor.
A predictive survey is a plan on paper: a floor map, wall types, and a model that guesses where the signal lands. It is done before the radios are bought, and it is wrong in the details. A passive survey is a walk with a laptop or phone that only listens. It measures the real signal, including the neighbours, and it is the survey in this lesson's lab. An active survey is a walk with a device that joins the network and sends traffic, so it measures the real throughput and the real roam, not just the signal. The honest order is predictive, then passive, then active. The predictive survey buys the radios. The passive survey proves the coverage. The active survey proves the experience. A floor that skipped the passive survey is a floor that guessed.
The 2.4 GHz band has three clean channels and every device in the building wants one. The 5 GHz band has many clean channels, but the signal does not travel as far through walls. The 6 GHz band has wide clean channels and few neighbours, but only new devices speak it. The trade is simple to state and hard to live with: 2.4 GHz reaches far and carries little, 5 GHz carries more and reaches less far, 6 GHz carries the most and reaches the newest devices only. Band steering is the controller pushing a capable client onto 5 GHz or 6 GHz, so the 2.4 GHz band is left for the devices that cannot move. A floor that runs everything on 2.4 GHz is a floor that chose reach over speed for devices that did not need the reach.
At 10x the clients, one access point per room stops working. The plan becomes one radio per 25 to 40 clients, and the survey becomes a capacity plan, not a coverage plan. The controller stops being a web page and becomes config in a repository, because nobody changes a hundred radios by hand. At 100x, the floor becomes a campus, and the problems become wired ones: the uplinks from the access points, the DHCP scopes, and the controller that must survive a failure. The air does not change. The number of radios does, and the management of them becomes the job.
When this lesson was scoped, the audit warned about one false friend: LoRa. The letters look like a wireless topic, but the hits in this course were ML LoRA, a model fine-tuning method, not LoRa wireless. It is not coverage of this lesson, and this lesson does not claim it. The wireless taught here is 802.11, the Wi-Fi in the office, and nothing else with a similar name.
An office moved into a new floor on a Monday. The access points were mounted on the same plan as the old floor, three radios on channel 6 in the 2.4 GHz band, because the plan was copied from a drawing and nobody re-surveyed. By 10 AM the floor was slow. Video calls froze. The wired network was fine. The app servers were fine. The air was the problem, and the air had three radios shouting over each other on one channel while the 5 GHz radios sat quiet, because band steering was off and every laptop joined 2.4 GHz by default.
Two things. First, the rule. No floor is copied from another floor. Every floor gets its own survey, because walls and neighbours differ, and a channel plan that worked in the old building is a guess in the new one. Second, the check. The controller config is reviewed before a move, and band steering is on by default. The survey sheet for the new floor is kept with the floor plan, so the next move starts from measurements, not from memory.
Design the wireless network for a two-floor office. Each floor is 40 metres by 25 metres, open plan, with 120 people per floor. The meeting rooms sit along one wall. The far corner room is used for video calls all day. The client laptops are new and speak Wi-Fi 6E. The printers and a few old devices only speak 2.4 GHz. The office wants WPA3, and the security team wants no shared password. Say how many access points per floor, which channels, which bands, how roaming works, and what you survey before and after the install.
Count the radios first. 120 people is about 240 devices, and one radio holds 25 to 40 clients before the airtime hop grows. That is 6 to 10 radios per floor on capacity alone. Start with 8 per floor, one per 15 people, and let the survey move them. Coverage alone would say fewer, and coverage is the wrong count here. The floor is dense, so capacity decides.
Then the bands. The new laptops go on 6 GHz where the channels are wide and the neighbours are few, and on 5 GHz where 6 GHz does not reach. The printers and the old devices stay on 2.4 GHz, on channels 1, 6, and 11, one per radio, never two adjacent radios on the same channel. Band steering is on, so a capable client does not sit on 2.4 GHz by accident. The far corner meeting room gets its own access point, because video calls do not forgive a weak signal.
Then the security. WPA3-Enterprise with 802.1X, so each person logs in with their own credential. No shared password, so a leaving person is one account disabled. Protected management frames are required, so a forged deauthentication frame cannot kick a call off the network.
Then the roam. 802.11k, 802.11v, and 802.11r are all on, so a laptop walking between floors keeps its call. The minimum RSSI is -67 dBm, so a laptop clinging to a distant radio gets moved to the close one. Test it by walking between floors on a call before the office opens.
Then the surveys. A predictive survey buys the radios. A passive survey after the install proves the coverage, room by room, with the numbers written down. An active survey proves the experience: a call walked across both floors without dropping. If the active survey fails, the radios move. The survey sheet is kept with the floor plan, and the channel plan is re-run every month, because the neighbours change.
Trade-offs to name. Capacity versus cost is the big one. Eight radios per floor costs more than four, and four would cover the floor. Coverage is not the job here. Capacity is. The 6 GHz band versus reach is the second. 6 GHz is fast and short, so the far rooms need their own radios. Security versus convenience is the third. Enterprise login is one more step for a new joiner, and it is worth it, because a shared password is a secret everybody knows.
Saved in this browser, so the Lessons index can show your progress across the course.