In the space they serve, at ceiling height, away from metal and away from other access points - and not in a cupboard.

The ideal is a ceiling mount in the middle of the area, radiating downwards. Most indoor access points have antennas designed for exactly that orientation, and mounting them on a wall or above a suspended ceiling tile aims the pattern somewhere unintended.

Avoid metalwork: ducting, cable trays, structural steel and metal-backed insulation all reflect and block. A few centimetres of clearance matters more than it seems.

Corridors are a common mistake. An access point in a corridor serves the corridor well and the rooms either side poorly, because the signal has to pass through walls in every direction. Rooms are better served from within the rooms.

Above suspended ceilings is usually acceptable for tiles and plasterboard, but not where there is a metal grid or a foil-backed membrane.

And never a comms cupboard: a metal cabinet in a corner of the building is the single worst position available.

Efficiency in crowded environments far more than headline speed - which is what most networks actually need.

OFDMA divides a channel into smaller units so a single transmission can carry data for several clients at once. In a room full of devices each sending small amounts, that is a large improvement in real capacity, because the overhead of each separate transmission is avoided.

Target wake time lets devices agree scheduled times to transmit, so battery-powered devices sleep longer and contend for the channel less - better for the device and for everyone else on the channel.

Improved multi-user MIMO and better handling of dense deployments follow the same theme.

6E adds the 6 GHz band where regulations allow, which is a large amount of spectrum with no legacy devices in it. That is the biggest single improvement available, but only clients supporting 6 GHz can use it.

The benefit requires clients that support the standard, so it arrives gradually as devices are replaced.

Usually 802.3at as a minimum, and 802.3bt for tri-band and high-power units - and underpowering causes silent degradation rather than failure.

A modern access point with several radios can exceed the power available from 802.3af. When it does, most units do not simply fail. They disable a radio, reduce transmit power, or disable a second Ethernet port, and they report it only in a log nobody reads.

The result is a network that underperforms for months while every access point appears to be online.

So check the required class per model, confirm the switch supports it, and check the switch's total budget rather than its per-port figure - a 48-port switch with 48 high-power access points is well beyond most switch budgets.

Cable length matters too: loss over a long run can drop a device below its class, and the effect is position-dependent, which makes it look like a faulty unit rather than a power problem.

For current high-end access points, yes - a gigabit uplink can become the constraint.

A tri-band access point with multiple spatial streams can aggregate more than a gigabit of client traffic under favourable conditions. Where that happens, the single gigabit uplink limits it, and the fault appears as wireless slowness that no amount of radio tuning will fix.

Multi-gigabit Ethernet - 2.5 or 5 Gb over existing structured cabling - was developed for exactly this, and it works over cabling already installed, which is why it is the practical answer rather than recabling.

Whether it matters depends on realistic load. An access point serving a busy lecture theatre or a dense office may approach it; one serving a quiet corridor will not come close.

For new installations that will be in service for years, specifying multi-gigabit switch ports at access point positions is inexpensive insurance. Retrofitting them later means changing the switch.

Enough that each serves a manageable number of active devices - which is a density calculation, not a coverage one.

Start from the number of concurrent active devices per area and what they are doing. A rough working figure is a few dozen active devices per access point for general office use, and considerably fewer where video calling is constant.

Then check that the resulting layout does not create interference. If the density required means access points are close together, transmit power must come down so cells stay small and channels can be reused - which is why high-density designs use many low-power access points rather than a few strong ones.

Allow for the building. Solid walls contain signal helpfully; open plan spreads it and makes reuse harder.

And design for where people are. Meeting rooms, canteens and training rooms concentrate devices far beyond the average for their floor area, and are the positions where under-provisioning is noticed first.

They coexist on the same network, but they cannot be managed together or coordinate their radios - so mixing within one coverage area is a poor idea.

The wireless standards are interoperable from the client's point of view: a device connects to any compliant access point, and roaming between different manufacturers technically works.

What does not work is coordination. Radio resource management, assisted roaming, load balancing and single-console monitoring are all vendor-specific. Two manufacturers' access points in the same space each optimise independently and can work against each other, particularly on channel and power selection.

Roaming quality also suffers, because the fast-roaming features depend on the access points sharing state.

Mixing between separate areas - one manufacturer per building or per floor, with the boundary at a place people do not roam across - is workable.

Within one space, standardise. The operational cost of two management systems usually exceeds any hardware saving.

A validation survey, client experience with real devices, channel and power assignment, and roaming through the areas people actually walk.

A post-installation survey measures what was built rather than what was designed - signal strength, signal to noise, and coverage overlap at the height devices are used.

Test with the actual client devices in use. Access points report their own view of the world; a five-year-old handheld scanner sees something quite different from a current laptop, and it is the worst device that defines the experience.

Check the channel plan the controller has chosen. Automatic assignment is usually good but occasionally puts two adjacent access points on the same channel, which is easily corrected once seen.

Walk the roaming paths - corridors, stairwells, between buildings - with something that shows the association, particularly if voice or scanning is used.

Record the results. They are the baseline against which the first complaint is assessed.